The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping

Clinical applications of radiomics and deep learning in breast and lung cancer: A narrative literature review on current evidence and future perspectives

Radiomics, analysing quantitative features from medical imaging, has rapidly become an emerging field in translational oncology. Radiomics has been investigated in several neoplastic malignancies as it might allow for a non-invasive tumour characterization and for the identification of predictive and prognostic biomarkers. Over the last few years, evidence has been accumulating regarding potential clinical applications of machine learning in many crucial moments of cancer patients' history. However, the incorporation of radiomics in clinical decision-making process is still limited by low data reproducibility and study variability. Moreover, the need for prospective validations and standardizations is emerging. In this narrative review, we summarize current evidence regarding radiomic applications in high-incidence cancers (breast and lung) for screening, diagnosis, staging, treatment choice, response, and clinical outcome evaluation. We also discuss pro and cons of the radiomic approach, suggesting possible solutions to critical issues which might invalidate radiomics studies and propose future perspectives.

Peritumoral edema enhances MRI-based deep learning radiomic model for axillary lymph node metastasis burden prediction in breast cancer

To investigate whether peritumoral edema (PE) could enhance deep learning radiomic (DLR) model in predicting axillary lymph node metastasis (ALNM) burden in breast cancer. Invasive breast cancer patients with preoperative MRI were retrospectively enrolled and categorized into low (< 2 lymph nodes involved (LNs+)) and high (≥ 2 LNs+) burden groups based on surgical pathology. PE was evaluated on T2WI, and intra- and peri-tumoral radiomic features were extracted from MRI-visible tumors in DCE-MRI. Deep learning models were developed for LN burden prediction in the training cohort and validated in an independent cohort. The incremental value of PE was evaluated through receiver operating characteristic (ROC) analysis, confirming the improvement in the area under the curve (AUC) using the Delong test. This was complemented by net reclassification improvement (NRI) and integrated discrimination improvement (IDI) metrics. The deep learning combined model, incorporating PE with selected radiomic features, demonstrated significantly higher AUC values compared to the MRI model and the DLR model in the training cohort (n = 177) (AUC: 0.953 vs. 0.849 and 0.867, p < 0.05) and the validation cohort (n = 111) (AUC: 0.963 vs. 0.883 and 0.882, p < 0.05). The complementary analysis demonstrated that PE significantly enhances the prediction performance of the DLR model (Categorical NRI: 0.551, p < 0.001; IDI = 0.343, p < 0.001). These findings were confirmed in the validation cohort (Categorical NRI: 0.539, p < 0.001; IDI = 0.387, p < 0.001). PE improved preoperative ALNM burden prediction of DLR model, facilitating personalized axillary management in breast cancer patients.

A dual-radiomics model for overall survival prediction in early-stage NSCLC patient using pre-treatment CT images

Introduction

Radiation therapy (RT) is one of the primary treatment options for early-stage non-small cell lung cancer (ES-NSCLC). Therefore, accurately predicting the overall survival (OS) rate following radiotherapy is crucial for implementing personalized treatment strategies. This work aims to develop a dual-radiomics (DR) model to (1) predict 3-year OS in ES-NSCLC patients receiving RT using pre-treatment CT images, and (2) provide explanations between feature importanceand model prediction performance.

Methods

The publicly available TCIA Lung1 dataset with 132 ES-NSCLC patients received RT were studied: 89/43 patients in the under/over 3-year OS group. For each patient, two types of radiomic features were examined: 56 handcrafted radiomic features (HRFs) extracted within gross tumor volume, and 512 image deep features (IDFs) extracted using a pre-trained U-Net encoder. They were combined as inputs to an explainable boosting machine (EBM) model for OS prediction. The EBM's mean absolute scores for HRFs and IDFs were used as feature importance explanations. To evaluate identified feature importance, the DR model was compared with EBM using either (1) key or (2) non-key feature type only. Comparison studies with other models, including supporting vector machine (SVM) and random forest (RF), were also included. The performance was evaluated by the area under the receiver operating characteristic curve (AUCROC), accuracy, sensitivity, and specificity with a 100-fold Monte Carlo cross-validation.

Results

The DR model showed highestperformance in predicting 3-year OS (AUCROC=0.81 ± 0.04), and EBM scores suggested that IDFs showed significantly greater importance (normalized mean score=0.0019) than HRFs (score=0.0008). The comparison studies showed that EBM with key feature type (IDFs-only demonstrated comparable AUCROC results (0.81 ± 0.04), while EBM with non-key feature type (HRFs-only) showed limited AUCROC (0.64 ± 0.10). The results suggested that feature importance score identified by EBM is highly correlated with OS prediction performance. Both SVM and RF models were unable to explain key feature type while showing limited overall AUCROC=0.66 ± 0.07 and 0.77 ± 0.06, respectively. Accuracy, sensitivity, and specificity showed a similar trend.

Discussion

In conclusion, a DR model was successfully developed to predict ES-NSCLC OS based on pre-treatment CT images. The results suggested that the feature importance from DR model is highly correlated to the model prediction power.

Habitat Imaging With Tumoral and Peritumoral Radiomics for Prediction of Lung Adenocarcinoma Invasiveness on Preoperative Chest CT: A Multicenter Study

Background: Tumors' growth processes result in spatial heterogeneity, with the development of tumor subregions (i.e., habitats) having unique biologic characteristics. Objective: To develop and validate a habitat model combining tumor and peritumoral radiomics features on chest CT for predicting invasiveness of lung adenocarcinoma. Methods: This retrospective study included 1156 patients (mean age, 57.5 years; 464 male, 692 female) from three centers and a public dataset, who underwent chest CT before lung adenocarcinoma resection (variable date ranges across datasets). Patients from one center formed training (n=500) and validation (n=215) sets; patients from the other sources formed three external test sets (n=249, 113, 79). For each patient, a single nodule was manually segmented on chest CT. The nodule segmentation was combined with an automatically generated 4-mm peritumoral region into a whole-volume volume-of-interest (VOI). A Gaussian mixture model (GMM) identified voxel clusters with similar first-order energy across patients. GMM results were used to divide each patient's whole-volume VOI into multiple habitats, defined consistently across patients. Radiomic features were extracted from each habitat. After feature selection, a habitat model was developed for predicting invasiveness, using pathologic assessment as a reference. An integrated model was constructed, combining features extracted from habitats and whole-volume VOIs. Model performance was evaluated, including in subgroups based on nodule density (pure ground-glass, part-solid, solid). Results: Invasive cancer was diagnosed in 625/1156 patients. GMM identified four as the optimal number of voxel clusters and thus of per-patient tumor habitats. The habitat model had AUC of 0.932 in the validation set, and 0.881, 0.880, and 0.764 in the three external test sets. The integrated model had AUC of 0.947 in the validation set and 0.936, 0.908, and 0.800 in the three external test sets. In the three external test sets combined, across nodule densities, AUCs for the habitat model were 0.836-0.969 and for the integrated model were 0.846-0.917. Conclusions: Habitat imaging combining tumoral and peritumoral radiomic features could help predict lung adenocarcinoma invasiveness. Prediction is improved when combining information on tumor subregions and the tumor overall. Clinical Impact: The findings may aid personalized preoperative assessments to guide clinical decision-making in lung adenocarcinoma.

An approach leveraging radiomics and model checking for the automatic early diagnosis of adhesive capsulitis

Adhesive Capsulitis of the shoulder is a painful pathology limiting shoulder movements, commonly known as "Frozen Shoulder". Since this pathology limits movement, it is important to make an early diagnosis. Diagnosing capsulitis relies on clinical assessment, although diagnostic imaging, such as Magnetic Resonance Imaging, can provide predictive or supportive information for specific characteristic signs. However, its diagnosis is not so simple nor so immediate, indeed it remains a difficult topic for many general radiologists and expert musculoskeletal radiologists. This study aims to investigate whether it is possible to use disease signs within a medical image to automatically diagnose Adhesive Capsulitis. To this purpose, we propose an automatic Model Checking-based approach to quickly diagnose the Adhesive Capsulitis taking as input the radiomic feature values from the medical images. Furthermore, we compare the performance achieved by our method with diagnostic results obtained by professional radiologists with different levels of experience. To the best of our knowledge, this is the first method for the automatic diagnosis of Adhesive Capsulitis of the Shoulder.

Pilot study: radiomic analysis for predicting treatment response to whole-brain radiotherapy combined temozolomide in lung cancer brain metastases

Objective

The objective of this study is to assess the viability of utilizing radiomics for predicting the treatment response of lung cancer brain metastases (LCBM) to whole-brain radiotherapy (WBRT) combined with temozolomide (TMZ).

Methods

Fifty-three patients diagnosed with LCBM and undergoing WBRT combined with TMZ were enrolled. Patients were divided into responsive and non-responsive groups based on the RANO-BM criteria. Radiomic features were extracted from contrast-enhanced the whole brain tissue CT images. Feature selection was performed using t-tests, Pearson correlation coefficients, and Least Absolute Shrinkage And Selection (LASSO) regression. Logistic regression was employed to construct the radiomics model, which was then integrated with clinical data to develop the nomogram model. Model performance was evaluated using receiver operating characteristic (ROC) curves, and clinical utility was assessed using decision curve analysis (DCA).

Results

A total of 1834 radiomic features were extracted from each patient's images, and 3 features with predictive value were selected. Both the radiomics and nomogram models exhibited satisfactory predictive performance and clinical utility, with the nomogram model demonstrating superior predictive value. The ROC analysis revealed that the AUC of the radiomics model in the training and testing sets were 0.776 and 0.767, respectively, while the AUC of the nomogram model were 0.799 and 0.833, respectively. DCA curves demonstrated that both models provided benefits to patients across various thresholds.

Conclusion

Radiomic-defined image biomarkers can effectively predict the treatment response of WBRT combined with TMZ in patients with LCBM, offering potential to optimize treatment decisions for this condition.

Value of Dynamic Contrast-Enhanced MRI for Grade Group Prediction in Prostate Cancer: A Radiomics Pilot Study

RATIONALE AND OBJECTIVES

To determine the role of dynamic contrast-enhanced (DCE) MRI-radiomics in predicting the International Society of Urological Pathology Grade Group (ISUP-GG) in therapy-naïve prostate cancer (PCa) patients.

MATERIALS AND METHODS

In this ethics review board-approved retrospective study on two prospective clinical trials between 2017 and 2020, 73 men with suspected/confirmed PCa were included. All participants underwent multiparametric MRI. On MRI, dominant lesions (per PI-RADS) were identified. DCE-MRI radiomic features were extracted from the segmented volumes following the image biomarker standardisation initiative (IBSI) guidelines through 14 time points. Histopathology evaluation on the cognitive-fusion targeted biopsies was set as the reference standard. Univariate regression was done to evaluate potential predictors across all calculated features. Random forest imputation was used for multivariate modelling.

