Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder that affects behavior, attention, and learning. Current diagnoses rely heavily on subjective assessments, underscoring the need for objective imaging-based methods. This study aims to explore whether structural connectivity networks derived from MRI can reveal alterations associated with ADHD and support data-driven understanding. We collected brain MRI data from 947 individuals (aged 7-26 years; 590 males, 356 females, 1 unspecified) across eight centers, sourced from the Neuro Bureau ADHD-200 preprocessed dataset. Transformer-based deep learning models were used to learn relationships between different brain regions and construct structural connectivity networks. To prepare input for the model, each region was transformed into a standardized data sequence using four different strategies. The strength of connectivity between brain regions was then measured to identify structural differences related to ADHD. Five-fold cross-validation and statistical analyses were used to evaluate model robustness and group differences, respectively. Here we show that the proposed method performs well in distinguishing ADHD individuals from healthy controls, with accuracy reaching 71.9 percent and an area under curve of 0.74. The structural networks also reveal significant differences in connectivity patterns (paired t-test: P = 0.81 × 10^-6), particularly involving regions responsible for motor and executive function. Notably, the importance rankings of several brain regions, including the thalamus and caudate, differ markedly between groups. This study shows that ADHD may be associated with connectivity alterations in multiple brain regions. Our findings suggest that brain structural connectivity networks built using Transformer-based methods offer a promising tool for both diagnosis and further research into brain structure.

Sparse-view computed tomography (CT) holds promise for reducing radiation exposure and enabling novel system designs. Traditional reconstruction algorithms, including Filtered Backprojection (FBP) and Model-Based Iterative Reconstruction (MBIR), often produce artifacts in sparse-view data. Deep Learning Reconstruction (DLR) offers potential improvements, but task-based evaluations of DLR in sparse-view CT remain limited. This study employs an Artificial Intelligence (AI) observer to evaluate the diagnostic accuracy of FBP, MBIR, and DLR for intracranial hemorrhage detection and classification, offering a cost-effective alternative to human radiologist studies. A public brain CT dataset with labeled intracranial hemorrhages was used to train an AI observer model. Sparse-view CT data were simulated, with reconstructions performed using FBP, MBIR, and DLR. Reconstruction quality was assessed using metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). Diagnostic utility was evaluated using Receiver Operating Characteristic (ROC) analysis and Area Under the Curve (AUC) values for One-vs-Rest and One-vs-One classification tasks. DLR outperformed FBP and MBIR in all quality metrics, demonstrating reduced noise, improved structural similarity, and fewer artifacts. The AI observer achieved the highest classification accuracy with DLR, while FBP surpassed MBIR in task-based accuracy despite inferior image quality metrics, emphasizing the value of task-based evaluations. DLR provides an effective balance of artifact reduction and anatomical detail in sparse-view CT brain imaging. This proof-of-concept study highlights AI observer models as a viable, cost-effective alternative for evaluating CT reconstruction techniques.

Fat and muscle mass are potential biomarkers of wellbeing and disease in oncology, but clinical measurement methods vary considerably. Here we evaluate the accuracy, precision and ability to track change for multiple deep learning (DL) models that quantify fat and muscle mass from abdominal MRI. Specifically, subcutaneous fat (SF), intra-abdominal fat (VF), external muscle (EM) and psoas muscle (PM) were evaluated using 15 convolutional neural network (CNN)-based and 4 transformer-based deep learning model architectures. There was negligible difference in the accuracy of human observers and all deep learning models in delineating SF or EM. Both of these tissues had excellent repeatability of their delineation. VF was measured most accurately by the human observers, then by CNN-based models, which outperformed transformer-based models. In distinction, PM delineation accuracy and repeatability was poor for all assessments. Repeatability limits of agreement determined when changes measured in individual patients were due to real change rather than test-retest variation. In summary, DL model accuracy and precision of delineating fat and muscle volumes varies between CNN-based and transformer-based models, between different tissues and in some cases with gender. These factors should be considered when investigators deploy deep learning methods to estimate biomarkers of fat and muscle mass.

