Accurate prediction of treatment response and selection of optimal treatments remain challenging in epilepsy management. With no reliable surrogate biomarkers for treatment response, the current process of selecting an antiseizure medication remains largely a trial-and-error approach. Other non-pharmacological treatment options, such as epilepsy surgery, are viable alternatives for patients with drug-resistant epilepsy. Statistical and machine learning techniques have been used to predict seizure outcomes associated with antiseizure medications and epilepsy surgery. Recent breakthroughs in deep learning have unveiled new pathways and opportunities, potentially revolutionising personalised treatment selection in health care. In this Review, we explore a broad range of studies that have used various statistical and machine learning methodologies, with particular emphasis on state-of-the-art deep learning techniques to predict the outcomes of both pharmaceutical and surgical treatments for epilepsy. We also review potential future research trajectories and address the inherent challenges of incorporating machine learning into the clinical management of epilepsy.

Accurate fetal movement (FM) detection is essential for assessing prenatal health, as abnormal movement patterns can indicate underlying complications such as placental dysfunction or fetal distress. Traditional methods, including maternal perception and cardiotocography (CTG), suffer from subjectivity and limited accuracy. To address these challenges, we propose Contrastive Ultrasound Video Representation Learning (CURL), a novel self-supervised learning framework for FM detection from extended fetal ultrasound video recordings. Our approach leverages a dual-contrastive loss, incorporating both spatial and temporal contrastive learning, to learn robust motion representations. Additionally, we introduce a task-specific sampling strategy, ensuring the effective separation of movement and non-movement segments during self-supervised training, while enabling flexible inference on arbitrarily long ultrasound recordings through a probabilistic fine-tuning approach. Evaluated on an in-house dataset of 92 subjects, each with 30-minute ultrasound sessions, CURL achieves a sensitivity of 78.01

Generalized Category Discovery (GCD) aims to identify both known and novel classes from unlabeled data with the aid of labeled examples. While promising, most existing GCD methods rely on simultaneous access to labeled and unlabeled datasets—an assumption often impractical in real-world deployments. Continual Category Discovery (CCD) relaxes this requirement by adapting a pre-trained model to streaming unlabeled data, yet it typically assumes domain-consistent data distributions. This places a strong limitation on its applicability. In this work, we study Open Continual Category Discovery (OCCD), where the model must robustly discover previously unseen concepts from real-world data streams that may originate from heterogeneous and shifting domains. To address this, we propose an adaptive framework built on three key ideas. First, we propose a weight-aware separation module, which leverages partial unbalanced optimal transport for instance probability modeling and employs binary response spectrum quantization to generate cues for distinguishing known and unknown categories, enabling automatic sample separation. Second, for known categories, we introduce a cross-domain semantic alignment module that incorporates adversarial learning to perform adaptive prototype matching, thereby enhancing robustness against domain shifts. Finally, for unknown categories, we design a category topology consistency constraint that preserves semantic relationships between known and novel classes during distribution shifts. Experiments show our approach excels at discovering new categories while maintaining strong performance on known ones in evolving domains.

Medical vision-language models (VLMs) are strong zero-shot recognizers for medical imaging, but their reliability under domain shift hinges on calibrated uncertainty with guarantees. Split conformal prediction (SCP) offers finite-sample coverage, yet prediction sets often become large (low efficiency) and class-wise coverage unbalanced-high class-conditioned coverage gap (CCV), especially in few-shot, imbalanced regimes; moreover, naively adapting to calibration labels breaks exchangeability and voids guarantees. We propose (Laplacian-Assisted Transductive Adaptation), a training- and label-free refinement that operates on the joint calibration and test pool by smoothing zero-shot probabilities over an image-image k-NN graph using a small number of CCCP mean-field updates, preserving SCP validity via a deterministic transform. We further introduce a failure-aware conformal score that plugs into the vision-language uncertainty (ViLU) framework, providing instance-level difficulty and label plausibility to improve prediction set efficiency and class-wise balance at fixed coverage. is black-box (no VLM updates), compute-light (windowed transduction, no backprop), and includes an optional prior knob that can run strictly label-free or, if desired, in a label-informed variant using calibration marginals once. Across three medical VLMs and nine downstream tasks, consistently reduces set size and CCV while matching or tightening target coverage, outperforming prior transductive baselines and narrowing the gap to label-using methods, while using far less compute. Comprehensive ablations and qualitative analyses show that sharpens zero-shot predictions without compromising exchangeability.