RESULTS

73 index lesions were reviewed. Histopathology revealed 28, 16, 13 and 16 lesions with ISUP-GG-Negative/1/2, ISUP-GG-3, ISUP-GG-4 and ISUP-GG-5, respectively. From the extracted features, total lesion enhancement (TLE), minimum enhancement intensity and Grey-Level Run Length Matrix (GLRLM) were the most significantly different parameters among ISUP-GGs (Neg/1/2 vs 3/4 vs 5). 16 features with significant cross-sectional associations with ISUP-GGs entered the multivariate analysis. The final DCE partitioning model used only four features (lesion sphericity, TLE, GLRLM and Grey-Level Zone Length Matrix). For the binarized diagnosis (ISUP-GG≤2 vs ISUP-GG>2), the accuracy reached 81%.

CONCLUSION

DCE-MRI radiomics might be used as a non-invasive tool for aiding pathological grade group prediction in therapy-naïve PCa patients, potentially adding complementary information to PI-RADS for supporting tailored diagnostic pathways and treatment planning.

FAST (fast analytical simulator of tracer)-PET: an accurate and efficient PET analytical simulation tool

Objective.Simulation of positron emission tomography (PET) images is an essential tool in the development and validation of quantitative imaging workflows and advanced image processing pipelines. Existing Monte Carlo or analytical PET simulators often compromise on either efficiency or accuracy. We aim to develop and validate fast analytical simulator of tracer (FAST)-PET, a novel analytical framework, to simulate PET images accurately and efficiently.Approach. FAST-PET simulates PET images by performing precise forward projection, scatter, and random estimation that match the scanner geometry and statistics. Although the same process should be applicable to other scanner models, we focus on the Siemens Biograph Vision-600 in this work. Calibration and validation of FAST-PET were performed through comparison with an experimental scan of a National Electrical Manufacturers Association (NEMA) Image Quality (IQ) phantom. Further validation was conducted between FAST-PET and Geant4 Application for Tomographic Emission (GATE) quantitatively in clinical image simulations in terms of intensity-based and texture-based features and task-based tumor segmentation.Main results.According to the NEMA IQ phantom simulation, FAST-PET's simulated images exhibited partial volume effects and noise levels comparable to experimental images, with a relative bias of the recovery coefficient RC within 10% for all spheres and a coefficient of variation for the background region within 6% across various acquisition times. FAST-PET generated clinical PET images exhibit high quantitative accuracy and texture comparable to GATE (correlation coefficients of all features over 0.95) but with ∼100-fold lower computation time. The tumor segmentation masks comparison between both methods exhibited significant overlap and shape similarity with high concordance CCC > 0.97 across measures.Significance.FAST-PET generated PET images with high quantitative accuracy comparable to GATE, making it ideal for applications requiring extensive PET image simulations such as virtual imaging trials, and the development and validation of image processing pipelines.

Radiomics nomogram based on CT radiomics features and clinical factors for prediction of Ki-67 expression and prognosis in clear cell renal cell carcinoma: a two-center study

OBJECTIVES

To develop and validate a radiomics nomogram combining radiomics features and clinical factors for preoperative evaluation of Ki-67 expression status and prognostic prediction in clear cell renal cell carcinoma (ccRCC).

METHODS

Two medical centers of 185 ccRCC patients were included, and each of them formed a training group (n = 130) and a validation group (n = 55). The independent predictor of Ki-67 expression status was identified by univariate and multivariate regression, and radiomics features were extracted from the preoperative CT images. The maximum relevance minimum redundancy (mRMR) and the least absolute shrinkage and selection operator algorithm (LASSO) were used to identify the radiomics features that were most relevant for high Ki-67 expression. Subsequently, clinical model, radiomics signature (RS), and radiomics nomogram were established. The performance for prediction of Ki-67 expression status was validated using area under curve (AUC), calibration curve, Delong test, decision curve analysis (DCA). Prognostic prediction was assessed by survival curve and concordance index (C-index).

RESULTS

Tumour size was the only independent predictor of Ki-67 expression status. Five radiomics features were finally identified to construct the RS (AUC: training group, 0.821; validation group, 0.799). The radiomics nomogram achieved a higher AUC (training group, 0.841; validation group, 0.814) and clinical net benefit. Besides, the radiomics nomogram provided a highest C-index (training group, 0.841; validation group, 0.820) in predicting prognosis for ccRCC patients.

CONCLUSIONS

The radiomics nomogram can accurately predict the Ki-67 expression status and exhibit a great capacity for prognostic prediction in patients with ccRCC and may provide value for tailoring personalized treatment strategies and facilitating comprehensive clinical monitoring for ccRCC patients.

Multiparametric Characterization of Focal Cortical Dysplasia Using 3D MR Fingerprinting

OBJECTIVES

To develop a multiparametric machine-learning (ML) framework using high-resolution 3 dimensional (3D) magnetic resonance (MR) fingerprinting (MRF) data for quantitative characterization of focal cortical dysplasia (FCD).

MATERIALS

We included 119 subjects, 33 patients with focal epilepsy and histopathologically confirmed FCD, 60 age- and gender-matched healthy controls (HCs), and 26 disease controls (DCs). Subjects underwent whole-brain 3 Tesla MRF acquisition, the reconstruction of which generated T1 and T2 relaxometry maps. A 3D region of interest was manually created for each lesion, and z-score normalization using HC data was performed. We conducted 2D classification with ensemble models using MRF T1 and T2 mean and standard deviation from gray matter and white matter for FCD versus controls. Subtype classification additionally incorporated entropy and uniformity of MRF metrics, as well as morphometric features from the morphometric analysis program (MAP). We translated 2D results to individual probabilities using the percentage of slices above an adaptive threshold. These probabilities and clinical variables were input into a support vector machine for individual-level classification. Fivefold cross-validation was performed and performance metrics were reported using receiver-operating-characteristic-curve analyses.

RESULTS

FCD versus HC classification yielded mean sensitivity, specificity, and accuracy of 0.945, 0.980, and 0.962, respectively; FCD versus DC classification achieved 0.918, 0.965, and 0.939. In comparison, visual review of the clinical magnetic resonance imaging (MRI) detected 48% (16/33) of the lesions by official radiology report. In the subgroup where both clinical MRI and MAP were negative, the MRF-ML models correctly distinguished FCD patients from HCs and DCs in 98.3% of cross-validation trials for the magnetic resonance imaging negative group and MAP negative group. Type II versus non-type-II classification exhibited mean sensitivity, specificity, and accuracy of 0.835, 0.823, and 0.83, respectively; type IIa versus IIb classification showed 0.85, 0.9, and 0.87. In comparison, the transmantle sign was present in 58% (7/12) of the IIb cases.

INTERPRETATION

The MRF-ML framework presented in this study demonstrated strong efficacy in noninvasively classifying FCD from normal cortex and distinguishing FCD subtypes. ANN NEUROL 2024.

Primary tumor heterogeneity on pre-treatment [68Ga]Ga-PSMA PET/CT for the prediction of biochemical recurrence in prostate cancer

PURPOSE

The aim of this study is to research the value of the texture analysis of primary tumors in pre-treatment [68Ga]Ga-PSMA PET in the prediction of the development of biochemical recurrence (BCR) in prostate cancer patients who underwent definitive therapies.

METHODS

51 patients with prostate adenocarcinoma who had a pre-treatment [68Ga]Ga-PSMA-11 PET/CT and underwent definitive radiotherapy (RT) or radical prostatectomy (RP) were included in the study. Demographics, clinicopathologic features, the presence of BCR, and the last follow-up date of patients were recorded. Textural and conventional PET parameters (maximum standardized uptake value (SUVmax), total lesion-PSMA (TL-PSMA), and PSMA-tumor volume (PSMA-TV)) were obtained from PET/CT images using LifeX program. Parameters were grouped using the Youden index in ROC analysis. Factors predicting the BCR were determined using Cox regression analyses.

RESULTS

29 (56.9%) patients have received primary curative RT, while the remaining 22 (43.1%) patients have undergone RP. 5 (22.7%) patients with RP and 3 (10.3%) patients with curative RT have developed BCR during the follow-up. INTENSITY-BASED-minimum grey level (P=.050), GLCM-sum variance (P=.019), and GLCM-cluster prominence (P=.050) were associated with BCR in univariate analysis. INTENSITY-BASED-minimum grey level (P=.009) and GLCM-sum variance (P=.004) were found as independent predictors of BCR in the multivariate analysis.

CONCLUSION

Tumor heterogeneity on pre-treatment [68Ga]Ga-PSMA PET is associated with a high risk of BCR in PCa patients who underwent definitive therapies.

Multiparametric MRI radiomics for predicting disease-free survival and high-risk histopathological features for tumor recurrence in endometrial cancer

Background

Current preoperative imaging is insufficient to predict survival and tumor recurrence in endometrial cancer (EC), necessitating invasive procedures for risk stratification.

Purpose

To establish a multiparametric MRI radiomics model for predicting disease-free survival (DFS) and high-risk histopathologic features in EC.

Methods

This retrospective study included 71 patients with histopathology-proven EC and preoperative MRI over a 10-year period. Clinicopathology data were extracted from health records. Manual MRI segmentation was performed on T2-weighted (T2W), apparent diffusion coefficient (ADC) maps and dynamic contrast-enhanced T1-weighted images (DCE T1WI). Radiomic feature (RF) extraction was performed with PyRadiomics. Associations between RF and histopathologic features were assessed using logistic regression. Associations between DFS and RF or clinicopathologic features were assessed using the Cox proportional hazards model. All RF with univariate analysis p-value < 0.2 were included in elastic net analysis to build radiomic signatures.

Results

The 3-year DFS rate was 68% (95% CI = 57%-80%). There were no significant clinicopathologic predictors for DFS, whilst the radiomics signature was a strong predictor of DFS (p<0.001, HR 3.62, 95% CI 1.94, 6.75). From 107 RF extracted, significant predictive elastic net radiomic signatures were established for deep myometrial invasion (p=0.0097, OR 4.81, 95% CI 1.46, 15.79), hysterectomy grade (p=0.002, OR 5.12, 95% CI 1.82, 14.45), hysterectomy histology (p=0.0061, OR 18.25, 95% CI 2.29,145.24) and lymphovascular space invasion (p<0.001, OR 5.45, 95% CI 2.07, 14.36).

Conclusion

Multiparametric MRI radiomics has the potential to create a non-invasive a priori approach to predicting DFS and high-risk histopathologic features in EC.

Volumetric and textural analysis of PET/CT in patients with diffuse large B-cell lymphoma highlights the importance of novel MTVrate feature

OBJECTIVES

To investigate the prognostic value of clinical, volumetric, and radiomics-based textural parameters in baseline [18F]FDG-PET/CT scans of diffuse large B-cell lymphoma (DLBCL) patients.