Accurate diagnosis of pancreatic cystic lesions (PCLs) is essential to guide appropriate management and reduce unnecessary surgeries. Despite multiple guidelines in PCL management, a substantial proportion of patients still undergo major resections for benign cysts, and a majority of resected intraductal papillary mucinous neoplasms (IPMNs) show only low-grade dysplasia, leading to significant clinical, financial, and psychological burdens. This review highlights emerging endoscopic approaches that enhance diagnostic accuracy and support organ-sparing, minimally invasive management of PCLs. Recent studies suggest that endoscopic ultrasound (EUS) and its accessory techniques, such as contrast-enhanced EUS and needle-based confocal laser endomicroscopy, as well as next-generation sequencing analysis of cyst fluid, not only accurately characterize PCLs but are also well tolerated and cost-effective. Additionally, emerging therapeutics such as EUS-guided radiofrequency ablation (RFA) and EUS-chemoablation are promising as minimally invasive treatments for high-risk mucinous PCLs in patients who are not candidates for surgery. Accurate diagnosis of PCLs remains challenging, leading to many patients undergoing unnecessary surgery. Emerging endoscopic imaging biomarkers, artificial intelligence analysis, and molecular biomarkers enhance diagnostic precision. Additionally, novel endoscopic ablative therapies offer safe, minimally invasive, organ-sparing treatment options, thereby reducing the healthcare resource burdens associated with overtreatment.

Heart failure with preserved ejection fraction (HFpEF) poses significant diagnostic and prognostic challenges due to its clinical heterogeneity. This study proposes a multi-modal, explainable machine learning framework that integrates clinical variables and cardiac magnetic resonance (CMR)-derived features, particularly epicardial adipose tissue (EAT) volume, to improve risk stratification and outcome prediction in patients with HFpEF. A retrospective cohort of 301 participants (171 in the HFpEF group and 130 in the control group) was analyzed. Baseline characteristics, CMR-derived EAT volume, and laboratory biomarkers were integrated into machine learning models. Model performance was evaluated using accuracy, precision, recall, and F1-score. Additionally, receiver operating characteristic area under the curve (ROC-AUC) and precision-recall area under the curve (PR-AUC) were employed to assess discriminative power across varying decision thresholds. Hyperparameter optimization and ensemble techniques were applied to enhance predictive performance. HFpEF patients exhibited significantly higher EAT volume (70.9±27.3 vs. 41.9±18.3 mL, p<0.001) and NT-proBNP levels (1574 [963,2722] vs. 33 [10,100] pg/mL, p<0.001), along with a greater prevalence of comorbidities. The voting classifier demonstrated the highest accuracy for HFpEF diagnosis (0.94), with a precision of 0.96, recall of 0.94, and an F1-score of 0.95. For prognostic tasks, AdaBoost, XGBoost and Random Forest yielded superior performance in predicting adverse clinical outcomes, including rehospitalization and all-cause mortality (accuracy: 0.95). Key predictive features identified included EAT volume, right atrioventricular groove (Right AVG), tricuspid regurgitation velocity (TRV), and metabolic syndrome. Explainable models combining clinical and CMR-derived features, especially EAT volume, improve support for HFpEF diagnosis and outcome prediction. These findings highlight the value of a data-driven, interpretable approach to characterizing HFpEF phenotypes and may facilitate individualized risk assessment in selected populations.

This study aimed to evaluate the diagnostic performance of ChatGPT-4o, a large language model developed by OpenAI, in challenging cases of oral and maxillofacial diseases presented in the <i>Clinicopathologic Conference</i> section of the journal <i>Oral Surgery</i>, <i>Oral Medicine</i>, <i>Oral Pathology</i>, <i>Oral Radiology</i>. A total of 123 diagnostically challenging oral and maxillofacial cases published in the aforementioned journal were retrospectively collected. The case presentations, which included detailed clinical, radiographic, and sometimes histopathologic descriptions, were input into ChatGPT-4o. The model was prompted to provide a single most likely diagnosis for each case. These outputs were then compared to the final diagnoses established by expert consensus in each original case report. The accuracy of ChatGPT-4o was calculated based on exact diagnostic matches. ChatGPT-4o correctly diagnosed 96 out of 123 cases, achieving an overall diagnostic accuracy of 78%. Nevertheless, even in cases where the exact diagnosis was not provided, the model often suggested one of the clinically reasonable differential diagnoses. ChatGPT-4o demonstrates a promising ability to assist in the diagnostic process of complex maxillofacial conditions, with a relatively high accuracy rate in challenging cases. While it is not a replacement for expert clinical judgment, large language models may offer valuable decision support in oral and maxillofacial radiology, particularly in educational or consultative contexts. Not applicable.

