Radiomics: the process and the challenges

Prediction of the efficacy of first transarterial chemoembolization for advanced hepatocellular carcinoma via a clinical-radiomics model

BACKGROUND

Hepatocellular carcinoma (HCC) is a common tumor with a poor prognosis. Early intervention is essential; thus, good prognostic markers to identify patients who benefit from first transarterial chemoembolization (TACE) are needed.

AIM

To investigate the efficacy of computed tomography (CT) radiomics in predicting the success of the first TACE in patients with advanced HCC and to develop an early prediction model based on clinical radiomics features.

METHODS

Data from 122 patients with advanced HCC treated with TACE were analyzed. Intratumoral and peritumoral areas on arterial and venous CT images were selected to extract radiomic features, which were screened in the training cohort using the minimum redundancy maximum correlation. Then, support vector machines were used to construct the model. To construct a receiver operating characteristic curve, the predictive efficacy of each model was evaluated on the basis of the area under the curve (AUC).

RESULTS

Among the 122 patients, 72 patients were effectively treated via TACE, and in 50 patients, this treatment was ineffective. In the radiomics model, the areas under the curve of the venous phase model were 0.867 (95%CI: 0.790-0.940) in the training cohort and 0.755 (0.600-0.910) in the validation cohort, indicating good predictive efficacy. The multivariate logistic regression results indicated that preoperative alpha-fetoprotein levels (P = 0.01) were a risk factor for TACE. The screened clinical features were combined with the radiomic features to construct a combined model. This combined model had an AUC of 0.92 (0.87-0.95) in the training cohort and 0.815 (0.67-0.95) in the validation cohort.

CONCLUSION

CT radiomics has good value in predicting the efficacy of the first TACE treatment in patients with HCC. The combined model was a better tool for predicting the first TACE efficacy in patients with advanced HCC and could provide an efficient predictive tool to help with the selection of patients for TACE.

Non-radiomics imaging (US-CEUS) features and clinical text features: correlation with microvascular invasion and tumor grading in hepatocellular carcinoma

OBJECTIVES

To predict microvascular invasion (MVI) status and tumor grading of hepatocellular carcinoma (HCC) by evaluating preoperative non-radiomics ultrasound and contrast-enhanced ultrasound (US-CEUS) features and determine the influences of MVI/tumor grading on the category of CEUS LI-RADS for HCC.

METHODS

A total of 506 HCC patients who underwent preoperative US-CEUS examinations from 8 hospitals between July 2020 and June 2023 were enrolled. According to the MVI status, all the patients were classified, and HCC differentiation was assessed by using Edmondson-Steiner (ES) grading: MVI-negative (M0) and low-grade ES (GI/II) (MN-L, n = 297) and MVI-positive (M1/M2) and/or high-grade ES (GIII/IV) (MP-H, n = 209). Stratified analysis was performed based on fibrosis stage and tumor size.

RESULTS

The results proved that MN-L HCC was more frequently classified into the LR-5 category (p = 0.034), while MP-H HCC was more frequently classified into the LR-TIV (p = 0.010). The heterogeneously arterial phase hyperenhancement (APHE) is significantly correlated with MVI(+)/high grade-ES (p = 0.003). Compared with MN-L HCC, the onset of washout was earlier, washout rate was higher, and tumor-invasion border was larger (all p < 0.01) in MP-H HCC. In addition, fibrosis stage and tumor size significantly influenced the onset of washout and washout rate of HCC (all p < 0.01). The tumor-invasion border was only positively correlated with tumor size (p < 0.001) rather than fibrosis stage (p > 0.05).

CONCLUSIONS

MVI status and tumor grading influence the classification of LR-5 and LR-TIV. Heterogeneous APHE, higher washout rate, earlier onset of washout (≤65 s), larger tumor-invasion border (≥3 mm) and higher alpha fetoprotein level indicate the presence of MVI and/or high-grade ES.

Effects of parametric feature maps on the reproducibility of radiomics from different fields of view in cardiac magnetic resonance cine images- a clinical and experimental study setting

In cardiac MRI, the field of view (FOV) is adapted to the individual patient's size, influencing spatial resolution and myocardial radiomics. This study aimed to investigate the effects of parametric feature maps on radiomics derived from cine images acquired with different FOV sizes on individuals without myocardial pathologies. In the clinical setting, cardiac MRI scans from clinical care were screened retrospectively for patients without pathological findings, neither in the MRI nor the medical history or follow-up, resulting in 61 included patients. In the experimental setting, 12 healthy volunteers were prospectively examined on a 1.5 Tesla MRI scanner with cine images acquired with three different FOVs (256 × 329 mm, 279 × 359 mm, 302 × 390 mm). One midventricular end-diastolic short-axis slice of the non-enhanced cine images was extracted for healthy volunteers and patients. The left ventricular myocardium was encompassed with regions of interest (ROIs). Ninety-three features were extracted using PyRadiomics. Images were converted to parametric radiomic feature maps using pretested software. ROIs were copied to the maps to retrieve the feature quantity. The variability of features across the different FOVs from the original images and feature maps was assessed with coefficients of variation (COVs) and rated stable at up to 10%. When derived from the original images, out of the 93 extracted features, only 24 (patients) and 29 (volunteers) revealed COVs < 10%. When extracted from the parametric maps, the number of stable features increased by 63% and 66%, with 39 (patients) and 48 (volunteers) features showing COVs < 10%, respectively. Software-computed parametric feature maps improve the reproducibility of radiomics across different FOVs in cardiac cine images of individuals without myocardial pathologies. Prospective investigations with different FOVs of a patient collective with myocardial pathologies could enhance the generalizability of the findings.

Enhancing radiomics features via a large language model for classifying benign and malignant breast tumors in mammography

BACKGROUND AND OBJECTIVES

Radiomics is widely used to assist in clinical decision-making, disease diagnosis, and treatment planning for various target organs, including the breast. Recent advances in large language models (LLMs) have helped enhance radiomics analysis.

MATERIALS AND METHODS

Herein, we sought to improve radiomics analysis by incorporating LLM-learned clinical knowledge, to classify benign and malignant tumors in breast mammography. We extracted radiomics features from the mammograms based on the region of interest and retained the features related to the target task. Using prompt engineering, we devised an input sequence that reflected the selected features and the target task. The input sequence was fed to the chosen LLM (LLaMA variant), which was fine-tuned using low-rank adaptation to enhance radiomics features. This was then evaluated on two mammogram datasets (VinDr-Mammo and INbreast) against conventional baselines.

RESULTS

The enhanced radiomics-based method performed better than baselines using conventional radiomics features tested on two mammogram datasets, achieving accuracies of 0.671 for the VinDr-Mammo dataset and 0.839 for the INbreast dataset. Conventional radiomics models require retraining from scratch for an unseen dataset using a new set of features. In contrast, the model developed in this study effectively reused the common features between the training and unseen datasets by explicitly linking feature names with feature values, leading to extensible learning across datasets. Our method performed better than the baseline method in this retraining setting using an unseen dataset.

CONCLUSIONS

Our method, one of the first to incorporate LLM into radiomics, has the potential to improve radiomics analysis.

Predictive classification of lung cancer pathological based on PET/CT radiomics

OBJECTIVES

To develop and validate a combined clinical and radiomics model for non-invasive prediction of lung cancer (LC) pathological types (lung adenocarcinoma, lung squamous cell carcinoma, and small cell lung cancer) based on patients' pre-treatment FDG PET/CT images and clinical data, as a complementary tool to aid in the diagnosis of LC pathological histological classification.

METHODS

In total, 896 patients with pathological confirmation of lung cancer were part of this retrospective study. The training and test groups included 819 patients who underwent scanning using scanner 1. The independent validation group included 77 patients who using scanner 2. The optimal features were retained by least absolute shrinkage and selection operator algorithm dimensionality reduction screening of the collected radiomics features, clinical parameters, and PET metabolic parameters. Five models were established to predict the lung cancer pathological types by the k-nearest neighbor classification (KNN) algorithm. The performance of the prediction model was assessed by calculating the area under the curve (AUC) from the receiver operator characteristic curve (ROC).

RESULTS

Of all five predictive models (the PET-only radiomics model, the CT-only radiomics model, the PET/CT radiomics model, the clinical-only model and the combined clinical and PET/CT radiomics model), the clinical combined PET/CT radiomics model exhibited best performance. The macro-AUC for the training, test and independent validation groups were 0.974, 0.931, 0.960, the micro-AUC were 0.976, 0.940, 0.970, and the accuracy were 0.963, 0.914, and 0.961, respectively.

CONCLUSIONS

Our model combined radiomics and clinical data and showed higher performance in non-invasively predicting the LC pathological types, which suggesting that PET/CT radiomics may be a promising technique for predicting LC histopathology.

Developing a Reproducible Radiomics Model for Diagnosis of Active Crohn's Disease on CT Enterography Across Annotation Variations and Acquisition Differences

To systematically identify radiomics features on CT enterography (CTE) scans which can accurately diagnose active Crohn's disease across multiple sources of variation. Retrospective study of CTE scans curated between 2013 and 2015, comprising 164 subjects (65 male, 99 female; all patients were over the age of 18) with endoscopic confirmation for the presence or absence of active Crohn's disease. All patients had three distinct sets of scans available (full and reduced dose, where the latter had been reconstructed via two different methods), acquired on a single scanner at a single institution. Radiomics descriptors from annotated terminal ileum regions were individually and systematically evaluated for resilience to different imaging variations (changes in dose/reconstruction, batch effects, and simulated annotation differences) via multiple reproducibility measures. Multiple radiomics models (by accounting for each source of variation) were evaluated in terms of classifier area under the ROC curve (AUC) for identifying patients with active Crohn's disease, across separate discovery and hold-out validation cohorts. Radiomics descriptors selected based on resiliency to multiple sources of imaging variation yielded the highest overall classification performance in the discovery cohort (AUC = 0.79 ± 0.04) which also best generalized in hold-out validation (AUC = 0.81). Performance was maintained across multiple doses and reconstructions while also being significantly better (p < 0.001) than non-resilient descriptors or descriptors only resilient to a single source of variation. Radiomics features can accurately diagnose active Crohn's disease on CTE scans across multiple sources of imaging variation via systematic analysis of reproducibility measures. Clinical utility and translatability of radiomics features for diagnosis and characterization of Crohn's disease on CTE scans will be contingent on their reproducibility across multiple types and sources of imaging variation.

