Search Results (923)

Background/Objectives: Malignant salivary gland tumors (MSGTs) are rare, biologically heterogeneous neoplasms in which histopathologic diagnosis and classification are challenging and subject to interobserver variability. Artificial intelligence (AI) approaches using radiologic, histopathologic, and molecular data, including radiomics, deep learning, and biomarker-based models, have been proposed as adjunctive diagnostic tools. This systematic review aimed to identify and critically appraise AI/ML models across radiologic, histopathologic, and molecular domains for distinct diagnostic tasks in MSGTs, and to integrate their diagnostic roles through a functional meta-synthesis. Methods: We conducted a PRISMA 2020-compliant systematic review. Embase, PubMed/MEDLINE, and Scopus were searched from inception to February 2026. Eligible studies developed or validated AI/ML diagnostic or classification models in human salivary gland tumor cohorts and reported extractable performance metrics. Results: From 1265 records, eight studies (1922 participants) met the inclusion criteria, spanning CT/MRI radiomics or deep learning (n = 4), whole-slide histopathology deep learning (n = 3), and DNA methylation-based classification (n = 1). External validation was reported in two CT-based benign–malignant discrimination studies, with AUCs of 0.890 (95% CI 0.844–0.937) and 0.745 (95% CI 0.699–0.791). Heterogeneity in model construction, outcome definitions, and validation strategies precluded meta-analysis. Risk of bias was frequently high in QUADAS-2/PROBAST assessments, driven by retrospective sampling, limited blinding, and analysis-related concerns, while calibration and utility were rarely assessed. Conclusions: AI/ML models for MSGTs demonstrate promising diagnostic performance, particularly for preoperative benign–malignant discrimination, but the current evidence base is limited by heterogeneity, predominantly internal validation, and high risk of bias. The functional meta-synthesis identified three convergent diagnostic domains: malignancy discrimination, histopathologic subtype classification, and molecular/epigenetic taxonomy refinement. Full article

Background: Thyroid cancer incidence continues to rise globally, yet current diagnostic methods, reliant on ultrasound-guided fine-needle aspiration, suffer from substantial inter-observer variability and indeterminate results. Objective: This review explores the transformative potential of integrated artificial intelligence (AI) foundation models in thyroid cancer management. We propose a paradigm shift using foundation models—large-scale, multimodal architectures pre-trained on diverse datasets—to bridge the gap between initial pixels and long-term prognostic prediction. Proposed Models: We introduce two integrated conceptual frameworks: ThyroSight-Prognos for high-precision assessment in specialized tertiary settings and SonoPredict-AI for cost-effective screening in primary care. Key Innovations: By synthesizing data from ultrasound, pathology (WSI), genomics, and clinical parameters through explainable AI (XAI), these models aim to reduce unnecessary surgeries and personalize treatment pathways. Challenges and Outlook: This paper addresses critical implementation challenges, including data heterogeneity, hardware requirements, and regulatory trust, ultimately providing a strategic blueprint for future multi-center prospective clinical validation to revolutionize thyroid care through precision oncology. Full article

Artificial intelligence (AI) in cardiovascular pathology involves the use of computational models, including machine learning and deep learning (DL), to analyse complex and heterogeneous data. These data include histopathological whole-slide images, cardiovascular imaging techniques such as cardiac magnetic resonance, echocardiography, computed tomography (CT), clinical parameters, and molecular information. The integration of these multimodal data sources allows AI to overcome the limitations of single-modality analysis, improving diagnostic accuracy, prognostic stratification, and personalised clinical decision-making while reducing inter-observer variability. Cardiovascular disease remains the leading cause of mortality worldwide, highlighting the need for more precise and timely diagnostic tools. AI has shown significant promise, particularly in digital pathology, where the digitisation of histological slides combined with advanced algorithms enables improved diagnosis, prognostic assessment, and translational research. This review summarises current AI applications in cardiovascular pathology, focusing on heart transplant rejection, cardiomyopathies, myocarditis, and atherosclerotic and valvular diseases. Automated methods offer important advantages, including diagnostic standardisation, quantitative histological analysis, and improved reproducibility. However, several challenges remain, such as the need for large, well-annotated shared datasets, limited interpretability of AI models, and ethical and legal issues related to clinical implementation. AI represents a promising tool for advancing cardiovascular pathology and personalised medicine, although robust multicentre validation is required before routine clinical adoption. Full article