Dementia diagnosis requires integrating multi-modal clinical assessments from diverse informants and clinicians under incomplete and heterogeneous data conditions. Yet most AI-driven approaches remain Alzheimer's disease (AD)-centric, framing the problem as binary AD detection or three-stage AD progression modeling within well-curated research settings. This pathology-driven paradigm overlooks the broader, syndrome-level nature of dementia, which spans multiple stages, phenotypes, and etiologies. In this paper, we propose Dementia-Agents, a clinically aligned multi-agent framework for real-world dementia staging and phenotyping. The framework follows a three-step workflow: (1) a data agent translates structured clinical records into semantically faithful textual representations that preserve missing-data signals and routes them to domain-aligned experts; (2) five fine-tuned expert agents generate domain-level predictions; and (3) a coordinator agent performs probabilistic aggregation to produce final staging and phenotyping decisions. We develop and evaluate Dementia-Agents on a real-world clinical cohort of 1,066 patients from two cognitive neurology services. Compared with monolithic multi-modal large language models (MLLMs) and prior medical multi-agent systems, our approach achieves consistent improvements in diagnostic performance for real-world syndrome-level dementia staging and phenotyping, while preserving domain-level interpretability.

The application of artificial intelligence (AI) in clinical practice presents numerous ethical concerns. However, the attitudes of stakeholders toward its ethical impact have yet to be reviewed. We aimed to review the attitudes of stakeholders toward the ethical impact of applying AI in clinical practice. We undertook a literature search of Ovid Medline and Scopus. We included empirical studies of clinicians, patients, and caregivers that investigated their attitudes toward applying AI in clinical practice. We developed a methodology based on the four principles of bioethics—beneficence, non-maleficence, autonomy, and justice—plus explainability to determine if a study investigated ethical impact. 103 studies were included. Themes related to beneficence included improved efficiency, improved decision-making and health outcomes, and more patient-centered care. Themes related to non-maleficence included inefficiency, diminished decision-making and worse health outcomes, less patient-centered care, de-skilling, and data insecurity. Themes related to autonomy included patient consent, sharing AI-generated information, and respecting patient preferences. Themes related to justice included bias, healthcare access, and responsibility. Themes related to explainability included improved decision-making and better health outcomes as well as de-skilling. While all of the included studies queried at least one theme related to ethics, very few had the explicit objective of studying ethical attitudes. Moreover, few studies queried attitudes toward explainability. Further research is needed to address these gaps. Studies often reported conflicting attitudes, with stakeholders reporting that AI could harbor both ethical advantages and disadvantages for clinical practice. Further research is needed to address these ethical trade-offs.

Hypothesis. Artificial general intelligence is, at its core, a compression problem. Effective compression demands resonance: deep learning scales best when its architecture aligns with the fundamental structure of the data. These are the fundamental principles. Yet, modern vision architectures have strayed from these truths: visual signals are highly redundant, while discriminative information, the surprise, is sparse. Current models process dense pixel grids uniformly, wasting vast compute on static background rather than focusing on the predictive residuals that define motion and meaning. We argue that to solve visual understanding, we must align our architectures with the information-theoretic principles of video, i.e., Codecs. Method. OneVision-Encoder encodes video by compressing predictive visual structure into semantic meaning. By adopting Codec Patchification, OV-Encoder abandons uniform computation to focus exclusively on the 3.1 Evidence. The results validate our core hypothesis: efficiency and accuracy are not a trade-off; they are positively correlated. When integrated into LLM, it consistently outperforms strong vision backbones such as Qwen3-ViT and SigLIP2 across 16 image, video, and document understanding benchmarks, despite using substantially fewer visual tokens and pretraining data. Notably, on video understanding tasks, OV-Encoder achieves an average improvement of 4.1

