Computational Radiomics System to Decode the Radiographic Phenotype

Urethral tissue characterization using multiparametric ultrasound imaging

A decrease in urethral closure pressure is one of the primary causes of stress urinary incontinence in women. Atrophy of the urethral muscles is a primary factor in the 15 % age-related decline in urethral closure pressure per decade. Incontinence not only affects the well-being of women but is also a leading cause of nursing home admission. The objective of this research was to develop a noninvasive test to assess urethral tissue microenvironmental changes using multiparametric ultrasound (mpUS) imaging technique. Transperineal B-scan ultrasound (US) data were captured using clinical scanners equipped with curvilinear or linear transducers. Imaging was performed on volunteers from our institution medical center (n = 15, 22 to 76 y.o.) during Valsalva maneuvers. After expert delineation of the region of interest in each frame, the central axis of the urethra was automatically defined to determine the angle between the urethra and the US beam for further analysis. By integrating angle-dependent backscatter with radiomic texture feature analysis, a mpUS technique was developed to identify biomarkers that reflect subtle microstructural changes expected within the urethral tissue. The process was repeated when the urethra and US beam were at a fixed angle. Texture selection was conducted for both angle-dependent and angle-independent results to remove redundancies. Ultimately, a distinct biomarker was derived using a random forest regression model to compute the urethra score based on features selected from both processes. Angle-dependent backscatter analysis shows that the calculated slope of US mean image intensity decreased by 0.89 (±0.31) % annually, consistent with the expected atrophic disorganization of urethral tissue structure and the associated reduction in urethral closure pressure with age. Additionally, textural analysis performed at a specific angle (i.e., 40 degrees) revealed changes in gray level nonuniformity, skewness, and correlation by 0.08 (±0.04) %, -2.16 (±1.14) %, and -0.32 (±0.35) % per year, respectively. The urethra score was ultimately determined by combining data selected from both angle-dependent and angle-independent analysis strategies using a random forest regression model with age, yielding an R2 value of 0.96 and a p-value less than 0.001. The proposed mpUS tissue characterization technique not only holds promise for guiding future urethral tissue characterization studies without the need for tissue biopsies or invasive functional testing but also aims to minimize observer-induced variability. By leveraging mpUS imaging strategies that account for angle dependence, it provides more accurate assessments. Notably, the urethra score, calculated from US images that reflect tissue microstructural changes, serves as a potential biomarker providing clinicians with deeper insight into urethral tissue function and may aid in diagnosing and managing related conditions while helping to determine the causes of incontinence.

A retrospective study differentiating nontuberculous mycobacterial pulmonary disease from pulmonary tuberculosis on computed tomography using radiomics and machine learning algorithms

OBJECTIVE

To evaluate the effectiveness of a machine learning based on computed tomography (CT) radiomics to distinguish nontuberculous mycobacterial pulmonary disease (NTM-PD) from pulmonary tuberculosis (PTB).

METHODS

In this retrospective analysis, medical records of 99 individuals afflicted with NTM-PD and 285 individuals with PTB in Zhejiang Chinese and Western Medicine Integrated Hospital were examined. Random numbers generated by a computer were utilized to stratify the study cohort, with 80% designated as the training cohort and 20% as the validation cohort. A total of 2153 radiomics features were extracted using Python (Pyradiomics package) to analyse the CT characteristics of the large disease areas. The identification of significant factors was conducted through the least absolute shrinkage and selection operator (LASSO) regression. The following four supervised learning classifier models were developed: random forest (RF), support vector machine (SVM), logistic regression (LR), and extreme gradient boosting (XGBoost). For assessment and comparison of the predictive performance among these models, receiver-operating characteristic (ROC) curves and the areas under the ROC curves (AUCs) were employed.

RESULTS

The Student's t-test, Levene test, and LASSO algorithm collectively selected 23 optimal features. ROC analysis was then conducted, with the respective AUC values of the XGBoost, LR, SVM, and RF models recorded to be 1, 0.9044, 0.8868, and 0.7982 in the training cohort. In the validation cohort, the respective AUC values of the XGBoost, LR, SVM, and RF models were 0.8358, 0.8085, 0.87739, and 0.7759. The DeLong test results noted the lack of remarkable variation across the models.

CONCLUSION

The CT radiomics features can help distinguish between NTM-PD and PTB. Among the four classifiers, SVM showed a stable performance in effectively identifying these two diseases.

Established the prediction model of early-stage non-small cell lung cancer spread through air spaces (STAS) by radiomics and genomics features

BACKGROUND

This study was aimed to establish a prediction model for spread through air spaces (STAS) in early-stage non-small cell lung cancer based on imaging and genomic features.

METHODS

We retrospectively collected 204 patients (47 STAS+ and 157 STAS-) with non-small cell lung cancer who underwent surgical treatment in the Jinling Hospital from January 2021 to December 2021. Their preoperative CT images, genetic testing data (including next-generation sequencing data from other hospitals), and clinical data were collected. Patients were randomly divided into training and testing cohorts (7:3).

RESULTS

The study included a total of 204 eligible patients. STAS were found in 47 (23.0%) patients, and no STAS were found in 157 (77.0%) patients. The receiver operating characteristic curve showed that radiomics model, clinical genomics model, and mixed model had good predictive performance (area under the curve [AUC] = 0.85; AUC = 0.70; AUC = 0.85).

CONCLUSIONS

The prediction model based on radiomics and genomics features has a good prediction performance for STAS.

Textural heterogeneity of liver lesions in CT imaging - comparison of colorectal and pancreatic metastases

PURPOSE

Tumoral heterogeneity poses a challenge for personalized cancer treatments. Especially in metastasized cancer, it remains a major limitation for successful targeted therapy, often leading to drug resistance due to tumoral escape mechanisms. This work explores a non-invasive radiomics-based approach to capture textural heterogeneity in liver lesions and compare it between colorectal cancer (CRC) and pancreatic cancer (PDAC).

MATERIALS AND METHODS

In this retrospective single-center study 73 subjects (42 CRC, 31 PDAC) with 1291 liver metastases (430 CRC, 861 PDAC) were segmented fully automated on contrast-enhanced CT images by a UNet for medical images. Radiomics features were extracted using the Python package Pyradiomics. The mean coefficient of variation (CV) was calculated patient-wise for each feature to quantify the heterogeneity. An unpaired t-test identified features with significant differences in feature variability between CRC and PDAC metastases.

RESULTS

In both colorectal and pancreatic liver metastases, interlesional heterogeneity in imaging can be observed using quantitative imaging features. 75 second-order features were extracted to compare the varying textural characteristics. In total, 18 radiomics features showed a significant difference (p < 0.05) in their expression between the two malignancies. Out of these, 16 features showed higher levels of variability within the cohort of pancreatic metastases, which, as illustrated in a radar plot, suggests greater textural heterogeneity for this entity.

CONCLUSIONS

Radiomics has the potential to identify the interlesional heterogeneity of CT texture among individual liver metastases. In this proof-of-concept study for the quantification and comparison of imaging-related heterogeneity in liver metastases a variation in the extent of heterogeneity levels in CRC and PDAC liver metastases was shown.

MRI-based machine learning radiomics for prediction of HER2 expression status in breast invasive ductal carcinoma

Background

Human epidermal growth factor receptor 2 (HER2) is a tumor biomarker with significant prognostic and therapeutic implications for invasive ductal breast carcinoma (IDC).

Objective

This study aimed to explore the effectiveness of a multisequence magnetic resonance imaging (MRI)-based machine learning radiomics model in classifying the expression status of HER2, including HER2-positive, HER2-low, and HER2 completely negative (HER2-zero), among patients with IDC.

Methods

A total of 402 female patients with IDC confirmed through surgical pathology were enrolled and subsequently divided into a training group (n = 250, center I) and a validation group (n = 152, center II). Radiomics features were extracted from the preoperative MRI. A simulated annealing algorithm was used for key feature selection. Two classification tasks were performed: task 1, the classification of HER2-positive vs. HER2-negative (HER2-low and HER2-zero), and task 2, the classification of HER2-low vs. HER2-zero. Logistic regression, random forest (RF), and support vector machine were conducted to establish radiomics models. The performance of the models was evaluated using the area under the curve (AUC) of the operating characteristics (ROC).

Results

In total, 4506 radiomics features were extracted from multisequence MRI. A radiomics model for prediction of expression state of HER2 was successfully developed. Among the three classification algorithms, RF achieved the highest performance in classifying HER2-positive from HER2-negative and HER2-low from HER2-zero, with AUC values of 0.777 and 0.731, respectively.

Conclusions

Machine learning-based MRI radiomics may aid in the non-invasive prediction of the different expression status of HER2 in IDC.

Utilizing multiclassifier radiomics analysis of ultrasound to predict high axillary lymph node tumour burden in node-positive breast cancer patients: a multicentre study

BACKGROUND

The tumor burden within the axillary lymph nodes (ALNs) constitutes a pivotal factor in breast cancer, serving as the primary determinant for treatment decisions and exhibiting a close correlation with prognosis.

OBJECTIVE

This study aimed to investigate the potential of ultrasound-based radiomics and clinical characteristics in non-invasively distinguishing between low tumor burden (1-2 positive nodes) and high tumor burden (more than 2 positive nodes) in patients with node-positive breast cancer.

METHODS

A total of 215 patients with node-positive breast cancer, who underwent preoperative ultrasound examinations, were enrolled in this study. Among these patients, 144 cases were allocated to the training set, 37 cases to the validation set, and 34 cases to the testing set. Postoperative histopathology was used to determine the status of ALN tumor burden. The region of interest for breast cancer was delineated on the ultrasound image. Nine models were developed to predict high ALN tumor burden, employing a combination of three feature screening methods and three machine learning classifiers. Ultimately, the optimal model was selected and tested on both the validation and testing sets. In addition, clinical characteristics were screened to develop a clinical model. Furthermore, Shapley additive explanations (SHAP) values were utilized to provide explanations for the machine learning model.

RESULTS

During the validation and testing sets, the models demonstrated area under the curve (AUC) values ranging from 0.577 to 0.733 and 0.583 to 0.719, and accuracies ranging from 64.9% to 75.7% and 64.7% to 70.6%, respectively. Ultimately, the Boruta_XGB model, comprising five radiomics features, was selected as the final model. The AUC values of this model for distinguishing low from high tumor burden were 0.828, 0.715, and 0.719 in the training, validation, and testing sets, respectively, demonstrating its superiority over the clinical model.

CONCLUSIONS

The developed radiomics models exhibited a significant level of predictive performance. The Boruta_XGB radiomics model outperformed other radiomics models in this study.

Breast MRI radiomics and machine learning-based predictions of response to neoadjuvant chemotherapy - How are they affected by variations in tumor delineation?