Understandings 18 FDG PET radiomics and its application to lymphoma

The early identification of lymphoma patients who fail front-line treatment is crucial for optimizing disease management. Positron emission tomography, a well-established tool for staging and response evaluation in lymphoma, is typically assessed visually or semiquantitatively, leaving much of its latent information unexploited. Radiomic analysis, which employs mathematical descriptors, can enable the extraction of quantitative features from baseline images that correlate with the disease's biological characteristics. Emerging radiomic features such as metabolic tumour volume, total lesion glycolysis and markers of disease dissemination and metabolic heterogeneity are proving to be powerful prognostic biomarkers in lymphoma. Texture analysis, the most advanced area of radiomics, offers highly complex features that require further standardization and validation before being adopted as reliable biomarkers. Combining radiomic features with clinical risk factors and genomic data holds promising potential for improving clinical risk prediction. This review explores the current state of radiomic analysis, progress towards its standardization and its incorporation into clinical practice and trial designs. The integration of radiomic markers with circulating tumour DNA may provide a comprehensive approach to developing baseline and dynamic risk scores, facilitating the testing of novel treatments and advancing personalized treatment of aggressive lymphomas.

Impact of harmonization on predicting complications in head and neck cancer after radiotherapy using MRI radiomics and machine learning techniques

BACKGROUND

Variations in medical images specific to individual scanners restrict the use of radiomics in both clinical practice and research. To create reproducible and generalizable radiomics-based models for outcome prediction and assessment, data harmonization is essential.

PURPOSE

This study aims to investigate the impact of harmonization in performance of machine learning-based radiomics model toward the prediction of radiotherapy-induced toxicity (early and late sticky saliva and xerostomia) in head and neck cancer (HNC) patients after radiation therapy usingT 1 $T_1$ andT 2 $T_2$ -weighted magnetic resonance (MR) images.

METHODS

A total of 85 HNC patients who underwent radiotherapy was studied. Radiomic features were extracted fromT 1 $T_1$ andT 2 $T_2$ -weighted MR images with standardized protocols. Data harmonization was performed using ComBat algorithm to reduce inter-center variability. Besides imaging features, both dosimetric and demographic features were extracted and used in our model. Recursive feature elimination was employed as feature selection method to identify the most important variables. Ten classification algorithms, including eXtreme Gradient Boosting (XGBoost), multilayer perceptron (MLP), support vector machines (SVM), random forest (RF), k-nearest neighbor (KNN), Naive Bayes (NB), logistic regression (LR), and decision tree (DT), boosted generalized linear model (GLMB), and stack learning (SL) were utilized and compared to develop predictive models. This evaluation comparisons were performed before and after harmonization to demonstrate its significance.

RESULTS

Our results indicate that harmonization consistently enhances predictive performance across various complications and imaging modalities. In early and late sticky saliva prediction usingT 1 $T_1$ -weighted images, the SVM and RF models achieved an impressive area under the curve (AUC) of 0.88 ± $\pm$ 0.09 and 0.97 ± $\pm$ 0.05 with harmonization versus 0.42 ± $\pm$ 0.12 and 0.83 ± $\pm$ 0.08 without harmonization, respectively. Similarly, in early and late xerostomia prediction, the model attained an AUC of 0.79 ± $\pm$ 0.15 and 0.61 ± $\pm$ 0.14 with harmonization and 0.55 ± $\pm$ 0.17 and 0.46 ± $\pm$ 0.14 without harmonization.

CONCLUSION

Our study highlights the importance of harmonization techniques in improving the performance of predictive models utilizing magnetic resonance imaging radiomics features. While harmonization consistently enhanced performance for sticky saliva and early xerostomia usingT 1 $T_1$ -weighted features, the prediction of early and late xerostomia usingT 2 $T_2$ -weighted features remains challenging. These findings try to develop accurate and reliable predictive models in medical imaging, that contribute to improve patient care and treatment outcomes.

Using Machine Learning on MRI Radiomics to Diagnose Parotid Tumours Before Comparing Performance with Radiologists: A Pilot Study

The parotid glands are the largest of the major salivary glands. They can harbour both benign and malignant tumours. Preoperative work-up relies on MR images and fine needle aspiration biopsy, but these diagnostic tools have low sensitivity and specificity, often leading to surgery for diagnostic purposes. The aim of this paper is (1) to develop a machine learning algorithm based on MR images characteristics to automatically classify parotid gland tumours and (2) compare its results with the diagnoses of junior and senior radiologists in order to evaluate its utility in routine practice. While automatic algorithms applied to parotid tumours classification have been developed in the past, we believe that our study is one of the first to leverage four different MRI sequences and propose a comparison with clinicians. In this study, we leverage data coming from a cohort of 134 patients treated for benign or malignant parotid tumours. Using radiomics extracted from the MR images of the gland, we train a random forest and a logistic regression to predict the corresponding histopathological subtypes. On the test set, the best results are given by the random forest: we obtain a 0.720 accuracy, a 0.860 specificity, and a 0.720 sensitivity over all histopathological subtypes, with an average AUC of 0.838. When considering the discrimination between benign and malignant tumours, the algorithm results in a 0.760 accuracy and a 0.769 AUC, both on test set. Moreover, the clinical experiment shows that our model helps to improve diagnostic abilities of junior radiologists as their sensitivity and accuracy raised by 6 % when using our proposed method. This algorithm may be useful for training of physicians. Radiomics with a machine learning algorithm may help improve discrimination between benign and malignant parotid tumours, decreasing the need for diagnostic surgery. Further studies are warranted to validate our algorithm for routine use.

Radiomic Features Prognosticate Treatment Response in CAR-T Cell Therapy

Background: Diffuse large B-cell lymphomas (DLBCLs) are the most common, aggressive disease form that accounts for 30% of all lymphoma cases. Identifying patients who will respond to these advanced cell-based therapies is an unaddressed challenge. Methods: We propose to develop a radiomics- (quantitative image metric) based signature on the patients' imaging scans (positron emission tomography/computed tomography, PET/CT) and use these metrics to prognosticate response to axi-cel (axicabtagene ciloleucel), autologous CD19 chimeric antigen receptor (CAR) T-cell (CAR-T) therapy. We curated a cohort of 155 patients with relapsed/refractory (R/R) DLBCL who were treated with axi-cel. Using their baseline image scan (PET/CT), the largest lesions related to nodal/extra-nodal disease were identified and characterized using imaging metrics (radiomics). We used principal component (PC) analysis to reduce the dimensionality of these features across the functional categories (size, shape, and texture). We evaluated the prognostic ability of radiomic-based PC to treatment response (1-year), measured by overall survival (OS) and progression-free survival (PFS). Results: We found that radiomic PC was prognostic of overall survival (Shape-PC, q < 0.013/0.0108, Size-PC, q < 0.003/0.0088), in CT/PET, respectively. In comparison, the metabolic tumor volume (MTV) was prognostic (q < 0.0002/0.0007). The radiomic PCs across the functional categories showed moderate to weak correlation with MTV, Spearman's ρ of 0.44/0.35/0.27, and 0.45/0.36/0.55 for Size/Shape/Texture-PC1 obtained on PET and CT, respectively. Conclusions: We found radiomic PC based on size and shape metrics that are able to prognosticate treatment response to CAR-T therapy.

Prediction of clinical stages of cervical cancer via machine learning integrated with clinical features and ultrasound-based radiomics

To investigate the prediction of a model constructed by combining machine learning (ML) with clinical features and ultrasound radiomics in the clinical staging of cervical cancer. General clinical and ultrasound data of 227 patients with cervical cancer who received transvaginal ultrasonography were retrospectively analyzed. The region of interest (ROI) radiomics profiles of the original image and derived image were retrieved and profile screening was performed. The chosen profiles were employed in radiomics model and Radscore formula construction. Prediction models were developed utilizing several ML algorithms by Python based on an integrated dataset of clinical features and ultrasound radiomics. Model performances were evaluated via AUC. Plot calibration curves and clinical decision curves were used to assess model efficacy. The model developed by support vector machine (SVM) emerged as the superior model. Integrating clinical characteristics with ultrasound radiomics, it showed notable performance metrics in both the training and validation datasets. Specifically, in the training set, the model obtained an AUC of 0.88 (95% Confidence Interval (CI): 0.83-0.93), alongside a 0.84 accuracy, 0.68 sensitivity, and 0.91 specificity. When validated, the model maintained an AUC of 0.77 (95% CI: 0.63-0.88), with 0.77 accuracy, 0.62 sensitivity, and 0.83 specificity. The calibration curve aligned closely with the perfect calibration line. Additionally, based on the clinical decision curve analysis, the model offers clinical utility over wide-ranging threshold possibilities. The clinical- and radiomics-based SVM model provides a noninvasive tool for predicting cervical cancer stage, integrating ultrasound radiomics and key clinical factors (age, abortion history) to improve risk stratification. This approach could guide personalized treatment (surgery vs. chemoradiation) and optimize staging accuracy, particularly in resource-limited settings where advanced imaging is scarce.

Integrating Radiogenomics and Machine Learning in Musculoskeletal Oncology Care

Musculoskeletal tumors present a diagnostic challenge due to their rarity, histological diversity, and overlapping imaging features. Accurate characterization is essential for effective treatment planning and prognosis, yet current diagnostic workflows rely heavily on invasive biopsy and subjective radiologic interpretation. This review explores the evolving role of radiogenomics and machine learning in improving diagnostic accuracy for bone and soft tissue tumors. We examine integrating quantitative imaging features from MRI, CT, and PET with genomic and transcriptomic data to enable non-invasive tumor profiling. AI-powered platforms employing convolutional neural networks (CNNs) and radiomic texture analysis show promising results in tumor grading, subtype differentiation (e.g., Osteosarcoma vs. Ewing sarcoma), and predicting mutation signatures (e.g., TP53, RB1). Moreover, we highlight the use of liquid biopsy and circulating tumor DNA (ctDNA) as emerging diagnostic biomarkers, coupled with point-of-care molecular assays, to enable early and accurate detection in low-resource settings. The review concludes by discussing translational barriers, including data harmonization, regulatory challenges, and the need for multi-institutional datasets to validate AI-based diagnostic frameworks. This article synthesizes current advancements and provides a forward-looking view of precision diagnostics in musculoskeletal oncology.