Medical foundation models have shown promise in controlled benchmarks, yet widespread deployment remains hindered by reliance on task-specific fine-tuning. Here, we introduce DermFM-Zero, a dermatology vision-language foundation model trained via masked latent modelling and contrastive learning on over 4 million multimodal data points. We evaluated DermFM-Zero across 20 benchmarks spanning zero-shot diagnosis and multimodal retrieval, achieving state-of-the-art performance without task-specific adaptation. We further evaluated its zero-shot capabilities in three multinational reader studies involving over 1,100 clinicians. In primary care settings, AI assistance enabled general practitioners to nearly double their differential diagnostic accuracy across 98 skin conditions. In specialist settings, the model significantly outperformed board-certified dermatologists in multimodal skin cancer assessment. In collaborative workflows, AI assistance enabled non-experts to surpass unassisted experts while improving management appropriateness. Finally, we show that DermFM-Zero's latent representations are interpretable: sparse autoencoders unsupervisedly disentangle clinically meaningful concepts that outperform predefined-vocabulary approaches and enable targeted suppression of artifact-induced biases, enhancing robustness without retraining. These findings demonstrate that a foundation model can provide effective, safe, and transparent zero-shot clinical decision support.

BACKGROUND AND PURPOSE:Predicting the final location and volume of lesions in acute ischemic stroke is crucial for clinical management. While CTP is routinely used for estimating lesion outcomes, conventional threshold-based methods have limitations. We developed specialized outcome-prediction deep learning models that predict infarct core in successful reperfusion cases and the combined core-penumbra region in unsuccessful reperfusion cases. MATERIALS AND METHODS:We developed single-modal and multimodal deep learning models using CTP parameter maps to predict the final infarct lesion on follow-up DWI. Using a multicenter data set from multiple sites, we developed deep learning models and evaluated them separately for patients with complete recanalization (successful reperfusion [CR], n = 350) and no recanalization (unsuccessful reperfusion [NR], n = 138) after treatment. The CR model was designed to predict the infarct core region, while the NR model predicted the expanded, hypoperfused tissue encompassing both the core and penumbra regions. Five-fold cross-validation was performed for robust evaluation. RESULTS:The multimodal 3D nnU-Net model demonstrated superior performance, achieving mean Dice scores of 35.36% in patients with CR and 50.22% in those with NR. This model substantially outperformed the current clinically used method, providing more accurate outcome estimates than the conventional single-technique threshold-based measures, which yielded Dice scores of 15.73% and 39.71% for CR and NR groups, respectively. CONCLUSIONS:Our approach offered both successful reperfusion and unsuccessful reperfusion estimations for potential treatment outcomes, enabling clinicians to better evaluate treatment eligibility for reperfusion therapies and assess potential treatment benefits. This advancement facilitates more personalized treatment recommendations and has the potential to substantially enhance clinical decision-making in acute ischemic stroke management by providing more accurate tissue outcome predictions than conventional single-technique threshold-based approaches.

Cracks are critical indicators for monitoring the health of critical infrastructure and assessing material degradation. However, achieving accurate crack segmentation is particularly challenging because cracks typically exhibit stochastic branching trajectories and are embedded within nonuniform material textures. These characteristics result in diverse crack morphologies and significant interference from complex background noise. To tackle this problem, we uncover the intricate interactions by analyzing different frequency-domain characteristics of crack images, and propose a Frequency Difference-Aware Mamba (FD-Mamba). FD-Mamba fully integrates the advantages of frequency difference priors and Mamba, effectively addressing the issues of spatial structure disruption and noise interference during state space modeling. Specifically, FD-Mamba consists of three key components: the Frequency Difference Prior Extraction (FDPE) module, the Spatial Dynamic Fusion (SDF) module, and the Frequency Difference-Enhanced (FDE) module. First, the FDPE module applies the Discrete Wavelet Transform to extract multi-scale frequency-difference priors. These priors are then progressively injected by the SDF module to refine global spatial modeling, thereby improving the continuity of Mamba when handling complex crack morphologies. Furthermore, the FDE module utilizes the Fast Fourier Transform to extract the frequency difference features from the encoder, and incorporates nonlinear intensity remapping along with a gating mechanism to effectively suppress background noise. Extensive experiments across three diverse crack datasets validate the effectiveness of FD-Mamba, showcasing its superior performance in segmenting cracks with enhanced spatial coherence and noise resilience.