Manual delineation of volumes of interest (VOIs) by experts is considered the gold-standard method in radiomics analysis. However, it suffers from inter- and intra-operator variability. A quantitative assessment of the impact of variations in these delineations on the performance of the radiomics predictors is required to develop robust radiomics based prediction models. In this study, we developed radiomics models for the prediction of pathological complete response to neoadjuvant chemotherapy in patients with two different breast cancer subtypes based on contrast-enhanced magnetic resonance imaging acquired prior to treatment (baseline MRI scans). Different mathematical operations such as erosion, smoothing, dilation, randomization, and ellipse fitting were applied to the original VOIs delineated by experts to simulate variations of segmentation masks. The effects of such VOI modifications on various steps of the radiomics workflow, including feature extraction, feature selection, and prediction performance, were evaluated. Using manual tumor VOIs and radiomics features extracted from baseline MRI scans, an AUC of up to 0.96 and 0.89 was achieved for human epidermal growth receptor 2 positive and triple-negative breast cancer, respectively. For smoothing and erosion, VOIs yielded the highest number of robust features and the best prediction performance, while ellipse fitting and dilation lead to the lowest robustness and prediction performance for both breast cancer subtypes. At most 28% of the selected features were similar to manual VOIs when different VOI delineation data were used. Differences in VOI delineation affect different steps of radiomics analysis, and their quantification is therefore important for development of standardized radiomics research.

Using machine learning to predict carotid artery symptoms from CT angiography: A radiomics and deep learning approach

Purpose

To assess radiomics and deep learning (DL) methods in identifying symptomatic Carotid Artery Disease (CAD) from carotid CT angiography (CTA) images. We further compare the performance of these novel methods to the conventional calcium score.

Methods

Carotid CT angiography (CTA) images from symptomatic patients (ischaemic stroke/transient ischaemic attack within the last 3 months) and asymptomatic patients were analysed. Carotid arteries were classified into culprit, non-culprit and asymptomatic. The calcium score was assessed using the Agatston method. 93 radiomic features were extracted from regions-of-interest drawn on 14 consecutive CTA slices. For DL, convolutional neural networks (CNNs) with and without transfer learning were trained directly on CTA slices. Predictive performance was assessed over 5-fold cross validated AUC scores. SHAP and GRAD-CAM algorithms were used for explainability.

Results

132 carotid arteries were analysed (41 culprit, 41 non-culprit, and 50 asymptomatic). For asymptomatic vs symptomatic arteries, radiomics attained a mean AUC of 0.96(± 0.02), followed by DL 0.86(± 0.06) and then calcium 0.79(± 0.08). For culprit vs non-culprit arteries, radiomics achieved a mean AUC of 0.75(± 0.09), followed by DL 0.67(± 0.10) and then calcium 0.60(± 0.02). For multi-class classification, the mean AUCs were 0.95(± 0.07), 0.79(± 0.05), and 0.71(± 0.07) for radiomics, DL and calcium, respectively. Explainability revealed consistent patterns in the most important radiomic features.

Conclusions

Our study highlights the potential of novel image analysis techniques in extracting quantitative information beyond calcification in the identification of CAD. Though further work is required, the transition of these novel techniques into clinical practice may eventually facilitate better stroke risk stratification.

Prediction of preoperative lymph-vascular space invasion and survival outcomes of cervical squamous cell carcinoma by utilizing 18 F-FDG PET/CT imaging at early stage

OBJECTIVE

To establish nomograms for predicting preoperative lymph-vascular space invasion (LVSI) and survival outcomes of cervical squamous cell carcinoma (CSCC) based on PET/CT radiomics.

METHODS

One hundred and twenty-three patients with CSCC and LVSI status were enrolled retrospectively. Independent predictors of LVSI were identified through clinicopathological factors and PET/CT metabolic parameters. We extracted 1316 features from PET and CT volume of interest, respectively. Additionally, four models (PET-RS: radiomic signature of PET only; CT-RS: radiomic signature of CT only; PET/CT-RS + clinical data; PET/CT-RS: radiomic signature of PET and CT) were established to predict LVSI status. Calculation of radiomics scores of PET/CT was executed for assessment of the survival outcomes, followed by development of nomograms with radiomics (NR) or without radiomics (NWR).

RESULTS

One hundred and twenty-three patients with pathologically confirmed CSCC had been categorized into two sets (training and testing sets). It was found that only maximum standardized uptake value (SUV max ) and squamous cell carcinoma antigen were independent predictors of LVSI. Meanwhile, the PET/CT-RS + clinical data outperformed the other three models in the training set [area under the curve (AUC): 0.91 vs. 0.861 vs. 0.81 vs. 0.814] and the testing set (AUC: 0.885 vs. 0.857 vs. 0.783 vs. 0.798). Additionally, SUV max and LVSI had been demonstrated to be independent prognostic indicators for progression-free survival and overall survival. Decision curve analysis and calibration curve indicated that NRs were superior to NWRs. The survival outcomes were assessed.

CONCLUSION

PET/CT-based radiomic signature nomogram enables a new method for preoperative prediction of LVSI and survival prognosis for patients with CSCC.

Multi-task Bayesian model combining FDG-PET/CT imaging and clinical data for interpretable high-grade prostate cancer prognosis

We propose a fully automatic multi-task Bayesian model, named Bayesian Sequential Network (BSN), for predicting high-grade (Gleason  ≥  8) prostate cancer (PCa) prognosis using pre-prostatectomy FDG-PET/CT images and clinical data. BSN performs one classification task and five survival tasks: predicting lymph node invasion (LNI), biochemical recurrence-free survival (BCR-FS), metastasis-free survival, definitive androgen deprivation therapy-free survival, castration-resistant PCa-free survival, and PCa-specific survival (PCSS). Experiments are conducted using a dataset of 295 patients. BSN outperforms widely used nomograms on all tasks except PCSS, leveraging multi-task learning and imaging data. BSN also provides automated prostate segmentation, uncertainty quantification, personalized feature-based explanations, and introduces dynamic predictions, a novel approach that relies on short-term outcomes to refine long-term prognosis. Overall, BSN shows great promise in its ability to exploit imaging and clinicopathological data to predict poor outcome patients that need treatment intensification with loco-regional or systemic adjuvant therapy for high-risk PCa.

Completely non-invasive prediction of IDH mutation status based on preoperative native CT images

The isocitrate dehydrogenase (IDH) mutation status is one of the most important markers according to the 2021 WHO classification of CNS tumors. Preoperatively, this information is usually obtained based on invasive biopsies, contrast-enhanced MR images or PET images generated using radioactive tracers. However, the completely non-invasive determination of IDH mutation status using routinely acquired preoperative native CT images has hardly been investigated to date. In our study, we show that radiomics-based machine learning allows to determine IDH mutation status based on preoperative native CT images both with very high accuracy and completely non-invasively. Based on independent test data, we are able to correctly identify 91.1% of cases with an IDH mutation. Our final model, containing only six features, exhibits a high area under the curve of 0.847 and an excellent area under the precision-recall curve of 0.945. In the future, such models may be used for a completely non-invasive prediction of important genetic markers, potentially allowing treating physicians to reduce the number of biopsies and speed up further treatment planning.

Optimizing radiomics for prostate cancer diagnosis: feature selection strategies, machine learning classifiers, and MRI sequences

OBJECTIVES

Radiomics-based analyses encompass multiple steps, leading to ambiguity regarding the optimal approaches for enhancing model performance. This study compares the effect of several feature selection methods, machine learning (ML) classifiers, and sources of radiomic features, on models' performance for the diagnosis of clinically significant prostate cancer (csPCa) from bi-parametric MRI.

METHODS

Two multi-centric datasets, with 465 and 204 patients each, were used to extract 1246 radiomic features per patient and MRI sequence. Ten feature selection methods, such as Boruta, mRMRe, ReliefF, recursive feature elimination (RFE), random forest (RF) variable importance, L1-lasso, etc., four ML classifiers, namely SVM, RF, LASSO, and boosted generalized linear model (GLM), and three sets of radiomics features, derived from T2w images, ADC maps, and their combination, were used to develop predictive models of csPCa. Their performance was evaluated in a nested cross-validation and externally, using seven performance metrics.

RESULTS

In total, 480 models were developed. In nested cross-validation, the best model combined Boruta with Boosted GLM (AUC = 0.71, F1 = 0.76). In external validation, the best model combined L1-lasso with boosted GLM (AUC = 0.71, F1 = 0.47). Overall, Boruta, RFE, L1-lasso, and RF variable importance were the top-performing feature selection methods, while the choice of ML classifier didn't significantly affect the results. The ADC-derived features showed the highest discriminatory power with T2w-derived features being less informative, while their combination did not lead to improved performance.

CONCLUSION

The choice of feature selection method and the source of radiomic features have a profound effect on the models' performance for csPCa diagnosis.

CRITICAL RELEVANCE STATEMENT

This work may guide future radiomic research, paving the way for the development of more effective and reliable radiomic models; not only for advancing prostate cancer diagnostic strategies, but also for informing broader applications of radiomics in different medical contexts.

KEY POINTS

Radiomics is a growing field that can still be optimized. Feature selection method impacts radiomics models' performance more than ML algorithms. Best feature selection methods: RFE, LASSO, RF, and Boruta. ADC-derived radiomic features yield more robust models compared to T2w-derived radiomic features.

Morphological similarity and white matter structural mapping of new daily persistent headache: a structural connectivity and tract-specific study

BACKGROUND

New daily persistent headache (NDPH) is a rare primary headache disorder characterized by daily and persistent sudden onset headaches. Specific abnormalities in gray matter and white matter structure are associated with pain, but have not been well studied in NDPH. The objective of this work is to explore the fiber tracts and structural connectivity, which can help reveal unique gray and white matter structural abnormalities in NDPH.

METHODS

The regional radiomics similarity networks were calculated from T1 weighted (T1w) MRI to depict the gray matter structure. The fiber connectivity matrices weighted by diffusion metrics like fractional anisotropy (FA), mean diffusivity (MD) and radial diffusivity (RD) were built, meanwhile the fiber tracts were segmented by anatomically-guided superficial fiber segmentation (Anat-SFSeg) method to explore the white matter structure from diffusion MRI. The considerable different neuroimaging features between NDPH and healthy controls (HC) were extracted from the connectivity and tract-based analyses. Finally, decision tree regression was used to predict the clinical scores (i.e. pain intensity) from the above neuroimaging features.

RESULTS

T1w and diffusion MRI data were available in 51 participants after quality control: 22 patients with NDPH and 29 HCs. Significantly decreased morphological similarity was found between the right superior frontal gyrus and right hippocampus. The superficial white matter (SWM) showed significantly decreased FA in fiber tracts including the right superficial-frontal, left superficial-occipital, bilateral superficial-occipital-temporal (Sup-OT) and right superficial-temporal, meanwhile significant increased RD was found in the left Sup-OT. For the fiber connectivity, NDPH showed significantly decreased FA in the bilateral basal ganglion and temporal lobe, increased MD in the right frontal lobe, and increased RD in the right frontal lobe and left temporal-occipital lobe. Clinical scores could be predicted dominantly by the above significantly different neuroimaging features through decision tree regression.