Influence of Fetal-Type Posterior Cerebral Artery on Morphological Characteristics and Rupture Risk of Posterior Communicating Artery Aneurysms: A Radiomics Approach

Background/Objectives: The fetal-type posterior cerebral artery (fetal PCA) is an anatomical variant that alters hemodynamics and may influence posterior communicating artery (PCoA) aneurysm rupture risk. Aneurysm shape and size irregularity are key rupture predictors. This study investigates the impact of fetal PCA on PCoA aneurysm morphology and rupture risk using a radiomics-based approach. Methods: We retrospectively analyzed 87 patients with PCoA aneurysms (39 ruptured, 48 unruptured) treated at a tertiary center (January 2017-December 2022). Seventeen morphological parameters and 18 radiomic features were extracted per aneurysm. Patients were grouped by fetal PCA presence. Logistic regression and receiver operating characteristic (ROC) analyses identified rupture predictors. Results: Of 87 aneurysms, 38 had fetal PCA (24 ruptured, 14 unruptured), and 49 did not (15 ruptured, 34 unruptured). Fetal PCA was significantly associated with rupture (odds ratio [OR]: 3.28, p = 0.018). A higher non-sphericity index (NSI) correlated with rupture risk (OR: 3.35, p = 0.016). In non-fetal PCA aneurysms, size-related parameters such as height (6.83 ± 3.54 vs. 4.88 ± 2.57 mm, p = 0.034) and area (190.84 ± 167.08 vs. 107.94 ± 103.10 mm2, p = 0.046) were key rupture predictors. In fetal PCA aneurysms, flow-related parameters like vessel angle (55.78 ± 31.39 vs. 38.51 ± 24.71, p = 0.035) were more influential. ROC analysis showed good discriminatory power, with an area under the curve: 0.726 for fetal PCA and 0.706 for NSI. Conclusions: Fetal PCA influences PCoA aneurysm rupture risk and morphology. NSI is a reliable rupture marker. Integrating morphological and anatomical data may improve rupture risk assessment and clinical decision-making.

MRI radiomics in diagnosing high and low grade meningiomas through systematic review and meta analysis

To evaluate the diagnostic value of magnetic resonance imaging (MRI) radiomics in distinguishing high-grade meningiomas (HGM) from low-grade meningiomas (LGM). A systematic search was conducted in PubMed, EMbase, Web of Science, and The Cochrane Library databases up to December 31, 2023. Two researchers independently screened studies, extracted data, and assessed risk of bias and quality of included studies as well. Meta-analysis was performed using Stata 14 software to calculate pooled sensitivity (SEN), specificity (SPE), positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC). A total of 21 studies with 2253 patients were included (607 HGM, 1646 LGM). Meta-analysis showed an overall SEN of 0.82 (95% CI 0.74-0.88) and SPE of 0.85 (95% CI 0.81-0.89). The PLR and NLR were 5.64 (95% CI 4.17-7.64) and 0.21 (95% CI 0.14-0.31), respectively, with a pooled DOR of 26.66 (95% CI 14.42-49.27) and an AUC of 0.91 (95% CI 0.88-0.93), indicating high diagnostic accuracy. Although additional research is required to validate suitable techniques, MRI radiomics shows strong potential as an accurate tool for meningioma grading. Standardizing radiomics application could enhance diagnostic precision and clinical decision-making for meningioma grading in the future.Trial Registration: CRD42024500086.

Radiomics of Dynamic Contrast-Enhanced MRI for Predicting Radiation-Induced Hepatic Toxicity After Intensity Modulated Radiotherapy for Hepatocellular Carcinoma: A Machine Learning Predictive Model Based on the SHAP Methodology

Objective

To develop an interpretable machine learning (ML) model using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) radiomic data, dosimetric parameters, and clinical data for predicting radiation-induced hepatic toxicity (RIHT) in patients with hepatocellular carcinoma (HCC) following intensity-modulated radiation therapy (IMRT).

Methods

A retrospective analysis of 150 HCC patients was performed, with a 7:3 ratio used to divide the data into training and validation cohorts. Radiomic features from the original MRI sequences and Delta-radiomic features were extracted. Seven ML models based on radiomics were developed: logistic regression (LR), random forest (RF), support vector machine (SVM), eXtreme Gradient Boosting (XGBoost), adaptive boosting (AdaBoost), decision tree (DT), and artificial neural network (ANN). The predictive performance of the models was evaluated using receiver operating characteristic (ROC) curve analysis and calibration curves. Shapley additive explanations (SHAP) were employed to interpret the contribution of each variable and its risk threshold.

Results

Original radiomic features and Delta-radiomic features were extracted from DCE-MRI images and filtered to generate Radiomics-scores and Delta-Radiomics-scores. These were then combined with independent risk factors (Body Mass Index (BMI), V5, and pre-Child-Pugh score(pre-CP)) identified through univariate and multivariate logistic regression and Spearman correlation analysis to construct the ML models. In the training cohort, the AUC values were 0.8651 for LR, 0.7004 for RF, 0.6349 for SVM, 0.6706 for XGBoost, 0.7341 for AdaBoost, 0.6806 for Decision Tree, and 0.6786 for ANN. The corresponding accuracies were 84.4%, 65.6%, 75.0%, 65.6%, 71.9%, 68.8%, and 71.9%, respectively. The validation cohort further confirmed the superiority of the LR model, which was selected as the optimal model. SHAP analysis revealed that Delta-radiomics made a substantial positive contribution to the model.

Conclusion

The interpretable ML model based on radiomics provides a non-invasive tool for predicting RIHT in patients with HCC, demonstrating satisfactory discriminative performance.

An MRI-based deep transfer learning radiomics nomogram for predicting meningioma grade

The aim of this study was to establish a nomogram based on clinical, radiomics, and deep transfer learning (DTL) features to predict meningioma grade. Three hundred forty meningiomas from one hospital composed the training set, and 102 meningiomas from another hospital composed the test set. The enhanced T1 WI images were used for analysis. The clinical, radiomics and DTL features were selected to construct the model. Radiomics and DTL scores were calculated. The deep transfer learning radiomics (DTLR) nomogram was developed on the basis of selected clinical features, radiomics scores and DTL scores. Receiver operating characteristic (ROC) curves and decision curve analysis (DCA) curves were drawn. The clinical features of sex, shape, indistinct margin and peritumoral edema were selected and used to construct the clinical model. Thirty-two radiomics features and 28 DTL features were selected for model construction. The clinical model had an AUC of 0.788. (95% CI: 0.6996-0.8756), with an accuracy of 0.745, a sensitivity of 0.941, and a specificity of 0.549 in the test set. The DTLR nomogram had the highest AUC of 0.866 (95% CI: 0.7984-0.9340), with an accuracy of 0.804, a sensitivity of 0.745, and a specificity of 0.863 in the test set. Compared with the other models, the DTLR nomogram had the greatest net benefit according to the DCA. There was a significant difference between the DTLR nomogram and the clinical model, no significant difference between the rest models in DeLong test.The DTLR nomogram has superior predictive value in DCA and could be a valuable method in clinical decision-making. Given the results of DeLong test, only the radiomics model is sufficient and there is no need to add DTL features. As a new attempt, the DTLR nomogram needs to be improved in the future study.

Radiomics prediction of surgery in ulcerative colitis refractory to medical treatment

BACKGROUND

The surgeries in drug-resistant ulcerative colitis are determined by complex factors. This study evaluated the predictive performance of radiomics analysis on the basis of whether patients with ulcerative colitis in hospital were in the surgical or medical treatment group by discharge from hospital.

METHODS

This single-center retrospective cohort study used CT at admission of patients with US admitted from 2015 to 2022. The target of prediction was whether the patient would undergo surgery by the time of discharge. Radiomics features were extracted using the rectal wall at the level of the tailbone tip of the CT as the region of interest. CT data were randomly classified into a training cohort and a validation cohort, and LASSO regression was performed using the training cohort to create a formula for calculating the radiomics score.

RESULTS

A total of 147 patients were selected, and data from 184 CT scans were collected. Data from 157 CT scans matched the selection criteria and were included. Five features were used for the radiomics score. Univariate logistic regression analysis of clinical information detected a significant influence of severity (p < 0.001), number of drugs used until surgery (p < 0.001), Lichtiger score (p = 0.024), and hemoglobin (p = 0.010). Using a nomogram combining these items, we found that the discriminatory power in the surgery and medical treatment groups was AUC 0.822 (95% confidence interval (CI) 0.841-0.951) for the training cohort and AUC 0.868 (95% CI 0.729-1.000) for the validation cohort, indicating a good ability to discriminate the outcomes.

CONCLUSIONS

Radiomics analysis of CT images of patients with US at the time of admission, combined with clinical data, showed high predictive ability regarding a treatment strategy of surgery or medical treatment.

Multiparametric MRI-Based Radiomics for Identifying Primary Central Nervous System Diffuse Large B-cell Lymphomas' Pathological Subtypes

RATIONALE AND OBJECTIVES

To explore the predictive potential of radiomics features extracted from preoperative multiparametric magnetic resonance imaging (MRI) for identifying pathological subtypes of primary central nervous system diffuse large B-cell lymphomas (PCNS-DLBCL).

METHODS

This study recruited 186 patients with PCNS-DLBCL, including 55 with germinal center B-cell-like (GCB) subtype and 131 with non-GCB subtype. The largest abnormal signal regions of the tumor were automatically segmented in T1-weighted images (T1WI), T2-weighted images, T2 fluid-attenuated inversion recovery, contrast-enhanced T1-weighted (CE-T1WI), and apparent diffusion coefficient (ADC) maps, respectively. Radiomics features were extracted from preprocessed multiparameter preoperative MRI images. To identify GCB and non-GCB subtypes, radiomics models were constructed based on each MRI sequence and combinations of sequences. Clinical models and models combining radiomics and clinical features were also constructed to compare performance.

RESULTS

Radiomics models combining multiple sequences generally outperformed single-sequence radiomics models. The ADC+CE-T1WI model exhibited superior discriminative power, with an area under the curve of 0.867 (95% CI, 0.745-0.988). Models incorporating more sequences (3-5 sequences) did not demonstrate better performance. The performance of the model combining radiomics features with clinical features showed no improvement.

