Search Results (84)

Objective: To develop machine learning models by integrating transvaginal ultrasound (TVUS) with clinical indicators, conduct visual analysis of the models, and systematically assess their diagnostic efficacy in differentiating early cervical neoplastic lesions. Methods: A total of 144 eligible patients (84 cases of early cervical cancer and 60 cases of cervical myoma) admitted to the First Affiliated Hospital of Chongqing Medical University from January 2018 to August 2025 were retrospectively enrolled in this study. Their clinical data, human papillomavirus (HPV) test results, Thinprep Cytologic Test (TCT) findings, TVUS images and magnetic resonance (MR) imaging data were collected and subjected to comprehensive statistical analysis. Univariate and multivariate Logistic Regression analyses were performed to identify independent differentiating factors for lesion classification. Eleven machine learning models were subsequently constructed, and their diagnostic performance was evaluated using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), and the DeLong test. Finally, a nomogram was developed based on the optimal-performing model for clinical visualization. Results: The TVUS–clinical indicator integration model identified five independent differentiating factors: HPV status, TCT findings, menopausal status, ultrasonic tumor blood supply, and ultrasonic tumor morphology. In contrast, the MR–clinical indicator integration model screened out three independent factors: HPV status, TCT findings, and intratumoral signal intensity on MR T2-weighted imaging (T2WI). The TVUS integration model demonstrated marginally superior diagnostic performance, with a sensitivity of 0.988, specificity of 0.983, and an area under the ROC curve (AUC) of 0.991, compared with the MR integration model (sensitivity: 0.952, specificity: 0.950, AUC: 0.975); however, this difference in AUC values was not statistically significant (p = 0.911). Among the 11 machine learning models, the Logistic Regression model exhibited optimal classification performance and stability. DCA curves confirmed that all constructed models outperformed single-index diagnostic strategies in clinical decision-making for lesion differentiation. A nomogram was further established based on the Logistic Regression model for intuitive clinical application. Conclusions: Multiple machine learning models integrating TVUS with clinical indicators are successfully developed, and a corresponding nomogram is constructed in this study. Full article

Background/Objectives: We developed and validated a multimodal magnetic resonance imaging (MRI) framework combining deep learning segmentation with radiomics to predict local recurrence in nasopharyngeal carcinoma (NPC). Methods: This retrospective two-center study included 1074 NPC patients treated between 2015 and 2019. Center 1 cases were split 8:2 into training and internal test sets, while Center 2 served for external validation. A multimodal Swin UNet model automatically segmented tumors from pretreatment T1-weighted, T2-weighted, and contrast-enhanced T1 (CET1) images. Radiomics features were extracted from expert-reviewed regions of interest, selected, and modeled using extreme gradient boosting for recurrence prediction. Results: The multimodal segmentation model maintained consistent but moderate Dice similarity coefficients (0.737, 0.666, and 0.726 for T1WI, T2WI, and CET1 in external validation). These values reflect the moderate overlap typical for nasopharyngeal carcinoma, given its highly infiltrative growth and ill-defined boundaries along complex anatomic interfaces. For local recurrence prediction, single-modality models reached external AUCs between 0.754 and 0.781. Importantly, the multimodal fusion model demonstrated numerical improvement over single modalities in the external validation set (e.g., vs. T1WI, p = 0.141), achieving an AUC of 0.910, accuracy of 0.908, sensitivity of 0.805, specificity of 0.946, and F1-score of 0.825. Conclusions: The multimodal MRI radiomics model, developed alongside a deep learning segmentation module, demonstrated favorable multicenter performance for evaluating NPC recurrence risk. The primary prognostic analysis was based on expert-reviewed regions of interest; a supplementary analysis using fully automatic segmentation masks yielded comparable, non-significantly different performance across all cohorts (Training AUC: 0.887; Internal Test AUC: 0.892; External Validation AUC: 0.885 vs. 0.910, p = 0.145), supporting the feasibility of future end-to-end deployment. Fusing multimodal features yielded numerical improvements over single-sequence models in external validation, providing a basis for post-treatment surveillance planning. Full article