CONCLUSIONS

Our research indicates the structural abnormalities of SWM and the neural pathways projected between regions like right hippocampus and left caudate nucleus, along with morphological similarity changes between the right superior frontal gyrus and right hippocampus, constitute the pathological features of NDPH. The decision tree regression demonstrates correlations between these structural changes and clinical scores.

Both intra- and peri-tumoral radiomics signatures can be used to predict lymphatic vascular space invasion and lymphatic metastasis positive status from endometrial cancer MR imaging

OBJECTIVES

To identify lymphatic vascular space invasion (LVSI) and lymphatic node metastasis (LNM) status of endometrial cancer (EC) patients, using radiomics based on MRI images.

METHODS

Five hundred and ninety-eight EC patients between January 2015 and September 2020 from two institutions were retrospectively included. Tumoral regions on DWI, T1CE, and T2W images were manually outlined. Radiomics features were extracted from tumor region and peri-tumor region of different thicknesses. We established sub-models to select features from each smaller category. Using this method, we separately constructed radiomic signatures for intra-tumoral and peri-tumoral images using different sequences. We constructed intra-tumoral and peri-tumoral models by combining their features, and a multi-sequence model by combining logits. Models were trained with 397 patients and validated with 170 internal and 31 external patients.

RESULTS

For LVSI positive/LNM positive status identification, the multi-parameter MRI radiomics model achieved the area under curve (AUC) values of 0.771 (95%CI: [0.692-0.849])/0.801 (95%CI: [0.704, 0.898]) and 0.864 (95%CI: [0.728-1.000])/0.976 (95%CI: [0.919, 1.000]) in internal and external test cohorts, respectively.

CONCLUSIONS

Intra-tumoral and peri-tumoral radiomics signatures based on mpMRI can both be used to identify LVSI or LNM status in EC patients non-invasively. Further studies on LVSI and LNM should pay attention to both of them.

Sexual Dimorphism of Radiomic Features in the Brain: An Exploratory Study Using 700 μm MP2RAGE MRI at 7 T

OBJECTIVES

The aim of this study was to determine whether MRI radiomic features of key cerebral structures differ between women and men, and whether detection of such differences depends on the image resolution.

MATERIALS AND METHODS

Ultrahigh resolution (UHR) 3D MP2RAGE (magnetization-prepared 2 rapid acquisition gradient echo) T1-weighted MR images (voxel size, 0.7 × 0.7 × 0.7 mm 3 ) of the brain of 30 subjects (18 women and 12 men; mean age, 39.0 ± 14.8 years) without abnormal findings on MRI were retrospectively included. MRI was performed on a whole-body 7 T MR system. A convolutional neural network was used to segment the following structures: frontal cortex, frontal white matter, thalamus, putamen, globus pallidus, caudate nucleus, and corpus callosum. Eighty-seven radiomic features were extracted respectively: gray-level histogram (n = 18), co-occurrence matrix (n = 24), run-length matrix (n = 16), size-zone matrix (n = 16), and dependence matrix (n = 13). Feature extraction was performed at UHR and, additionally, also after resampling to 1.4 × 1.4 × 1.4 mm 3 voxel size (standard clinical resolution). Principal components (PCs) of radiomic features were calculated, and independent samples t tests with Cohen d as effect size measure were used to assess differences in PCs between women and men for the different cerebral structures.

RESULTS

At UHR, at least a single PC differed significantly between women and men in 6/7 cerebral structures: frontal cortex ( d = -0.79, P = 0.042 and d = -1.01, P = 0.010), frontal white matter ( d = -0.81, P = 0.039), thalamus ( d = 1.43, P < 0.001), globus pallidus ( d = 0.92, P = 0.020), caudate nucleus ( d = -0.83, P = 0.039), and corpus callosum ( d = -0.97, P = 0.039). At standard clinical resolution, only a single PC extracted from the corpus callosum differed between sexes ( d = 1.05, P = 0.009).

CONCLUSIONS

Nonnegligible differences in radiomic features of several key structures of the brain exist between women and men, and need to be accounted for. Very high spatial resolution may be required to uncover and further investigate the sexual dimorphism of brain structures on MRI.

Test-retest repeatability of myocardial radiomic features from quantitative cardiac magnetic resonance T1 and T2 mapping

Radiomics of cardiac magnetic resonance (MR) imaging has proved to be potentially useful in the study of various myocardial diseases. Therefore, assessing the repeatability degree in radiomic features measurement is of fundamental importance. The aim of this study was to assess test-retest repeatability of myocardial radiomic features extracted from quantitative T1 and T2 maps. A representative group of 24 subjects (mean age 54 ± 18 years) referred for clinical cardiac MR imaging were enrolled in the study. For each subject, T1 and T2 mapping through MOLLI and T2-prepared TrueFISP acquisition sequences, respectively, were performed at 1.5 T. Then, 98 radiomic features of different classes (shape, first-order, second-order) were extracted from a region of interest encompassing the whole left ventricle myocardium in a short axis slice. The repeatability was assessed performing different and complementary analyses: intraclass correlation coefficient (ICC) and limits of agreement (LOA) (i.e., the interval within which 95% of the percentage differences between two repeated measures are expected to lie). Radiomic features were characterized by a relatively wide range of repeatability degree in terms of both ICC and LOA. Overall, 44.9% and 38.8% of radiomic features showed ICC values > 0.75 for T1 and T2 maps, respectively, while 25.5% and 23.4% of radiomic features showed LOA between ±10%. A subset of radiomic features for T1 (Mean, Median, 10Percentile, 90Percentile, RootMeanSquared, Imc2, RunLengthNonUniformityNormalized, RunPercentage and ShortRunEmphasis) and T2 (MaximumDiameter, RunLengthNonUniformityNormalized, RunPercentage, ShortRunEmphasis) maps presented both ICC > 0.75 and LOA between ±5%. Overall, radiomic features extracted from T1 maps showed better repeatability performance than those extracted from T2 maps, with shape features characterized by better repeatability than first-order and textural features. Moreover, only a limited subset of 9 and 4 radiomic features for T1 and T2 maps, respectively, showed high repeatability degree in terms of both ICC and LOA. These results confirm the importance of assessing test-retest repeatability degree in radiomic feature estimation and might be useful for a more effective/reliable use of myocardial T1 and T2 mapping radiomics in clinical or research studies.

Cell-free DNA assay for malignancy classification of high-risk lung nodules

OBJECTIVE

Although low-dose computed tomography has been proven effective to reduce lung cancer-specific mortality, a considerable proportion of surgically resected high-risk lung nodules were still confirmed pathologically benign. There is an unmet need of a novel method for malignancy classification in lung nodules.

METHODS

We recruited 307 patients with high-risk lung nodules who underwent curative surgery, and 247 and 60 cases were pathologically confirmed malignant and benign lung lesions, respectively. Plasma samples from each patient were collected before surgery and performed low-depth (5×) whole-genome sequencing. We extracted cell-free DNA characteristics and determined radiomic features. We built models to classify the malignancy using our data and further validated models with 2 independent lung nodule cohorts.

RESULTS

Our models using one type of profile were able to distinguish lung cancer and benign lung nodules at an area under the curve metrics of 0.69 to 0.91 in the study cohort. Integrating all the 5 base models using cell-free DNA profiles, the cell-free DNA-based ensemble model achieved an area under the curve of 0.95 (95% CI, 0.92-0.97) in the study cohort and 0.98 (95% CI, 0.96-1.00) in the validation cohort. At a specificity of 95.0%, the sensitivity reached 80.0% in the study cohort. With the same threshold, the specificity and sensitivity had similar performances in both validation cohorts. Furthermore, the performance of area under the curve reached 0.97 in both the study and validation cohorts when considering the radiomic profile.

CONCLUSIONS

The cell-free DNA profiles-based method is an efficient noninvasive tool to distinguish malignancies and high-risk but pathologically benign lung nodules.

An Integrated Nomogram Combining Deep Learning and Radiomics for Predicting Malignancy of Pulmonary Nodules Using CT-Derived Nodules and Adipose Tissue: A Multicenter Study

BACKGROUND

Correctly distinguishing between benign and malignant pulmonary nodules can avoid unnecessary invasive procedures. This study aimed to construct a deep learning radiomics clinical nomogram (DLRCN) for predicting malignancy of pulmonary nodules.

METHODS

One thousand and ninety-eight patients with 6-30 mm pulmonary nodules who received histopathologic diagnosis at 3 centers were included and divided into a primary cohort (PC), an internal test cohort (I-T), and two external test cohorts (E-T1, E-T2). The DLRCN was built by integrating adipose tissue radiomics features, intranodular and perinodular deep learning features, and clinical characteristics for diagnosing malignancy of pulmonary nodules. The least absolute shrinkage and selection operator (LASSO) was used for feature selection. The performance of DLRCN was assessed with respect to its calibration curve, area under the curve (AUC), and decision curve analysis (DCA). Furthermore, we compared it with three radiologists. The net reclassification improvement (NRI), integrated discrimination improvement (IDI), and subgroup analysis were also taken into account.

RESULTS

The incorporation of adipose tissue radiomics features led to significant NRI and IDI (NRI = 1.028, p < 0.05, IDI = 0.137, p < 0.05). In the I-T, E-T1, and E-T2, the AUCs of DLRCN were 0.946 (95% CI: 0.936, 0.955), 0.948 (95% CI: 0.933, 0.963) and 0.962 (95% CI: 0.945, 0.979), The calibration curve revealed good predictive accuracy between the actual probability and predicted probability (p > 0.05). DCA showed that the DLRCN was clinically useful. Under equal specificity, the sensitivity of DLRCN increased by 8.6% compared to radiologist assessments. The subgroup analysis conducted on adipose tissue radiomics features further demonstrated their supplementary value in determining the malignancy of pulmonary nodules.

CONCLUSION

The DLRCN demonstrated good performance in predicting the malignancy of pulmonary nodules, which was comparable to radiologist assessments. The adipose tissue radiomics features have notably enhanced the performance of DLRCN.

Radiomic study of antenatal prediction of severe placenta accreta spectrum from MRI

OBJECTIVES

We previously demonstrated the potential of radiomics for the prediction of severe histological placenta accreta spectrum (PAS) subtypes using T2-weighted MRI. We aim to validate our model using an additional dataset. Secondly, we explore whether the performance is improved using a new approach to develop a new multivariate radiomics model.

METHODS

Multi-centre retrospective analysis was conducted between 2018 and 2023. Inclusion criteria: MRI performed for suspicion of PAS from ultrasound, clinical findings of PAS at laparotomy and/or histopathological confirmation. Radiomic features were extracted from T2-weighted MRI. The previous multivariate model was validated. Secondly, a 5-radiomic feature random forest classifier was selected from a randomized feature selection scheme to predict invasive placenta increta PAS cases. Prediction performance was assessed based on several metrics including area under the curve (AUC) of the receiver operating characteristic curve (ROC), sensitivity, and specificity.