Despite the impressive progress of recent MLLMs on spatio-temporal video grounding (STVG), existing evaluations and training data focus primarily on simple queries. They largely overlook the compositional queries prevalent in real-world scenarios, where a target must be disambiguated by jointly reasoning about its attributes and relations to other entities. To bridge this gap, we propose Compositional Spatio-Temporal Video Grounding (CompSTVG), a task that requires models to process complex textual queries where every intertwined attribute and relational cue is essential for disambiguation. To facilitate this task at scale, we build a synthetic data engine that leverages a spatio-temporal scene graph as a difficulty measure and casts difficulty-controlled query synthesis as a constraint programming problem, producing difficulty-graded data for both evaluation and training. Built on this engine, we introduce STVG-CompBench, a benchmark stratified by explicit difficulty levels that jointly capture temporal complexity and spatial interference. Evaluating 11 representative STVG models on STVG-CompBench reveals that current models perform poorly on compositional queries, exhibiting a sharp performance drop that is typically obscured by overall dataset-level averages. We further construct synthetic training data and propose CurrSTVG, a curriculum reinforcement learning framework that delivers consistent gains, with the largest improvements observed on the most challenging compositional queries.

Medical image segmentation plays a crucial role in assisting clinicians with diagnosing critical medical conditions. In deep learning, few-shot learning methods aim to replicate human learning by leveraging fewer examples for determining a prediction for a novel class. Researchers in the medical imaging community have also explored novel methods for few-shot medical image segmentation, leveraging meta-learning, foundation models and self-supervised learning (SSL). Acknowledging this growing interest, we review the literature on few-shot medical image segmentation from 2020 to early 2025, focusing on architectural modifications, loss-inspired learning strategies, and meta-learning frameworks. We further divide each category into fine-grained deep learning-oriented solutions, including self-supervised learning, contrastive learning, regularization, and foundation models providing in-depth discussions on architectural improvements and representation learning strategies. Additionally, we present preliminary results from several few-shot segmentation models across both medical and computer vision domains, evaluating their strengths and limitations for medical image applications. Finally, based on the limitations observed, advancements from the natural image domain, and empirical findings, we outline future research directions, providing specific insights into data-efficient learning, rapid adaptation of foundation models and generalization. The code is available here.

Learning from noisy labels remains a major challenge in medical image analysis, where annotation demands expert knowledge and substantial inter-observer variability often leads to inconsistent or erroneous labels. Despite extensive research on learning with noisy labels (LNL), the robustness of existing methods in medical imaging has not been systematically assessed. To address this gap, we introduce LNMBench, a comprehensive benchmark for Label Noise in Medical imaging. LNMBench encompasses 10 representative methods evaluated across 7 datasets, 6 imaging modalities, and 3 noise patterns, establishing a unified and reproducible framework for robustness evaluation under realistic conditions. Comprehensive experiments reveal that the performance of existing LNL methods degrades substantially under high and real-world noise, highlighting the persistent challenges of class imbalance and domain variability in medical data. Motivated by these findings, we further propose a simple yet effective improvement to enhance model robustness under such conditions. The LNMBench codebase is publicly released to facilitate standardized evaluation, promote reproducible research, and provide practical insights for developing noise-resilient algorithms in both research and real-world medical applications.The codebase is publicly available on https://github.com/myyy777/LNMBench.

Epilepsy affects approximately 1% of the global population and is one of the most common neurological disorders worldwide. Using electroencephalogram (EEG) data, deep learning (DL) has been proposed to automatically detect seizures, assisting clinicians in diagnosing and managing seizures in patients. However, EEG signals often present challenges such as variability in seizure patterns and artefacts. Other biosignals can be useful in seizure detection, yet most studies have primarily focused on EEG or single biosignals, largely ignoring the potential performance improvements offered by integrating multiple biosignals. In this work, we proposed a DL algorithm that utilises multi-biosignals, including EEG, electrocardiogram (ECG), electromyography (EMG), and respiratory signals. The algorithm employs multi-scaled window lengths with convolutional cross-attention on different biosignals for the final prediction of seizures. The cross-attention mechanism facilitates the interaction of information from short-term and long-term signals across multi-biosignals. Our study, evaluated on 63 patients, demonstrated an improvement in seizure detection, achieving an AUC of 0.90, an accuracy of 0.90, a specificity of 0.91, and a sensitivity of 0.70, compared to the EEG-only algorithm which achieved an AUC of 0.86 and sensitivity of 0.51. Notably, our algorithm outperformed an earlier study using the identical dataset by a margin of 0.25 in AUC. Furthermore, we conducted validation studies with other models, showing that our algorithm outperformed commonly used DL algorithms for seizure detection.