Background/Objectives: Breast MRI surveillance for women with a personal history of breast cancer (PHBC) is often limited by costs and acquisition times. This study aims to identify the optimal abbreviated breast MR (ABMR) protocol for this population by assessing the diagnostic performance of different sequence additions. Methods: This retrospective study included 1002 women with PHBC who underwent postoperative breast MRI with ultrafast sequences. Propensity score matching using 12 variables yielded recurrence (n = 21) and nonrecurrence (n = 42) groups with balanced characteristics. Four ABMR protocols were simulated by sequentially combining sequences: Step 1 (FAST protocol) included precontrast T1-weighted imaging (T1WI), early-phase T1WI, and subtracted maximal intensity projection (MIP). Step 2 added ultrafast MIP; Step 3 incorporated delayed-phase T1WI; and Step 4 included T2WI and diffusion weighted imaging (DWI). Three expert breast radiologists independently reviewed MRIs. Sensitivity, specificity, accuracy, and area under the curve (AUC) were assessed. Results: Sensitivity, specificity, and accuracy for ABMR protocols ranged from 76.2% to 90.5%, 88.1% to 92.9%, and 85.7% to 90.5%, respectively. The FAST protocol alone provided reliable performance (sensitivity: 81%; specificity: 88.1–90.5%; accuracy: 85.7–87.3%). Additional sequences yielded modest improvements, but no statistically significant differences were observed across all 3 readers (p > 0.05). ABMR protocols demonstrated equivalent diagnostic performance for PHBC surveillance. Conclusions: The FAST protocol alone provided reliable results, indicating its potential as a primary ABMR protocol. While additional sequences slightly improved specificity, they did not significantly enhance diagnostic accuracy. Full article

Background/Objectives: This study aims to develop a comprehensive Clinical Decision Support System (CDSS) that integrates multi-sequence sacroiliac joint (SIJ) MRIs with rheumatological, clinical, and laboratory findings into the decision-making process for the early diagnosis of axial spondyloarthritis (axSpA), incorporating segmentation-supported explainability. Methods: Multi-sequence SIJ MRI data (T1-WI, T2-WI, STIR, and PD-WI) were analysed from 367 participants (n = 193 axSpA; n = 174 non-axSpA controls). Sequence-based classification was performed using VGG16, ResNet50, DenseNet121, and InceptionV3 models; additionally, a lightweight and parameter-efficient SacroNet architecture was developed. Slice-level probability scores were converted to patient-level scores using the Dynamic Top-K Averaging method. Image-based scores were combined with a logistic regression-based clinical risk score using weighted linear integration (0.60 image/0.40 clinical) and a conservative threshold (τ = 0.70). Grad-CAM was applied for visual interpretability. Furthermore, to support the diagnostic outcomes with precise spatial data, active inflammation in STIR and T2-WI sequences was segmented. For this purpose, the MDC-UNet model was employed and compared with baseline U-Net derivatives. Results: Sequence-specific analysis showed VGG16 performing best on T1-WI (AUC = 0.920; Accuracy = 0.878) and DenseNet121 on STIR (AUC = 0.793; Accuracy = 0.771). The SacroNet architecture provided competitive classification performance at the patient level despite its low number of parameters (~110 K). Furthermore, MDC-UNet successfully segmented active inflammation, yielding Dice scores of 0.752 (HD95: 19.25) for STIR and 0.682 (HD95: 26.21) for T2-WI. Conclusions: The findings demonstrate that patient-level decision integration based on multi-sequence MRI, when used in conjunction with clinical risk scoring and segmentation-assisted interpretability, can provide a feasible and interpretable DSS framework for the early diagnosis of axSpA. Full article