RESULTS

We present 100 women [mean age 34.6 (±3.9) with PAS], 64 of whom had placenta increta. Firstly, we validated the previous multivariate model and found that a support vector machine classifier had a sensitivity of 0.620 (95% CI: 0.068; 1.0), specificity of 0.619 (95% CI: 0.059; 1.0), an AUC of 0.671 (95% CI: 0.440; 0.922), and accuracy of 0.602 (95% CI: 0.353; 0.817) for predicting placenta increta. From the new multivariate model, the best 5-feature subset was selected via the random subset feature selection scheme comprised of 4 radiomic features and 1 clinical variable (number of previous caesareans). This clinical-radiomic model achieved an AUC of 0.713 (95% CI: 0.551; 0.854), accuracy of 0.695 (95% CI 0.563; 0.793), sensitivity of 0.843 (95% CI 0.682; 0.990), and specificity of 0.447 (95% CI 0.167; 0.667).

CONCLUSION

We validated our previous model and present a new multivariate radiomic model for the prediction of severe placenta increta from a well-defined, cohort of PAS cases.

ADVANCES IN KNOWLEDGE

Radiomic features demonstrate good predictive potential for identifying placenta increta. This suggests radiomics may be a useful adjunct to clinicians caring for women with this high-risk pregnancy condition.

Development and validation of survival prediction models for patients with hepatocellular carcinoma treated with transcatheter arterial chemoembolization plus tyrosine kinase inhibitors

BACKGROUND

Due to heterogeneity of molecular biology and microenvironment, therapeutic efficacy varies among hepatocellular carcinoma (HCC) patients treated with transcatheter arterial chemoembolization (TACE) and tyrosine kinase inhibitors (TKIs). We examined combined models using clinicoradiological characteristics, mutational burden of signaling pathways, and radiomics features to predict survival prognosis.

METHODS

Two cohorts comprising 111 patients with HCC were used to build prognostic models. The training and test cohorts included 78 and 33 individuals, respectively. Mutational burden was calculated based on 17 cancer-associated signaling pathways. Radiomic features were extracted and selected from computed tomography images using a pyradiomics system. Models based on clinicoradiological indicators, mutational burden, and radiomics score (rad-score) were built to predict overall survival (OS) and progression-free survival (PFS).

RESULTS

Eastern Cooperative Oncology Group performance status, Child-Pugh class, peritumoral enhancement, PI3K_AKT and hypoxia mutational burden, and rad-score were used to create a combined model predicting OS. C-indices were 0.805 (training cohort) and 0.768 (test cohort). The areas under the curve (AUCs) were 0.889, 0.900, and 0.917 for 1-year, 2-year, and 3-year OS, respectively. To predict PFS, alpha-fetoprotein level, tumor enhancement pattern, hypoxia and receptor tyrosine kinase mutational burden, and rad-score were used. C-indices were 0.782 (training cohort) and 0.766 (test cohort). AUCs were 0.885 and 0.925 for 6-month and 12-month PFS, respectively. Calibration and decision curve analyses supported the model's accuracy and clinical potential.

CONCLUSIONS

The nomogram models are hopeful to predict OS and PFS in patients with intermediate-advanced HCC treated with TACE plus TKIs, offering a promising tool for treatment decisions and monitoring patient progress.

Magnetic resonance imaging (MRI)-based intratumoral and peritumoral radiomics for prognosis prediction in glioma patients

AIM

The purpose of this study was to identify robust radiological features from intratumoral and peritumoral regions, evaluate MRI protocols, and machine learning methods for overall survival stratification of glioma patients, and explore the relationship between radiological features and the tumour microenvironment.

MATERIAL AND METHODS

A retrospective analysis was conducted on 163 glioma patients, divided into a training set (n=113) and a testing set (n=50). For each patient, 2135 features were extracted from clinical MRI. Feature selection was performed using the Minimum Redundancy Maximum Relevance method and the Random Forest (RF) algorithm. Prognostic factors were assessed using the Cox proportional hazards model. Four machine learning models (RF, Logistic Regression, Support Vector Machine, and XGBoost) were trained on clinical and radiological features from tumour and peritumoral regions. Model evaluations on the testing set used receiver operating characteristic curves.

RESULTS

Among the 163 patients, 96 had an overall survival (OS) of less than three years postsurgery, while 67 had an OS of more than three years. Univariate Cox regression in the validation set indicated that age (p=0.003) and tumour grade (p<0.001) were positively associated with the risk of death within three years postsurgery. The final predictive model incorporated 13 radiological and 7 clinical features. The RF model, combining intratumor and peritumor radiomics, achieved the best predictive performance (AUC = 0.91; ACC = 0.86), outperforming single-region models.

CONCLUSION

Combined intratumoral and peritumoral radiomics can improve survival prediction and have potential as a practical imaging biomarker to guide clinical decision-making.

Deep learning Radiomics Based on Two-Dimensional Ultrasound for Predicting the Efficacy of Neoadjuvant Chemotherapy in Breast Cancer

We investigate the predictive value of a comprehensive model based on preoperative ultrasound radiomics, deep learning, and clinical features for pathological complete response (pCR) after neoadjuvant chemotherapy (NAC) for the breast cancer. We enrolled 155 patients with pathologically confirmed breast cancer who underwent NAC. The patients were randomly divided into the training set and the validation set in the ratio of 7:3. The deep learning and radiomics features of pre-treatment ultrasound images were extracted, and the random forest recursive elimination algorithm and the least absolute shrinkage and selection operator were used for feature screening and DL-Score and Rad-Score construction. According to multifactorial logistic regression, independent clinical predictors, DL-Score, and Rad-Score were selected to construct the comprehensive prediction model DLRC. The performance of the model was evaluated in terms of its predictive effect, and clinical practicability. Compared to the clinical, radiomics (Rad-Score), and deep learning (DL-Score) models, the DLRC accurately predicted the pCR status, with an area under the curve (AUC) of 0.937 (95%CI: 0.895-0.970) in the training set and 0.914 (95%CI: 0.838-0.973) in the validation set. Moreover, decision curve analysis confirmed that the DLRC had the highest clinical value among all models. The comprehensive model DLRC based on ultrasound radiomics, deep learning, and clinical features can effectively and accurately predict the pCR status of breast cancer after NAC, which is conducive to assisting clinical personalized diagnosis and treatment plan.

Xerostomia prediction in patients with nasopharyngeal carcinoma during radiotherapy using segmental dose distribution in dosiomics and radiomics models

OBJECTIVES

This study aimed to integrate radiomics and dosiomics features to develop a predictive model for xerostomia (XM) in nasopharyngeal carcinoma after radiotherapy. It explores the influence of distinct feature extraction methods and dose ranges on the performance.

MATERIALS AND METHODS

Data from 363 patients with nasopharyngeal carcinoma were retrospectively analyzed. We pioneered a dose-segmentation strategy, where the overall dose distribution (OD) was divided into four segmental dose distributions (SDs) at intervals of 15 Gy. Features were extracted using manual definition and deep learning, applying OD or SD and integrating radiomics and dosiomics, yielding corresponding feature scores (manually defined radiomics, MDR; manually defined dosiomics, MDD; deep learning-based radiomics, DLR; deep learning-based dosiomics, DLD). Subsequently, 18 models were developed by combining features and model types (random forest and support vector machine).

RESULTS AND CONCLUSION

Under OD, O(DLR_DLD) demonstrated exceptional performance, with an optimal area under the curve (AUC) of 0.81 and an average AUC of 0.71. Within SD, S(DLR_DLD) surpassed the other models, achieving an optimal AUC of 0.90 and an average AUC of 0.85. Therefore, the integration of dosiomics into radiomics can augment predictive efficacy. The dose-segmentation strategy can facilitate the extraction of more profound information. This indicates that ScoreDLR and ScoreMDR were negatively associated with XM, whereas ScoreDLD, derived from SD exceeding 15 Gy, displayed a positive association with XM. For feature extraction, deep learning was superior to manual definition.

Diffusion Tensor and Kurtosis MRI-Based Radiomics Analysis of Kidney Injury in Type 2 Diabetes

BACKGROUND

Diffusion tensor imaging (DTI) and diffusion kurtosis imaging (DKI) can provide quantitative parameters that show promise for evaluation of diabetic kidney disease (DKD). The combination of radiomics with DTI and DKI may hold potential clinical value in detecting DKD.

PURPOSE

To investigate radiomics models of DKI and DTI for predicting DKD in type 2 diabetes mellitus (T2DM) and evaluate their performance in automated renal parenchyma segmentation.

STUDY TYPE

Prospective.

POPULATION

One hundred and sixty-three T2DM patients (87 DKD; 63 females; 27-80 years), randomly divided into training cohort (N = 114) and validation cohort (N = 49).

FIELD STRENGTH/SEQUENCE

1.5-T, diffusion spectrum imaging (DSI) with 9 different b-values.

ASSESSMENT

The images of DSI were processed to generate DKI and DTI parameter maps, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). The Swin UNETR model was trained with 5-fold cross-validation using 100 samples for renal parenchyma segmentation. Subsequently, radiomics features were automatically extracted from each parameter map. The performance of the radiomics models on the validation cohort was evaluated by utilizing the receiver operating characteristic (ROC) curve.

STATISTICAL TESTS

Mann-Whitney U test, Chi-squared test, Pearson correlation coefficient, least absolute shrinkage and selection operator (LASSO), dice similarity coefficient (DSC), decision curve analysis (DCA), area under the curve (AUC), and DeLong's test. The threshold for statistical significance was set at P < 0.05.

RESULTS

The DKI_MD achieved the best segmentation performance (DSC, 0.925 ± 0.011). A combined radiomics model (DTI_FA, DTI_MD, DKI_FA, DKI_MD, and DKI_RD) showed the best performance (AUC, 0.918; 95% confidence interval [CI]: 0.820-0.991). When the threshold probability was greater than 20%, the combined model provided the greatest net benefit. Among the single parameter maps, the DTI_FA exhibited superior diagnostic performance (AUC, 887; 95% CI: 0.779-0.972).

DATA CONCLUSION

The radiomics signature constructed based on DKI and DTI may be used as an accurate and non-invasive tool to identify T2DM and DKD.