METHODS

We retrospectively investigated baseline PET/CT scans and collected clinical data of fifty DLBCL patients. PET images were segmented semiautomatically to determine metabolic tumor volume (MTV), then the largest segmented lymphoma volume of interest (VOI) was used to extract first-, second-, and high-order textural features. A novel value, MTVrate was introduced as the quotient of the largest lesion's volume and total body MTV. Receiver operating characteristics (ROC) analyses were performed and 24-months progression-free survival (PFS) of low- and high-risk cohorts were compared by log-rank analyses. A machine learning algorithm was used to build a prognostic model from the available clinical, volumetric, and textural data based on logistic regression.

RESULTS

The area-under-the-curve (AUC) on ROC analysis was the highest of MTVrate at 0.74, followed by lactate-dehydrogenase, MTV, and skewness, with AUCs of 0.68, 0.63, and 0.55, respectively which parameters were also able to differentiate the PFS. A combined survival analysis including MTV and MTVrate identified a subgroup with particularly low PFS at 38%. In the machine learning-based model had an AUC of 0.83 and the highest relative importance was attributed to three textural features and both MTV and MTVrate as important predictors of PFS.

CONCLUSION

Individual evaluation of different biomarkers yielded only limited prognostic data, whereas a machine learning-based combined analysis had higher effectivity. MTVrate had the highest prognostic ability on individual analysis and, combined with MTV, it identified a patient group with particularly poor prognosis.

Automatic detection of cognitive impairment in patients with white matter hyperintensity and causal analysis of related factors using artificial intelligence of MRI

PURPOSE

White matter hyperintensity (WMH) is a common feature of brain aging, often linked with cognitive decline and dementia. This study aimed to employ deep learning and radiomics to develop models for detecting cognitive impairment in WMH patients and to analyze the causal relationships among cognitive impairment and related factors.

MATERIALS AND METHODS

A total of 79 WMH patients from hospital 1 were randomly divided into a training set (62 patients) and a testing set (17 patients). Additionally, 29 patients from hospital 2 were included as an independent testing set. All participants underwent formal neuropsychological assessments to determine cognitive status. Automated identification and segmentation of WMH were conducted using VB-net, with extraction of radiomics features from cortex, white matter, and nuclei. Four machine learning classifiers were trained on the training set and validated on the testing set to detect cognitive impairment. Model performances were evaluated and compared. Causal analyses were conducted among cortex, white matter, nuclei alterations, and cognitive impairment.

RESULTS

Among the models, the logistic regression (LR) model based on white matter features demonstrated the highest performance, achieving an AUC of 0.819 in the external test dataset. Causal analyses indicated that age, education level, alterations in cortex, white matter, and nuclei were causal factors of cognitive impairment.

CONCLUSION

The LR model based on white matter features exhibited high accuracy in detecting cognitive impairment in WMH patients. Furthermore, the possible causal relationships among alterations in cortex, white matter, nuclei, and cognitive impairment were elucidated.

Adaptive Machine Learning Approach for Importance Evaluation of Multimodal Breast Cancer Radiomic Features

Breast cancer holds the highest diagnosis rate among female tumors and is the leading cause of death among women. Quantitative analysis of radiological images shows the potential to address several medical challenges, including the early detection and classification of breast tumors. In the P.I.N.K study, 66 women were enrolled. Their paired Automated Breast Volume Scanner (ABVS) and Digital Breast Tomosynthesis (DBT) images, annotated with cancerous lesions, populated the first ABVS+DBT dataset. This enabled not only a radiomic analysis for the malignant vs. benign breast cancer classification, but also the comparison of the two modalities. For this purpose, the models were trained using a leave-one-out nested cross-validation strategy combined with a proper threshold selection approach. This approach provides statistically significant results even with medium-sized data sets. Additionally it provides distributional variables of importance, thus identifying the most informative radiomic features. The analysis proved the predictive capacity of radiomic models even using a reduced number of features. Indeed, from tomography we achieved AUC-ROC 89.9 % using 19 features and 92.1 % using 7 of them; while from ABVS we attained an AUC-ROC of 72.3 % using 22 features and 85.8 % using only 3 features. Although the predictive power of DBT outperforms ABVS, when comparing the predictions at the patient level, only 8.7% of lesions are misclassified by both methods, suggesting a partial complementarity. Notably, promising results (AUC-ROC ABVS-DBT 71.8 % - 74.1 % ) were achieved using non-geometric features, thus opening the way to the integration of virtual biopsy in medical routine.

RadShap: An Explanation Tool for Highlighting the Contributions of Multiple Regions of Interest to the Prediction of Radiomic Models

Explaining the decisions made by a radiomic model is of significant interest, as it can provide valuable insights into the information learned by complex models and foster trust in well-performing ones, thereby facilitating their clinical adoption. Promising radiomic approaches that aggregate information from multiple regions within an image currently lack suitable explanation tools that could identify the regions that most significantly influence their decisions. Here we present a model- and modality-agnostic tool (RadShap, https://github.com/ncaptier/radshap), based on Shapley values, that explains the predictions of multiregion radiomic models by highlighting the contribution of each individual region. Methods: The explanation tool leverages Shapley values to distribute the aggregative radiomic model's output among all the regions of interest of an image, highlighting their individual contribution. RadShap was validated using a retrospective cohort of 130 patients with advanced non-small cell lung cancer undergoing first-line immunotherapy. Their baseline PET scans were used to build 1,000 synthetic tasks to evaluate the degree of alignment between the tool's explanations and our data generation process. RadShap's potential was then illustrated through 2 real case studies by aggregating information from all segmented tumors: the prediction of the progression-free survival of the non-small cell lung cancer patients and the classification of the histologic tumor subtype. Results: RadShap demonstrated strong alignment with the ground truth, with a median frequency of 94% for consistently explained predictions in the synthetic tasks. In both real-case studies, the aggregative models yielded superior performance to the single-lesion models (average [±SD] time-dependent area under the receiver operating characteristic curve was 0.66 ± 0.02 for the aggregative survival model vs. 0.55 ± 0.04 for the primary tumor survival model). The tool's explanations provided relevant insights into the behavior of the aggregative models, highlighting that for the classification of the histologic subtype, the aggregative model used information beyond the biopsy site to correctly classify patients who were initially misclassified by a model focusing only on the biopsied tumor. Conclusion: RadShap aligned with ground truth explanations and provided valuable insights into radiomic models' behaviors. It is implemented as a user-friendly Python package with documentation and tutorials, facilitating its smooth integration into radiomic pipelines.

Prediction of HER2 Expression in Gastric Adenocarcinoma Based On Preoperative Noninvasive Multimodal 18F-FDG PET/CT Imaging

RATIONALE AND OBJECTIVES

This study aims to investigate the role of a flourine-18 fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT) multimodal radiomics model in predicting the status of human epidermal growth factor receptor 2 (HER2) expression preoperatively in cases of gastric adenocarcinoma.

MATERIALS AND METHODS

This retrospective study included 133 patients with gastric adenocarcinoma who were classified into training (n = 93) and validation (n = 40) cohorts in a ratio of 7:3. Features were selected using Least Absolute Shrinkage and Selection Operator and Extreme Gradient Boosting (XGBoost) methods; further, prediction models were constructed using logistic regression and XGBoost. These models were evaluated and validated using area under the curve (AUC), decision curves, and calibration curves to select the best-performing model.

RESULTS

Six different models were established to predict HER2 expression. Among these, the comprehensive model, which integrates seven clinical features, one CT feature, and five PET features, demonstrated AUC values of 0.95 (95% confidence interval [CI]: 0.89-1.00) and 0.76 (95% CI: 0.52-1.00) in the training and validation cohorts, respectively. Compared with other models, this model exhibited a superior net benefit on the decision curve and demonstrated good alignment agreement with the observed values on the calibration curve. Based on these findings, we constructed a nomogram for visualizing the model, providing a noninvasive preoperative method for predicting HER2 expression.

CONCLUSION

The preoperative 18F-FDG PET/CT multimodal radiomics model can effectively predict HER2 expression in patients with gastric adenocarcinoma, thereby guiding clinical decision-making and advancing the field of precision medicine.

Short-term mortality prediction in acute pulmonary embolism: Radiomics values of skeletal muscle and intramuscular adipose tissue

BACKGROUND

Acute pulmonary embolism (APE) is a potentially life-threatening disorder, emphasizing the importance of accurate risk stratification and survival prognosis. The exploration of imaging biomarkers that can reflect patient survival holds the potential to further enhance the stratification of APE patients, enabling personalized treatment and early intervention. Therefore, in this study, we develop computed tomography pulmonary angiography (CTPA) radiomic signatures for the prognosis of 7- and 30-day all-cause mortality in patients with APE.

METHODS

Diagnostic CTPA images from 829 patients with APE were collected. Two hundred thirty-four features from each skeletal muscle (SM), intramuscular adipose tissue (IMAT) and both tissues combined (SM + IMAT) were calculated at the level of thoracic vertebra 12. Radiomic signatures were derived using 10 times repeated three-fold cross-validation on the training data for SM, IMAT and SM + IMAT for predicting 7- and 30-day mortality independently. The performance of the radiomic signatures was then evaluated on held-out test data and compared with the simplified pulmonary embolism severity index (sPESI) score, a well-established biomarker for risk stratification in APE. Predictive accuracy was assessed by the area under the receiver operating characteristic curve (AUC) with a 95% confidence interval (CI), sensitivity and specificity.

RESULTS

The radiomic signatures based on IMAT and a combination of SM and IMAT (SM + IMAT) achieved moderate performance for the prediction of 30-day mortality on test data (IMAT: AUC = 0.68, 95% CI [0.57-0.78], sensitivity = 0.57, specificity = 0.73; SM + IMAT: AUC = 0.70, 95% CI [0.60-0.79], sensitivity = 0.74, specificity = 0.54). Radiomic signatures developed for predicting 7-day all-cause mortality showed overall low performance. The clinical signature, that is, sPESI, achieved slightly better performance in terms of AUC on test data compared with the radiomic signatures for the prediction of both 7- and 30-day mortality on the test data (7 days: AUC = 0.73, 95% CI [0.67-0.79], sensitivity = 0.92, specificity = 0.16; 30 days: AUC = 0.74, 95% CI [0.66-0.82], sensitivity = 0.97, specificity = 0.16).

CONCLUSIONS

We developed and tested radiomic signatures for predicting 7- and 30-day all-cause mortality in APE using a multicentric retrospective dataset. The present multicentre work shows that radiomics parameters extracted from SM and IMAT can predict 30-day all-cause mortality in patients with APE.

Machine-Learning and Radiomics-Based Preoperative Prediction of Ki-67 Expression in Glioma Using MRI Data

BACKGROUND

Gliomas are the most common primary brain tumours and constitute approximately half of all malignant glioblastomas. Unfortunately, patients diagnosed with malignant glioblastomas typically survive for less than a year. In light of this circumstance, genotyping is an effective means of categorising gliomas. The Ki67 proliferation index, a widely used marker of cellular proliferation in clinical contexts, has demonstrated potential for predicting tumour classification and prognosis. In particular, magnetic resonance imaging (MRI) plays a vital role in the diagnosis of brain tumours. Using MRI to extract glioma-related features and construct a machine learning model offers a viable avenue to classify and predict the level of Ki67 expression.