Background: MRI-based radiomics has shown promise in men with prostate cancer (PCa); however, successful clinical implementation is contingent upon on reproducible measurements. Purpose: We assessed the reproducibility of radiomics features extracted from bi-parametric prostate MRI (bpMRI) in prostate lesions and non-tumoral prostate tissue in men with PCa undergoing active surveillance (AS). Methods: This retrospective study included 47 men with biopsy-proven PCa undergoing AS (mean 68.9 ± 8.2 years, mean PSA density [PSAD] 0.08 ± 0.03 ng/mL/mL) who underwent two bpMRI approximately 12 months apart (range, 10–14 months; December 2018 to April 2020). The reproducibility of radiomics measurements was assessed using the same MRI platform (3T Skyra, Siemens Healthineers; inter-platform) (n = 37), different MRI vendors (Skyra, Siemens Healthineers; 3T Discovery MR750, GE Healthcare; inter-platform) (n = 10), and between observers (n = 10). Shape/1st-/2nd-order radiomics features were extracted from regions of interest on axial T2-weighted (T2-WI), diffusion-weighted imaging (DWI, b1600), and apparent diffusion coefficient (ADC) maps on prostate lesions, non-tumoral peripheral zones (PZs), and transition zones (TZs) using software. Reproducibility was evaluated by calculating the intraclass correlation coefficient (ICC) and coefficient of variation (CV). Associations of clinical variables and prostate volume were assessed. Results: PCa diagnoses included Gleason grade groups 1 (n = 46) and 2 (n = 1)]. Thirty-seven lesions (mean size 0.9 ± 0.4 cm) in 31 patients had PI-RADS v2.1 scores of 2 (n = 3)/3 (n = 12)/4 (n = 21)/5 (n = 1); 16 patients demonstrated diffuse PI-RADS 2 changes. Lesion radiomics features from T2-WI yielded a high proportion of good/moderate ICCs (intra-platform, 77.8%; inter-platform, 56.5%), whereas most DWI/ADC features yielded poor reproducibility. Similar results were observed for non-tumoral PZ/TZ. Intra-platform CVs were lowest for T2-WI lesion features (13.6%) and background PZ/TZ (<13.3%), while DWI/ADC exceeded 20%. Inter-platform CVs were lowest for lesions on T2-WI and were <18% for DWI/ADC; all background PZ/TZ CVs were < 16.4%. Inter-observer analyses showed good/moderate ICCs across all sequences and regions (57.4–92.6%). The distribution of ICC and CV values did not differ between intra- and inter-platform analyses (p > 0.05). Higher reproducibility (ICC > 0.5) was associated with larger prostate volume (intra-platform diagnostic odds ratio [DOR] = 2.58, 95% confidence interval [95%CI], 1.35–3.80, p = 0.01; inter-platform DOR = 3.48, 95%CI 1.79–5.17, p = 0.01) and older age (inter-platform DOR = 5.30, 95%CI 3.75–6.85, p < 0.01). Conclusions: Radiomics measurements from T2-WI demonstrated better intra-/inter-platform reproducibility than DWI/ADC for prostate lesions and non-tumoral tissue. Patient factors (larger prostate volumes and older age) influence radiomics stability. The optimization of diffusion-based radiomics features is needed to improve reproducibility given the essential role of DWI in prostate MRI. Full article

Background: Fulminant type 1 diabetes (FT1D) is a rare but life-threatening subtype of type 1 diabetes. The concurrence of FT1D with myocarditis is uncommon and attracts further clinical attention. Case Presentation: A 33-year-old female was transferred by a local hospital to West China Hospital because of altered consciousness, abrupt onset of hyperglycemia with ketoacidosis, significantly increased cardiac biomarkers, and ST segment elevations. Her random blood glucose at the local hospital was 50.19 mmol/L. Insulin infusion and fluid resuscitation were started immediately before referral. On admission, her random blood glucose was 14.17 mmol/L. HbA1C and glycosylated albumin (GA) were 6.3% and 21.45%, respectively. Her fasting C-peptide level was 0.022 nmol/L. Anti-Glutamic Acid Decarboxylase (anti-GAD) antibody was 25.06 IU/mL. FT1D was diagnosed based on the 2012 New Diagnosis Criteria of FT1D. Electrocardiogram showed significant ST segment elevation in leads II, III, aVF, and V3-V6. Echocardiography revealed a mildly reduced left ventricular ejection fraction (LVEF) of 46%. Coronary angiography displayed no abnormality. Cardiac magnetic resonance imaging revealed areas of increased signal intensity in the interventricular septum, basal and mid inferolateral walls, and apical inferior wall and subepicardial late gadolinium enhancement (LGE), particularly in the lateral aspects of the left ventricle on T2-weighted imaging (T2WI). Acute myocarditis was diagnosed based on the European Society of Cardiology 2013 Task Force Criteria. She was treated with insulin, fluid resuscitation, and supportive care, leading to rapid recovery of ketoacidosis and cardiac function. At the four-month follow-up, she remained on insulin therapy with good glycemic control but persistent low C-peptide levels. Conclusion: This case report raises awareness about FT1D, determines the differential diagnosis of acute cardiac presentations in an FT1D patient, and highlights clinical reasoning so that clinicians can recognize and manage similar presentations on time. Full article