CONCLUSION

Radiomics based on multiparametric MRI have independent value in predicting the pathological subtypes of PCNS-DLBCL patients.

Harmonizing CT scanner acquisition variability in an anthropomorphic phantom: A comparative study of image-level and feature-level harmonization using GAN, ComBat, and their combination

PURPOSE

Radiomics allows for the quantification of medical images and facilitates precision medicine. Many radiomic features derived from computed tomography (CT) are sensitive to variations across scanners, reconstruction settings, and acquisition protocols. In this phantom study, eight different CT reconstruction parameters were varied to explore image- and feature-level harmonization approaches to improve tissue classification.

METHODS

Varying reconstructions of an anthropomorphic radiopaque phantom containing three lesion categories (metastasis, hemangioma, and benign cyst) and normal liver tissue were used for evaluating two harmonization methods and their combination: (i) generative adversarial networks (GANs) at the image level; (ii) ComBat at the feature level, and (iii) a combination of (i) and (ii). A total of 93 texture and intensity features were extracted from each tissue class before and after image-level harmonization and were also harmonized at the feature level. Reproducibility and stability were assessed via the Concordance Correlation Coefficient (CCC) and pairwise comparisons using paired stability tests. The ability of features to discriminate between tissue classes was assessed by measuring the area under the receiver operating characteristic curve. The global reproducibility and discriminative power were assessed by averaging over the entire dataset and across all tissue types.

RESULTS

ComBat improved reproducibility by 31.58% and stability by 5.24%, while GAN increased reproducibility by 8% it reduced stability by 4.33%. Classification analysis revealed that ComBat increased average AUC by 15.19%, whereas GAN decreased AUC by 2.56%.

CONCLUSION

While GAN qualitatively enhances image harmonization, ComBat provides superior statistical improvements in feature stability and classification performance, highlighting the importance of robust feature-level harmonization in radiomics.

Longitudinal stability of HyperSightTM-CBCT based radiomic features in patients with CT guided adaptive SBRT for prostate cancer

CT-guided adaptive radiotherapy (aRT) based on HyperSightTM-CBCT provides high-quality imaging, allowing quantitative radiomic feature analysis as a monitoring tool. This study comprehensively evaluates the stability of radiomic features, as potential imaging biomarkers, in pelvic structures of prostate cancer patients treated with adaptive stereotactic body radiation therapy (SBRT). Between December 2023 and July 2024, 32 patients with localized prostate cancer underwent adaptive SBRT at the Ethos® linear accelerator (Varian, Siemens Healthineers) with HyperSight-CBCT imaging. Longitudinal stability was assessed by intraclass correlation coefficient (ICC) over five fractions of aRT for target structures and non-hollow organs at risk. In pooled organs at risk, 93.0% of features showed very high stability (ICC > 0.9) compared to 67.4% in pooled target structures, indicating significantly lower stability for target structures (p = 0.00009129). Second-order features demonstrated greater stability than conventional and shape-based features (p = 0.0433, p = 0.0252). Fraction number significantly affected longitudinal prostate feature variability (p = 0.0135). This study comprehensively analyzed HyperSight-CBCT imaging to evaluate longitudinal stability of radiomic features during adaptive SBRT for prostate cancer. The trends observed will provide a framework for future CT-guided aRT studies, facilitating quantitative imaging analysis of radiological biomarkers for clinical translation and improving personalized treatment.

Development and validation of a CT-based radiomics nomogram for predicting progression-free survival in patients with small cell lung cancer

PURPOSE

Small cell lung cancer (SCLC) is a highly aggressive form of lung cancer, representing about 15% of cases worldwide. Despite advances in imaging, such as low-dose CT, which have increased diagnostic rates, survival outcomes for SCLC patients have remained stagnant. Recent studies have only focused on radiomics, which extracts detailed quantitative features from imaging, with clinical risk factors to improve prognostic models. Therefore, this study aimed to develop a clinical-radiomics fusion nomogram based on computed tomography (CT) to estimate progression-free survival (PFS) in patients diagnosed with SCLC. By integrating radiomics features extracted from CT with clinical data, this model provides personalized prognostic assessment for clinicians. Its clinical utility lies in aiding treatment decision-making by offering more accurate prognostic evaluation, optimizing therapeutic strategies, and identifying high-risk patients at an early stage, ultimately improving overall survival and quality of life.

METHODS

To develop the nomogram model, 95 patients diagnosed with pathologically confirmed SCLC between January 1, 2013, and December 31, 2023, were included in the study cohort. Participants were randomly divided into training and validation cohorts in a 7:3 ratio. Radiomics features associated with PFS were generated using the least absolute shrinkage and selection operator (LASSO) along with univariate and multivariate analyses. Additionally, in the training cohort, both univariate and multivariate analyses using Cox regression were conducted to identify the significant clinical risk factors influencing PFS. The predictive performance of the clinical and clinical-radiomics fusion nomogram were evaluated using the concordance index, calibration plots, and decision curve analysis (DCA).

RESULTS

Five radiomics features were selected and used to calculate the radiomics score (Rad-score). The radiomics features were significantly associated with PFS (hazard ratio: 0.5765, 95% confidence interval: 0.3641-0.9128, p < 0.05). Three clinical risk factors significantly associated with PFS were identified: neuron-specific enolase (NSE), carbohydrate antigen 125 levels (CA125), and treatment type, such as surgery. The clinical-radiomics fusion nomogram model (C-index:0.744) demonstrated superior performance compared to the clinical nomogram model (C-index: 0.718) in the training cohort. DCA indicated that the clinical-radiomics fusion nomogram outperformed the clinical nomogram in terms of clinical usefulness.

CONCLUSIONS

A CT-based clinical-radiomics fusion nomogram was developed to predict PFS in patients with SCLC, which is useful in providing individualized information.

ADVANCES IN KNOWLEDGE

A clinical-radiomics fusion nomogram was constructed to estimate the probability of PFS based on clinical risk factors and the rad-score.

Vascular and glymphatic dysfunction as drivers of cognitive impairment in Alzheimer's disease: Insights from computational approaches

Alzheimer's disease (AD) is driven by complex interactions between vascular dysfunction, glymphatic system impairment, and neuroinflammation. Vascular aging, characterized by arterial stiffness and reduced cerebral blood flow (CBF), disrupts the pulsatile forces necessary for glymphatic clearance, exacerbating amyloid-beta (Aβ) accumulation and cognitive decline. This review synthesizes insights into the mechanistic crosstalk between these systems and explores their contributions to AD pathogenesis. Emerging machine learning (ML) tools, such as DeepLabCut and Motion sequencing (MoSeq), offer innovative solutions for analyzing multimodal data and enhancing diagnostic precision. Integrating ML with imaging and behavioral analyses bridges gaps in understanding vascular-glymphatic dysfunction. Future research must prioritize these interactions to develop early diagnostics and targeted interventions, advancing our understanding of neurovascular health in AD.

Assessment of CT-to-physical density table for multiple image reconstruction functions with a large-bore scanner for radiotherapy treatment planning

PURPOSE

To evaluate the performance of the Aquilion Exceed LB computed tomography (CT) scanner for radiotherapy treatment planning, this study examined the effect of different combinations of the image reconstruction function (IRF) (AiCE and AIDR) and scan parameters on the CT-to-physical density (CT-PD) table and radiation dose in the phantom, and the effect of different object positions on CT values.

METHODS

To investigate IRF's influence on each material, we calculated CT values by varying tube current, pitch, field of view (FOV), and phantom position for each IRF, comparing them with reference values using filtered back projection (FBP). Furthermore, we evaluated changes in depth dose values due to IRF differences using a solid phantom.

RESULTS

In the combinations of changes in IRF and scan parameters the change in CT value (ΔHU) of each material was within ±10 HU, except for most conditions. The change in physical density (ΔPD) was within ±0.02 g/cm3 for all combinations. For changes in phantom position, ΔHU was within ±25 HU for changes within the scan FOV, except for Bone 200 mg/cc and 1250 mg/cc. In areas outside the scan FOV with an expanded FOV, ΔHU was significantly larger than within the scan FOV. Variations in depth dose for different IRFs using solid phantoms were within ±0.5 %, except at material boundaries.

CONCLUSION

Our evaluations of the CT values and dose calculations suggested no need to change the CT-PD table, even with multiple IRFs.

T1WI Radiomics Analysis of Anterior Scalene Muscle: A Preliminary Application in Neurogenic Thoracic Outlet Syndrome

AIM

This study aimed to analyze the differences in radiomic features of the anterior scalene muscle and evaluate the diagnostic performance of MRI-based radiomics model for neurogenic thoracic outlet syndrome (NTOS).

MATERIALS AND METHODS

Imaging data of patients with NTOS who underwent preoperative brachial plexus magnetic resonance neurography were collected and were randomly divided into training and test groups. The anterior scalene muscle area was sliced in the T1WI sequence as the region of interest for the extraction of radiomics features. The most significant features were identified using feature selection and dimensionality-reduction methods. Various machine learning algorithms were applied to construct regression models. Model performance was evaluated using area under the receiver operating characteristic curve (AUROC).

RESULTS

Totally, 267 radiomics features were extracted, of which 57 showed significant differences ( P  ≤ 0.05) between the abnormal and normal anterior scalene muscle groups. The least absolute shrinkage and selection operator regression model identified 13 optimal radiomic features with nonzero coefficients for constructing the model. In the training set, the AUROCs of diagnostic models built by different machine learning algorithms, ranked from highest to lowest, were as follows: support vector machine (SVM), 0.953; multilayer perception (MLP), 0.936; logistic regression (LR), 0.926; light gradient boosting machine (LightGBM), 0.906; and K-nearest neighbors (KNN), 0.813. In the testing set, the rankings were as follows: LR, 0.933; SVM, 0.886; KNN, 0.843; LightGBM, 0.824; and MLP, 0.706.

CONCLUSIONS

NTOS is attributed to anterior scalene muscle abnormalities and exhibits distinct radiomic features. Integrating these features with machine learning can improve traditional manual image interpretation, offering further clarity in NTOS diagnosis.