METHODS

This study retrospectively collected MRI data and postoperative immunohistochemical results from 613 glioma patients from the First Affliated Hospital of Nanjing Medical University. Subsequently, we performed registration and skull stripping on the four MRI modalities: T1-weighted (T1), T2-weighted (T2), T1-weighted with contrast enhancement (T1CE), and Fluid Attenuated Inversion Recovery (FLAIR). Each modality's segmentation yielded three distinct tumour regions. Following segmentation, a comprehensive set of features encompassing texture, first-order, and shape attributes were extracted from these delineated regions. Feature selection was conducted using the least absolute shrinkage and selection operator (LASSO) algorithm with subsequent sorting to identify the most important features. These selected features were further analysed using correlation analysis to finalise the selection for machine learning model development. Eight models: logistic regression (LR), naive bayes, decision tree, gradient boosting tree, and support vector classification (SVM), random forest (RF), XGBoost, and LightGBM were used to objectively classify Ki67 expression.

RESULTS

In total, 613 patients were enroled in the study, and 24,455 radiomic features were extracted from each patient's MRI. These features were eventually reduced to 36 after LASSO screening, RF importance ranking, and correlation analysis. Among all the tested machine learning models, LR and linear SVM exhibited superior performance. LR achieved the highest area under the curve score of 0.912 ± 0.036, while linear SVM obtained the top accuracy with a score of 0.884 ± 0.031.

CONCLUSION

This study introduced a novel approach for classifying Ki67 expression levels using MRI, which has been proven to be highly effective. With the LR model at its core, our method demonstrated its potential in signalling a promising avenue for future research. This innovative approach of predicting Ki67 expression based on MRI features not only enhances our understanding of cell activity but also represents a significant leap forward in brain glioma research. This underscores the potential of integrating machine learning with medical imaging to aid in the diagnosis and prognosis of complex diseases.

Temporary treatment cessation compared with continuation of tyrosine kinase inhibitors for adults with renal cancer: the STAR non-inferiority RCT

Background

There is interest in using treatment breaks in oncology, to reduce toxicity without compromising efficacy.

Trial design

A Phase II/III multicentre, open-label, parallel-group, randomised controlled non-inferiority trial assessing treatment breaks in patients with renal cell carcinoma.

Methods

Patients with locally advanced or metastatic renal cell carcinoma, starting tyrosine kinase inhibitor as first-line treatment at United Kingdom National Health Service hospitals.

Interventions

At trial entry, patients were randomised (1 : 1) to a drug-free interval strategy or a conventional continuation strategy. After 24 weeks of treatment with sunitinib/pazopanib, drug-free interval strategy patients took up a treatment break until disease progression with additional breaks dependent on disease response and patient choice. Conventional continuation strategy patients continued on treatment. Both trial strategies continued until treatment intolerance, disease progression on treatment, withdrawal or death.

Objective

To determine if a drug-free interval strategy is non-inferior to a conventional continuation strategy in terms of the co-primary outcomes of overall survival and quality-adjusted life-years.

Co-primary outcomes

For non-inferiority to be concluded, a margin of ≤ 7.5% in overall survival and ≤ 10% in quality-adjusted life-years was required in both intention-to-treat and per-protocol analyses. This equated to the 95% confidence interval of the estimates being above 0.812 and -0.156, respectively. Quality-adjusted life-years were calculated using the utility index of the EuroQol-5 Dimensions questionnaire.

Results

Nine hundred and twenty patients were randomised (461 conventional continuation strategy vs. 459 drug-free interval strategy) from 13 January 2012 to 12 September 2017. Trial treatment and follow-up stopped on 31 December 2020. Four hundred and eighty-eight (53.0%) patients [240 (52.1%) vs. 248 (54.0%)] continued on trial post week 24. The median treatment-break length was 87 days. Nine hundred and nineteen patients were included in the intention-to-treat analysis (461 vs. 458) and 871 patients in the per-protocol analysis (453 vs. 418). For overall survival, non-inferiority was concluded in the intention-to-treat analysis but not in the per-protocol analysis [hazard ratio (95% confidence interval) intention to treat 0.97 (0.83 to 1.12); per-protocol 0.94 (0.80 to 1.09) non-inferiority margin: 95% confidence interval ≥ 0.812, intention to treat: 0.83 > 0.812 non-inferior, per-protocol: 0.80 < 0.812 not non-inferior]. Therefore, a drug-free interval strategy was not concluded to be non-inferior to a conventional continuation strategy in terms of overall survival. For quality-adjusted life-years, non-inferiority was concluded in both the intention-to-treat and per-protocol analyses [marginal effect (95% confidence interval) intention to treat -0.05 (-0.15 to 0.05); per-protocol 0.04 (-0.14 to 0.21) non-inferiority margin: 95% confidence interval ≥ -0.156]. Therefore, a drug-free interval strategy was concluded to be non-inferior to a conventional continuation strategy in terms of quality-adjusted life-years.

Limitations

The main limitation of the study is the fewer than expected overall survival events, resulting in lower power for the non-inferiority comparison.

Future work

Future studies should investigate treatment breaks with more contemporary treatments for renal cell carcinoma.

Conclusions

Non-inferiority was shown for the quality-adjusted life-year end point but not for overall survival as pre-defined. Nevertheless, despite not meeting the primary end point of non-inferiority as per protocol, the study suggested that a treatment-break strategy may not meaningfully reduce life expectancy, does not reduce quality of life and has economic benefits. Although the treating clinicians' perspectives were not formally collected, the fact that clinicians recruited a large number of patients over a long period suggests support for the study and provides clear evidence that a treatment-break strategy for patients with renal cell carcinoma receiving tyrosine kinase inhibitor therapy is feasible.

Trial registration

This trial is registered as ISRCTN06473203.

Funding

This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment Programme (NIHR award ref: 09/91/21) and is published in full in Health Technology Assessment; Vol. 28, No. 45. See the NIHR Funding and Awards website for further award information.

Impact of quantitative CT texture analysis on the outcome of CT-guided bone biopsy

Texture analysis can provide new imaging-based biomarkers. Texture analysis derived from computed tomography (CT) might be able to better characterize patients undergoing CT-guided percutaneous bone biopsy. The present study evaluated this and correlated texture features with bioptic outcome in patients undergoing CT-guided bone biopsy. Overall, 123 patients (89 female patients, 72.4 %) were included into the present study. All patients underwent CT-guided percutaneous bone biopsy with an 11 Gauge coaxial needle. Clinical parameters and quantitative imaging features were investigated. Random forest classifier was used to predict a positive biopsy result. Overall, 69 patients had osteolytic metastasis (56.1 %) and 54 had osteoblastic metastasis (43.9 %). The overall positive biopsy rate was 72 %. The developed radiomics model demonstrated a prediction accuracy of a positive biopsy result with an AUC of 0.75 [95 %CI 0.65 - 0.85]. In a subgroup of breast cancer patients, the model achieved an AUC of 0.85 [95 %CI 0.73 - 0.96]. In the subgroup of non-breast cancer patients, the signature achieved an AUC of 0.80 [95 %CI 0.60 - 0.99]. Quantitative CT imaging findings comprised of conventional and texture features can aid to predict the bioptic result of CT-guided bone biopsies. The developed radiomics signature aids in clinical decision-making, and could identify patients at risk for a negative biopsy.

Radiomics for clinical decision support in radiation oncology

Radiomics is a promising tool for the development of quantitative biomarkers to support clinical decision-making. It has been shown to improve the prediction of response to treatment and outcome in different settings, particularly in the field of radiation oncology by optimising the dose delivery solutions and reducing the rate of radiation-induced side effects, leading to a fully personalised approach. Despite the promising results offered by radiomics at each of these stages, standardised methodologies, reproducibility and interpretability of results are still lacking, limiting the potential clinical impact of these tools. In this review, we briefly describe the principles of radiomics and the most relevant applications of radiomics at each stage of cancer management in the framework of radiation oncology. Furthermore, the integration of radiomics into clinical decision support systems is analysed, defining the challenges and offering possible solutions for translating radiomics into a clinically applicable tool.

Delta radiomics: an updated systematic review

BACKGROUND

Radiomics can provide quantitative features from medical imaging that can be correlated with various biological features and diverse clinical endpoints. Delta radiomics, on the other hand, consists in the analysis of feature variation at different acquisition time points, usually before and after therapy. The aim of this study was to provide a systematic review of the different delta radiomics approaches.

METHODS

Eligible articles were searched in Embase, Pubmed, and ScienceDirect using a search string that included free text and/or Medical Subject Headings (MeSH) with 3 key search terms: 'radiomics,' 'texture,' and 'delta.' Studies were analyzed using QUADAS-2 and the RQS tool.

RESULTS

Forty-eight studies were finally included. The studies were divided into preclinical/methodological (5 studies, 10.4%); rectal cancer (6 studies, 12.5%); lung cancer (12 studies, 25%); sarcoma (5 studies, 10.4%); prostate cancer (3 studies, 6.3%), head and neck cancer (6 studies, 12.5%); gastrointestinal malignancies excluding rectum (7 studies, 14.6%) and other disease sites (4 studies, 8.3%). The median RQS of all studies was 25% (mean 21% ± 12%), with 13 studies (30.2%) achieving a quality score < 10% and 22 studies (51.2%) < 25%.

CONCLUSIONS

Delta radiomics shows potential benefit for several clinical endpoints in oncology, such asdifferential diagnosis, prognosis and prediction of treatment response, evaluation of side effects. Nevertheless, the studies included in this systematic review suffer from the bias of overall low methodological rigor, so that the conclusions are currently heterogeneous, not robust and hardly replicable. Further research with prospective and multicenter studies is needed for the clinical validation of delta radiomics approaches.

MRI-based deep learning and radiomics for prediction of occult cervical lymph node metastasis and prognosis in early-stage oral and oropharyngeal squamous cell carcinoma: a diagnostic study

INTRODUCTION

The incidence of occult cervical lymph node metastases (OCLNM) is reported to be 20-30% in early-stage oral cancer and oropharyngeal cancer. There is a lack of an accurate diagnostic method to predict occult lymph node metastasis and to help surgeons make precise treatment decisions.

AIM

To construct and evaluate a preoperative diagnostic method to predict OCLNM in early-stage oral and oropharyngeal squamous cell carcinoma (OC and OP SCC) based on deep learning features (DLFs) and radiomics features.

METHODS

A total of 319 patients diagnosed with early-stage OC or OP SCC were retrospectively enrolled and divided into training, test and external validation sets. Traditional radiomics features and DLFs were extracted from their MRI images. The least absolute shrinkage and selection operator (LASSO) analysis was employed to identify the most valuable features. Prediction models for OCLNM were developed using radiomics features and DLFs. The effectiveness of the models and their clinical applicability were evaluated using the area under the curve (AUC), decision curve analysis (DCA), and survival analysis.