Background: Radiomics has emerged as a promising approach to non-invasively characterize the molecular landscape of gliomas, providing quantitative, high-dimensional data derived from routine MRI. Given the recent shift toward molecularly driven classification, radiomics may support precision oncology by predicting key genomic, epigenetic, and phenotypic alterations without the need for invasive tissue sampling. This systematic review aimed to synthesize current radiomics applications for the non-invasive prediction of molecular biomarkers in gliomas, evaluating methodological trends, performance metrics, and translational readiness. Methods: This review followed the PRISMA 2020 guidelines. A systematic search was conducted in PubMed, Ovid MEDLINE, and Scopus on 10 January 2025, and updated on 1 February 2025, using predefined MeSH terms and keywords related to glioma, radiomics, machine learning, deep learning, and molecular biomarkers. Eligible studies included original research using MRI-based radiomics to predict molecular alterations in human gliomas, with reported performance metrics. Data extraction covered study design, cohort size, MRI sequences, segmentation approaches, feature extraction software, computational methods, biomarkers assessed, and diagnostic performance. Methodological quality was evaluated using the Radiomics Quality Score (RQS), Image Biomarker Standardization Initiative (IBSI) criteria, and Newcastle–Ottawa Scale (NOS). Due to heterogeneity, no meta-analysis was performed. Results: Of 744 screened records, 70 studies met the inclusion criteria. A total of 10,324 patients were included across all studies (mean 140 patients/study, range 23–628). The most frequently employed MRI sequences were T2-weighted (59 studies, 84.3%), contrast-enhanced T1WI (53 studies, 75.7%), T1WI (50 studies, 71.4%), and FLAIR (48 studies, 68.6%); diffusion-weighted imaging was used in only 7 studies (12.8%). Manual segmentation predominated (52 studies, 74.3%), whereas automated approaches were used in 13 studies (18.6%). Common feature extraction platforms included 3D Slicer (20 studies, 28.6%) and MATLAB-based tools (17 studies, 24.3%). Machine learning methods were applied in 47 studies (67.1%), with support vector machines used in 29 studies (41.4%); deep learning models were implemented in 27 studies (38.6%), primarily convolutional neural networks (20 studies, 28.6%). IDH mutation was the most frequently predicted biomarker (49 studies, 70%), followed by ATRX (27 studies, 38.6%), MGMT methylation (8 studies, 11,4%), and 1p/19q codeletion (7 studies, 10%). Reported AUC values ranged from 0.80 to 0.99 for IDH, approximately 0.71–0.953 for 1p/19q, 0.72–0.93 for MGMT, and 0.76–0.97 for ATRX, with deep learning or hybrid pipelines generally achieving the highest performance. RQS values highlighted substantial methodological variability, and IBSI adherence was inconsistent. NOS scores indicated high-quality methodology in a limited subset of studies. Conclusions: Radiomics demonstrates strong potential for the non-invasive prediction of key glioma molecular biomarkers, achieving high diagnostic performance across diverse computational approaches. However, widespread clinical translation remains hindered by heterogeneous imaging protocols, limited standardization, insufficient external validation, and variable methodological rigor. Full article

This study explored radiomic features that help identify non-contrast-enhancing tumors (nCET) by analyzing regions where contrast-enhancing tumors (CET) transformed into nCET after Bevacizumab (BEV) treatment. The BEV cohort included 24 recurrent GBM (rGBM) patients treated with BEV, showing reduced contrast-enhancement on gadolinium-enhanced T1-weighted imaging (T1Gd) imaging. The 11C-methionine positron emission tomography (Met-PET) cohort consisted of 24 newly diagnosed GBM (nGBM) patients with available Met-PET data. VOIs were created from T2WI, FLAIR, T1Gd, and Met-PET to analyze nCET and T2/FLAIR lesions. After significant radiomic features were identified, a prediction model for nCET was developed in the BEV cohort and subsequently evaluated in the Met-PET cohort. A total of 37 and 46 significant radiomic features were found in the BEV and Met-PET cohorts, respectively. The key feature, T2WI_whole_GLCMcorrelation_1, was selected for predictive modeling. The model demonstrated high accuracy (AUC = 0.93, p < 0.0001) in the BEV cohort, with sensitivity and specificity of 0.91, while the Met-PET cohort showed moderate accuracy (AUC = 0.74, p = 0.0053). Image reconstruction using these features also effectively visualized nCET in nGBM. These findings suggest that radiomic features in CET regions transforming to nCET after BEV treatment harbors valuable information for identifying nCET in GBM. Full article