The Value of Whole-Volume Radiomics Machine Learning Model Based on Multiparametric MRI in Predicting Triple-Negative Breast Cancer

OBJECTIVE

This study aimed to investigate the value of radiomics analysis in the precise diagnosis of triple-negative breast cancer (TNBC) based on breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and apparent diffusion coefficient (ADC) maps.

METHODS

This retrospective study included 326 patients with pathologically proven breast cancer (TNBC: 129, non-TNBC: 197). The lesions were segmented using the ITK-SNAP software, and whole-volume radiomics features were extracted using a radiomics platform. Radiomics features were obtained from DCE-MRI and ADC maps. The least absolute shrinkage and selection operator regression method was employed for feature selection. Three prediction models were constructed using a support vector machine classifier: Model A (based on the selected features of the ADC maps), Model B (based on the selected features of DCE-MRI), and Model C (based on the selected features of both combined). Receiver operating characteristic curves were used to evaluate the diagnostic performance of the conventional MR image model and the 3 radiomics models in predicting TNBC.

RESULTS

In the training dataset, the AUCs for the conventional MR image model and the 3 radiomics models were 0.749, 0.801, 0.847, and 0.896. The AUCs for the conventional MR image model and 3 radiomics models in the validation dataset were 0.693, 0.742, 0.793, and 0.876, respectively.

CONCLUSIONS

Radiomics based on the combination of whole volume DCE-MRI and ADC maps is a promising tool for distinguishing between TNBC and non-TNBC.

Radiomics and Artificial Intelligence Landscape for [18F]FDG PET/CT in Multiple Myeloma

[18F]FDG PET/CT is a powerful imaging modality of high performance in multiple myeloma (MM) and is considered the appropriate method for assessing treatment response in this disease. On the other hand, due to the heterogeneous and sometimes complex patterns of bone marrow infiltration in MM, the interpretation of PET/CT can be particularly challenging, hampering interobserver reproducibility and limiting the diagnostic and prognostic ability of the modality. Although many approaches have been developed to address the issue of standardization, none can yet be considered a standard method for interpretation or objective quantification of PET/CT. Therefore, advanced diagnostic quantification approaches are needed to support and potentially guide the management of MM. In recent years, radiomics has emerged as an innovative method for high-throughput mining of image-derived features for clinical decision making, which may be particularly helpful in oncology. In addition, machine learning and deep learning, both subfields of artificial intelligence (AI) closely related to the radiomics process, have been increasingly applied to automated image analysis, offering new possibilities for a standardized evaluation of imaging modalities such as CT, PET/CT and MRI in oncology. In line with this, the initial but steadily growing literature on the application of radiomics and AI-based methods in the field of [18F]FDG PET/CT in MM has already yielded encouraging results, offering a potentially reliable tool towards optimization and standardization of interpretation in this disease. The main results of these studies are presented in this review.

A nomogram-based survival prediction model for non-small cell lung cancer patients based on clinical risk factors and multiregion radiomics features

AIM

This study focuses on developing a nomogram-based overall survival (OS) prediction model for non-small cell lung cancer (NSCLC) patients by integrating clinical factors with multiregion radiomics features extracted from pretreatment CT images. The proposed nomogram aims to assist clinicians in stratifying patients into high- and low-risk groups for personalised treatment strategies.

MATERIALS AND METHODS

From 2008 to 2018, 77 NSCLC patients were included. The radiomics feature was extracted from the internal and peripheral tumour region of pretreatment computed tomography (CT) images. The least absolute shrinkage and selection operator (LASSO) and the univariable Cox regression model were used to select the radiomics features. The Rad-score was defined as a linear combination of the selected radiomics features and the Cox proportional hazards regression coefficients. The combined model was constructed based on the clinicopathological factors and the Rad-score. The discrimination capacity of the prediction model was evaluated by Harrell's concordance index (C-index), the calibration curve, and the Kaplan-Meier survival curve.

RESULTS

We found that nine radiomics features and histology were independent predictors. The combined model showed the best performance (C-index: 0.799 [95% CI: 0.726-0.872]) compared with the clinical model (C-index: 0.692 [95% CI: 0.625-0.759]) and Rad-score (C-index: 0.663 [95% CI: 0.580-0.746]), and could significantly stratify into high-risk and low-risk NSCLC patients. The calibration curve also showed good consistency between the observation and the prediction.

CONCLUSIONS

The multregion radiomics features have the potential for predicting OS in NSCLC patients. The nomogram-based survival prediction model demonstrates significant potential in guiding clinical decision-making, allowing for precise and personalised treatment for NSCLC patients.

The Role of AI in the Evaluation of Neuroendocrine Tumors: Current State of the Art

Advancements in Artificial Intelligence (AI) are driving a paradigm shift in the field of medical diagnostics, integrating new developments into various aspects of the clinical workflow. Neuroendocrine neoplasms are a diverse and heterogeneous group of tumors that pose significant diagnostic and management challenges due to their variable clinical presentations and biological behavior. Innovative approaches are essential to overcome these challenges and improve the current standard of care. AI-driven applications, particularly in imaging workflows, hold promise for enhancing tumor detection, classification, and grading by leveraging advanced radiomics and deep learning techniques. This article reviews the current and emerging applications of AI computer vision in the care of neuroendocrine neoplasms, focusing on its integration into imaging workflows, diagnostics, prognostic modeling, and therapeutic planning.

Predicting telomerase reverse transcriptase promoter mutation status in glioblastoma by whole-tumor multi-sequence magnetic resonance texture analysis

OBJECTIVE

This study aimed to determine the feasibility of preoperative multi-sequence magnetic resonance texture analysis (MRTA) for predicting TERT promoter mutation status in IDH-wildtype glioblastoma (IDHwt GB).

METHODS

The clinical and imaging data of 111 patients with IDHwt GB at our hospital between November 2018 and June 2023 were retrospectively analyzed as the training set, and those of 23 patients with IDHwt GB between July 2023 and November 2023 were interpreted as the validation set. We used molecular sequencing results to classify the training set into TERT promoter mutation and wildtype groups. Textural features of the whole-tumor volume were extracted, including T2-weighted imaging (T2WI), T2-fluid-attenuated inversion recovery, apparent diffusion coefficient (ADC) map, and contrast-enhanced T1-weighted imaging (CE-T1). All textural features were obtained using open-source pyradiomics. After feature selection, logistic regression was used to build prediction models, and a nomogram was generated. Finally, the model was validated using validation cohort.

RESULTS

The CE-T1_Model (AUC 0.704) had a better predictive ability than the T2_Model (AUC 0.684) and ADC_Model (AUC 0.624). The MRI_Combined_Model (CE-T1, T2, and ADC texture features) (AUC 0.780) had a better predictive ability than the Clinical_Model (AUC 0.758). The Combined_Model (CE-T1, T2, ADC texture features, and clinical features) had the best predictive performance (AUC 0.871), with a sensitivity, specificity, and accuracy of 82.60 %, 83.30 %, and 80.18 %, respectively. The AUC, sensitivity, specificity, and accuracy in the validation cohort were 0.775, 86.70 %, 75.00 %, and 69.57 %, respectively.

CONCLUSIONS

Whole-tumor multi-sequence MRTA can be used as non-invasive quantitative parameters to assist in the preoperative clinical prediction of TERT promoter mutation status in IDHwt GB.

Radiomics for Differentiating Pancreatic Mucinous Cystic Neoplasm from Serous Cystic Neoplasm: Systematic Review and Meta-Analysis

BACKGROUND

As pancreatic cystic neoplasms (PCN) differ in current standard of care, and these treatments can affect quality of life to varying degrees, a definitive preoperative diagnosis must be reliable. Current diagnostic approaches, specifically traditional cross-sectional imaging techniques, face certain limitations. But radiomics has been shown to have high diagnostic accuracy across a range of diseases. Objective to conduct a comprehensive review of the literature on the use of radiomics to differentiate Mucinous Cystic Neoplasm (MCN) from Serous Cystic Neoplasm (SCN).

METHODS

This study was comprehensively searched in Pubmed, Scopus and Web of Science databases for meta-analysis of studies that used radiomics to distinguish MCN from SCN. Risk of bias was assessed using the diagnostic accuracy study quality assessment method and combined with sensitivity, specificity, diagnostic odds ratio, and summary receiver operating characteristic (SROC)curve analysis.

RESULTS

A total of 884 patients from 8 studies were included in this analysis, including 365 MCN and 519 SCN. The Meta-analysis found that radiomics identified MCN and SCN with high sensitivity and specificity, with combined sensitivity and specificity of 0.84(0.82-0.87) and 0.82(0.79-0.84). The positive likelihood ratio (PLR) and the negative likelihood ratio (NLR) are 5.61(3.72, 8.47) and 0.14(0.09-0.26). In addition, the area under the SROC curve (AUC) was drawn at 0.93. No significant risk of publication bias was detected through the funnel plot analysis. The performances of feature extraction from the volume of interest (VOI) or Using AI classifier in the radiomics models were superior to those of protocols employing region of interest (ROI) or absence of AI classifier.

CONCLUSION

This meta-analysis demonstrates that radiomics exhibits high sensitivity and specificity in distinguishing between MCN and SCN, and has the potential to become a reliable diagnostic tool for their identification.

Predicting lateral pelvic lymph node metastasis in rectal cancer patients using MRI radiomics: a multicenter retrospective study

MRI has relatively low sensitivity and specificity in detecting lymph node metastases. This study aimed to develop and validate an MRI radiomics-based model for predicting lateral pelvic lymph node (LPLN) metastasis in rectal cancer patients who underwent LPLN dissection, and to compare its performance with that of radiologists. This multicenter retrospective study included 336 rectal cancer patients (199 men; mean age, 58.9 years ± 11.1 [standard deviation]) who underwent LPLN dissection. Patients were divided into development (n = 190) and validation (n = 146) cohorts. Radiomics features were extracted from MR images, and the Least Absolute Shrinkage and Selection Operator regression was used to construct radiomics and clinical-radiomics models. Model performance was compared with radiologists using receiver operating characteristic (ROC) analysis. Malignant LPLN was diagnosed in 32.4% of the development cohort (65/190) and 32.9% of the validation cohort (48/146) (P = 0.798). Seven radiomics features and two clinical features were selected. The radiomics and clinical-radiomics models demonstrated area under the curves (AUCs) of 0.819 and 0.830 in the development cohort and 0.821 and 0.829 in the validation cohort, respectively. The optimal cut-off (- 0.47) yielded sensitivities of 72.3% and 45.8% and specificities of 82.4% and 87.8% in the development and validation cohorts, respectively. Decision curve analysis indicated no additional net benefit from the clinical-radiomics model compared to the radiomics-only model. Radiologists' AUCs were significantly lower than that of the radiomics model (0.842) and improved with radiomics probability scores (0.734 vs. 0.801; 0.668 vs. 0.791). The MRI-based radiomics model significantly improves the prediction of LPLN metastasis in rectal cancer, outperforming conventional criteria used by radiologists.Trial registration: Retrospectively registered.