Glioma, the most common primary intracranial tumor, poses significant challenges to precision diagnosis and treatment due to its heterogeneity and invasiveness. With the introduction of the 2021 WHO classification standard based on molecular biomarkers, the role of imaging in non-invasive subtyping and therapeutic monitoring of gliomas has become increasingly crucial. While conventional MRI shows limitations in assessing metabolic status and differentiating tumor recurrence, positron emission tomography (PET) combined with radiomics and artificial intelligence technologies offers a novel paradigm for precise diagnosis and treatment monitoring through quantitative extraction of multimodal imaging features (e.g., intensity, texture, dynamic parameters). This review systematically summarizes the technical workflow of PET radiomics (including tracer selection, image segmentation, feature extraction, and model construction) and its applications in predicting molecular subtypes (such as IDH mutation and MGMT methylation), distinguishing recurrence from treatment-related changes, and prognostic stratification. Studies demonstrate that amino acid tracers (e.g., ¹⁸F-FET, ¹¹C-MET) combined with multimodal radiomics models significantly outperform traditional parametric analysis in diagnostic efficacy. Nevertheless, current research still faces challenges including data heterogeneity, insufficient model interpretability, and lack of clinical validation. Future advancements require multicenter standardized protocols, open-source algorithm frameworks, and multi-omics integration to facilitate the transformative clinical translation of PET radiomics from research to practice.

Traditional echocardiographic parameters such as ejection fraction (EF) and global longitudinal strain (GLS) have limitations in the early detection of cardiac dysfunction. EF often remains normal despite underlying pathology, and GLS is influenced by load conditions and vendor variability. There is a growing need for reproducible, interpretable, and operator-independent parameters that capture subtle and global cardiac functional alterations. We introduce the Acoustic Index, a novel AI-derived echocardiographic parameter designed to quantify cardiac dysfunction from standard ultrasound views. The model combines Extended Dynamic Mode Decomposition (EDMD) based on Koopman operator theory with a hybrid neural network that incorporates clinical metadata. Spatiotemporal dynamics are extracted from echocardiographic sequences to identify coherent motion patterns. These are weighted via attention mechanisms and fused with clinical data using manifold learning, resulting in a continuous score from 0 (low risk) to 1 (high risk). In a prospective cohort of 736 patients, encompassing various cardiac pathologies and normal controls, the Acoustic Index achieved an area under the curve (AUC) of 0.89 in an independent test set. Cross-validation across five folds confirmed the robustness of the model, showing that both sensitivity and specificity exceeded 0.8 when evaluated on independent data. Threshold-based analysis demonstrated stable trade-offs between sensitivity and specificity, with optimal discrimination near this threshold. The Acoustic Index represents a physics-informed, interpretable AI biomarker for cardiac function. It shows promise as a scalable, vendor-independent tool for early detection, triage, and longitudinal monitoring. Future directions include external validation, longitudinal studies, and adaptation to disease-specific classifiers.

Deep learning has revolutionized neuroimage analysis by delivering unprecedented speed and accuracy. However, the narrow scope of many training datasets constrains model robustness and generalizability. This challenge is particularly acute in magnetic resonance imaging (MRI), where image appearance varies widely across pulse sequences and scanner hardware. A recent domain-randomization strategy addresses the generalization problem by training deep neural networks on synthetic images with randomized intensities and anatomical content. By generating diverse data from anatomical segmentation maps, the approach enables models to accurately process image types unseen during training, without retraining or fine-tuning. It has demonstrated effectiveness across modalities including MRI, computed tomography, positron emission tomography, and optical coherence tomography, as well as beyond neuroimaging in ultrasound, electron and fluorescence microscopy, and X-ray microtomography. This tutorial paper reviews the principles, implementation, and potential of the synthesis-driven training paradigm. It highlights key benefits, such as improved generalization and resistance to overfitting, while discussing trade-offs such as increased computational demands. Finally, the article explores practical considerations for adopting the technique, aiming to accelerate the development of generalizable tools that make deep learning more accessible to domain experts without extensive computational resources or machine learning knowledge.

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic interpretability, particularly through the use of sparse autoencoders (SAEs), offers a promising approach for uncovering human-interpretable features within large transformer-based models. In this study, we apply Matryoshka-SAE to the radiology-specialised multimodal large language model, MAIRA-2, to interpret its internal representations. Using large-scale automated interpretability of the SAE features, we identify a range of clinically relevant concepts - including medical devices (e.g., line and tube placements, pacemaker presence), pathologies such as pleural effusion and cardiomegaly, longitudinal changes and textual features. We further examine the influence of these features on model behaviour through steering, demonstrating directional control over generations with mixed success. Our results reveal practical and methodological challenges, yet they offer initial insights into the internal concepts learned by MAIRA-2 - marking a step toward deeper mechanistic understanding and interpretability of a radiology-adapted multimodal large language model, and paving the way for improved model transparency. We release the trained SAEs and interpretations: https://huggingface.co/microsoft/maira-2-sae.