Background and Objectives: Differentiation between cystic pituitary adenoma and Rathke’s cleft cyst (RCC) is clinically important because these lesions require different treatment strategies. This study aimed to develop and internally validate an MRI-based multivariable diagnostic prediction model to differentiate cystic pituitary adenoma from Rathke’s cleft cyst before treatment. Materials and Methods: A retrospective analysis was performed on 56 adult patients (27 cystic pituitary adenomas and 29 RCCs) who underwent pituitary MRI between 2019 and 2021. MRI examinations were independently evaluated for ten predefined imaging features. Diagnoses were established using histopathology or a validated clinical–radiological diagnostic algorithm. Interobserver agreement and diagnostic performance were analyzed using multivariable logistic regression, with internal validation performed using bootstrap resampling. Results: Interobserver agreement was excellent (κ (kappa) = 0.81–1.0). Fluid–fluid level, hypointense rim on T2-weighted images, septation, and paramedian coronal location were significantly associated with cystic pituitary adenoma. In contrast, spontaneous T1-weighted hyperintensity, intracystic nodule, and midline sagittal location were more frequently observed in RCC. The final multivariable model demonstrated excellent discrimination (AUC = 0.91), with stable performance after bootstrap validation (optimism-corrected AUC = 0.88). Conclusions: The proposed MRI-based multivariable prediction model demonstrated high discrimination and provides a structured approach for estimating the probability of cystic pituitary adenoma using routinely available MRI features. Such an approach may help reduce unnecessary surgical interventions in patients with Rathke’s cleft cyst while facilitating appropriate treatment planning for cystic pituitary adenomas. However, external validation in larger cohorts is required before routine clinical implementation. Full article

COVID-19 still poses a global public health challenge, exerting pressure on radiology services. Chest X-ray (CXR) imaging is widely used for respiratory assessment due to its accessibility and cost-effectiveness. However, its interpretation is often challenging because of subtle radiographic features and inter-observer variability. Although recent deep learning (DL) approaches have shown strong performance in automated CXR classification, their black-box nature limits interpretability. This study proposes an explainable deep learning framework for COVID-19 detection from chest X-ray images. The framework incorporates anatomically guided preprocessing, including lung-region isolation, contrast-limited adaptive histogram equalization (CLAHE), bone suppression, and feature enhancement. A novel four-channel input representation was constructed by combining lung-isolated soft-tissue images with frequency-domain opacity maps, vessel enhancement maps, and texture-based features. Classification was performed using a modified Xception-based convolutional neural network, while Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to provide visual explanations and enhance interpretability. The framework was evaluated on the publicly available COVID-19 Radiography Database, achieving an accuracy of 95.3%, an AUC of 0.983, and a Matthews Correlation Coefficient of approximately 0.83. Threshold optimisation improved sensitivity, reducing missed COVID-19 cases while maintaining high overall performance. Explainability analysis showed that model attention was primarily focused on clinically relevant lung regions. Full article

Medical image segmentation represents a fundamental task in medical image analysis, serving as a critical component for accurate diagnosis, treatment planning, and disease monitoring. The emergence of Denoising Diffusion Probabilistic Models (DDPMs) has revolutionized the landscape of generative modeling and recently gained significant attention in medical image analysis. This comprehensive review examines the current state of the art in diffusion models for medical image segmentation, covering theoretical foundations, methodological innovations, computational efficiency strategies, and clinical applications. We analyze recent advances in latent diffusion frameworks, transformer-based architectures, and ambiguous segmentation modeling while addressing the practical challenges of implementing these models in clinical environments. The review encompasses applications across multiple medical imaging modalities including Magnetic Resonance Imaging (MRI), Computed Tomography (CT), ultrasound, and X-ray imaging, providing insights into performance achievements and identifying future research directions. Through systematic analysis of publications mostly from 2019 to 2025, we demonstrate that diffusion models have achieved remarkable progress in addressing fundamental challenges including data scarcity, inter-observer variability, and uncertainty quantification. Notable achievements include inference time being reduced from 91.23 s to 0.34 s for echocardiogram segmentation (LDSeg, Echo dataset), DSC scores up to 0.96 for knee cartilage MRI segmentation, and a +13.87% DSC improvement over baseline methods for breast ultrasound segmentation. This review serves as a comprehensive resource for researchers and clinicians interested in leveraging diffusion models for medical image segmentation, providing a roadmap for future research and clinical translation. Full article