RESULTS

Seventeen prediction models were constructed. The Resnet50 deep learning (DL) model based on the combination of radiomics and DL features achieves the optimal performance, with AUC values of 0.928 (95% CI: 0.881-0.975), 0.878 (95% CI: 0.766-0.990), 0.796 (95% CI: 0.666-0.927), and 0.834 (95% CI: 0.721-0.947) in the training, test, external validation set1, and external validation set2, respectively. Moreover, the Resnet50 model has great prediction value of prognosis in patients with early-stage OC and OP SCC.

CONCLUSION

The proposed MRI-based Resnet50 DL model demonstrated high capability in diagnosis of OCLNM and prognosis prediction in the early-stage OC and OP SCC. The Resnet50 model could help refine the clinical diagnosis and treatment of the early-stage OC and OP SCC.

Fully automated hybrid approach on conventional MRI for triaging clinically significant liver fibrosis: A multi-center cohort study

Establishing reliable noninvasive tools to precisely diagnose clinically significant liver fibrosis (SF, ≥F2) remains an unmet need. We aimed to build a combined radiomics-clinic (CoRC) model for triaging SF and explore the additive value of the CoRC model to transient elastography-based liver stiffness measurement (FibroScan, TE-LSM). This retrospective study recruited 595 patients with biopsy-proven liver fibrosis at two centers between January 2015 and December 2021. At Center 1, the patients before December 2018 were randomly split into training (276) and internal test (118) sets, the remaining were time-independent as a temporal test set (96). Another data set (105) from Center 2 was collected for external testing. Radiomics scores were built with selected features from Deep learning-based (ResUNet) automated whole liver segmentations on MRI (T2FS and delayed enhanced-T1WI). The CoRC model incorporated radiomics scores and relevant clinical variables with logistic regression, comparing routine approaches. Diagnostic performance was evaluated by the area under the receiver operating characteristic curve (AUC). The additive value of the CoRC model to TE-LSM was investigated, considering necroinflammation. The CoRC model achieved AUCs of 0.79 (0.70, 0.86), 0.82 (0.73, 0.89), and 0.81 (0.72-0.91), outperformed FIB-4, APRI (all p < 0.05) in the internal, temporal, and external test sets and maintained the discriminatory power in G0-1 subgroups (AUCs range, 0.85-0.86; all p < 0.05). The AUCs of joint CoRC-LSM model were 0.86 (0.79-0.94), and 0.81 (0.72-0.90) in the internal and temporal sets (p = 0.01). The CoRC model was useful for triaging SF, and may add value to TE-LSM.

Treatment Response Biomarkers: Working Toward Personalized Radiotherapy for Lung Cancer

Owing to major advances in the field of radiation oncology, patients with lung cancer can now receive technically individualized radiotherapy treatments. Nevertheless, in the era of precision oncology, radiotherapy-based treatment selection needs to be improved as many patients do not benefit or are not offered optimum therapies. Cost-effective robust biomarkers can address this knowledge gap and lead to individuals being offered more bespoke treatments leading to improved outcome. This narrative review discusses some of the current achievements and challenges in the realization of personalized radiotherapy delivery in patients with lung cancer.

Atherosclerotic carotid artery disease Radiomics: A systematic review with meta-analysis and radiomic quality score assessment

BACKGROUND

Stroke, a leading global cause of mortality and neurological disability, is often associated with atherosclerotic carotid artery disease. Distinguishing between symptomatic and asymptomatic carotid artery disease is crucial for appropriate treatment decisions. Radiomics, a quantitative image analysis technique, and ML have emerged as promising tools in medical imaging, including neuroradiology. This systematic review and meta-analysis aimed to evaluate the methodological quality of studies employing radiomics for atherosclerotic carotid artery disease analysis and ML algorithms for culprit plaque identification using CT or MRI.

MATERIALS AND METHODS

Pubmed, WoS and Scopus databases were searched for relevant studies published from January 2005 to May 2023. RQS assessed methodological quality of studies included in the review. QUADAS-2 assessed the risk of bias. A meta-analysis and three meta regressions were conducted on study performance based on model type, imaging modality and segmentation method.

RESULTS

RQS assessed methodological quality, revealing an overall low score and consistent findings with other radiology domains. QUADAS-2 indicated an overall low risk, except for a single study with high bias. The meta-analysis demonstrated that radiomics-based ML models for predicting culprit plaques had a satisfactory performance, with an AUC of 0.85, surpassing clinical models. However, combining radiomics with clinical features yielded the highest AUC of 0.89. Meta-regression analyses confirmed these findings. MRI-based models slightly outperformed CT-based ones, but the difference was not significant.

CONCLUSION

In conclusion, radiomics and ML hold promise for assessing carotid plaque vulnerability, aiding in early cerebrovascular event prediction. Combining radiomics with clinical data enhances predictive performance.

Impact of harmonization on the reproducibility of MRI radiomic features when using different scanners, acquisition parameters, and image pre-processing techniques: a phantom study

This study investigated the impact of ComBat harmonization on the reproducibility of radiomic features extracted from magnetic resonance images (MRI) acquired on different scanners, using various data acquisition parameters and multiple image pre-processing techniques using a dedicated MRI phantom. Four scanners were used to acquire an MRI of a nonanatomic phantom as part of the TCIA RIDER database. In fast spin-echo inversion recovery (IR) sequences, several inversion durations were employed, including 50, 100, 250, 500, 750, 1000, 1500, 2000, 2500, and 3000 ms. In addition, a 3D fast spoiled gradient recalled echo (FSPGR) sequence was used to investigate several flip angles (FA): 2, 5, 10, 15, 20, 25, and 30 degrees. Nineteen phantom compartments were manually segmented. Different approaches were used to pre-process each image: Bin discretization, Wavelet filter, Laplacian of Gaussian, logarithm, square, square root, and gradient. Overall, 92 first-, second-, and higher-order statistical radiomic features were extracted. ComBat harmonization was also applied to the extracted radiomic features. Finally, the Intraclass Correlation Coefficient (ICC) and Kruskal-Wallis's (KW) tests were implemented to assess the robustness of radiomic features. The number of non-significant features in the KW test ranged between 0-5 and 29-74 for various scanners, 31-91 and 37-92 for three times tests, 0-33 to 34-90 for FAs, and 3-68 to 65-89 for IRs before and after ComBat harmonization, with different image pre-processing techniques, respectively. The number of features with ICC over 90% ranged between 0-8 and 6-60 for various scanners, 11-75 and 17-80 for three times tests, 3-83 to 9-84 for FAs, and 3-49 to 3-63 for IRs before and after ComBat harmonization, with different image pre-processing techniques, respectively. The use of various scanners, IRs, and FAs has a great impact on radiomic features. However, the majority of scanner-robust features is also robust to IR and FA. Among the effective parameters in MR images, several tests in one scanner have a negligible impact on radiomic features. Different scanners and acquisition parameters using various image pre-processing might affect radiomic features to a large extent. ComBat harmonization might significantly impact the reproducibility of MRI radiomic features.

A deep learning model based on contrast-enhanced computed tomography for differential diagnosis of gallbladder carcinoma

BACKGROUND

Gallbladder carcinoma (GBC) is highly malignant, and its early diagnosis remains difficult. This study aimed to develop a deep learning model based on contrast-enhanced computed tomography (CT) images to assist radiologists in identifying GBC.

METHODS

We retrospectively enrolled 278 patients with gallbladder lesions (> 10 mm) who underwent contrast-enhanced CT and cholecystectomy and divided them into the training (n = 194) and validation (n = 84) datasets. The deep learning model was developed based on ResNet50 network. Radiomics and clinical models were built based on support vector machine (SVM) method. We comprehensively compared the performance of deep learning, radiomics, clinical models, and three radiologists.

RESULTS

Three radiomics features including LoG_3.0 gray-level size zone matrix zone variance, HHL first-order kurtosis, and LHL gray-level co-occurrence matrix dependence variance were significantly different between benign gallbladder lesions and GBC, and were selected for developing radiomics model. Multivariate regression analysis revealed that age ≥ 65 years [odds ratios (OR) = 4.4, 95% confidence interval (CI): 2.1-9.1, P < 0.001], lesion size (OR = 2.6, 95% CI: 1.6-4.1, P < 0.001), and CA-19-9 > 37 U/mL (OR = 4.0, 95% CI: 1.6-10.0, P = 0.003) were significant clinical risk factors of GBC. The deep learning model achieved the area under the receiver operating characteristic curve (AUC) values of 0.864 (95% CI: 0.814-0.915) and 0.857 (95% CI: 0.773-0.942) in the training and validation datasets, which were comparable with radiomics, clinical models and three radiologists. The sensitivity of deep learning model was the highest both in the training [90% (95% CI: 82%-96%)] and validation [85% (95% CI: 68%-95%)] datasets.

CONCLUSIONS

The deep learning model may be a useful tool for radiologists to distinguish between GBC and benign gallbladder lesions.

Prediction of breast cancer and axillary positive-node response to neoadjuvant chemotherapy based on multi-parametric magnetic resonance imaging radiomics models

PURPOSE

Accurate identification of primary breast cancer and axillary positive-node response to neoadjuvant chemotherapy (NAC) is important for determining appropriate surgery strategies. We aimed to develop combining models based on breast multi-parametric magnetic resonance imaging and clinicopathologic characteristics for predicting therapeutic response of primary tumor and axillary positive-node prior to treatment.

MATERIALS AND METHODS

A total of 268 breast cancer patients who completed NAC and underwent surgery were enrolled. Radiomics features and clinicopathologic characteristics were analyzed through the analysis of variance and the least absolute shrinkage and selection operator algorithm. Finally, 24 and 28 optimal features were selected to construct machine learning models based on 6 algorithms for predicting each clinical outcome, respectively. The diagnostic performances of models were evaluated in the testing set by the area under the curve (AUC), sensitivity, specificity, and accuracy.

RESULTS

Of the 268 patients, 94 (35.1 %) achieved breast cancer pathological complete response (bpCR) and of the 240 patients with clinical positive-node, 120 (50.0 %) achieved axillary lymph node pathological complete response (apCR). The multi-layer perception (MLP) algorithm yielded the best diagnostic performances in predicting apCR with an AUC of 0.825 (95 % CI, 0.764-0.886) and an accuracy of 77.1 %. And MLP also outperformed other models in predicting bpCR with an AUC of 0.852 (95 % CI, 0.798-0.906) and an accuracy of 81.3 %.

CONCLUSIONS

Our study established non-invasive combining models to predict the therapeutic response of primary breast cancer and axillary positive-node prior to NAC, which may help to modify preoperative treatment and determine post-NAC surgery strategy.

