Search Results (1,971)

Radiomics has the potential to improve risk stratification in head and neck squamous cell carcinoma (HNSCC), but clinical adoption is limited by inconsistent performance across institutions. A key source of variability is how radiomic features are generated, preprocessed, and selected prior to model development. This multicenter study evaluated how radiomics parameterization and feature selection strategies affect external model performance, feature stability, and time-to-event risk stratification. We studied pre-treatment CT scans from 752 patients with primary HNSCC from three hospitals. For each scan, 1648 radiomic features were computed using 20 different preparation methods that varied in scaling, outlier removal, and gray-level bin width. We compared five feature selection methods: Graph-FS with connected components, Boruta, Lasso, RFE-RF, and mRMR. The classification models used were Random Forest, XGBoost, CatBoost, and Logistic Regression. We measured performance using external ROC-AUC, bootstrap confidence intervals, Brier score, and RobustScore. Stability of feature selection was assessed using the Kuncheva and Jaccard indices. Cox proportional hazards models confirmed time-to-event results, and consensus SHAP analysis helped explain the models. Radiomics parameterization influenced model performance, and no single configuration was optimal across all analyses. Radiomics-only models outperformed clinical-only models, while clinical–radiomics models achieved the highest overall performance. mRMR and Lasso produced the highest average external AUCs, while Graph-FS showed the greatest stability. The best classification model achieved an external AUC of 0.817. In Cox validation, the best clinical–radiomics configuration achieved an external C-index of 0.662 and separated high- and low-risk patients in the external cohort. Full article

Background/Objectives: Oral tongue squamous cell carcinoma (OTSCC) remains clinically challenging because conventional clinicopathological markers do not fully explain variability in recurrence and survival. This systematic review and functional meta-synthesis aimed to identify and critically appraise studies using preoperative magnetic resonance imaging (MRI)-based radiomics, artificial intelligence (AI), deep learning, or quantitative MRI-derived models to predict recurrence and prognostic outcomes in OTSCC. Methods: PubMed, Scopus, and Embase were searched from inception to March 2026. Eligible studies included prognostic model investigations in adults with OTSCC or primary tongue cancer without reported base-of-tongue/oropharyngeal involvement, undergoing preoperative MRI and surgery, with recurrence- or survival-related follow-up. The primary synthesis was a functional meta-synthesis; pooling was not performed because studies were not sufficiently comparable. Results: Seven retrospective studies were included, with a summed descriptive sample of 1287 participants. The evidence base was heterogeneous in MRI sequences, segmentation workflows, model architecture, validation strategy, and endpoint definition. Functional meta-synthesis identified four domains: direct recurrence-oriented modeling, broader prognostic stratification, reported incremental or complementary value over clinical frameworks, and translational maturity/technical implementation. Several studies reported associations between MRI-derived signatures and recurrence- or survival-related outcomes, but findings were interpreted narratively because of differences in primary endpoints, imaging features, model design, validation methods, and outcome definitions. Most studies were judged at high overall risk of bias, and certainty of evidence ranged from low to very low. Conclusions: MRI-based radiomics and AI show preliminary promise for prognostic stratification in OTSCC, particularly recurrence-related risk refinement, but current evidence remains limited by retrospective design, heterogeneity, sparse external validation, and low certainty. Full article

Pathological complete response (pCR) after neoadjuvant chemotherapy (NACT) provides an endpoint for treatment evaluation in breast cancer. Multi-sequence breast MRI can support pCR prediction, but routine examinations may lack usable T1-weighted or T2-weighted sequences. Many models merge radiomic and deep features by concatenation, leaving the interaction between handcrafted descriptors and learned representations weakly specified. We developed a radiomics-guided framework for pCR prediction from multi-sequence breast MRI. The model uses a multi-branch 2.5D encoder for sequence-specific features, radiomics-guided channel recalibration, and masked token fusion to aggregate available sequence tokens. We evaluated the framework on 157 patients from the I-SPY1 Trial cohort with patient-level five-fold cross-validation, fixed sequence-combination analysis, and slice-window sensitivity analysis. The full model achieved 78.4% accuracy and 0.809 AUC, compared with 75.8% accuracy and 0.788 AUC for the strongest channel-concatenation baseline. In this cohort, radiomics-guided multi-sequence learning was feasible, with external validation required before clinical interpretation. Full article