An MRI-based fusion model for preoperative prediction of perineural invasion status in patients with intrahepatic cholangiocarcinoma

BACKGROUND

To develop and validate an MRI-based fusion model for preoperative prediction of perineural invasion (PNI) status in patients with intrahepatic cholangiocarcinoma (ICC).

METHODS

A retrospective collection of 192 ICC patients from three medical centers (training set: n = 147; external test set: n = 45) was performed. Patients were classified into the PNI-positive and PNI-negative groups based on postoperative pathological results. After image preprocessing, a total of 1,197 features were extracted from T2-weighted imaging (T2WI). Feature selection was performed, and a radiomics model was constructed using machine learning algorithms, followed by SHapley Additive exPlanations (SHAP) visualization. Subsequently, a deep learning model was constructed based on the pre-trained ResNet101, with Gradient-weighted Class Activation Mapping (Grad-CAM) used for visualization. Finally, a fusion model incorporating deep learning, radiomics, and clinical features was developed using logistic regression, and visualization was performed with a nomogram. The predictive performance of the model was evaluated based on the area under the curve (AUC), calibration curves, and decision curve analysis (DCA).

RESULTS

The fusion model, which integrates deep learning signature, radiomics signature, and two clinical features, demonstrated strong discrimination for PNI status. In the training set, the AUC was 0.905, with an accuracy of 0.823; in the external test set, the AUC was 0.760, with an accuracy of 0.778. Visualization methods provided support for the practical application of the model.

CONCLUSION

The fusion model aids in the preoperative identification of PNI status in patients with ICC, and may help guide clinical decision-making regarding preoperative staging and adjuvant therapy.

Infrared thermography-based radiomics for early detection of metabolic syndrome

Radiomics is increasingly utilized in medical image analysis. This study evaluates the use of infrared thermography, a technique well-suited for radiomic analysis, in diagnosing metabolic syndrome (MS). Facial and palmar thermographs from 200 males (100 healthy controls and 100 MS patients) were analyzed. The dataset was split into a training cohort (n = 140) and a validation cohort (n = 60). All participants underwent laboratory testing on the same day as infrared thermography imaging. A total of 1656 radiomic features were extracted from each participant's thermographs and refined using Pearson correlation coefficients, two-sample t-tests, and LASSO regression. A binary random forest (RF) classification model was then constructed and evaluated based on its calibration, discrimination, and clinical utility. The RF model demonstrated strong diagnostic performance, achieving an AUC of 0.91 in the validation cohort. Calibration and decision curve analyses confirmed the model's clinical applicability. Infrared thermography-based radiomics offers a promising, non-invasive method for early screening of MS, highlighting its potential clinical utility.

HER2 (2+)/SISH-positive vs. HER2 (3+) Breast Cancer: Pre-treatment MRI Differences and Accuracy of pCR Prediction on Post-treatment MRI

RATIONALE AND OBJECTIVES

To evaluate whether HER2 (human epidermal growth factor receptor 2) (2+)/SISH (silver-enhanced in situ hybridization)+ and HER2 (3+) breast cancers exhibit distinct imaging characteristics on pre-treatment MRI and assess differences in pCR (pathologic complete response) prediction accuracy on post-treatment MRI, considering interobserver variability.

METHODS

This retrospective study included 301 HER2-positive breast cancer patients (mean age, 54 ± 10 years) who underwent NAC and surgery. Pre-treatment MRI features were analyzed in consensus. Two radiologists independently assessed post-treatment MRI for shrinkage patterns and response according to RECIST v1.1, further categorizing complete responses into rCR (radiologic complete response) and near-rCR. Interobserver agreement was measured (Cohen's kappa), and pCR was defined as no residual invasive or in situ tumor in the breast (ypT0) on the final pathology report. Sensitivity, specificity, and AUC were used to evaluate pCR prediction.

RESULTS

Fifty-four patients had HER2 (2+)/SISH+ and 247 had HER2 (3+) tumors. pCR rates were significantly higher in HER2 (3+) (58.7% vs. 18.5%, p < 0.001). On pre-treatment MRI, HER2 (2+)/SISH+ tumors more often appeared as single masses, while HER2 (3+) tumors showed more NME (non-mass enhancement) (44.5% vs. 16.7%, p < 0.001) and mass with NME (33.6% vs. 9.3%, p = 0.005). Post-treatment MRI showed simple concentric shrinkage in HER2 (2+)/SISH+ and no enhancement in HER2 (3+). Agreement was moderate (κ = 0.541-0.588). For pCR prediction, rCR alone yielded AUCs ranging from 0.659 to 0.756. Adding near-rCR improved specificity but reduced sensitivity, with a significant AUC increase for one reader (p = 0.011).

CONCLUSION

Pre-treatment MRI revealed distinct imaging characteristics between subgroups. While pCR rates were higher in HER2 (3+), MRI-based pCR prediction showed similar performance, though near-rCR reduced sensitivity.

Exploring the in silico adaptation of the Nephroblastoma Oncosimulator to MRI scans, treatment data, and histological profiles of patients from different risk groups

Introduction

Nephroblastoma or Wilms' tumor is the most prevalent type of renal tumor in pediatric oncology. Although the overall survival rate for this condition is excellent today (∼90%), there have been no significant improvements over the past two decades. In silico models aim to simulate tumor progression and treatment responses over time; they hold immense potential for enhancing the predictive accuracy and optimizing treatment protocols as they are inspired by the digital twin paradigm.

Methods

The present study uses T2-weighted magnetic resonance images, chemotherapy treatment plans, and post-surgical histological profiles from three patients enrolled in the SIOP 2001/GPOH clinical trial, where each patient represents a distinct clinically assessed risk group. We investigated the clinical adaptation of the Nephroblastoma Oncosimulator to the datasets from these patients with the goal of deriving appropriate value distributions of the model input parameters that enable accurate prediction of tumor volume reduction in response to preoperative chemotherapy.

Results

Our primary focus was on the total cell kill ratio as a parameter reflecting treatment effectiveness. We derived the distribution of this parameter for one patient from each risk group: low (Mdn = 0.875, IQR [0.750, 0.875], n = 178), intermediate (Mdn = 0.875, IQR [0.750, 0.875], n = 175), and high (Mdn = 0.485, IQR [0.438, 0.532], n = 103). Statistically significant differences were observed between the high-risk group and both the low- and intermediate-risk groups (p < 0.001).

Discussion

The present work establishes a foundation for further studies using available retrospective datasets and additional patients per risk group. These efforts are expected to help validate the findings, advance model development, and extend this mechanistic multiscale discretized cancer model. However, clinical validation is ultimately required to assess the potential uses of the model in clinical decision-support systems.

Preoperative CT-based radiomics nomogram for progression-free survival prediction in pediatric posterior mediastinal malignancies

Background

Progression-free survival (PFS) prediction plays a pivotal role in developing personalized treatment strategies and ensuring favorable long-term outcomes in pediatric posterior mediastinal malignant tumors. This study developed and validated the first preoperative contrast-enhanced computed tomography (CT)-based radiomics nomogram to forecast PFS in posterior mediastinal malignancies patients. The aim was to provide a clinically applicable prognostic tool to stratify high-risk populations.

Methods

Medical data from 306 patients with posterior mediastinal malignancies were analyzed retrospectively and randomly divided into training (n = 215) and test sets (n = 91). The clinical model was built using conventional clinical data and CT signs. Selection of the radiomic features was performed using maximum relevance minimum redundancy and the least absolute shrinkage and selection operator. To overcome class imbalance, the synthetic minority over-sampling technique was used in the training set. Radiomics signature was derived using logistic regression algorithm, and we developed a nomogram by integrating the clinical model and the radiomics signature. The predictive efficiency of the nomogram was assessed using the area under the curve (AUC), brier score (BS), decision curve analysis, and calibration.

Results

The Ki-67 index and metastasis were identified as independent predictors, with the test set achieving an AUC of 0.82 (0.647-0.964) and a BS of 0.21 (0.181-0.239). Nineteen radiomics features most relevant to PFS were retained, with the logistic regression algorithm achieving an AUC of 0.77 (0.589-0.896) and a BS of 0.26 (0.215-0.292) in the test set. The radiomics nomogram demonstrated best predictive capability in the test set, achieving an AUC of 0.87 (0.733-0.968) and a BS of 0.22 (0.177-0.255), compared with remaining prediction models. Both calibration curves and decision curve analysis demonstrated good fit and clinical benefit.

Conclusions

Our contrast-enhanced CT-based radiomics nomogram may be a dependable, precise, and noninvasive prognostic tool to predict PFS in pediatric posterior mediastinal malignancies preoperatively.

Incorporating frequency domain features into radiomics for improved prognosis of esophageal cancer

Esophageal cancer is a highly aggressive gastrointestinal malignancy with a poor prognosis, making accurate prognostic assessment essential for patient care. The performance of the esophageal cancer prognosis model based on conventional radiomics is limited, as it mainly characterizes the spatial features such as texture of the tumor area, and cannot fully describe the complexity of esophageal cancer tumors. Therefore, we incorporate the frequency domain features into radiomics to improve the prognostic ability of esophageal cancer. Three hundred fifteen esophageal cancer patients participated in the death risk prediction experiment, with 80% being the training set and 20% being the testing set. We use fivefold cross validation for training, and fuse the 5 trained models through voting to obtain the final prognostic model for testing. The CatBoost achieved the best performance compared to machine learning methods such as random forests and decision tree. The experimental results showed that the combination of frequency domain and radiomics features achieved the highest performance in death predicting esophageal cancer (accuracy: 0.7423, precision: 0.7470, recall: 0.7375, specification: 0.8030, AUC: 0.8487), which was significantly better than the performance of frequency domain or radiomics features alone. The results of Kaplan-Meier survival analysis validated the performance of our method in death predicting esophageal cancer. The proposed method provides technical support for accurate prognosis of esophageal cancer.