LEVEL OF EVIDENCE

2 TECHNICAL EFFICACY: Stage 2.

Radiomic Prediction of CCND1 Expression Levels and Prognosis in Low-grade Glioma Based on Magnetic Resonance Imaging

OJECTIVES

Low-grade glioma (LGG) is associated with increased mortality owing to recrudescence and the tendency for malignant transformation. Therefore, it is imperative to discover novel prognostic biomarkers as existing traditional prognostic biomarkers of glioma, including clinicopathological features and imaging examinations, are unable to meet the clinical demand for precision medicine. Accordingly, we aimed to evaluate the prognostic value of cyclin D1 (CCND1) expression levels and construct radiomic models to predict these levels in patients with LGG MATERIALS AND METHODS: A total of 412 LGG cases from The Cancer Genome Atlas (TCGA) were used for gene-based prognostic analysis. Using magnetic resonance imaging (MRI) images stored in The Cancer Imaging Archive with genomic data from TCGA, 149 cases were selected for radiomics feature extraction and model construction. After feature extraction, the radiomic signature was constructed using logistic regression (LR) and support vector machine (SVM) analyses.

RESULTS

CCND1 was identified as a prognosis-related gene with differential expression in tumor and normal samples and plays a role in regulating both the cell cycle and immune response. Landmark analysis revealed that high-expression levels of CCND1 were beneficial for survival (P < 0.05) in advanced LGG. Four optimal radiomics features were selected to construct radiomics models. The performance of LR and SVM achieved areas under the curve of 0.703 and 0.705, as well as 0.724 and 0.726 in the training and validation sets, respectively.

CONCLUSION

Elevated levels of CCND1 expression could impact the prognosis of patients with LGG. MRI-based radiomics, especially the AUC values, can serve as a novel tool for predicting CCND1 expression and understanding the correlation between elevated CCND1 expression and prognosis.

AVAILABILITY OF DATA AND MATERIALS

The datasets analyzed during the current study are available in the TCGA, TCIA, UCSC XENA and GTEx repository, https://portal.gdc.cancer.gov/, https://www.cancerimagingarchive.net/, https://xenabrowser.net/datapages/, https://www.gtexportal.org/home/.

Computational Hemodynamics-Based Growth Prediction for Small Abdominal Aortic Aneurysms: Laminar Simulations Versus Large Eddy Simulations

Prior studies have shown that computational fluid dynamics (CFD) simulations help assess patient-specific hemodynamics in abdominal aortic aneurysms (AAAs); patient-specific hemodynamic stressors are frequently used to predict an AAA's growth. Previous studies have utilized both laminar and turbulent simulation models to simulate hemodynamics. However, the impact of different CFD simulation models on the predictive modeling of AAA growth remains unknown and is thus the knowledge gap that motivates this study. Specifically, CFD simulations were performed for 70 AAA models derived from 70 patients' computed tomography angiography (CTA) data with known growth status (i.e., fast-growing [> 5 mm/yr] or slowly growing [< 5 mm/yr]). We used laminar and large eddy simulation (LES) models to obtain hemodynamic parameters to predict AAAs' growth status. Predicting the growth status of AAAs was based on morphological, hemodynamic, and patient health parameters in conjunction with three classical machine learning (ML) classifiers, namely, support vector machine (SVM), K-nearest neighbor (KNN), and generalized linear model (GLM). Our preliminary results estimated aneurysmal flow stability and wall shear stress (WSS) were comparable in both laminar and LES flow simulations. Moreover, computed WSS and velocity-related hemodynamic variables obtained from the laminar and LES simulations showed comparable abilities in differentiating the growth status of AAAs. More importantly, the predictive modeling performance of the three ML classifiers mentioned above was similar, with less than a 2% difference observed (p-value > 0.05). In closing, our findings suggest that two different flow simulations investigated did not significantly affect outcomes of computational hemodynamics and predictive modeling of AAAs' growth status, given the data investigated.

Estimating lumbar bone mineral density from conventional MRI and radiographs with deep learning in spine patients

PURPOSE

This study aimed to develop machine learning methods to estimate bone mineral density and detect osteopenia/osteoporosis from conventional lumbar MRI (T1-weighted and T2-weighted images) and planar radiography in combination with clinical data and imaging parameters of the acquisition protocol.

METHODS

A database of 429 patients subjected to lumbar MRI, radiographs and dual-energy x-ray absorptiometry within 6 months was created from an institutional database. Several machine learning models were trained and tested (373 patients for training, 86 for testing) with the following objectives: (1) direct estimation of the vertebral bone mineral density; (2) classification of T-score lower than - 1 or (3) lower than - 2.5. The models took as inputs either the images or radiomics features derived from them, alone or in combination with metadata (age, sex, body size, vertebral level, parameters of the imaging protocol).

RESULTS

The best-performing models achieved mean absolute errors of 0.15-0.16 g/cm2 for the direct estimation of bone mineral density, and areas under the receiver operating characteristic curve of 0.82 (MRIs) - 0.80 (radiographs) for the classification of T-scores lower than - 1, and 0.80 (MRIs) - 0.65 (radiographs) for T-scores lower than - 2.5.

CONCLUSIONS

The models showed good discriminative performances in detecting cases of low bone mineral density, and more limited capabilities for the direct estimation of its value. Being based on routine imaging and readily available data, such models are promising tools to retrospectively analyse existing datasets as well as for the opportunistic investigation of bone disorders.

Tumor habitat-based MRI features assessing early response in locally advanced nasopharyngeal carcinoma

OBJECTIVE

The early response to concurrent chemoradiotherapy in patients with locally advanced nasopharyngeal carcinoma (LA-NPC) is closely correlated with prognosis. In this study, we aimed to predict early response using a combined model that combines sub-regional radiomics features from multi-sequence MRI with clinically relevant factors.

METHODS

A total of 104 patients with LA-NPC were randomly divided into training and test cohorts at a ratio of 3:1. Radiomic features were extracted from subregions within the tumor area using the K-means clustering method, and feature selection was performed using LASSO regression. Four models were established: a radiomics model, a clinical model, an Intratumor Heterogeneity (ITH) score-based model and a combined model that integrates the ITH score with clinical factors. The predictive performance of these models was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).

RESULTS

Among the models, the combined model incorporating the ITH score and clinical factors exhibited the highest predictive performance in the test cohort (AUC=0.838). Additionally, the models based on ITH score showed superior prognostic value in both the training cohort (AUC=0.888) and the test cohort (AUC=0.833).

CONCLUSION

The combined model that integrates the ITH score with clinical factors exhibited superior performance in predicting early response following concurrent chemoradiotherapy in patients with LA-NPC.

Multicenter investigation of preoperative distinction between primary central nervous system lymphomas and glioblastomas through interpretable artificial intelligence models

OBJECTIVE

Research into the effectiveness and applicability of deep learning, radiomics, and their integrated models based on Magnetic Resonance Imaging (MRI) for preoperative differentiation between Primary Central Nervous System Lymphoma (PCNSL) and Glioblastoma (GBM), along with an exploration of the interpretability of these models.

MATERIALS AND METHODS

A retrospective analysis was performed on MRI images and clinical data from 261 patients across two medical centers. The data were split into a training set (n = 153, medical center 1) and an external test set (n = 108, medical center 2). Radiomic features were extracted using Pyradiomics to build the Radiomics Model. Deep learning networks, including the transformer-based MobileVIT Model and Convolutional Neural Networks (CNN) based ConvNeXt Model, were trained separately. By applying the "late fusion" theory, the radiomics model and deep learning model were fused to produce the optimal Max-Fusion Model. Additionally, Shapley Additive exPlanations (SHAP) and Grad-CAM were employed for interpretability analysis.

RESULTS

In the external test set, the Radiomics Model achieved an Area under the receiver operating characteristic curve (AUC) of 0.86, the MobileVIT Model had an AUC of 0.91, the ConvNeXt Model demonstrated an AUC of 0.89, and the Max-Fusion Model showed an AUC of 0.92. The Delong test revealed a significant difference in AUC between the Max-Fusion Model and the Radiomics Model (P = 0.02).

CONCLUSION

The Max-Fusion Model, combining different models, presents superior performance in distinguishing PCNSL and GBM, highlighting the effectiveness of model fusion for enhanced decision-making in medical applications.

CLINICAL RELEVANCE STATEMENT

The preoperative non-invasive differentiation between PCNSL and GBM assists clinicians in selecting appropriate treatment regimens and clinical management strategies.

Prediction of early recurrence of adult-type diffuse gliomas following radiotherapy using multi-modal magnetic resonance images

BACKGROUND

Adult-type diffuse gliomas are among the central nervous system's most aggressive malignant primary neoplasms. Despite advancements in systemic therapies and technological improvements in radiation oncology treatment delivery, the survival outcome for these patients remains poor. Fast and accurate assessment of tumor response to oncologic treatments is crucial, as it can enable the early detection of recurrent or refractory gliomas, thereby allowing timely intervention with life-prolonging salvage therapies.

PURPOSE

Radiomics is a developing field with great potential to improve medical image interpretation. This study aims to apply a radiomics-based predictive model for classifying response to radiotherapy within the first 3 months post-treatment.

METHODS

Ninety-five patients were selected from the Burdenko Glioblastoma Progression Dataset. Tumor regions were delineated in the axial plane on contrast-enhanced T1(CE T1W) and T2 fluid-attenuated inversion recovery (T2_FLAIR) magnetic resonance imaging (MRI). Hand-crafted radiomic (HCR) features, including first- and second-order features, were extracted using PyRadiomics (3.7.6) in Python (3.10). Then, recursive feature elimination with a random forest (RF) classifier was applied for feature dimensionality reduction. RF and support vector machine (SVM) classifiers were built to predict treatment outcomes using the selected features. Leave-one-out cross-validation was employed to tune hyperparameters and evaluate the models.

RESULTS

For each segmented target, 186 HCR features were extracted from the MRI sequence. Using the top-ranked radiomic features from a combination of CE T1W and T2_FLAIR, an optimized classifier achieved the highest averaged area under the curve (AUC) of 0.829 ± 0.075 using the RF classifier. The HCR features of CE T1W produced the worst outcomes among all models (0.603 ± 0.024 and 0.615 ± 0.075 for RF and SVM classifiers, respectively).

CONCLUSIONS

We developed and evaluated a radiomics-based predictive model for early tumor response to radiotherapy, demonstrating excellent performance supported by high AUC values. This model, harnessing radiomic features from multi-modal MRI, showed superior predictive performance compared to single-modal MRI approaches. These results underscore the potential of radiomics in clinical decision support for this disease process.

Enhancing the differential diagnosis of small pulmonary nodules: a comprehensive model integrating plasma methylation, protein biomarkers, and LDCT imaging features

BACKGROUND

Accurate differentiation between malignant and benign pulmonary nodules, especially those measuring 5-10 mm in diameter, continues to pose a significant diagnostic challenge. This study introduces a novel, precise approach by integrating circulating cell-free DNA (cfDNA) methylation patterns, protein profiling, and computed tomography (CT) imaging features to enhance the classification of pulmonary nodules.