Background/Objectives: Medial meniscus extrusion (MME) is a quantitative marker of altered meniscal containment and load sharing. Although ultrasonography enables dynamic assessment under functional loading, it remains unclear whether generalized ligamentous hypermobility influences physiologic extrusion behavior in healthy knees. The aim of this study was to quantify load-dependent MME in healthy adults and to determine whether generalized hypermobility is associated with greater physiologic extrusion under progressive loading conditions. Methods: In this prospective observational study, 106 healthy adults aged 18–40 years were evaluated between October and December 2025. Generalized joint hypermobility was defined as a Beighton score ≥5. MME was measured by standardized ultrasonography on the dominant limb in three conditions: sitting (unloaded), bipedal stance, and unipedal stance. Patellar tendon shear-wave elastography (SWE) was recorded in kilopascals (kPa). Interobserver reliability was assessed in the first 25 participants using ICC (2,1). Group comparisons, multivariable linear regression for loading-related Δ-extrusion (Unipedal−Sitting and Bipedal−Sitting), and a linear mixed-effects model for repeated MME measures, including a Position × Hypermobility interaction, were performed. Results: Twenty-eight participants (26.4%) were classified as hypermobile. The hypermobile group showed significantly lower patellar tendon SWE than controls (23.8 ± 7.0 vs. 37.6 ± 9.7 kPa, p < 0.001). MME increased stepwise with loading in both groups and remained consistently higher in hypermobile participants across sitting, bipedal, and unipedal conditions (all p < 0.001). Loading-related extrusion was also greater in the hypermobile group for both Bipedal−Sitting (p = 0.037) and Unipedal−Sitting (p = 0.002). In multivariable regression, lower patellar tendon SWE independently predicted greater loading-related extrusion, whereas hypermobility status did not remain an independent predictor. In the mixed model, the Position × Hypermobility interaction was significant and was most pronounced during the unipedal stance. Conclusions: In healthy adults, medial meniscus extrusion increases stepwise from unloaded sitting to bipedal and unipedal weight bearing. Participants with generalized hypermobility demonstrated higher physiologic MME values and a more pronounced load-dependent pattern, particularly during unipedal stance. However, in adjusted analyses, lower patellar tendon stiffness on SWE, rather than hypermobility status itself, independently predicted greater loading-related extrusion. These findings support a contextual interpretation of ultrasound-based MME measurements in relation to loading condition and hypermobility phenotype. Full article

Background and Objectives: Brainstem infarctions remain challenging to identify due to their small size, complex anatomy, and known limitations of conventional axial diffusion-weighted imaging (DWI), particularly in the posterior fossa. Thin-section coronal DWI may improve lesion conspicuity by providing higher spatial resolution and an orthogonal imaging perspective. To evaluate whether 3 mm thin-section coronal DWI improves lesion visualization and delineation compared with standard 4 mm axial DWI in patients with MRI-confirmed acute brainstem infarction. Materials and Methods: In this retrospective single-center study, 125 consecutive patients with isolated brainstem infarction confirmed by MRI (January 2021–January 2024) were included. All patients underwent both axial and coronal DWI acquisitions. Lesions were classified by anatomical location and by the imaging plane providing better visualization (“coronal better” vs. “equal”). Lesion volumes were calculated using manual segmentation. Image interpretation was performed independently by two neuroradiologists. Interobserver agreement was assessed using Cohen’s kappa and intraclass correlation coefficient (ICC). Statistical analysis included both parametric and nonparametric tests, with confidence intervals reported. Results: Coronal DWI provided improved or equivalent lesion visualization in all cases. Improved visualization was most frequent in midbrain infarctions (100%) and in a subset of medullary lesions (26.7%). Lesions better visualized on coronal DWI were significantly smaller than those equally visualized (mean volume ~0.23 mL vs. ~0.55 mL, p < 0.0001). Twelve midbrain and eight medullary lesions were identified only on coronal DWI within the imaging protocol, all showing confirmation on ADC and/or FLAIR correlation. Interobserver agreement was substantial to excellent. Conclusions: Thin-section coronal DWI improves visualization and delineation of small brainstem infarctions, particularly in anatomically compact regions. These findings support its role as a complementary sequence rather than a replacement for standard axial imaging. Full article