Batch effects arising from variations in hardware, acquisition protocols, and reconstruction parameters present a critical challenge in radiomics, limiting the generalizability of models across multicentre studies. Existing harmonization methods, such as ComBat, CovBat, z-score normalization, and Generative Adversarial Networks, exhibit significant limitations when applied to high-dimensional radiomic data. ComBat assumes a linear feature space and tends to leave residual center-specific information recoverable by downstream classifiers. This paper introduces RiTex (Riemannian Texture Harmonization), a framework that solves a generalized eigenvalue problem between class-aware biological scatter and Ledoit–Wolf-regularized per-batch covariances, with the SPD-manifold Fréchet mean used as a principled averaging step. We evaluate RiTex on the 50-dataset radMLBench benchmark and on a new four-center head-and-neck benchmark with known center labels (n = 380 patients, k = 4 centers from TCIA: HGJ, MDACC, Maastro, QIN). On radMLBench, RiTex reduces the batch auto-detection AUC in 48/50 (96%) datasets, 42/50 (84%) reductions remain significant after Benjamini–Hochberg correction; the mean Batch AUC reduction is ΔBatch = −0.365 (95% bootstrap CI [−0.418, −0.312]), with no significant degradation in biological AUC (mean ΔBio = +0.018, 95% CI [−0.011, +0.047]). On the H&N benchmark with real center labels, RiTex reduces the Batch AUC from 0.74 to 0.59, while ComBat and CovBat leave it at ≈0.98. A component-wise ablation shows that the dominant source of empirical performance is the GEVD step, together with Ledoit–Wolf shrinkage. The SPD Fréchet mean acts as a theoretical scaffold with a negligible empirical contribution (ΔBatch AUC = −0.014 vs. arithmetic mean). Full article

Chemoradiotherapy plays an important role in the management of locally advanced head and neck squamous cell carcinoma. Unfortunately, a substantial fraction of patients experience treatment failure, while others suffer from significant treatment-related toxicity caused by intensive chemoradiotherapy regimens. This underscores the need for new biomarkers that can accurately capture the biological tumor heterogeneity and guide personalized therapy. Functional imaging combined with AI-based approaches such as radiomics and deep learning may offer a promising strategy for treatment stratification. However, a substantial number of challenges remain before clinical implementation can be achieved. Therefore, this review proposes a biology-driven framework for AI analysis of functional imaging in head and neck squamous cell carcinoma. In addition, it emphasizes the need for clinically oriented validation strategies to facilitate the translation of stratification models into clinical management. Full article

Advanced diffusion MRI techniques, particularly intravoxel incoherent motion (IVIM) and diffusion tensor imaging (DTI), have emerged as promising functional imaging tools for improving cervical cancer assessment beyond conventional anatomical MRI. This narrative review summarizes current evidence on the clinical utility of these diffusion-based techniques for tumor characterization, local staging, parametrial invasion, lymph node evaluation, treatment response monitoring, and emerging radiomics applications. Across studies, diffusion-related parameters, especially the apparent diffusion coefficient (ADC) and pure molecular diffusion coefficient (D), tend to be lower in malignant cervical tissues and correlate with increased cellularity, higher tumor grade, and more aggressive disease features. IVIM metrics appear especially useful for differentiating cervical cancer from normal tissue, predicting pelvic lymph node involvement, and detecting early treatment response to chemoradiotherapy or neoadjuvant chemotherapy before substantial morphological regression occurs. In contrast, DTI remains less extensively investigated; however, preliminary findings suggest potential value for evaluating parametrial invasion, stromal disruption, tumor grade, and lymph node metastasis, particularly when integrated with IVIM-derived indices. Although diffusion-derived radiomics may further support risk stratification and treatment-response prediction, the evidence base remains limited by small cohorts, single-center designs, methodological heterogeneity, and insufficient external validation. Overall, IVIM and DTI provide valuable non-invasive insight into cervical cancer biology, but standardized acquisition protocols, reproducible thresholds, and multicenter validation are needed before routine clinical implementation. Full article