METHODS

Blood samples were collected from 419 participants diagnosed with pulmonary nodules ranging from 5 to 30 mm in size, before any disease-altering procedures such as treatment or surgical intervention. High-throughput bisulfite sequencing was used to conduct DNA methylation profiling, while protein profiling was performed utilizing the Olink proximity extension assay. The dataset was divided into a training set and an independent test set. The training set included 162 matched cases of benign and malignant nodules, balanced for sex and age. In contrast, the test set consisted of 46 benign and 49 malignant nodules. By effectively integrating both molecular (DNA methylation and protein profiling) and CT imaging parameters, a sophisticated deep learning-based classifier was developed to accurately distinguish between benign and malignant pulmonary nodules.

RESULTS

Our results demonstrate that the integrated model is both accurate and robust in distinguishing between benign and malignant pulmonary nodules. It achieved an AUC score 0.925 (sensitivity = 83.7%, specificity = 82.6%) in classifying test set. The performance of the integrated model was significantly higher than that of individual methylation (AUC = 0.799, P = 0.004), protein (AUC = 0.846, P = 0.009), and imaging models (AUC = 0.866, P = 0.01). Importantly, the integrated model achieved a higher AUC of 0.951 (sensitivity = 83.9%, specificity = 89.7%) in 5-10 mm small nodules. These results collectively confirm the accuracy and robustness of our model in detecting malignant nodules from benign ones.

CONCLUSIONS

Our study presents a promising noninvasive approach to distinguish the malignancy of pulmonary nodules using multiple molecular and imaging features, which has the potential to assist in clinical decision-making.

TRIAL REGISTRATION

This study was registered on ClinicalTrials.gov on 01/01/2020 (NCT05432128). https://classic.

CLINICALTRIALS

gov/ct2/show/NCT05432128 .

A large scale multi institutional study for radiomics driven machine learning for meningioma grading

This study aims to develop and evaluate radiomics-based machine learning (ML) models for predicting meningioma grades using multiparametric magnetic resonance imaging (MRI). The study utilized the BraTS-MEN dataset's training split, including 698 patients (524 with grade 1 and 174 with grade 2-3 meningiomas). We extracted 4872 radiomic features from T1, T1 with contrast, T2, and FLAIR MRI sequences using PyRadiomics. LASSO regression reduced features to 176. The data was split into training (60%), validation (20%), and test (20%) sets. Five ML algorithms (TabPFN, XGBoost, LightGBM, CatBoost, and Random Forest) were employed to build models differentiating low-grade (grade 1) from high-grade (grade 2-3) meningiomas. Hyperparameter tuning was performed using Optuna, optimizing model-specific parameters and feature selection. The CatBoost model demonstrated the best performance, achieving an area under the receiver operating characteristic curve (AUROC) of 0.838 [95% confidence interval (CI): 0.689-0.935], precision of 0.492 (95% CI: 0.371-0.623), recall of 0.838 (95% CI: 0.689-0.935), F1 score of 0.620 (95% CI: 0.495-0.722), accuracy of 0.729 (95% CI: 0.650-0.800), an area under the precision-recall curve (AUPRC) of 0.620 (95% CI: 0.433-0.753), and Brier score of 0.156 (95% CI: 0.122-0.200). Other models showed comparable performance, with mean AUROCs ranging from 0.752 to 0.784. The radiomics-based ML approach presented in this study showcases the potential for non-invasive and pre-operative grading of meningiomas using multiparametric MRI. Further validation on larger and independent datasets is necessary to establish the robustness and generalizability of these findings.

Non-invasive assessment of programmed cell death ligand-1 expression using 18F-FDG PET-CT imaging in esophageal squamous cell carcinoma

Programmed cell death ligand-1 (PD-L1) is an ideal checkpoint for immunohistochemical detection. The method of obtaining PD-L1 expression through biopsy can impact the accurate assessment of PD-L1 expression due to the spatial and temporal heterogeneity of tumors. Because of the limited sample size, biopsies often give only a localized picture of the tumor. In this retrospective study, a total of 2,386 metabolic tumor volume (MTV) features were extracted from 18F-FDG PET-CT images. A radiomics model was developed to holistically and non-invasively assess PD-L1 expression in patients with esophageal squamous cell carcinoma by identifying seven independent factors through feature screening. The radiomics model shows effective discrimination, with an area under the receiver operating characteristic curve of 0.888 [95% confidence interval (CI): 0.831-0.945] and 0.889 (95% CI: 0.706-1.000) for the training and validation cohorts, respectively. The results of the decision curve analysis demonstrated that utilizing the radiation model to forecast PD-L1 expression levels yielded more net benefits at threshold probabilities below 0.669. The clinical impact curves demonstrate that when the threshold probability is less than 0.501, the loss-to-benefit ratio is less than one in all cases.

Myocardial perfusion SPECT radiomic features reproducibility assessment: Impact of image reconstruction and harmonization

BACKGROUND

Coronary artery disease (CAD) has one of the highest mortality rates in humans worldwide. Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) provides clinicians with myocardial metabolic information non-invasively. However, there are some limitations to interpreting SPECT images performed by physicians or automatic quantitative approaches. Radiomics analyzes images objectively by extracting quantitative features and can potentially reveal biological characteristics that the human eye cannot detect. However, the reproducibility and repeatability of some radiomic features can be highly susceptible to segmentation and imaging conditions.

PURPOSE

We aimed to assess the reproducibility of radiomic features extracted from uncorrected MPI-SPECT images reconstructed with 15 different settings before and after ComBat harmonization, along with evaluating the effectiveness of ComBat in realigning feature distributions.

MATERIALS AND METHODS

A total of 200 patients (50% normal and 50% abnormal) including rest and stress (without attenuation and scatter corrections) MPI-SPECT images were included. Images were reconstructed using 15 combinations of filter cut-off frequencies, filter orders, filter types, reconstruction algorithms, number of iterations and subsets resulting in 6000 images. Image segmentation was performed on the left ventricle in the first reconstruction for each patient and applied to 14 others. A total of 93 radiomic features were extracted from the segmented area, and ComBat was used to harmonize them. The intraclass correlation coefficient (ICC) and overall concordance correlation coefficient (OCCC) tests were performed before and after ComBat to examine the impact of each parameter on feature robustness and to assess harmonization efficiency. The ANOVA and the Kruskal-Wallis tests were performed to evaluate the effectiveness of ComBat in correcting feature distributions. In addition, the Student's t-test, Wilcoxon rank-sum, and signed-rank tests were implemented to assess the significance level of the impacts made by each parameter of different batches and patient groups (normal vs. abnormal) on radiomic features.

RESULTS

Before applying ComBat, the majority of features (ICC: 82, OCCC: 61) achieved high reproducibility (ICC/OCCC ≥ 0.900) under every batch except Reconstruction. The largest and smallest number of poor features (ICC/OCCC < 0.500) were obtained by IterationSubset and Order batches, respectively. The most reliable features were from the first-order (FO) and gray-level co-occurrence matrix (GLCM) families. Following harmonization, the minimum number of robust features increased (ICC: 84, OCCC: 78). Applying ComBat showed that Order and Reconstruction were the least and the most responsive batches, respectively. The most robust families, in a descending order, were found to be FO, neighborhood gray-tone difference matrix (NGTDM), GLCM, gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), and gray-level dependence matrix (GLDM) under Cut-off, Filter, and Order batches. The Wilcoxon rank-sum test showed that the number of robust features significantly differed under most batches in the Normal and Abnormal groups.

CONCLUSION

The majority of radiomic features show high levels of robustness across different OSEM reconstruction parameters in uncorrected MPI-SPECT. ComBat is effective in realigning feature distributions and enhancing radiomic features reproducibility.

Deep Learning With Optical Coherence Tomography for Melanoma Identification and Risk Prediction

Malignant melanoma is the most severe skin cancer with a rising incidence rate. Several noninvasive image techniques and computer-aided diagnosis systems have been developed to help find melanoma in its early stages. However, most previous research utilized dermoscopic images to build a diagnosis model, and only a few used prospective datasets. This study develops and evaluates a convolutional neural network (CNN) for melanoma identification and risk prediction using optical coherence tomography (OCT) imaging of mice skin. Longitudinal tests are performed on four animal models: melanoma mice, dysplastic nevus mice, and their respective controls. The CNN classifies melanoma and healthy tissues with high sensitivity (0.99) and specificity (0.98) and also assigns a risk score to each image based on the probability of melanoma presence, which may facilitate early diagnosis and management of melanoma in clinical settings.

Association Between Pericoronary Fat Attenuation Index as Evaluated by Coronary Artery CT Angiography and Clinical Interventions in Lipid Management Among Patients with Coronary Artery Disease

Objective

This study aims to evaluate the relationship between the pericoronary fat attenuation index (FAI), derived from coronary artery computed tomography angiography, and post-lipid management levels of low-density lipoprotein cholesterol in patients with coronary artery disease (CAD). Additionally, the study investigates coronary inflammation across different lipid management strategies.

Methods

We selected a cohort comprising 521 CAD patients who met the inclusion criteria. Patients were categorized into well-managed (LDL-C<2.6 mmol/L) and poorly managed (LDL-C≥2.6 mmol/L) groups based on lipid management efficacy. We collected anthropometric measures (height, weight, body mass index, and body surface area) and clinical indicators, including Gensini score, and FAI-related parameters for coronary atherosclerotic lesions. We analyzed the interrelations along these parameters and lipid management using statistical methods and assessed diagnostic value via receiver operating characteristic (ROC) curve analysis of these parameters was assessed through.

Results

The poorly managed group exhibited significantly higher levels of total cholesterol, triglycerides, and lower levels of high-density lipoprotein compared to the well-managed group (P < 0.05). Significant differences were observed between the groups in terms of lesion length in the proximal segment of the left anterior descending artery, FAI value in the proximal segment of lesions in the right coronary artery (RCA), volume thickness in the middle segment of RCA lesions, and lesion length in the distal segment of RCA (P < 0.05). ROC curve analysis revealed areas under the curve ranging from 0.484 to 0.660 for the parameters, indicating limited diagnostic efficacy.

Conclusion

The FAI in the RCA varies with lipid management strategies, suggesting it as a valuable metric for monitoring both perivascular inflammation and lipid status in CAD patients. However, its current diagnostic efficacy is limited, indicating the need for further research to improve its clinical utility.

ESR Essentials: radiomics-practice recommendations by the European Society of Medical Imaging Informatics

Radiomics is a method to extract detailed information from diagnostic images that cannot be perceived by the naked eye. Although radiomics research carries great potential to improve clinical decision-making, its inherent methodological complexities make it difficult to comprehend every step of the analysis, often causing reproducibility and generalizability issues that hinder clinical adoption. Critical steps in the radiomics analysis and model development pipeline-such as image, application of image filters, and selection of feature extraction parameters-can greatly affect the values of radiomic features. Moreover, common errors in data partitioning, model comparison, fine-tuning, assessment, and calibration can reduce reproducibility and impede clinical translation. Clinical adoption of radiomics also requires a deep understanding of model explainability and the development of intuitive interpretations of radiomic features. To address these challenges, it is essential for radiomics model developers and clinicians to be well-versed in current best practices. Proper knowledge and application of these practices is crucial for accurate radiomics feature extraction, robust model development, and thorough assessment, ultimately increasing reproducibility, generalizability, and the likelihood of successful clinical translation. In this article, we have provided researchers with our recommendations along with practical examples to facilitate good research practices in radiomics. KEY POINTS: Radiomics' inherent methodological complexity should be understood to ensure rigorous radiomic model development to improve clinical decision-making. Adherence to radiomics-specific checklists and quality assessment tools ensures methodological rigor. Use of standardized radiomics tools and best practices enhances clinical translation of radiomics models.