Background: Pancreatic hamartoma (PH) is an exceptionally rare, benign, mass-forming lesion accounting for less than 1% of all pancreatic tumors. Its rarity and non-neoplastic nature contribute to significant diagnostic challenges, often leading to misclassification as malignant disease. This study presents a case of PH and a systematic review of all reported cases, with emphasis on histopathological and imaging characteristics. Methods: A comprehensive electronic search of PubMed, Scopus, and Web of Science was conducted up to 1 April 2025, to identify eligible case reports and series. Results: We describe a 37-year-old woman with a cystic lesion of the pancreatic tail, ultimately confirmed histologically as a cystic pancreatic hamartoma following distal pancreatectomy with splenectomy, with an uneventful postoperative course. Of 687 screened studies, 51 met the inclusion criteria, comprising 77 cases (68 adults, 9 pediatric). PHs occurred most frequently in males (52.9%), with a mean age of 59.5 ± 12.9 years, and were often asymptomatic (57.4%). The pancreatic head was the most common site (52.9%). On MRI, PHs typically exhibited low T1-weighted and high T2-weighted signal intensity, with no FDG uptake (82%) and moderate or no restriction on DWI, distinguishing them from neuroendocrine tumors (NETs). Histologically, most lesions were solid (64.7%) or solid–cystic (35.3%), with low spindle cell cellularity and absent Langerhans islets. Conclusions: Low T1WI signal and moderate DWI signal are the key features distinguishing PHs from NETs. Incorporating these findings with EUS-FNA and immunohistochemistry can support a provisional diagnosis and help avoid unnecessary radical surgery. Full article

Background/Objectives: Asymptomatic meningiomas require frequent follow-up using MR imaging, with the standard of care being contrast-enhanced T1-weighted imaging (CE-T1WI) with Gadolinium-Based Contrast Agents (GBCAs). Limiting GBCA exposure reduces the environmental impact and limits possible gadolinium deposition in the brain. Therefore, the research objective was investigating the diagnostic accuracy of T2WI for evaluating significant meningioma growth (≥10% per year), using the CE-T1WI as reference standard. Methods: A total of 99 asymptomatic patients with the radiological diagnosis of meningioma and a minimum follow-up period of 11 months were retrospectively identified. Patients were scanned with various scanners in multiple hospitals. The maximum tumor diameter was measured in the transverse plane. Tumor growth was calculated in changes in millimeters and converted to percentages in the longest tumor diameter in the transverse plane. A paired-sample t-test was used to compare tumor growth on T2WI to CE-T1WI. The diagnostic accuracy of T2WI was determined by calculation of sensitivity, specificity, and positive and negative predictive values. Results: Mean follow-up time was 1.9 years. Significant tumor growth was found in 16 patients using T2WI, compared to 10 patients using CE-T1WI, which was not statistically significant. T2WI had a sensitivity of 80%, specificity of 90%, a positive predictive value of 47%, and negative predictive value of 98% for prediction of significant meningioma growth. Conclusions: T2WI was not inferior to CE-T1WI in the detection of significant tumor growth in asymptomatic meningioma and therefore can be used in follow-up to reduce gadolinium exposure. Full article

Objectives: Brucella spondylitis (BS) and tuberculosis spondylitis (TS) are two causes of infection that share overlapping clinical and imaging features, complicating diagnoses. Early differentiation is critical, as treatment regimens differ significantly. This study aims to develop a deep learning framework using multimodal computed tomography (CT) and magnetic resonance imaging (MRI) data to accurately distinguish between these two conditions, improving diagnostic accuracy and patient outcomes. Methods: In this study, imaging data were acquired from two centers using different MRI and CT protocols. Sagittal T1-weighted (T1WI) and T2-weighted imaging (T2WI), fat-suppression sequences (T2WI FSE), and sagittal CT data were collected. Image preprocessing included region of interest (ROI) segmentation, and normalization and augmentation techniques were used. A deep learning model, based on pre-trained GoogleNet architectures, was trained and evaluated against human radiologists using metrics including accuracy, sensitivity, and AUC to assess diagnostic performance. Results: In this study, the GoogleNet deep learning model outperformed other architectures in classifying TS and BS, achieving AUCs of 95.97%, 91.24%, and 81.25% across training, test, and external validation datasets, respectively. In contrast, ResNet, DenseNet, and EfficientNet models showed lower AUC values. GoogleNet also demonstrated high accuracy (90.77% training, 83.04% test) and 90.91% sensitivity and 61.11% specificity in external validation. When compared to three radiologists, GoogleNet outperformed in diagnostic accuracy and speed, achieving an AUC of 88.01% and processing cases in 0.001 min. These findings highlight the potential of AI to enhance diagnostic performance and efficiency. Lastly, the explanation provided by the Grad-Cam model precisely localized major lesions. Conclusions: This multimodal-imaging-based deep learning model could well differentiate TS and BS. Deep learning does not need manual feature extraction, selection, or model development, and has great potential in daily clinical practice. Full article