Artificial intelligence (AI) is reshaping ultrasound diagnosis by converting operator-dependent grayscale, Doppler, elastography, contrast-enhanced, automated-volume, and video data into reproducible decision support. In breast ultrasound, the most mature evidence involves benign–malignant lesion classification, BI-RADS risk stratification, reduction in unnecessary biopsy in selected low-risk lesions, assistance for less experienced readers, automated breast volume scanning, video-based assessment, axillary staging, and prediction of biologic markers such as molecular subtype, HER2 status, Ki-67 expression, lymphovascular invasion, and nodal metastasis. AI does not replace sonographers, radiologists, pathologists, or clinical judgment; rather, it can standardize feature extraction, prompt second-reader review, quantify uncertainty, and integrate imaging with clinical context. This review summarizes current clinical applications of AI in ultrasound diagnosis, which has a strong recent multicenter evidence base. It also discusses implementation requirements, including standardized acquisition, external validation, calibration, imaging–pathology concordance, workflow integration, data security, and equity across scanners and patient populations. Full article

Background/Objectives: Although the Spinal Instability Neoplastic Score (SINS) is widely used to estimate spinal metastases fracture risk and guide decisions on stabilisation procedures, prior studies have demonstrated mixed results. Patients with the same score exhibit clinically heterogeneous outcomes, with some SINS criteria correlating less well with the estimated fracture risk than others. There are also barriers to implementation such as the time burden required for manual calculation and interobserver variability associated with qualitative morphological criteria. SINS also lacks sensitivity for detecting latent structural compromise in treatment-naive patients and those susceptible to the iatrogenic effects of stereotactic body radiation therapy. This review aims to evaluate emerging imaging, biomechanical, and microstructural markers with the potential to improve fracture risk stratification and prognostication for spinal oncology patients. Methods: We synthesise evidence across three innovative frontiers: (1) biomechanical modelling, including CT-derived finite element analysis and failure-load pattern models; (2) radiomics, utilizing radiomics features from radiological imaging to develop a predictive model; and (3) microstructural MRI biomarkers, exploring the translatability of the Vertebral Bone Quality score, fat fraction, and paraspinal muscle atrophy from osteoporosis to the metastatic spine. Results: Emerging biomechanical, radiomic and microstructural imaging markers show potential in addressing some limitations of traditional SINS criteria for fracture risk stratification across the spinal oncology treatment continuum, from initial diagnosis to post-radiation surveillance, thereby facilitating more precise risk assessment. However, current evidence remains largely retrospective and heterogeneous, and further validation is required before clinical adoption. Conclusions: We propose a framework that shifts the paradigm from conventional morphological scoring toward a multiparametric assessment of spinal stability. Full article

Background/Objectives: Differentiating mediastinal lymphadenopathy associated with lung cancer from sarcoidosis remains a clinical challenge because of overlapping imaging findings. This study evaluated whether CT-derived radiomic features, alone and in combination with clinical variables, could support the non-invasive differentiation of these two entities. Methods: In this retrospective single-center study, 204 histopathologically confirmed mediastinal lymph nodes were analyzed. A total of 107 radiomic features were extracted from manually segmented contrast-enhanced thoracic CT images. Multiple feature selection methods, dimensionality reduction techniques, and machine learning classifiers were evaluated using patient-level five-fold cross-validation. Additional clinical-only, combined clinical–radiomic, one-node-per-patient sensitivity, and exploratory interobserver feature stability analyses were performed. Results: Among radiomics-only models, LASSO achieved the highest ROC–AUC of 0.9108, whereas ElasticNet achieved the highest accuracy of 81.20%. The clinical-only ensemble model using age, sex, and smoking status showed strong performance, with an accuracy of 94.92% and an ROC–AUC of 0.9733. Selected combined clinical–radiomic models showed numerically higher performance; the combined correlation-filtered ensemble model achieved the highest accuracy of 97.78% and an ROC–AUC of 1.0000. Clinical integration also yielded more compact feature subsets in some methods, as combined LASSO selected 9.6 variables while improving ROC–AUC from 0.9108 to 0.9667 compared with radiomics-only LASSO. In the one-node-per-patient sensitivity analysis, clinical-only and combined models retained high performance, whereas radiomics-only LASSO showed reduced performance. Exploratory interobserver analysis showed moderate reproducibility for only a subset of radiomic features. Conclusions: CT-derived radiomic features may provide complementary information for differentiating mediastinal lymphadenopathy associated with lung cancer from that associated with sarcoidosis. However, clinical variables were highly informative, and the incremental value of radiomics should be interpreted cautiously. Further multicenter studies with external validation, standardized segmentation protocols, and clinically balanced cohorts are required before routine clinical implementation can be recommended. Full article