Coupling Habitat Radiomic Analysis with the Diversification of the Tumor ecosystem: Illuminating New Strategy in the Assessment of Postoperative Recurrence of Non-Muscle Invasive Bladder Cancer

RATIONALE AND OBJECTIVES

Non-muscle-invasive bladder cancer (NMIBC) is highly recurrent, with each recurrence potentially progressing to muscle-invasive cancer, affecting patient prognosis. Intratumoral heterogeneity plays a crucial role in NMIBC recurrence. This study investigated a novel habitat-based radiomic analysis for stratifying NMIBC recurrence risk.

MATERIALS AND METHODS

A retrospective collection of 382 NMIBC patients between 2015 and 2021 from two medical institutions was carried out. Patients' CT images were collected across three phases, with tumor sites delineated within the bladder. Intratumoral habitats were identified using K-means clustering on 19 texture features of the tumor sites, followed by the extraction of 107 radiomic features per habitat with PyRadiomics. These features were integrated into machine learning algorithms to develop a habitat-based model (HBM) for predicting two-year recurrence of NMIBC patients. The clinical and multiphase radiomic models were also constructed for comparison, with the Delong test comparing their diagnostic efficiency. The impact of HMB on patients' recurrence-free survival and the correlation between HBM and tumor-stroma ratio were further analyzed.

RESULTS

Three distinct habitats were identified within NMIBC. The HBM showed an AUC of 0.932 (95% CI: 0.906 - 0.958) in the training cohort and 0.782 (95% CI: 0.674 - 0.890) in the validation cohort for predicting two-year recurrence. With comparison between different models, The HBM is demonstrated to possess superior diagnostic efficacy to the clinical model (p < 0.001) in the training cohort. However, no significant difference was noted between the multiphase and clinical models (p = 0.130) in the training cohort. The HBM score effectively distinguished the recurrence-free survival of NIMBC patients and demonstrated a significant correlation with the tumor-stroma ratio.

CONCLUSIONS

Habitat-based radiomics, coupled with machine learning, efficiently predicts NMIBC recurrence. Further research on habitat-based radiomics offers potential improvement in clinical management of NMIBC.

Report on the AAPM grand challenge on deep generative modeling for learning medical image statistics

BACKGROUND

The findings of the 2023 AAPM Grand Challenge on Deep Generative Modeling for Learning Medical Image Statistics are reported in this Special Report.

PURPOSE

The goal of this challenge was to promote the development of deep generative models for medical imaging and to emphasize the need for their domain-relevant assessments via the analysis of relevant image statistics.

METHODS

As part of this Grand Challenge, a common training dataset and an evaluation procedure was developed for benchmarking deep generative models for medical image synthesis. To create the training dataset, an established 3D virtual breast phantom was adapted. The resulting dataset comprised about 108 000 images of size 512  × $\times$  512. For the evaluation of submissions to the Challenge, an ensemble of 10 000 DGM-generated images from each submission was employed. The evaluation procedure consisted of two stages. In the first stage, a preliminary check for memorization and image quality (via the Fréchet Inception Distance [FID]) was performed. Submissions that passed the first stage were then evaluated for the reproducibility of image statistics corresponding to several feature families including texture, morphology, image moments, fractal statistics, and skeleton statistics. A summary measure in this feature space was employed to rank the submissions. Additional analyses of submissions was performed to assess DGM performance specific to individual feature families, the four classes in the training data, and also to identify various artifacts.

RESULTS

Fifty-eight submissions from 12 unique users were received for this Challenge. Out of these 12 submissions, 9 submissions passed the first stage of evaluation and were eligible for ranking. The top-ranked submission employed a conditional latent diffusion model, whereas the joint runners-up employed a generative adversarial network, followed by another network for image superresolution. In general, we observed that the overall ranking of the top 9 submissions according to our evaluation method (i) did not match the FID-based ranking, and (ii) differed with respect to individual feature families. Another important finding from our additional analyses was that different DGMs demonstrated similar kinds of artifacts.

CONCLUSIONS

This Grand Challenge highlighted the need for domain-specific evaluation to further DGM design as well as deployment. It also demonstrated that the specification of a DGM may differ depending on its intended use.

Differentiating MYCN-amplified RB1 wild-type retinoblastoma from biallelic RB1 mutant retinoblastoma using MR-based radiomics: a retrospective multicenter case-control study

MYCN-amplified RB1 wild-type (MYCNampRB1+/+) retinoblastoma is a rare and aggressive subtype, often resistant to standard therapies. Identifying unique MRI features is crucial for diagnosing this subtype, as biopsy is not recommended. This study aimed to differentiate MYCNampRB1+/+ from the most prevalent RB1-/- retinoblastoma using pretreatment MRI and radiomics. Ninety-eight unilateral retinoblastoma patients (19 MYCN cases and 79 matched controls) were included. Tumors on T2-weighted MR images were manually delineated and validated by experienced radiologists. Radiomics analysis extracted 120 features per tumor. Several combinations of feature selection methods, oversampling techniques and machine learning (ML) classifiers were evaluated in a repeated fivefold cross-validation machine learning pipeline to yield the best-performing prediction model for MYCN. The best model used univariate feature selection, data oversampling (duplicating MYCN cases), and logistic regression classifier, achieving a mean AUC of 0.78 (SD 0.12). SHAP analysis highlighted lower sphericity, higher flatness, and greater gray-level heterogeneity as predictive for MYCNampRB1+/+ status, yielding an AUC of 0.81 (SD 0.11). This study shows the potential of MRI-based radiomics to distinguish MYCNampRB1+/+ and RB1-/- retinoblastoma subtypes.

Lung nodule classification using radiomics model trained on degraded SDCT images

BACKGROUND AND OBJECTIVE

Low-dose computed tomography (LDCT) screening has shown promise in reducing lung cancer mortality; however, it suffers from high false positive rates and a scarcity of available annotated datasets. To overcome these challenges, we propose a novel approach using synthetic LDCT images generated from standard-dose CT (SDCT) scans from the LIDC-IDRI dataset. Our objective is to develop and validate an interpretable radiomics-based model for distinguishing likely benign from likely malignant pulmonary nodules.

METHODS

From a total of 1010 CT images (695 SDCTs and 315 LDCTs), we degraded SDCTs in the sinogram domain and obtained 1950 nodules as the training set. The 675 nodules from the LDCTs were stratified into 50%-50% partitions for validation and testing. Radiomic features were extracted from nodules, and three feature sets were assessed using: a) only shape and size (SS) features, b) all features but SS features, and c) all features. A systematic pipeline was developed to optimize the feature set and evaluate multiple machine learning models. Models were trained using degraded SDCT, validated and tested on the LDCT nodules.

RESULTS

Training a logistic regression model using three SS features yielded the most promising results, achieving on the test set mean balanced accuracy, sensitivity, specificity, and AUC-ROC scores of 0.81, 0.76, 0.85, and 0.87, respectively.

CONCLUSIONS

Our study demonstrates the feasibility and effectiveness of using synthetic LDCT images for developing a relatively accurate radiomics-based model in lung nodule classification. This approach addresses challenges associated with LDCT screening, offering potential implications for improving lung cancer detection and reducing false positives.

Proteogenomic analysis of air-pollution-associated lung cancer reveals prevention and therapeutic opportunities

Air pollution significantly impacts lung cancer progression, but there is a lack of a comprehensive molecular characterization of clinical samples associated with air pollution. Here, we performed a proteogenomic analysis of lung adenocarcinoma (LUAD) in 169 female never-smokers from the Xuanwei area (XWLC cohort), where coal smoke is the primary contributor to the high lung cancer incidence. Genomic mutation analysis revealed XWLC as a distinct subtype of LUAD separate from cases associated with smoking or endogenous factors. Mutational signature analysis suggested that Benzo[a]pyrene (BaP) is the major risk factor in XWLC. The BaP-induced mutation hotspot, EGFR-G719X, was present in 20% of XWLC which endowed XWLC with elevated MAPK pathway activations and worse outcomes compared to common EGFR mutations. Multi-omics clustering of XWLC identified four clinically relevant subtypes. These subgroups exhibited distinct features in biological processes, genetic alterations, metabolism demands, immune landscape, and radiomic features. Finally, MAD1 and TPRN were identified as novel potential therapeutic targets in XWLC. Our study provides a valuable resource for researchers and clinicians to explore prevention and treatment strategies for air-pollution-associated lung cancers.

Early prediction of acute pancreatitis with acute kidney injury using abdominal contrast-enhanced CT features

Early prediction of acute pancreatitis (AP) with acute kidney injury (AKI) using abdominal contrast-enhanced CT could effectively reduce the mortality and the economic burden on patients and society. However, this challenge is limited by the imaging manifestations of early-stage AP that are not clearly visible to the naked eye. To address this, we developed a machine learning model using imperceptible variations in the structural changes of pancreas and peripancreatic region, extracted by radiomics and artificial intelligence technology, to screen and stratify the high-risk AP patients at the early stage of AP. The results demonstrate that the machine learning model could screen the high-risk AP with AKI patients with an area under the curve (AUC) of 0.82 for the external cohort, superior to the human radiologists. This finding confirms the significant potential of machine learning in the screening of acute pancreatitis and contributes to personalized treatment and management for AP patients.

Machine learning-based discrimination of benign and malignant breast lesions on US: The contribution of shear-wave elastography

PURPOSE

To build and validate a combined radiomics and machine learning (ML) approach using B-mode US and SWE images to differentiate benign from malignant solid breast lesions (BLs) and compare its performance with that of an expert radiologist.

METHODS

Patients with at least one BI-RADS 2-6 BL who performed breast US integrated with SWE were retrospectively included. B-mode US and SWE images were manually segmented to extract radiomics features. A multi-step feature selection process was performed and a predictive model built using the Logistic Regression algorithm. The diagnostic accuracy was evaluated with the AUC and Matthews Correlation Coefficient (MCC) metrics. The performance of the ML classifier was compared to that of an expert radiologist.