Radiomics Studies on Ischemic Stroke and Carotid Atherosclerotic Disease: A Reporting Quality Assessment

Objective: To assess the reporting quality of radiomics studies on ischemic stroke, intracranial and carotid atherosclerotic disease using the Image Biomarker Standardization Initiative (IBSI) reporting guidelines with the aim of finding avenues of improvement for future publications. Method: PubMed database was searched to identify relevant radiomics studies. Of 560 articles, 41 original research articles were included in this analysis. Based on IBSI radiomics reporting guidelines, checklists for CT-based and MRI-based studies were created to allow a structured and comprehensive evaluation of each study's adherence to these guidelines. Results: The main topics covered included radiomics studies were ischemic stroke, intracranial artery disease, and carotid atherosclerotic disease. The reporting checklist median score was 17/40 for the 20 CT-based radiomics studies and 22.5/50 for the 20 MRI-based studies. Basic items like imaging modality, region of interest, and image biomarker set utilized were included in all studies. However, details regarding image acquisition and reconstruction, post-acquisition image processing, and image biomarkers computation were inconsistently detailed across studies. Conclusion: The overall reporting quality of the included radiomics studies was suboptimal. These findings underscore a pressing need for improved reporting practices in radiomics research, to ensure validation and reproducibility of results. Our study provides insights into current reporting standards and highlights specific areas where adherence to IBSI guidelines could be significantly improved.

A novel assessment of whole-mount Gleason grading in prostate cancer to identify candidates for radical prostatectomy: a machine learning-based multiomics study

Purpose: This study aims to assess whole-mount Gleason grading (GG) in prostate cancer (PCa) accurately using a multiomics machine learning (ML) model and to compare its performance with biopsy-proven GG (bxGG) assessment. Materials and Methods: A total of 146 patients with PCa recruited in a pilot study of a prospective clinical trial (NCT02659527) were retrospectively included in the side study, all of whom underwent 68Ga-PSMA-11 integrated positron emission tomography (PET) / magnetic resonance (MR) before radical prostatectomy (RP) between May 2014 and April 2020. To establish a multiomics ML model, we quantified PET radiomics features, pathway-level genomics features from whole exome sequencing, and pathomics features derived from immunohistochemical staining of 11 biomarkers. Based on the multiomics dataset, five ML models were established and validated using 100-fold Monte Carlo cross-validation. Results: Among five ML models, the random forest (RF) model performed best in terms of the area under the curve (AUC). Compared to bxGG assessment alone, the RF model was superior in terms of AUC (0.87 vs 0.75), specificity (0.72 vs 0.61), positive predictive value (0.79 vs 0.75), and accuracy (0.78 vs 0.77) and showed slightly decreased sensitivity (0.83 vs 0.89) and negative predictive value (0.80 vs 0.81). Among the feature categories, bxGG was identified as the most important feature, followed by pathomics, clinical, radiomics and genomics features. The three important individual features were bxGG, PSA staining and one intensity-related radiomics feature. Conclusion: The findings demonstrate a superior assessment of the developed multiomics-based ML model in whole-mount GG compared to the current clinical baseline of bxGG. This enables personalized patient management by identifying high-risk PCa patients for RP.

Radiomics in differential diagnosis of Wilms tumor and neuroblastoma with adrenal location in children

OBJECTIVES

Machine learning methods can be applied successfully to various medical imaging tasks. Our aim with this study was to build a robust classifier using radiomics and clinical data for preoperative diagnosis of Wilms tumor (WT) or neuroblastoma (NB) in pediatric abdominal CT.

MATERIAL AND METHODS

This is a single-center retrospective study approved by the Institutional Ethical Board. CT scans of consecutive patients diagnosed with WT or NB admitted to our hospital from January 2005 to December 2021 were evaluated. Three distinct datasets based on clinical centers and CT machines were curated. Robust, non-redundant, high variance, and relevant radiomics features were selected using data science methods. Clinically relevant variables were integrated into the final model. Dice score for similarity of tumor ROI, Cohen's kappa for interobserver agreement among observers, and AUC for model selection were used.

RESULTS

A total of 147 patients, including 90 WT (mean age 34.78 SD: 22.06 months; 43 male) and 57 NB (mean age 23.77 SD:22.56 months; 31 male), were analyzed. After binarization at 24 months cut-off, there was no statistically significant difference between the two groups for age (p = .07) and gender (p = .54). CT clinic radiomics combined model achieved an F1 score of 0.94, 0.93 accuracy, and an AUC 0.96.

CONCLUSION

In conclusion, the CT-based clinic-radiologic-radiomics combined model could noninvasively predict WT or NB preoperatively. Notably, that model correctly predicted two patients, which none of the radiologists could correctly predict. This model may serve as a noninvasive preoperative predictor of NB/WT differentiation in CT, which should be further validated in large prospective models.

CLINICAL RELEVANCE STATEMENT

CT-based clinic-radiologic-radiomics combined model could noninvasively predict Wilms tumor or neuroblastoma preoperatively.

KEY POINTS

• CT radiomics features can predict Wilms tumor or neuroblastoma from abdominal CT preoperatively. • Integrating clinic variables may further improve the performance of the model. • The performance of the combined model is equal to or greater than human readers, depending on the lesion size.

Development and benchmarking of a Deep Learning-based MRI-guided gross tumor segmentation algorithm for Radiomics analyses in extremity soft tissue sarcomas

BACKGROUND

Volume of interest (VOI) segmentation is a crucial step for Radiomics analyses and radiotherapy (RT) treatment planning. Because it can be time-consuming and subject to inter-observer variability, we developed and tested a Deep Learning-based automatic segmentation (DLBAS) algorithm to reproducibly predict the primary gross tumor as VOI for Radiomics analyses in extremity soft tissue sarcomas (STS).

METHODS

A DLBAS algorithm was trained on a cohort of 157 patients and externally tested on an independent cohort of 87 patients using contrast-enhanced MRI. Manual tumor delineations by a radiation oncologist served as ground truths (GTs). A benchmark study with 20 cases from the test cohort compared the DLBAS predictions against manual VOI segmentations of two residents (ERs) and clinical delineations of two radiation oncologists (ROs). The ROs rated DLBAS predictions regarding their direct applicability.

RESULTS

The DLBAS achieved a median dice similarity coefficient (DSC) of 0.88 against the GTs in the entire test cohort (interquartile range (IQR): 0.11) and a median DSC of 0.89 (IQR 0.07) and 0.82 (IQR 0.10) in comparison to ERs and ROs, respectively. Radiomics feature stability was high with a median intraclass correlation coefficient of 0.97, 0.95 and 0.94 for GTs, ERs, and ROs, respectively. DLBAS predictions were deemed clinically suitable by the two ROs in 35% and 20% of cases, respectively.

CONCLUSION

The results demonstrate that the DLBAS algorithm provides reproducible VOI predictions for radiomics feature extraction. Variability remains regarding direct clinical applicability of predictions for RT treatment planning.

Development and validation of a machine learning-based 18F-fluorodeoxyglucose PET/CT radiomics signature for predicting gastric cancer survival

BACKGROUND

Survival prognosis of patients with gastric cancer (GC) often influences physicians' choice of their follow-up treatment. This study aimed to develop a positron emission tomography (PET)-based radiomics model combined with clinical tumor-node-metastasis (TNM) staging to predict overall survival (OS) in patients with GC.

METHODS

We reviewed the clinical information of a total of 327 patients with pathological confirmation of GC undergoing 18 F-fluorodeoxyglucose (18 F-FDG) PET scans. The patients were randomly classified into training (n = 229) and validation (n = 98) cohorts. We extracted 171 PET radiomics features from the PET images and determined the PET radiomics scores (RS) using the least absolute shrinkage and selection operator (LASSO) and random survival forest (RSF). A radiomics model, including PET RS and clinical TNM staging, was constructed to predict the OS of patients with GC. This model was evaluated for discrimination, calibration, and clinical usefulness.

RESULTS

On multivariate COX regression analysis, the difference between age, carcinoembryonic antigen (CEA), clinical TNM, and PET RS in GC patients was statistically significant (p < 0.05). A radiomics model was developed based on the results of COX regression. The model had the Harrell's concordance index (C-index) of 0.817 in the training cohort and 0.707 in the validation cohort and performed better than a single clinical model and a model with clinical features combined with clinical TNM staging. Further analyses showed higher PET RS in patients who were older (p < 0.001) and those who had elevated CEA (p < 0.001) and higher clinical TNM (p < 0.001). At different clinical TNM stages, a higher PET RS was associated with a worse survival prognosis.

CONCLUSIONS

Radiomics models based on PET RS, clinical TNM, and clinical features may provide new tools for predicting OS in patients with GC.

A novel endoscopic ultrasomics-based machine learning model and nomogram to predict the pathological grading of pancreatic neuroendocrine tumors

Objectives

This research aimed to retrospectively construct and authenticate ultrasomics models using endoscopic ultrasonography (EUS) images for forecasting the pathological grading of pancreatic neuroendocrine tumors (PNETs).

Methods

After confirmation through pathological examination, a retrospective analysis of 79 patients was conducted, including 49 with grade 1 PNETs and 30 with grade 2/3 PNETs. These patients were randomized to the training or test cohort in a 6:4 proportion. The least absolute shrinkage and selection operator (LASSO) algorithm was used to reduce the dimensionality of ultrasomics features derived from standard EUS images. These nonzero coefficient features were retained and applied to construct prediction models via eight machine-learning algorithms. The optimum ulstrasomics model was determined, followed by creating and evaluating a nomogram.

Results

Ultrasomics features of 107 were extracted, and only those with coefficients greater than zero were retained. The XGboost ultrasomics model performed exceptionally well, achieving AUCs of 0.987 and 0.781 in the training and test cohorts, respectively. Furthermore, an effective nomogram was developed and visually represented. Finally, the calibration curves, decision curve analysis (DCA) plots, and clinical impact curve (CIC) displayed in the ulstrasomics model and nomogram demonstrated high accuracy. They provided significant net benefits for clinical decision-making.

Conclusions

A novel ulstrasomics model and nomogram were created and certified to predict the pathological grading of PNETs using EUS images. This study has the potential to provide valuable insights that improve the clinical applicability and efficacy of EUS in predicting the grading of PNETs.

Computed tomography-based radiomics machine learning models for differentiating enchondroma and atypical cartilaginous tumor in long bones

To explore the value of CT-based radiomics machine learning models for differentiating enchondroma from atypical cartilaginous tumor (ACT) in long bones and methods to improve model performance.59 enchondromas and 53 ACTs in long bones confirmed by pathology were collected retrospectively. The features were extracted from preoperative CT images of these patients, and least absolute shrinkage and selection operator (LASSO) regression was used for feature selection and dimensionality reduction. The selected features were used to construct classification models by thirteen machine learning algorithms. The data set was randomly divided into a training set and a test set at a proportion of 7:3 by ten-fold cross-validation to evaluate the performance of these models.A total of 1199 features were extracted, 9 features were selected, and 13 radiomics machine learning models were constructed. The area under the curve (AUC) of 11 models was more than 0.8, and that of 3 models was more than 0.9. The Extremely Randomized Trees model achieved the best performance (AUC = 0.9375 ± 0.0884), followed by the Adaptive Boosting model (AUC = 0.9188 ± 0.1010) and the Linear Discriminant Analysis model (AUC = 0.9062 ± 0.1459).CT-based radiomics machine learning models had great ability to distinguish enchondroma and ACT in long bones. By using filters to deeply mine high-order features in the original image and selecting appropriate machine learning algorithms, the performance of the model can be improved. · CT-based radiomics machine learning models can distinguish enchondroma and ACT in long bones.. · Using filters and selecting advanced machine learning algorithms can improve model performance.. · Clinical features have limited utility in distinguishing enchondroma and ACT in long bones.. · Hong R, Li Q, Ma J et al. Computed tomography-based radiomics machine learning models for differentiating enchondroma and atypical cartilaginous tumor in long bones. Fortschr Röntgenstr 2024; DOI 10.1055/a-2344-5398.