Orofacial pain (OFP) includes a broad spectrum of odontogenic and non-odontogenic conditions with overlapping clinical features that often limit diagnostic accuracy, driving increasing interest in artificial intelligence (AI) as a tool to enhance diagnostic precision and support clinical decision-making. A narrative review was conducted using PubMed/MEDLINE, Scopus, and Web of Science to identify studies (2016–2026) applying AI to the diagnosis, classification, or prediction of OFP in adults. Eligible studies reported at least two diagnostic performance metrics and were thematically grouped into odontogenic and non-odontogenic categories, the latter including musculoskeletal, neurovascular, and neuropathic pain. Twenty studies were included. Neurovascular pain, particularly migraine, showed the most consistent and highest diagnostic performance, likely due to the greater availability of structured clinical data and standardized diagnostic criteria. Musculoskeletal pain, especially temporomandibular disorders, also demonstrated high and reproducible performance. In contrast, odontogenic pain showed lower and more heterogeneous performance, with better results mainly in imaging-based models, while signal- and behavior-based approaches were less robust. Neuropathic pain exhibited moderate to high performance in selected radiomics studies, but overall results remained inconsistent due to phenotypic variability and limited objective biomarkers. Currently, AI shows promising potential in OFP diagnosis, especially for neurovascular and musculoskeletal pain, but clinical translation is limited by data heterogeneity and lack of validation. Progress in clinical practice depends on multimodal datasets and multicenter studies to ensure robust, generalizable tools. Full article

Background/Objectives: In this study, we develop and validate an interpretable machine learning (ML) model that integrates a hybrid swarm intelligence (SI)-based feature selection method with multiparametric magnetic resonance imaging (MRI)-derived RFs to estimate overall survival (OS) in glioblastoma multiforme (GBM) patients. Methods: A cohort of 276 GBM patients with open-access pre-treatment MRI data was used to perform comprehensive radiomic analysis. In the training (discovery) dataset, we employed five-fold cross-validation combined with bootstrapping to ensure robust methodological validation. Model evaluation covered the concordance index (C-index) with 95% confidence intervals (CIs). Additionally, survival stratification was performed using Kaplan–Meier curves and the log-rank test to separate patients into low- and high-risk groups for OS. The final survival model integrates patient age and ten independent RFs. Results: The model’s performance in the holdout test dataset was evaluated by a C-index of 0.71 (95% CI: 0.63–0.80), exhibiting statistically significant risk stratification (p = 3 × 10⁻⁴). Upon external validation, the model achieved a C-index of 0.67, maintaining statistical significance (p = 1 × 10⁻²). Conclusions: The research combined a traditional regularized Cox regression (Cox-LASSO) model with a new SI-based LASSO-PSO method, yielding significant stratification. To our knowledge, the present study offers one of the first studies to document the use of an interpretable ML model with an SI-based approach for successful risk stratification based on OS. Full article

Lower back pain (LBP) is a leading cause of disability globally, characterized by multifactorial origins that complicate accurate diagnosis and effective treatment planning. Artificial intelligence (AI), including machine learning (ML), deep learning (DL), and radiomics, has shown promise for improving the reproducibility and quantitative assessment of spine neuroimaging. This narrative review synthesizes current evidence on AI-derived imaging biomarkers in magnetic resonance imaging (MRI) and computed tomography (CT), with emphasis on disc degeneration, spinal stenosis, endplate signal abnormalities, paraspinal muscle composition, vertebral fractures, and spinal alignment. AI-based reconstruction, segmentation, and classification methods may reduce reader variability and enable standardized quantification of imaging features. However, the current evidence base remains dominated by technical and retrospective validation studies, and high diagnostic performance should not be interpreted as proof of improved patient-centered outcomes. The present review distinguishes technical feasibility, diagnostic assistance, prognostic association, and clinical utility, and highlights the persistent efficacy-effectiveness gap in AI-based spine imaging. Although multimodal models integrating imaging, clinical, biomechanical, and patient-reported data may improve future risk stratification, clinical translation remains constrained by heterogeneous datasets, limited external validation, incomplete interpretability, and evolving regulatory frameworks. Prospective multicenter validation and outcome-linked evaluation are required before AI-derived imaging biomarkers can be considered established tools for routine LBP management. Full article

