Coronary artery disease (CAD) remains a leading cause of global mortality, where precise plaque characterization is crucial for risk-stratification and intervention planning. Intravascular optical coherence tomography (IVOCT) provides high-resolution visualization of vascular structures, yet interpretation challenges hinder its broad clinical adoption. Although previous automated plaque segmentation studies have shown promising performance, automated plaque quantification remains challenging because clinically relevant tissue components are sparse, anatomically nested, affected by imaging artefacts and dependent on boundary-derived measurements. Moreover, most artificial intelligence models are evaluated against expert image annotations rather than tissue-level reference standards. To address these challenges, we propose an Enhanced Dual-segmentation Multi-decoder Network (DMNet-E). DMNet combines a mutually-exclusive multi-class decoder for accurate segmentation of small structures and a parallel multi-label binary decoder to preserve topological integrity in overlapping areas. Two task-specific components, including PizzaMix data augmentation for geometry-aware mixed-sample augmentation and Lipid Boundary Regression (LBR) module to highlight TCFA, further improve segmentation accuracy and clinical relevance. Through these designs, the model can fully harness complementary supervision signals provided by multi-class, multi-label, and regression objectives. DMNet-E achieved DSC values of 0.850, 0.706, and 0.865 for fibrous, lipid, and calcified plaques, respectively. Comparative experiments and histopathological validation jointly confirm its effectiveness and clinical relevance. Moreover, comprehensive transfer experiments further demonstrate DMNet’s generalization potential within related IVOCT segmentation tasks and across different backbone settings.
Brace treatment is the standard non-operative treatment of moderate adolescent idiopathic scoliosis (AIS). Yet reliable quantitative prediction of mid-term treatment response remains limited. The mid-term effectiveness of bracing is crucial because the risk of progression begins to increase when the curve exceeds 30° after skeletal maturity has been achieved. This study aimed to develop and externally validate a deep learning regression model to predict the mid-term Cobb angle following brace treatment using baseline and early post-brace radiographs. A retrospective multicenter cohort study was conducted including 294 patients with AIS treated with bracing at two tertiary referral centers. Standing full-spine radiographs were obtained at baseline (pre-brace), immediately post-brace and follow-ups, along with clinical variables including age, sex, and Risser stage. In cross-validation, the model achieved MAEs of 4.06° (T curve; R2 = 0.811) and 3.75° (TL/L curve; R2 = 0.746). External validation produced MAEs of 5.13° (R2 = 0.678) and 3.89° (R2 = 0.723) for T and TL/L curves, respectively. This imaging-based deep learning regression model provides precise and interpretative predictions of mid-term bracing outcomes. This approach may assist clinicians in individualized risk stratification and shared decision-making regarding brace management.
This review examines retinal image analysis from the perspective of clinical translation rather than benchmark-oriented model comparison. Instead of organizing prior work solely by disease category or model family, we synthesize recent advances through four connected dimensions: imaging modality, task taxonomy, methodological paradigm, and translational bottleneck. We compare major tasks, including classification, detection, segmentation, grading, progression prediction, and treatment-response assessment, across fundus photography, OCT/OCTA, and angiographic imaging. We further review the roles of CNNs, U-Net variants, 3D models, Transformers, graph-based methods, hybrid architectures, foundation models, self-supervised pretraining, and multimodal learning under different data conditions and clinical constraints. Beyond technical progress, we analyze why many high-performing systems still fail to translate reliably into practice, highlighting challenges related to distribution shift, label inconsistency, limited external validation, image-quality control, calibration, fairness, and workflow integration. We argue that the next stage of retinal AI requires transferable representations, robust external evaluation, trustworthy uncertainty handling, and demonstrated value within real-world care pathways, rather than incremental architectural novelty alone.
Rapid developments in the Internet of Medical Things (IoMT) technology have revolutionized healthcare by enabling real-time patient monitoring and remote diagnosis. However, increased connectivity raises significant cybersecurity concerns, such as data breaches, unauthorized access, and potential tampering with medical devices. Techniques such as threat intelligence, log monitoring, and intrusion detection systems are used to identify and respond to cyberattacks on the IoMT. As attackers evolve their strategies, they are increasingly turning to artificial intelligence to improve attack detection accuracy. This shift highlights the urgent need for smart, scalable, and privacy-preserving IDS solutions that can protect life-critical medical systems with limited computational resources. This research investigates the use of artificial intelligence (AI), specifically machine learning (ML), deep learning (DL), and hybrid techniques, to improve intrusion detection systems (IDS) for IoMT security. It includes an overview of publicly available IoMT datasets, an analysis of existing IDS frameworks, and a discussion of the limitations of AI-powered security solutions. Unlike prior surveys that mainly focus on general IoT or isolated AI methods, this study introduces a novel IoMT-specific taxonomy that integrates threat models, device capabilities, and regulatory/ethical constraints, and proposes a unified evaluation framework that balances detection accuracy with latency, resource consumption, and privacy preservation. Furthermore, current IDS approaches do not fully address regulatory requirements, real-time constraints, and dataset heterogeneity, which remain major barriers to practical IoMT deployment. In addition, this study outlines key research challenges and highlights the importance of scalable, privacy-preserving, and resilient detection systems for securing IoMT ecosystems.
Contrast-enhanced abdominal computed tomography (CT) scans frequently detect renal cysts, some of which may be malignant. Early and precise localization improves patient management and enables quantitative image analysis. However, current segmentation methods often lack interpretability at both feature and pixel levels, highlighting the need for an explainable approach that can detect and correct model errors. We developed an interpretable 3D renal cyst segmentation framework and validated it on 1486 cysts from 568 patients across three centers and an open-source dataset. The framework first isolated the kidney and then segmented renal cysts, integrating segmentation quality control via uncertainty estimation. A Variational Autoencoder Generative Adversarial Network (VAE-GAN) learned latent representations of 3D patches and reconstructed inputs. Latent-space modifications, guided by segmentation model gradients, generated counterfactual explanations corresponding to varying Dice similarity coefficients (DSC). Radiomics features extracted from counterfactual images, using ground-truth masks, were analyzed for associations with segmentation performance. The framework achieved a DSC of 0.82 on two external test sets, with sensitivities of 0.95 and 0.97, respectively. DSCs for original and VAE-GAN-reconstructed images showed no significant differences. Counterfactual explanations revealed how changes in cyst appearance influenced model outputs and exposed segmentation discrepancies. Radiomics features positively and negatively correlated with DSCs were identified. Using uncertainty estimates to flag segmentations with DSC < 0.75 reduced poor segmentations from 17% to 5% when only 20% of cases were sent for manual correction. This combination of counterfactual explanations and uncertainty maps improved segmentation interpretability and can be generalized for other applications.