Impact of Preprocessing Parameters in Medical Imaging-Based Radiomic Studies: A Systematic Review

BACKGROUND

Lately, radiomic studies featuring the development of a signature to use in prediction models in diagnosis or prognosis outcomes have been increasingly published. While the results are shown to be promising, these studies still have many pitfalls and limitations. One of the main issues of these studies is that radiomic features depend on how the images are preprocessed before their computation. Since, in widely known and used software for radiomic features calculation, it is possible to set these preprocessing parameters before the calculation of the radiomic feature, there are ongoing studies assessing the stability and repeatability of radiomic features to find the most suitable preprocessing parameters for every used imaging modality.

MATERIALS AND METHODS

We performed a comprehensive literature search using four electronic databases: PubMed, Cochrane Library, Embase, and Scopus. Mesh terms and free text were modeled in search strategies for databases. The inclusion criteria were studies where preprocessing parameters' influence on feature values and model predictions was addressed. Records lacking information on image acquisition parameters were excluded, and any eligible studies with full-text versions were included in the review process, while conference proceedings and monographs were disregarded. We used the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) tool to investigate the risk of bias. We synthesized our data in a table divided by the imaging modalities subgroups.

RESULTS

After applying the inclusion and exclusion criteria, we selected 43 works. This review examines the impact of preprocessing parameters on the reproducibility and reliability of radiomic features extracted from multimodality imaging (CT, MRI, CBCT, and PET/CT). Standardized preprocessing is crucial for consistent radiomic feature extraction. Key preprocessing steps include voxel resampling, normalization, and discretization, which influence feature robustness and reproducibility. In total, 44% of the included works studied the effects of an isotropic voxel resampling, and most studies opted to employ a discretization strategy. From 2021, several studies started selecting the best set of preprocessing parameters based on models' best performance. As for comparison metrics, ICC was the most used in MRI studies in 58% of the screened works.

CONCLUSIONS

From our work, we highlighted the need to harmonize the use of preprocessing parameters and their values, especially in light of future studies of prospective studies, which are still lacking in the current literature.

Radiomic feature robustness in CT scans affected by fiducial marker induced streak artifacts for patients with hepatocellular carcinoma

BACKGROUND

Stereotactic body radiation therapy for hepatocellular carcinoma necessitates the implantation of gold fiducial markers in the liver, resulting in artifacts on computed tomography (CT) scans, which affect radiomic feature values.

PURPOSE

This report aims to assess the effect of these artifacts on radiomic features and how removing CT slices affects radiomic features extracted from 3D regions of interest (ROI).

METHODS

First, the range variation in 38 tumor contours unaffected by artifacts was assessed after sequentially and randomly removing 25%, 50%, 75% of slices. Subsequently, the agreement of feature values before and after removing ROI slices containing artifacts from 186 patients' CT scans was assessed with Lin's concordance correlation coefficient (CCC) and a Wilcoxon signed-rank test.

RESULTS

In artifact-free tumor volumes, at least 71% of features remain robust with up to 50% of slices removed, while 56% remains robust with up to 75% of slices removed. When comparing contours before and after removing slices containing artifacts, around a third of features in the tumor contour and surrounding area remain robust (CCC > 0.9), compared to 44% in the healthy liver. Concerning the tumor, 13% (Gray Level Size Zone Matrix) to 61% (first order) of the features remain robust (CCC > 0.9). Over 90% of features differ significantly as assessed by Wilcoxon signed-rank test, however.

CONCLUSIONS

This study demonstrates that removing slices containing artifacts is a feasible solution for the CT fiducial problem in this patient population and provides insight into which features are affected.

Robustness of radiomics among photon-counting detector CT and dual-energy CT systems: a texture phantom study

OBJECTIVES

To evaluate the robustness of radiomics features among photon-counting detector CT (PCD-CT) and dual-energy CT (DECT) systems.

METHODS

A texture phantom consisting of twenty-eight materials was scanned with one PCD-CT and four DECT systems (dual-source, rapid kV-switching, dual-layer, and sequential scanning) at three dose levels twice. Thirty sets of virtual monochromatic images at 70 keV were reconstructed. Regions of interest were delineated for each material with a rigid registration. Ninety-three radiomics were extracted per PyRadiomics. The test-retest repeatability between repeated scans was assessed by Bland-Altman analysis. The intra-system reproducibility between dose levels, and inter-system reproducibility within the same dose level, were evaluated by intraclass correlation coefficient (ICC) and concordance correlation coefficient (CCC). Inter-system variability among five scanners was assessed by coefficient of variation (CV) and quartile coefficient of dispersion (QCD).

RESULTS

The test-retest repeatability analysis presented that 97.1% of features were repeatable between scan-rescans. The mean ± standard deviation ICC and CCC were 0.945 ± 0.079 and 0.945 ± 0.079 for intra-system reproducibility, respectively, and 86.0% and 85.7% of features were with ICC > 0.90 and CCC > 0.90, respectively, between different dose levels. The mean ± standard deviation ICC and CCC were 0.157 ± 0.174 and 0.157 ± 0.174 for inter-system reproducibility, respectively, and none of the features were with ICC > 0.90 or CCC > 0.90 within the same dose level. The inter-system variability suggested that 6.5% and 12.8% of features were with CV < 10% and QCD < 10%, respectively, among five CT systems.

CONCLUSION

The radiomics features were non-reproducible with significant variability in values among different CT techniques.

CLINICAL RELEVANCE STATEMENT

Radiomics features are non-reproducible with significant variability in values among photon-counting detector CT and dual-energy CT systems, necessitating careful attention to improve the cross-system generalizability of radiomic features before implementation of radiomics analysis in clinical routine.

KEY POINTS

CT radiomics stability should be guaranteed before the implementation in the clinical routine. Radiomics robustness was on a low level among photon-counting detectors and dual-energy CT techniques. Limited inter-system robustness of radiomic features may impact the generalizability of models.

Predicting major adverse cardiac events using radiomics nomogram of pericoronary adipose tissue based on CCTA: A multi-center study

BACKGROUND

The evolution of coronary atherosclerotic heart disease (CAD) is intricately linked to alterations in the pericoronary adipose tissue (PCAT). In recent epochs, characteristics of the PCAT have progressively ascended as focal points of research in CAD risk stratification and individualized clinical decision-making. Harnessing radiomic methodologies allows for the meticulous extraction of imaging features from these adipose deposits. Coupled with machine learning paradigms, we endeavor to establish predictive models for the onset of major adverse cardiovascular events (MACE).

PURPOSE

To appraise the predictive utility of radiomic features of PCAT derived from coronary computed tomography angiography (CCTA) in forecasting MACE.

METHODS

We retrospectively incorporated data from 314 suspected or confirmed CAD patients admitted to our institution from June 2019 to December 2022. An additional cohort of 242 patients from two external institutions was encompassed for external validation. The endpoint under consideration was the occurrence of MACE after a 1-year follow-up. MACE was delineated as cardiovascular mortality, newly diagnosed myocardial infarction, hospitalization (or re-hospitalization) for heart failure, and coronary target vessel revascularization occurring more than 30 days post-CCTA examination. All enrolled patients underwent CCTA scanning. Radiomic features were meticulously extracted from the optimal diastolic phase axial slices of CCTA images. Feature reduction was achieved through a composite feature selection algorithm, laying the groundwork for the radiomic signature model. Both univariate and multivariate analyses were employed to assess clinical variables. A multifaceted logistic regression analysis facilitated the crafting of a clinical-radiological-radiomic combined model (or nomogram). Receiver operating characteristic (ROC) curves, calibration, and decision curve analyses (DCA) were delineated, with the area under the ROC curve (AUCs) computed to gauge the predictive prowess of the clinical model, radiomic model, and the synthesized ensemble.

RESULTS

A total of 12 radiomic features closely associated with MACE were identified to establish the radiomic model. Multivariate logistic regression results demonstrated that smoking, age, hypertension, and dyslipidemia were significantly correlated with MACE. In the integrated nomogram, which amalgamated clinical, imaging, and radiomic parameters, the diagnostic performance was as follows: 0.970 AUC, 0.949 accuracy (ACC), 0.833 sensitivity (SEN), 0.981 specificity (SPE), 0.926 positive predictive value (PPV), and 0.955 negative predictive value (NPV). The calibration curve indicated a commendable concordance of the nomogram, and the decision curve analysis underscored its superior clinical utility.

CONCLUSIONS

The integration of radiomic signatures from PCAT based on CCTA, clinical indices, and imaging parameters into a nomogram stands as a promising instrument for prognosticating MACE events.

Impact of MRI radiomic feature normalization for prognostic modelling in uterine endometrial and cervical cancers

Widespread clinical use of MRI radiomic tumor profiling for prognostication and treatment planning in cancers faces major obstacles due to limitations in standardization of radiomic features. The purpose of the current work was to assess the impact of different MRI scanning- and normalization protocols for the statistical analyses of tumor radiomic data in two patient cohorts with uterine endometrial-(EC) (n = 136) and cervical (CC) (n = 132) cancer. 1.5 T and 3 T, T1-weighted MRI 2 min post-contrast injection, T2-weighted turbo spin echo imaging, and diffusion-weighted imaging were acquired. Radiomic features were extracted from within manually segmented tumors in 3D and normalized either using z-score normalization or a linear regression model (LRM) accounting for linear dependencies with MRI acquisition parameters. Patients were clustered into two groups based on radiomic profile. Impact of MRI scanning parameters on cluster composition and prognostication were analyzed using Kruskal-Wallis tests, Kaplan-Meier plots, log-rank test, random survival forests and LASSO Cox regression with time-dependent area under curve (tdAUC) (α = 0.05). A large proportion of the radiomic features was statistically associated with MRI scanning protocol in both cohorts (EC: 162/385 [42%]; CC: 180/292 [62%]). A substantial number of EC (49/136 [36%]) and CC (50/132 [38%]) patients changed cluster when clustering was performed after z-score-versus LRM normalization. Prognostic modeling based on cluster groups yielded similar outputs for the two normalization methods in the EC/CC cohorts (log-rank test; z-score: p = 0.02/0.33; LRM: p = 0.01/0.45). Mean tdAUC for prognostic modeling of disease-specific survival (DSS) by the radiomic features in EC/CC was similar for the two normalization methods (random survival forests; z-score: mean tdAUC = 0.77/0.78; LRM: mean tdAUC = 0.80/0.75; LASSO Cox; z-score: mean tdAUC = 0.64/0.76; LRM: mean tdAUC = 0.76/0.75). Severe biases in tumor radiomics data due to MRI scanning parameters exist. Z-score normalization does not eliminate these biases, whereas LRM normalization effectively does. Still, radiomic cluster groups after z-score- and LRM normalization were similarly associated with DSS in EC and CC patients.