Objective: To develop and validate a radiomics-based MRI model for prediction of diffuse-type tenosynovial giant cell tumor (D-TGCT), which has higher postoperative recurrence and more aggressive behavior than localized-type (L-TGCT). The study was conducted under the hypothesis that MRI-based radiomics models can predict D-TGCT with diagnostic performance significantly greater than chance level, as measured by the area under the receiver operating characteristic (ROC) curve (AUC) (null hypothesis: AUC ≤ 0.5; alternative hypothesis: AUC > 0.5). Materials and Methods: This retrospective study included 84 patients with histologically confirmed TGCT (54 L-TGCT, 30 D-TGCT) who underwent preoperative MRI between January 2005 and December 2024. Tumor segmentation was manually performed on T2-weighted (T2WI) and contrast-enhanced T1-weighted images. After standardized preprocessing, 1691 radiomic features were extracted, and feature selection was performed using minimum redundancy maximum relevance. Multivariate logistic regression (MLR) and random forest (RF) classifiers were developed using a training cohort (n = 52) and tested in an independent test cohort (n = 32). Model performance was assessed AUC, sensitivity, specificity, and accuracy. Results: In the training set, D-TGCT prevalence was 32.6%; in the test set, it was 40.6%. The MLR model used three T2WI features: wavelet-LHL_glszm_GrayLevelNonUniformity, wavelet-HLL_gldm_LowGrayLevelEmphasis, and square_firstorder_Median. Training performance was high (AUC 0.94; sensitivity 75.0%; specificity 90.9%; accuracy 85.7%) but dropped in testing (AUC 0.60; sensitivity 62.5%; specificity 60.6%; accuracy 61.2%). The RF classifier demonstrated more stable performance [(training) AUC 0.85; sensitivity 43.8%; specificity 87.9%; accuracy 73.5% and (test) AUC 0.73; sensitivity 56.2%; specificity 72.7%; accuracy 67.3%]. Conclusions: Radiomics-based MRI models may help predict D-TGCT. While the MLR model overfitted, the RF classifier demonstrated relatively greater robustness and generalizability, suggesting that it may support clinical decision-making for D-TGCT in the future. Full article

Background: Radiomics offers quantitative, high-dimensional data from conventional imaging and holds promise for improving diagnosis and treatment of pituitary adenomas (PAs). This systematic review aimed to synthesize current clinical applications of radiomics in PAs, focusing on diagnostic, predictive, and prognostic modeling. Methods: This review followed the PRISMA 2020 guidelines. A systematic search was performed in PubMed, Scopus, and Web of Science on 10 January 2024, and updated on 5 March 2024, using predefined keywords and MeSH terms. Studies were included if they evaluated radiomics-based models using MRI for diagnosis, classification, consistency, invasiveness, treatment response, or recurrence in human PA populations. Data extraction included study design, sample size, MRI sequences, feature types, machine learning algorithms, and model performance metrics. Study quality was assessed via the Newcastle-Ottawa Scale. Descriptive statistics summarized study characteristics; no meta-analysis was performed due to heterogeneity. Results: Out of 341 identified articles, 49 studies met inclusion criteria, encompassing a total of more than 9350 patients. The majority were retrospective (43 studies, 88%). MRI sequences used included T2-weighted imaging (35 studies, 71%), contrast-enhanced T1WI (34 studies, 69%), and T1WI (21 studies, 43%). PyRadiomics was the most common feature extraction tool (20 studies, 41%). Machine learning was employed in 43 studies (88%), predominantly support vector machines (16 studies, 33%), random forests (9 studies, 18%), and logistic regression (9 studies, 18%). Deep learning methods were applied in 17 studies (35%). Regarding diagnostic performance, 22 studies (45%) reported an (AUC) ≥0.85 in test datasets. External validation was performed in only 6 studies (12%). Radiomics applications included histological subtype prediction (14 studies, 29%), surgical outcome prediction (13 studies, 27%), invasiveness assessment (7 studies, 15%), tumor consistency evaluation (8 studies, 16%), and response to medical or radiotherapy treatments (3 studies, 6%). One study (2%) addressed automated segmentation and volumetry. Conclusions: Radiomics enables high-performance, noninvasive prediction of PA subtypes, consistency, invasiveness, treatment response, and recurrence, with 22 studies (45%) reporting AUC ≥0.85. Despite promising results, clinical translation remains limited by methodological heterogeneity, low external validation (6 studies, 12%), and lack of standardization. Full article