Prediction of solid pseudopapillary tumor invasiveness of the pancreas based on multiphase contrast-enhanced CT radiomics nomogram

Objectives

To construct a multiphase contrast-enhanced CT-based radiomics nomogram that combines traditional CT features and radiomics signature for predicting the invasiveness of pancreatic solid pseudopapillary neoplasm (PSPN).

Methods

A total of 114 patients with surgical pathologic diagnoses of PSPN were retrospectively included and classified into training (n = 79) and validation sets (n = 35). Univariate and multivariate analyses were adopted for screening traditional CT features significantly associated with the invasiveness of PSPN as independent predictors, and a traditional CT model was established. Radiomics features were extracted from the contrast-enhanced CT images, and logistic regression analysis was employed to establish a machine learning model, including an unenhanced model (model U), an arterial phase model (model A), a venous phase model (model V), and a combined radiomics model (model U+A+V). A radiomics nomogram was subsequently constructed and visualized by combining traditional CT independent predictors and radiomics signature. Model performance was assessed through Delong's test and receiver operating characteristic (ROC) curve analysis. Decision curve analysis (DCA) was applied to assess the model's clinical utility.

Results

Multivariate analysis suggested that solid tumors (OR = 6.565, 95% CI: 1.238-34.816, P = 0.027) and ill-defined tumor margins (OR = 2.442, 95% CI: 1.038-5.741, P = 0.041) were independent predictors of the invasiveness of PSPN. The areas under the curve (AUCs) of the traditional CT model in the training and validation sets were 0.653 and 0.797, respectively. Among the four radiomics models, the model U+A+V exhibited the best diagnostic performance, with AUCs of 0.857 and 0.839 in the training and validation sets, respectively. In addition, the AUCs of the nomogram in the training and validation sets were 0.87 and 0.867, respectively, which were better than those of the radiomics model and the traditional CT model. The DCA results indicated that with the threshold probability being within the relevant range, the radiomics nomogram offered an increased net benefit to clinical decision making.

Conclusion

Multiphase contrast-enhanced CT radiomics can noninvasively predict the invasiveness of PSPN. In addition, the radiomics nomogram combining radiomics signature and traditional CT signs can further improve classification ability.

Diagnostic Performance of Radiomics Modeling in Predicting the Human Papillomavirus Status of Oropharyngeal Cancer: A Systematic Review and Meta-Analysis

In this review, we sought to assess the diagnostic performance and methodological quality of studies utilizing radiomics for the prediction of human papillomavirus (HPV) status in patients with oropharyngeal squamous cell carcinoma. A comprehensive literature search of PubMed, Ovid, Cochrane, Web of Science, and Scopus from inception until June 7, 2022, was performed to identify eligible studies. Strict inclusion and exclusion criteria were applied to the identified studies. Data collection was performed by two independent reviewers with disagreements resolved by consensus review with a third reviewer. In total, 14 articles were chosen, with a total of 15 radiomics models. Of the included studies, 12 models reported sensitivity, with a mean of 0.778 (standard deviation (SD) = 0.073). Similarly, 12 models reported specificity, with a mean of 0.751 (SD = 0.111). The area under the curve (AUC) was reported by all 15 models, with a mean of 0.814 (SD = 0.081). Finally, accuracy was reported by eight models, with a mean of 0.768 (SD = 0.044). A meta-analysis was performed on eight studies that reported AUCs with confidence intervals (CIs), returning a pooled AUC of 0.764 (95% CI = 0.758 to 0.770). The Radiomics Quality Score (RQS) was applied to each included study as a measure of quality. RQS ranged from -1 to 22, with a mean of 13.4 and an intraclass coefficient of 0.874. Radiomics modeling has shown promise in serving as a diagnostic indicator for HPV status in patients with oropharyngeal cancer. Nevertheless, the quality of research methodologies in this area is a limiting factor for its broader clinical application and highlights the need for enhanced funding to support further research efforts.

Current and Future Applications of PET Radiomics in Radiation Oncology

This review delves into the principles of PET imaging and radiomics, emphasizing their importance in detecting, staging, and monitoring various cancers. It highlights the clinical applications of PET radiomics in oncology, showcasing its impact on personalized cancer care. Additionally, the review addresses challenges such as standardizing PET radiomics, integrating multiomics data, and ethical concerns in clinical decision-making. Future directions are also discussed, including broader applications of PET radiomics in clinical trials, artificial intelligence integration for automated analysis, and incorporating multiomics data for a comprehensive understanding of tumor biology.

Improving Prediction of Intracranial Aneurysm Rupture Status Using Temporal Velocity-Informatics

This study uses a spatial pattern analysis of time-resolved aneurysmal velocity fields to enhance the characterization of intracranial aneurysms' (IA) rupture status. We name this technique temporal velocity-informatics (TVI). In this study, using imaging data obtained from 112 subjects harboring IAs with known rupture status, we reconstructed 3D models to get aneurysmal velocity data by performing computational fluid dynamics (CFD) simulations and morphological information. TVI analyses were conducted for time-resolved velocity fields to quantitatively obtain spatial and temporal flow disturbance. Lastly, we employed four machine learning (ML) methods (e.g., support vector machine [SVM]) to evaluate the prediction performance of the proposed TVI. Overall, the SVM's prediction with TVI performed the best: an area under the curve (AUC) value of 0.92 and a total accuracy of 86%. With TVI, the SVM classifier correctly identified 77 and 92% of ruptured and unruptured IAs, respectively.

Radiomics to predict PNI in ESCC

OBJECTIVE

This study aimed to investigate whether contrast-enhanced computed tomography (CECT) based radiomics analysis could noninvasively predict the perineural invasion (PNI) in esophageal squamous cell carcinoma (ESCC).

METHODS

398 patients with ESCC who underwent resection between February 2016 and March 2020 were retrospectively enrolled in this study. Patients were randomly divided into training and testing cohorts in a 7:3 ratio. Radiomics analysis was performed on the arterial phase images of CECT scans. From these images, 1595 radiomics features were initially extracted. The intraclass correlation coefficient (ICC), wilcoxon rank-sum test, spearman correlation analysis, and boruta algorithm were used for feature selection. Logistic regression (LR), random forest (RF), and support vector machine (SVM) models were established to preidict the PNI status. The performance of these radiomics models was assessed by the area under the receiver operating characteristic curve (AUC). Decision curve analysis (DCA) was conducted to evaluate their clinical utility.

RESULTS

Six radiomics features were retained to build the radiomics models. Among these models, the random forest (RF) model demonstrated superior performance. In the training cohort, the AUC value of the RF model was 0.773, compared to 0.627 for the logistic regression (LR) model and 0.712 for the support vector machine (SVM) model. Similarly, in the testing cohort, the RF model achieved an AUC value of 0.767, outperforming the LR model at 0.638 and the SVM model at 0.683. Decision curve analysis (DCA) suggested that the RF radiomics model exhibited the highest clinical utility.

CONCLUSIONS

CECT-based radiomics analysis, particularly utilizing the RF, can noninvasively predict the PNI in ESCC preoperatively. This novel approach could enhance patient management by providing personalized information, thereby facilitating the development of individualized treatment strategies for ESCC patients.

MRI-based Radiomic Features for Risk Stratification of Ductal Carcinoma in Situ in a Multicenter Setting (ECOG-ACRIN E4112 Trial)

Background Ductal carcinoma in situ (DCIS) is a nonlethal, preinvasive breast cancer for which breast MRI is best suited for accurate disease extent characterization. DCIS is often overtreated, necessitating robust models for improved risk stratification. Purpose To develop logistic regression models using clinical and MRI-based radiomic features of DCIS and to evaluate the performance of such models in predicting disease upstaging at surgery and DCIS score. Materials and Methods This study is a secondary analysis of dynamic contrast-enhanced (DCE) MRI data from the Eastern Cooperative Oncology Group-American College of Radiology Imaging Network, or ECOG-ACRIN, E4112 trial. Primary analysis focused on predicting disease upstaging (n = 295), and secondary analysis focused on predicting DCIS score (n = 174). Radiologist-drawn lesion segmentations and publicly available Cancer Phenomics Toolkit, or CaPTk, software was used to compute 65 radiomic features. Participants were clustered into groups based on their radiomic features (ie, radiomic phenotypes), and principal component analysis was used to summarize the feature space. Clinical information and qualitative MRI features were available. Associations were tested using χ2 and likelihood ratio tests. Data were split into training and test sets with a 3:2 ratio, and model performance was assessed on the test set using the area under the receiver operating characteristic curve (AUC). Results Data from 297 female participants with median age of 60 years (IQR, 51-67 years) were analyzed. Two radiomic phenotypes were identified that were associated with disease upstaging (P = .007). For predicting disease upstaging, the top three radiomic principal components combined with clinical and qualitative MRI predictors yielded the highest AUC of 0.77 (95% CI: 0.65, 0.88) among all tested models (P = .007), identifying 25% more true-negative (49 of 93 true-negative findings, 53% specificity) findings, compared with using clinical information alone (23 of 93 true-negative findings, 28% specificity). Radiomic models were not predictive of the DCIS score (P > .05). Conclusion In patients with DCIS, combining radiomic metrics with clinical information improved prediction of disease upstaging but not DCIS score. ClinicalTrials.gov Identifier: NCT02352883 Supplemental material is available for this article. ©RSNA, 2025 See also the editorial by Kim and Woo in this issue.

Artificial intelligence based on imaging data to predict rectal cancer recurrence: A meta-analysis

PURPOSE

The purpose of this study was to evaluate the diagnostic performance of artificial intelligence based on imaging data to predict rectal cancer recurrence using a meta-analysis system.

MATERIALS AND METHODS

Medline, Embase, Cochrane Library, Web of Science, and other databases were searched for all articles on artificial intelligence prediction of rectal cancer recurrence based on imaging data published publicly from the establishment of the library to December 31, 2023. The quality of the articles was assessed using Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2). Meta-analysis was performed by the software Revman 5.4 and Statistics data (Stata), and sensitivity analysis was used to detect potential sources of heterogeneity and test to assess the presence of publication bias. We evaluated how well imaging-based data can predict recurrence in patients with rectal cancer by analysing the pooled sensitivity, specificity, and area under the curve.