Interpretation of Artificial Intelligence Models in Healthcare: A Pictorial Guide for Clinicians

Artificial intelligence (AI) models can play a more effective role in managing patients with the explosion of digital health records available in the healthcare industry. Machine-learning (ML) and deep-learning (DL) techniques are two methods used to develop predictive models that serve to improve the clinical processes in the healthcare industry. These models are also implemented in medical imaging machines to empower them with an intelligent decision system to aid physicians in their decisions and increase the efficiency of their routine clinical practices. The physicians who are going to work with these machines need to have an insight into what happens in the background of the implemented models and how they work. More importantly, they need to be able to interpret their predictions, assess their performance, and compare them to find the one with the best performance and fewer errors. This review aims to provide an accessible overview of key evaluation metrics for physicians without AI expertise. In this review, we developed four real-world diagnostic AI models (two ML and two DL models) for breast cancer diagnosis using ultrasound images. Then, 23 of the most commonly used evaluation metrics were reviewed uncomplicatedly for physicians. Finally, all metrics were calculated and used practically to interpret and evaluate the outputs of the models. Accessible explanations and practical applications empower physicians to effectively interpret, evaluate, and optimize AI models to ensure safety and efficacy when integrated into clinical practice.

Radiomic Gradient in Peritumoural Tissue of Liver Metastases: A Biomarker for Clinical Practice? Analysing Density, Entropy, and Uniformity Variations with Distance from the Tumour

The radiomic analysis of the tissue surrounding colorectal liver metastases (CRLM) enhances the prediction accuracy of pathology data and survival. We explored the variation of the textural features in the peritumoural tissue as the distance from CRLM increases. We considered patients with hypodense CRLMs >10 mm and high-quality computed tomography (CT). In the portal phase, we segmented (1) the tumour, (2) a series of concentric rims at a progressively increasing distance from CRLM (from one to ten millimetres), and (3) a cylinder of normal parenchyma (Liver-VOI). Sixty-three CRLMs in 51 patients were analysed. Median peritumoural HU values were similar to Liver-VOI, except for the first millimetre around the CRLM. Entropy progressively decreased (from 3.11 of CRLM to 2.54 of Liver-VOI), while uniformity increased (from 0.135 to 0.199, p < 0.001). At 10 mm from CRLM, entropy was similar to the Liver-VOI in 62% of cases and uniformity in 46%. In small CRLMs (≤30 mm) and responders to chemotherapy, normalisation of entropy and uniformity values occurred in a higher proportion of cases and at a shorter distance. The radiomic analysis of the parenchyma surrounding CRLMs unveiled a wide halo of progressively decreasing entropy and increasing uniformity despite a normal radiological aspect. Underlying pathology data should be investigated.

Pretreatment CT-based machine learning radiomics model predicts response in unresectable hepatocellular carcinoma treated with lenvatinib plus PD-1 inhibitors and interventional therapy

BACKGROUND

Lenvatinib plus PD-1 inhibitors and interventional (LPI) therapy have demonstrated promising treatment effects in unresectable hepatocellular carcinoma (HCC). However, biomarkers for predicting the response to LPI therapy remain to be further explored. We aimed to develop a radiomics model to noninvasively predict the efficacy of LPI therapy.

METHODS

Clinical data of patients with HCC receiving LPI therapy were collected in our institution. The clinical model was built with clinical information. Nine machine learning classifiers were tested and the multilayer perceptron classifier with optimal performance was used as the radiomics model. The clinical-radiomics model was constructed by integrating clinical and radiomics scores through logistic regression analysis.

RESULTS

151 patients were enrolled in this study (2:1 randomization, 101 and 50 in the training and validation cohorts), of which three achieved complete response, 69 showed partial response, 46 showed stable disease, and 33 showed progressive disease. The objective response rate, disease control rate, and conversion resection rates were 47.7, 78.1 and 23.2%. 14 features were selected from the initially extracted 1223 for radiomics model construction. The area under the curves of the radiomics model (0.900 for training and 0.893 for validation) were comparable to that of the clinical-radiomics model (0.912 for training and 0.892 for validation), and both were superior to the clinical model (0.669 for training and 0.585 for validation). Meanwhile, the radiomics model can categorize participants into high-risk and low-risk groups for progression-free survival (PFS) and overall survival (OS) in the training (HR 1.913, 95% CI 1.121 to 3.265, p=0.016 for PFS; HR 4.252, 95% CI 2.051 to 8.816, p=0.001 for OS) and validation sets (HR 2.347, 95% CI 1.095 to 5.031, p=0.012 for PFS; HR 2.592, 95% CI 1.050 to 6.394, p=0.019 for OS).

CONCLUSION

The promising machine learning radiomics model was developed and validated to predict the efficacy of LPI therapy for patients with HCC and perform risk stratification, with comparable performance to clinical-radiomics model.

Enhancing prognostic prediction in hepatocellular carcinoma post-TACE: a machine learning approach integrating radiomics and clinical features

Objective

This study aimed to investigate the use of radiomics features and clinical information by four machine learning algorithms for predicting the prognosis of patients with hepatocellular carcinoma (HCC) who have been treated with transarterial chemoembolization (TACE).

Methods

A total of 105 patients with HCC treated with TACE from 2002 to 2012 were enrolled retrospectively and randomly divided into two cohorts for training (n = 74) and validation (n = 31) according to a ratio of 7:3. The Spearman rank, random forest, and univariate Cox regression were used to select the optimal radiomics features. Univariate Cox regression was used to select clinical features. Four machine learning algorithms were used to develop the models: random survival forest, eXtreme gradient boosting (XGBoost), gradient boosting, and the Cox proportional hazard regression model. The area under the curve (AUC) and C-index were devoted to assessing the performance of the models in predicting HCC prognosis.

Results

A total of 1,834 radiomics features were extracted from the computed tomography images of each patient. The clinical risk factors for HCC prognosis were age at diagnosis, TNM stage, and metastasis, which were analyzed using univariate Cox regression. In various models, the efficacy of the combined models generally surpassed that of the radiomics and clinical models. Among four machine learning algorithms, XGBoost exhibited the best performance in combined models, achieving an AUC of 0.979 in the training set and 0.750 in the testing set, demonstrating its strong prognostic prediction capability.

Conclusion

The superior performance of the XGBoost-based combined model underscores its potential as a powerful tool for enhancing the precision of prognostic assessments for patients with HCC.

Radiomics Model of Dynamic Contrast-Enhanced MRI for Evaluating Vessels Encapsulating Tumor Clusters and Microvascular Invasion in Hepatocellular Carcinoma

RATIONALE AND OBJECTIVES

To develop and validate a clinical-radiomics model of dynamic contrast-enhanced MRI (DCE-MRI) for the preoperative discrimination of Vessels encapsulating tumor clusters (VETC)- microvascular invasion (MVI) and prognosis of hepatocellular carcinoma (HCC).

MATERIALS AND METHODS

219 HCC patients from Institution 1 were split into internal training and validation groups, with 101 patients from Institution 2 assigned to external validation. Histologically confirmed VETC-MVI pattern categorizing HCC into VM-HCC+ (VETC+/MVI+, VETC-/MVI+, VETC+/MVI-) and VM-HCC- (VETC-/MVI-). The regions of intratumor and peritumor were segmented manually in the arterial, portal-venous and delayed phase (AP, PP, and DP, respectively) of DCE-MRI. Six radiomics models (intratumor and peritumor in AP, PP, and DP of DCE-MRI) and one clinical model were developed for assessing VM-HCC. Establishing intra-tumoral and peri-tumoral models through combining intratumor and peritumor features. The best-performing radiomics model and the clinical model were then integrated to create a Combined model.

RESULTS

In institution 1, pathological VM-HCC+ were confirmed in 88 patients (training set: 61, validation set: 27). In internal testing, the Combined model had an AUC of 0.85 (95% CI: 0.76-0.93), which reached an AUC of 0.75 (95% CI: 0.66-0.85) in external validation. The model's predictions were associated with early recurrence and progression-free survival in HCC patients.

CONCLUSIONS

The clinical-radiomics model offers a non-invasive approach to discern VM-HCC and predict HCC patients' prognosis preoperatively, which could offer clinicians valuable insights during the decision-making phase.

Evaluating the quality of radiomics-based studies for endometrial cancer using RQS and METRICS tools

OBJECTIVE

To assess the methodological quality of radiomics-based models in endometrial cancer using the radiomics quality score (RQS) and METhodological radiomICs score (METRICS).

METHODS

We systematically reviewed studies published by October 30th, 2023. Inclusion criteria were original radiomics studies on endometrial cancer using CT, MRI, PET, or ultrasound. Articles underwent a quality assessment by novice and expert radiologists using RQS and METRICS. The inter-rater reliability for RQS and METRICS among radiologists with varying expertise was determined. Subgroup analyses were performed to assess whether scores varied according to study topic, imaging technique, publication year, and journal quartile.

RESULTS

Sixty-eight studies were analysed, with a median RQS of 11 (IQR, 9-14) and METRICS score of 67.6% (IQR, 58.8-76.0); two different articles reached maximum RQS of 19 and METRICS of 90.7%, respectively. Most studies utilised MRI (82.3%) and machine learning methods (88.2%). Characterisation and recurrence risk stratification were the most explored outcomes, featured in 35.3% and 19.1% of articles, respectively. High inter-rater reliability was observed for both RQS (ICC: 0.897; 95% CI: 0.821, 0.946) and METRICS (ICC: 0.959; 95% CI: 0.928, 0.979). Methodological limitations such as lack of external validation suggest areas for improvement. At subgroup analyses, no statistically significant difference was noted.

CONCLUSIONS

Whilst using RQS, the quality of endometrial cancer radiomics research was apparently unsatisfactory, METRICS depicts a good overall quality. Our study highlights the need for strict compliance with quality metrics. Adhering to these quality measures can increase the consistency of radiomics towards clinical application in the pre-operative management of endometrial cancer.

CLINICAL RELEVANCE STATEMENT

Both the RQS and METRICS can function as instrumental tools for identifying different methodological deficiencies in endometrial cancer radiomics research. However, METRICS also reflected a focus on the practical applicability and clarity of documentation.

KEY POINTS

The topic of radiomics currently lacks standardisation, limiting clinical implementation. METRICS scores were generally higher than the RQS, reflecting differences in the development process and methodological content. A positive trend in METRICS score may suggest growing attention to methodological aspects in radiomics research.