RESULTS

427 Bls were included and divided into a training (286 BLs, of which 127 benign and 159 malignant) and a test set (141 BLs, of which 59 benign and 82 malignant). Of 1098 features extracted from B-mode US and SWE images, 13 were finally selected. The ML classifier showed an AUC of 0.768 and 0.746, and an MCC of 0.403 and 0.423 in the training and test sets, respectively. The performance was higher than that of the expert radiologist assessing only B-mode US images, but significantly lower when SWE images were also provided.

CONCLUSION

A ML approach based on B-mode US and SWE images may represent a potential tool in the characterization of BLs. SWE still gives its most relevant contribution in the clinical setting rather than included in a radiomics pipeline.

A machine learning based radiomics approach for predicting No. 14v station lymph node metastasis in gastric cancer

Purpose

To evaluate the potential of radiomics approach for predicting No. 14v station lymph node metastasis (14vM) in gastric cancer (GC).

Methods

The contrast enhanced CT (CECT) images with corresponding clinical information of 288 GC patients were retrospectively collected. Patients were separated into training set (n = 202) and testing set (n = 86). A total of 1,316 radiomics feature were extracted from portal venous phase images of CECT. Seven machine learning (ML) algorithms including naïve Bayes (NB), k-nearest neighbor (KNN), decision tree (DT), logistic regression (LR), random forest (RF), eXtreme gradient boosting (XGBoost) and support vector machine (SVM) were trained for development of optimal radiomics signature. A combined model was established by combining radiomics with important clinicopathological factors. The diagnostic ability of the signature and model were evaluated.

Results

LR algorithm was chosen for signature construction. The radiomics signature exhibited good discrimination accuracy of 14vM with AUCs of 0.83 in the training and 0.77 in the testing set. The risk of 14vM showed significant association with higher radiomics score. A combined model exhibited increased predictive ability and good agreement in the training (AUC = 0.87) and testing (AUC = 0.85) sets.

Conclusion

The ML-based radiomics model provided a promising image biomarker for preoperative detection of 14vM and may help the surgeon to decide whether to add 14v dissection to lymphadenectomy.

Prediction of lymphovascular invasion in esophageal squamous cell carcinoma by computed tomography-based radiomics analysis: 2D or 3D ?

BACKGROUND

To compare the performance between one-slice two-dimensional (2D) and whole-volume three-dimensional (3D) computed tomography (CT)-based radiomics models in the prediction of lymphovascular invasion (LVI) status in esophageal squamous cell carcinoma (ESCC).

METHODS

Two hundred twenty-four patients with ESCC (158 LVI-absent and 66 LVI-present) were enrolled in this retrospective study. The enrolled patients were randomly split into the training and testing sets with a 7:3 ratio. The 2D and 3D radiomics features were derived from the primary tumors' 2D and 3D regions of interest (ROIs) using 1.0 mm thickness contrast-enhanced CT (CECT) images. The 2D and 3D radiomics features were screened using inter-/intra-class correlation coefficient (ICC) analysis, Wilcoxon rank-sum test, Spearman correlation test, and the least absolute shrinkage and selection operator, and the radiomics models were built by multivariate logistic stepwise regression. The performance of 2D and 3D radiomics models was assessed by the area under the receiver operating characteristic (ROC) curve. The actual clinical utility of the 2D and 3D radiomics models was evaluated by decision curve analysis (DCA).

RESULTS

There were 753 radiomics features from 2D ROIs and 1130 radiomics features from 3D ROIs, and finally, 7 features were retained to construct 2D and 3D radiomics models, respectively. ROC analysis revealed that in both the training and testing sets, the 3D radiomics model exhibited higher AUC values than the 2D radiomics model (0.930 versus 0.852 and 0.897 versus 0.851, respectively). The 3D radiomics model showed higher accuracy than the 2D radiomics model in the training and testing sets (0.899 versus 0.728 and 0.788 versus 0.758, respectively). In addition, the 3D radiomics model has higher specificity and positive predictive value, while the 2D radiomics model has higher sensitivity and negative predictive value. The DCA indicated that the 3D radiomics model provided higher actual clinical utility regarding overall net benefit than the 2D radiomics model.

CONCLUSIONS

Both 2D and 3D radiomics features can be employed as potential biomarkers to predict the LVI in ESCC. The performance of the 3D radiomics model is better than that of the 2D radiomics model for the prediction of the LVI in ESCC.

Pathomics models for CD40LG expression and prognosis prediction in glioblastoma

Glioblastoma (GBM) is the most prevalent primary malignant tumor of the nervous system. In this study, we utilized pathomics analysis to explore the expression of CD40LG and its predictive value for the prognosis of GBM patients. We analyzed the expression differences of CD40LG in GBM tissue and normal brain tissue, along with performing survival prognosis analysis. Additionally, histopathological sections of GBM were used to screen for pathological features. Subsequently, SVM and LR pathomics models were constructed, and the models' performance was evaluated. The pathomics model was employed to predict CD40LG expression and patient prognosis. Furthermore, we investigated the potential molecular mechanisms through enrichment analysis, WGCNA analysis, immune correlation analysis, and immune checkpoint analysis. The expression level of CD40LG was significantly increased in GBM. Multivariate analysis demonstrated that high expression of CD40LG is a risk factor for overall survival (OS) in GBM patients. Five pathological features were identified, and SVM and LR pathomics models were constructed. Model evaluation showed promising predictive effects, with an AUC value of 0.779 for the SVM model, and the Hosmer-Lemeshow test confirmed the model's prediction probability consistency (P > 0.05). The LR model achieved an AUC value of 0.785, and the Hosmer-Lemeshow test indicated good agreement between the LR model's predicted probabilities and the true value (P > 0.05). Immune infiltration analysis revealed a significant correlation between the pathomics score (PS) and the degree of infiltration of activated DC cells, T cells CD4 naïve, Macrophages M2, Macrophages M1, and T cells CD4 memory resting. Our results demonstrate that the pathomics model exhibits predictability for CD40LG expression and GBM patient survival. These findings can be utilized to assist neurosurgeons in selecting optimal treatment strategies in clinical practice.

Evaluation of an Image-based Classification Model to Identify Glioma Subtypes Using Arterial Spin Labeling Perfusion MRI On the Publicly Available UCSF Glioma Dataset

PURPOSE

Glioma is a complex cancer comprising various subtypes and mutations, which may have different metabolic characteristics that can potentially be investigated and identified using perfusion imaging. Therefore, the aim of this work was to use radiomics and machine learning analysis of arterial spin labeling MRI data to automatically differentiate glioma subtypes and mutations.

METHODS

A total of 495 Arterial Spin Labeling (ASL) perfusion imaging datasets from the UCSF Glioma database were used in this study. These datasets were segmented to delineate the tumor volume and classified according to tumor grade, pathological diagnosis, and IDH status. Perfusion image data was obtained from a 3T MRI scanner using pseudo-continuous ASL. High level texture features were extracted for each ASL dataset using PyRadiomics after tumor volume segmentation and then analyzed using a machine learning framework consisting of ReliefF feature ranking and logistic model tree classification algorithms.

RESULTS

The results of the evaluation revealed balanced accuracies for the three endpoints ranging from 55.76% (SD = 4.28, 95% CI: 53.90-57.65) for the tumor grade using 25.4 ± 37.21 features, 62.53% (SD = 2.86, 95% CI: 61.27-63.78) for the mutation status with 23.3 ± 29.17 picked features, and 80.97% (SD = 1.83, 95% CI: 80.17-81.78) for the pathological diagnosis which used 47.3 ± 32.72 selected features.

CONCLUSIONS

Radiomics and machine learning analysis of ASL perfusion data in glioma patients hold potential for aiding in the diagnosis and treatment of glioma, mainly for discerning glioblastoma from astrocytoma, while performance for tumor grading and mutation status appears limited.

Radiomics approach for identifying radiation-induced normal tissue toxicity in the lung

The rapidly evolving field of radiomics has shown that radiomic features are able to capture characteristics of both tumor and normal tissue that can be used to make accurate and clinically relevant predictions. In the present study we sought to determine if radiomic features can characterize the adverse effects caused by normal tissue injury as well as identify if human embryonic stem cell (hESC) derived extracellular vesicle (EV) treatment can resolve certain adverse complications. A cohort of 72 mice (n = 12 per treatment group) were exposed to X-ray radiation to the whole lung (3 × 8 Gy) or to the apex of the right lung (3 × 12 Gy), immediately followed by retro-orbital injection of EVs. Cone-Beam Computed Tomography images were acquired before and 2 weeks after treatment. In total, 851 radiomic features were extracted from the whole lungs and < 20 features were selected to train and validate a series of random forest classification models trained to predict radiation status, EV status and treatment group. It was found that all three classification models achieved significantly high prediction accuracies on a validation subset of the dataset (AUCs of 0.91, 0.86 and 0.80 respectively). In the locally irradiated lung, a significant difference between irradiated and unirradiated groups as well as an EV sparing effect were observed in several radiomic features that were not seen in the unirradiated lung (including wavelet-LLH Kurtosis, wavelet HLL Large Area High Gray Level Emphasis, and Gray Level Non-Uniformity). Additionally, a radiation difference was not observed in a secondary comparison cohort, but there was no impact of imaging machine parameters on the radiomic signature of unirradiated mice. Our data demonstrate that radiomics has the potential to identify radiation-induced lung injury and could be applied to predict therapeutic efficacy at early timepoints.

Integrating Clinical Variables, Radiomics, and Tumor-derived Cell-free DNA for Enhanced Prediction of Resectable Esophageal Adenocarcinoma Outcomes

BACKGROUND

The value of integrating clinical variables, radiomics, and tumor-derived cell-free DNA (cfDNA) for the prediction of survival and response to chemoradiation of resectable esophageal adenocarcinoma (rEAC) patients is not yet known. Our aim was to investigate if radiomics and cfDNA metrics combined with clinical variables can improve personalized predictions.

METHODS

A cohort of 111 rEAC patients from two centers treated with neoadjuvant chemoradiotherapy was used for exploratory retrospective analyses. Models combining the clinical variables of the SOURCE survival model with radiomic features and cfDNA, were built using elastic net regression and internally validated using 5-fold cross validation. Model performance for overall survival (OS) and time to progression (TTP) were evaluated with the C-index and the area under the curve (AUC) for pathological complete response (pCR) RESULTS: The best performing baseline models for OS and TTP were based on the combination of SOURCE-cfDNA which reached a C-index of 0.55 and 0.59 compared to 0.44-0.45 with SOURCE alone. The addition of re-staging PET radiomics to SOURCE was the most promising addition for predicting OS (C-index: 0.65) and TTP (C-index: 0.60). Baseline risk-stratification was achieved for OS and TTP by combining SOURCE with radiomics or cfDNA, log-rank p<0.01. The best performing combination model for the prediction of pCR reached an AUC of 0.61 compared to 0.47 with SOURCE variables alone.

CONCLUSIONS

The addition of radiomics and cfDNA can improve the performance of an established survival model. External validity needs to be further assessed in future studies together with the optimization of radiomic pipelines.