RESULTS

Ten studies were included. The pooled sensitivity, specificity, and area under the curve of imaging-based data for recurrence in patients with rectal cancer were respectively 0.84 (95 % confidence interval [CI]: 0.74-0.91), 0.87 (95 % CI: 0.82-0.91) and 0.92 (95 % CI: 0.89-0.94). Based on QUADAS-2, the quality of the article is acceptable. We found the causes of heterogeneity through meta-regression: recurrence time predesign Lasso. Based on Deeks' funnel plot, no publication bias was detected.

CONCLUSION

Artificial intelligence based on imaging data has a high predictive ability for rectal cancer recurrence.

The application of artificial intelligence in upper gastrointestinal cancers

Upper gastrointestinal cancers, mainly comprising esophageal and gastric cancers, are among the most prevalent cancers worldwide. There are many new cases of upper gastrointestinal cancers annually, and the survival rate tends to be low. Therefore, timely screening, precise diagnosis, appropriate treatment strategies, and effective prognosis are crucial for patients with upper gastrointestinal cancers. In recent years, an increasing number of studies suggest that artificial intelligence (AI) technology can effectively address clinical tasks related to upper gastrointestinal cancers. These studies mainly focus on four aspects: screening, diagnosis, treatment, and prognosis. In this review, we focus on the application of AI technology in clinical tasks related to upper gastrointestinal cancers. Firstly, the basic application pipelines of radiomics and deep learning in medical image analysis were introduced. Furthermore, we separately reviewed the application of AI technology in the aforementioned aspects for both esophageal and gastric cancers. Finally, the current limitations and challenges faced in the field of upper gastrointestinal cancers were summarized, and explorations were conducted on the selection of AI algorithms in various scenarios, the popularization of early screening, the clinical applications of AI, and large multimodal models.

Predicting Hospitalization Length in Geriatric Patients Using Artificial Intelligence and Radiomics

(1) Background: Predicting hospitalization length for COVID-19 patients is crucial for optimizing resource allocation and patient management. Radiomics, combined with machine learning (ML), offers a promising approach by extracting quantitative imaging features from CT scans. The aim of the present study is to use and adapt machine learning (ML) architectures, exploiting CT radiomics information, and analyze algorithms' capability to predict hospitalization at the time of patient admission. (2) Methods: The original CT lung images of 168 COVID-19 patients underwent two segmentations, isolating the ground glass area of the lung parenchyma. After an isotropic voxel resampling and wavelet and Laplacian of Gaussian filtering, 92 intensity and texture radiomics features were extracted. Feature reduction was conducted by applying a last absolute shrinkage and selection operator (LASSO) to the radiomic features set. Three ML classification algorithms, linear support vector machine (LSVM), medium neural network (MNN), and ensemble subspace discriminant (ESD), were trained and validated through a 5-fold cross-validation technique. Model performance was assessed using accuracy, sensitivity, specificity, precision, F1-score, and the area under the receiver operating characteristic curve (AUC-ROC). (3) Results: The LSVM classifier achieved the highest predictive performance, with an accuracy of 86.0% and an AUC of 0.93. However, reliable outcomes are also registered when MNN and ESD architecture are used. (4) Conclusions: The study shows that radiomic features can be used to build a machine learning framework for predicting patient hospitalization duration. The findings suggest that radiomics-based ML models can accurately predict COVID-19 hospitalization length.

Predicting hemorrhagic transformation in acute ischemic stroke: a systematic review, meta-analysis, and methodological quality assessment of CT/MRI-based deep learning and radiomics models

Acute ischemic stroke (AIS) is a major cause of mortality and morbidity, with hemorrhagic transformation (HT) as a severe complication. Accurate prediction of HT is essential for optimizing treatment strategies. This review assesses the accuracy and utility of deep learning (DL) and radiomics in predicting HT through imaging, regarding clinical decision-making for AIS patients. A literature search was conducted across five databases (Pubmed, Scopus, Web of Science, Embase, IEEE) up to January 23, 2025. Studies involving DL or radiomics-based ML models for predicting HT in AIS patients were included. Data from training, validation, and clinical-combined models were extracted and analyzed separately. Pooled sensitivity, specificity, and AUC were calculated with a random-effects bivariate model. For the quality assessment of studies, the Methodological Radiomics Score (METRICS) and QUADAS-2 tool were used. 16 studies consisting of 3,083 individual participants were included in the meta-analysis. The pooled AUC for training cohorts was 0.87, sensitivity 0.80, and specificity 0.85. For validation cohorts, AUC was 0.87, sensitivity 0.81, and specificity 0.86. Clinical-combined models showed an AUC of 0.93, sensitivity 0.84, and specificity 0.89. Moderate to severe heterogeneity was noted and addressed. Deep-learning models outperformed radiomics models, while clinical-combined models outperformed deep learning-only and radiomics-only models. The average METRICS score was 62.85%. No publication bias was detected. DL and radiomics models showed great potential in predicting HT in AIS patients. However, addressing methodological issues-such as inconsistent reference standards and limited external validation-is essential for the clinical implementation of these models.

Improving differentiation of hemorrhagic brain metastases from non-neoplastic hematomas using radiomics and clinical feature fusion

OBJECTIVES

This study aimed to develop and validate a fusion model combining multi-sequence MRI radiomics and clinico-radiological features to distinguish hemorrhagic brain metastasis covered by hematoma (HBM.cbh) from non-neoplastic intracranial hematomas (nn-ICH).

METHODS

The data of 146 patients with pathologically or clinically proven HBM.cbh (n = 55) and nn-ICH (n = 91) were collected from two clinical institutions. Radiomics features were extracted from various regions (hemorrhage and/or edema) based on T2-weighted, T1-weighted, fluid-attenuated inversion-recovery, and T1 contrast-enhanced imaging. Synthetic minority over-sampling technique (SMOTE) was performed to balance the minority group (HBM.cbh). Logistic regression (LR) and k-nearest neighbors (KNN) were utilized to construct the models based on clinico-radiological factors (clinical model), radiomic features from various modalities of MRI (radiomics model), and their combination (fusion model). The area under the curve (AUC) values of different models on the external dataset were compared using DeLong's test.

RESULTS

The 4-sequence radiomics model based on the entire region performed the best in all radiomics models, with or without SMOTE, where the AUCs were 0.83 and 0.84, respectively. The AUC of clinical mode was 0.71 with SMOTE, and 0.62 without SMOTE. The fusion model demonstrated excellent predictive value with or without SMOTE (AUC: 0.93 and 0.90, respectively), outperforming both the radiomics and clinical model (0.93 vs. 0.83, 0.71, p < 0.05 and 0.90 vs. 0.84, 0.62, p < 0.05, respectively).

CONCLUSIONS

The multi-sequence radiomics model is an effective method for differentiating HBM.cbh from nn-ICH. It can yield the best diagnostic performance prediction model when combined with clinico-radiological features.

Do We Need to Add the Type of Treatment Planning System, Dose Calculation Grid Size, and CT Density Curve to Predictive Models?

Background: Generalizability and domain dependency are critical challenges in developing predictive models for healthcare, particularly in medical diagnostics and radiation oncology. Predictive models designed to assess tumor recurrence rely on comprehensive and high-quality datasets, encompassing treatment planning parameters, imaging protocols, and patient-specific data. However, domain dependency, arising from variations in dose calculation algorithms, computed tomography (CT) density conversion curves, imaging modalities, and institutional protocols, can significantly undermine model reliability and clinical utility. Methods: This study evaluated dose calculation differences in the head and neck cancer treatment plans of 19 patients using two treatment planning systems, Pinnacle 9.10 and RayStation 11, with similar dose calculation algorithms. Variations in the dose grid size and CT density conversion curves were assessed for their impact on domain dependency. Results: Results showed that dose grid size differences had a more significant influence within RayStation than Pinnacle, while CT curve variations introduced potential domain discrepancies. The findings underscore the critical role of precise and standardized treatment planning in enhancing the reliability of predictive modeling for tumor recurrence assessment. Conclusions: Incorporating treatment planning parameters, such as dose distribution and target volumes, as explicit features in model training can mitigate the impact of domain dependency and enhance prediction accuracy. Solutions such as multi-institutional data harmonization and domain adaptation techniques are essential to improve model generalizability and robustness. These strategies support the better integration of predictive modeling into clinical workflows, ultimately optimizing patient outcomes and personalized treatment strategies.

Multi-dimensional interpretable deep learning-radiomics based on intra-tumoral and spatial habitat for preoperative prediction of thymic epithelial tumours risk categorisation

BACKGROUND AND PURPOSE

This study aims to develop and compare combined models based on enhanced CT-based radiomics, multi-dimensional deep learning, clinical-conventional imaging and spatial habitat analysis to achieve accurate prediction of thymoma risk classification.

MATERIALS AND METHODS

205 consecutive patients with thymoma confirmed by surgical pathology were recruited from three medical centers. Venous phase enhanced CT images were used to delineate the tumor, and radiomics, 2D and 3D deep learning models based on the whole tumor were established and feature extraction was performed. The tumors were divided into different sub-regions by K-means clustering method and the corresponding features were obtained. The clinical-conventional imaging data of the patients were collected and evaluated, and the univariate and multivariate analysis were used for screening. The above types of features were fused with each other to construct a variety of combined models. Quantitative indicators such as area under the receiver operating characteristic (ROC) curve (AUC) were calculated to evaluate the performance of the model.

RESULTS

The AUC of RDLCSM developed based on LightGBM classifier was 0.953 in the training cohort, 0.930 in the internal validation cohort, 0.924 and 0.903 in the two external validation cohorts, respectively. RDLCSM performs better than RDLM (AUC range: 0.831-0.890) and 2DLCSM (AUC range: 0.785-0.916) based on KNN. In addition, RDLCSM had the highest accuracy (0.818-0.882) and specificity (0.926-1.000).

INTERPRETATION

The RDLCSM, which combines whole-tumor radiomics, 2D and 3D deep learning, clinical-visual radiology, and subregional omics, can be used as a non-invasive tool to predict thymoma risk classification.