Background: Endoscopic assessment is central to the management of inflammatory bowel disease (IBD), particularly within treat-to-target strategies. However, conventional high-definition white-light endoscopy (HD-WLE) is limited by interobserver variability and its inability to reliably reflect microscopic inflammation or predict long-term outcomes. Over the last decade, multiple technological innovations have reshaped the role of endoscopy in both disease activity monitoring and dysplasia surveillance. Methods: This narrative review provides a comprehensive and clinically oriented overview of emerging endoscopic technologies in IBD, including image-enhanced endoscopy, ultra-high-magnification techniques, artificial intelligence (AI), and molecular imaging. We discuss their diagnostic performance, prognostic implications, and potential integration into clinical practice. Results: Image-enhanced endoscopy improves visualization of subtle mucosal and vascular alterations and demonstrates stronger correlation with histological activity compared with HD-WLE alone. Confocal laser endomicroscopy and endocytoscopy enable in vivo microscopic assessment of epithelial architecture and barrier integrity, redefining remission beyond macroscopic healing. AI systems have shown expert-level performance in grading inflammatory severity in ulcerative colitis and high sensitivity in capsule endoscopy for Crohn’s disease, supporting objective and reproducible assessment. In surveillance, targeted high-definition inspection has replaced random biopsies, while adjunctive optical and AI-based tools enhance lesion detection and characterization. Molecular imaging introduces a predictive dimension by enabling visualization of drug–target engagement and dysplasia-specific pathways. Conclusions: Endoscopy in IBD is evolving from a descriptive modality toward a multimodal precision tool integrating enhanced imaging, AI-driven standardization, and molecular profiling. Although further validation and cost-effectiveness studies are required, these innovations have the potential to improve therapeutic stratification, surveillance strategies, and long-term patient outcomes. Full article

Background: Radiotherapy (RT) is a core modality for cancer treatment, yet it is plagued by inter-observer variability in target delineation, inefficient manual workflows, and challenges in fusing multi-type clinical data. Large language models (LLMs), with their superior semantic understanding and cross-modal fusion capabilities present novel solutions to these challenges. Scope: This narrative review provided a comprehensive overview of the current landscape and emerging trends of LLM applications across the entire RT workflow. Findings: LLMs demonstrated substantial clinical utility in key RT domains, including automated target volume delineation (e.g., Medformer, Radformer), dose prediction (e.g., DoseGNN), treatment planning automation (e.g., GPT-Plan), patient education, clinical decision support, medical information extraction, and prognosis assessment. These applications not only have the potential to enhance the accuracy and efficiency of RT but also facilitate the standardization of clinical pathways. However, widespread clinical adoption was impeded by critical limitations, including model hallucinations, insufficient generalizability, and unresolved issues regarding data privacy and ethical governance. Conclusions: LLMs possessed transformative potential to revolutionize radiation oncology. Future endeavors should prioritize technical refinements to mitigate model deficiencies, establish standardized evaluation benchmarks, and develop robust ethical frameworks. These concerted efforts are crucial for translating LLM research into clinical practice and advancing the era of intelligent, precision RT. Full article

The histopathological classification of ovarian carcinoma is fundamental for patient management. While microscopic evaluation by pathologists is the current diagnostic standard, it is known to be subject to interobserver variability, which can affect consistency in treatment decisions. This study addresses this clinical need by developing and validating a deep learning-based diagnostic support tool designed to enhance the objectivity and reproducibility of this classification. In this work, we address a key challenge in computational pathology—the tendency of attention mechanisms to overfit by concentrating on limited features—by systematically evaluating a direct regularization method within multiple instance learning (MIL) models. The models were trained and validated using 10-fold cross-validation on a public training set of 538 whole-slide images and further tested on an independent public dataset for the more challenging task of molecular subtype classification. We utilized features from a foundational model pre-trained on histopathology data to represent tissue morphology. Our findings demonstrate that directly regularizing the attention mechanism with a stochastic approach provides a statistically significant improvement in accuracy and generalization, highlighting its power as a robust technique to mitigate overfitting for this clinical task. In direct contrast to the reported variability in manual assessment, our final model achieved high consistency and accuracy, with a balanced accuracy of 0.854 and a Cohen’s Kappa of 0.791. The model also demonstrated strong generalization on the molecular classification task. Its attention mechanism provides visual heatmaps for pathologist review, fostering interpretability and trust. We have developed a highly accurate and generalizable artificial intelligence tool that directly addresses the challenge of interobserver variability in ovarian cancer classification. Its performance highlights the potential for artificial intelligence to serve as a decision support system, standardizing histopathological assessment. Full article