Alzheimer’s disease (AD) is a heterogeneous neurodegenerative disorder driven by complex interactions between genetic susceptibility, molecular pathways, and progressive brain alterations. Key genetic factors, including APOE, TREM2, and MAPT, contribute to pathological processes such as amyloid-β accumulation, tau aggregation, neuroinflammation, and synaptic dysfunction. Despite substantial advances in understanding these mechanisms, translating molecular insights into clinically accessible biomarkers remains a major challenge. Radiomics and machine learning (ML) have emerged as promising approaches for extracting high-dimensional quantitative features from medical imaging data and identifying complex patterns associated with disease processes. Radiomic features capture spatial heterogeneity and subtle characteristics of neurodegeneration that are not discernible using conventional imaging analysis. When integrated with ML, these features may serve as noninvasive surrogates of molecular activity, enabling the identification of imaging signatures associated with specific genetic backgrounds and biological pathways. This review aims to explore how radiomics and ML can bridge the gap between genetic and molecular mechanisms and in vivo imaging phenotypes in AD. We summarize current knowledge on genetic determinants and molecular pathways and discuss advances in molecular imaging, particularly tracers targeting amyloid and tau pathology. Furthermore, we analyze the emerging role of radiomics and ML in linking imaging phenotypes with underlying biological processes. This integrative framework may support improved disease stratification, early diagnosis, and prediction of therapeutic response, contributing to the development of precision medicine strategies and future theranostic approaches in Alzheimer’s disease. Full article

Background/Objectives: Computed tomography (CT)-based radiomics shows promise for non-small cell lung cancer (NSCLC) prognosis prediction, but model performance varies widely by feature selection and machine learning strategies. Optimal combinations remain unclear. This study aims to systematically compare feature selection methods and machine learning algorithms for 12-month overall survival prediction using CT radiomics in NSCLC patients. Methods: We analyzed 385 patients from The Cancer Imaging Archive (TCIA) NSCLC-Radiomics dataset. Radiomic features from primary tumor volumes were combined with clinical variables. Three feature selection methods—sequential forward selection (SFS), maximum relevance minimum redundancy (mRMR), and least absolute shrinkage and selection operator (LASSO)—were compared across five classifiers: k-nearest neighbors (KNN), support vector machine (SVM), random forest (RF), logistic regression (LR), and gradient boosting classifier (GBC). Performance was assessed using area under the receiver operating characteristic curve (AUC) and accuracy on independent test sets. Cox regression and Kaplan–Meier analyses evaluated survival risk stratification. Results: Logistic regression showed the most stable classification performance across feature selection strategies (test AUC 0.60–0.65, accuracy 0.72–0.73). The mRMR-LR model achieved highest AUC (0.65); LASSO-LR showed highest accuracy (0.73). For survival analysis, LASSO-based Cox modeling demonstrated superior risk stratification with significant separation between high- and low-risk groups in both training and testing sets (p = 0.0095). Conclusions: Simpler models like logistic regression provide robust performance in CT radiomics, while LASSO excels for survival risk stratification. As we employed single-dataset validation, clinical applicability remains limited because validation was performed within a single public dataset. Nevertheless, the findings provide methodological insights into the selection of feature selection and machine learning strategies for CT radiomics-based prognostic modeling in NSCLC. Full article

Background/Objectives: The aim of this study was to develop and validate a high-frequency ultrasound radiomics-based model for quantitative assessment of metacarpophalangeal (MCP) joint cartilage damage in early rheumatoid arthritis (RA). Methods: 656 MCP joints from 99 early RA patients and 65 healthy controls were prospectively enrolled and graded according to the Outcome Measures in Rheumatology (OMERACT) system. After radiomics feature extraction, five machine learning classifiers were evaluated. Radiomics, clinical, and combined models were constructed and assessed. Radiomics scores were compared among healthy grade 0 joints, early RA grade 0 joints stratified into two risk subgroups, and RA grade ≥ 1 joints. SHapley Additive exPlanations (SHAP) analysis was used for interpretation. Results: Eight stable radiomics features were selected. Among classifiers, support vector machine achieved the highest cross-validated performance and was selected as the final radiomics classifier (validation AUC = 0.804). The combined model, integrating radiomics features with age, disease duration, and Disease Activity Score in 28 joints, achieved the best diagnostic performance (AUC = 0.855), significantly outperforming both the radiomics and clinical models. Among OMERACT grade 0 joints, the high-risk subgroup demonstrated elevated radiomics-derived scores. SHAP analysis identified original_shape2D_PerimeterSurfaceRatio as the strongest contributor. Conclusions: High-frequency ultrasound radiomics combined with clinical features demonstrated strong performance in detecting MCP joint cartilage damage in early RA and may provide a quantitative extension to conventional semiquantitative assessment. Full article