Purpose: Predicting colorectal cancer liver metastasis (CRLM) is essential for prognostic assessment. This study aims to develop and validate an interpretable multimodal machine learning framework based on multiparametric MRI for predicting CRLM, and to enhance the clinical interpretability of the model through SHapley Additive exPlanations (SHAP) analysis and deep learning visualization. Methods: This multicenter retrospective study included 463 patients with pathologically confirmed colorectal cancer from two institutions, divided into training (n = 256), internal testing (n = 111), and external validation (n = 96) sets. Radiomics features were extracted from manually segmented regions on axial T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI). Deep learning features were obtained from a pretrained ResNet101 network using the same MRI inputs. A least absolute shrinkage and selection operator (LASSO) logistic regression classifier was developed for clinical, radiomics, deep learning, and combined models. Model performance was evaluated by AUC, sensitivity, specificity, and F1-score. SHAP was used to assess feature contributions, and Grad-CAM was applied to visualize deep feature attention. Results: The combined model integrating features across the three modalities achieved the highest performance across all datasets, with AUCs of 0.889 (training), 0.838 (internal test), and 0.822 (external validation), outperforming single-modality models. Decision curve analysis (DCA) revealed enhanced clinical net benefit from the integrated model, while calibration curves confirmed its good predictive consistency. SHAP analysis revealed that radiomic features related to T2WI texture (e.g., LargeDependenceLowGrayLevelEmphasis) and clinical biomarkers (e.g., CA19-9) were among the most predictive for CRLM. Grad-CAM visualizations confirmed that the deep learning model focused on tumor regions consistent with radiological interpretation. Conclusions: This study presents a robust and interpretable multiparametric MRI-based model for noninvasively predicting liver metastasis in colorectal cancer patients. By integrating handcrafted radiomics and deep learning features, and enhancing transparency through SHAP and Grad-CAM, the model provides both high predictive performance and clinically meaningful explanations. These findings highlight its potential value as a decision-support tool for individualized risk assessment and treatment planning in the management of colorectal cancer. Full article

Prostate cancer (PCa) is the most common malignancy in men. Precise grading is crucial for the effective treatment approaches of PCa. Machine learning (ML) applied to biparametric Magnetic Resonance Imaging (bpMRI) radiomics holds promise for improving PCa diagnosis and prognosis. This study investigated the efficiency of seven ML models to diagnose the different PCa grades, changing the input variables. Our studied sample comprised 214 men who underwent bpMRI in different imaging centers. Seven ML algorithms were compared using radiomic features extracted from T2-weighted (T2W) and diffusion-weighted (DWI) MRI, with and without the inclusion of Prostate-Specific Antigen (PSA) values. The performance of the models was evaluated using the receiver operating characteristic curve analysis. The models’ performance was strongly dependent on the input parameters. Radiomic features derived from T2WI and DWI, whether used independently or in combination, demonstrated limited clinical utility, with AUC values ranging from 0.703 to 0.807. However, incorporating the PSA index significantly improved the models’ efficiency, regardless of lesion location or degree of malignancy, resulting in AUC values ranging from 0.784 to 1.00. There is evidence that ML methods, in combination with radiomic analysis, can contribute to solving differential diagnostic problems of prostate cancers. Also, optimization of the analysis method is critical, according to the results of our study. Full article