Background: Glioblastoma (GBM) remains the most aggressive primary brain tumor, and intraoperative frozen section analysis is the current standard for rapid histopathological assessment. However, this approach is time-consuming and resource-intensive. Stimulated Raman scattering (SRS) imaging has emerged as a label-free technique enabling near real-time microscopic evaluation of fresh tissue. This study compares the visualization of selected histopathological features in a newly developed intraoperative SRS system with conventional hematoxylin–eosin (HE) staining in confirmed GBM. Methods: Tumor samples from 30 patients with neuropathologically confirmed GBM were analyzed. For each case, both HE-stained frozen sections and SRS-generated virtual HE-like images were prepared from separate portions of the specimen. Twelve neuropathologists with varying levels of experience assessed 60 images according to seven predefined GBM criteria, resulting in 720 image evaluations. Feature detection was analyzed using cluster-adjusted generalized estimating equation models, and interobserver agreement was assessed using Fleiss’ κ. Results: Descriptively, hypercellularity and hypervascularization were identified at similar frequencies in both modalities, whereas pleomorphism, endothelial proliferation, mitotic activity, and necrosis were more often recognized in HE images. In cluster-adjusted analyses, SRS showed significantly lower detection rates for hypercellularity, pleomorphism, endothelial proliferation, and mitotic activity, while no significant difference was observed for hypervascularization, necrosis, or pseudopalisading after false discovery rate correction. Interobserver agreement was feature-dependent and generally higher for HE than SRS, particularly for hypercellularity. Conclusions: In this feature-level analysis of neuropathologically confirmed GBM, SRS imaging provided rapid, label-free morphological information and showed comparable visualization of selected histopathological features, particularly hypervascularization. While conventional HE-stained frozen sections remained superior for certain WHO-defining features, SRS represents a promising intraoperative adjunct that may complement established neuropathological workflows. Further studies including non-tumor tissue and a broader range of glioma grades are needed to determine the full diagnostic accuracy and clinical applicability of this technique. Full article

Background/Objectives: Accurate bone fracture detection is essential for orthopedic diagnosis and trauma management. Manual interpretation of X-ray or CT images can be time-consuming and may lead to inter-observer variability, particularly in subtle or multi-site fracture cases. This study proposes an interpretable Hybrid Selective Feature Network (Hybrid SFNet) to improve multi-site bone fracture detection performance and boundary localization. Methods: The proposed Hybrid SFNet extends the original SFNet architecture by incorporating multi-scale convolutional feature extraction and a semantic flow mechanism to enhance structural representation and fracture boundary delineation. Preprocessing techniques, including Canny edge detection, normalization, and data augmentation, were applied to improve feature quality. Model interpretability was addressed using Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize regions contributing to predictions. The model was evaluated on publicly available multi-site fracture datasets using both standard and class-weighted loss configurations. Results: For binary fracture classification, the proposed model achieved 90 accuracy, 94% precision, 77% recall, and an F1-score of 85% for fractured cases. When class-weighted loss was applied, recall improved to 85%, reducing false negatives from 145 to 94 cases (approximately 35%). Under the weighted configuration, Cohen’s Kappa reached 0.79 and the Matthews Correlation Coefficient (MCC) reached 0.76. Conclusions: The proposed Hybrid SFNet provides an interpretable and effective framework for multi-site bone fracture detection. The integration of multi-scale feature extraction and semantic flow mechanisms enhances detection performance and boundary localization, while Grad-CAM supports clinical interpretability. These results indicate the model’s potential for supporting clinical decision-making in orthopedic imaging. Full article

Objectives: Interstitial lung disease (ILD) is a major cause of morbidity and mortality in patients with systemic sclerosis (SSc). This study aimed to evaluate interobserver variability and the relationship between lung ultrasonography (LUS) findings and histological fibrosis severity in a bleomycin (BLM)-induced mouse model of SSc. Materials and Methods: Twenty female BALB/c mice were randomly assigned to a control group (n = 10) or a BLM-treated group (n = 10). Pulmonary fibrosis was induced by daily subcutaneous administration of BLM for three weeks. Two blinded observers (a radiologist and a rheumatologist) performed LUS using a high-frequency linear probe and calculated scores based on B-line distribution. Lung fibrosis was evaluated by Masson’s trichrome staining and quantified using the Ashcroft scoring system. Interobserver agreement was assessed with Cohen’s kappa, and correlations were analyzed using Spearman’s rank test. Results: Control mice exhibited normal lung architecture, whereas all BLM-treated mice developed moderate to severe fibrosis, with significantly higher Ashcroft scores. LUS revealed multiple B-lines, pleural irregularities, and loss of A-lines in BLM-treated mice. LUS scores were considerably higher in the BLM group (p < 0.001). Radiologist-assessed scores showed a strong correlation with Ashcroft scores (ρ = 0.78), while rheumatologist-assessed scores demonstrated a moderate correlation (ρ ≈ 0.62). Interobserver agreement was moderate, with discrepancies mainly in intermediate fibrosis stages. Conclusions: LUS is a useful non-invasive method for semiquantitative assessment of pulmonary fibrosis in this SSc model. Its correlation with histological severity supports clinical relevance, while moderate interobserver variability highlights the need for standardized protocols and training. Full article