Ambivalence and hesitancy (A/H) are subtle behavioural states that may be expressed through language, voice, facial activity, and other non-verbal cues. The ABAW11 A/H Video Recognition Challenge asks systems to assign a binary A/H label to each naturalistic interview video. Performance is measured using Macro-F1 so that recognition of both A/H and No-A/H samples receives equal importance. We present CALM-AH, a multimodal ensemble that combines textual, acoustic, visual, and derived behavioural-statistical features. We construct 15 non-empty combinations of these feature branches. For each combination, we select the best of three classifier families using validation binary cross-entropy and optimise its decision threshold for validation Macro-F1. The resulting binary decisions are combined using fixed hard-voting weights transferred from BROTHER. We further introduce Reliability-Gated Multi-Expert Consensus(RG-MEC), an anchor-preserving decision-level ensemble that combines an initial prediction with three complementary correction experts: CALM-AH, AffectGPT, and a GPT-based semantic verifier. The initial system provides the default prediction. Its label is overridden only when all three correction experts unanimously support the same alternative class; otherwise, the anchor prediction is retained. This unanimity-gated design limits the influence of isolated expert errors while permitting bidirectional correction when task-specific, multimodal-affective, and semantic-pragmatic evidence are fully consistent. On the participant-disjoint ABAW11 dataset, CALM-AH achieves a Macro-F1 of 0.7525, and the complete RG-MEC system achieves 0.7771.

Culture serves as a fundamental determinant of human affective processing and profoundly shapes how individuals perceive and interpret emotional stimuli. Despite this intrinsic link extant evaluations regarding cultural alignment within Large Language Models primarily prioritize declarative knowledge such as geographical facts or established societal customs. These benchmarks remain insufficient to capture the subjective interpretative variance inherent to diverse sociocultural lenses. To address this limitation, we introduce CEDAR, a multimodal benchmark constructed entirely from scenarios capturing **C**ulturally **E**licited **D**istinct **A**ffective **R**esponses. To construct CEDAR, we implement a novel pipeline that leverages LLM-generated provisional labels to isolate instances yielding cross-cultural emotional distinctions, and subsequently derives reliable ground-truth annotations through rigorous human evaluation. The resulting benchmark comprises 10,962 instances across seven languages and 14 fine-grained emotion categories, with each language including 400 multimodal and 1,166 text-only samples. Comprehensive evaluations of 17 representative multilingual models reveal a dissociation between language consistency and cultural alignment, demonstrating that culturally grounded affective understanding remains a significant challenge for current models.

Orthopantomograms (OPGs) are the standard panoramic radiograph in dentistry, used for full-arch screening across multiple diagnostic tasks. While Vision Language Models (VLMs) now allow multi-task OPG analysis through natural language, they underperform task-specific models on most individual tasks. Agentic systems that orchestrate specialized tools offer a path to both versatility and accuracy, this approach remains unexplored in the field of dental imaging. To address this gap, we propose OPGAgent, a multi-tool agentic system for auditable OPG interpretation. OPGAgent coordinates specialized perception modules with a consensus mechanism through three components: (1) a Hierarchical Evidence Gathering module that decomposes OPG analysis into global, quadrant, and tooth-level phases with dynamically invoking tools, (2) a Specialized Toolbox encapsulating spatial, detection, utility, and expert zoos, and (3) a Consensus Subagent that resolves conflicts through anatomical constraints. We further propose OPG-Bench, a structured-report protocol based on (Location, Field, Value) triples derived from real clinical reports, which enables a comprehensive review of findings and hallucinations, extending beyond the limitations of VQA indicators. On our OPG-Bench and the public MMOral-OPG benchmark, OPGAgent outperforms current dental VLMs and medical agent frameworks across both structured-report and VQA evaluation. Code will be released upon acceptance.

Continual Category Discovery (CCD) aims to leverage models trained on known categories to automatically discover novel category concepts from continuously arriving streams of unlabeled data, while retaining the ability to recognize previously known classes. Despite recent progress, existing methods often assume that data across all stages are drawn from a single, stationary distribution—a condition rarely satisfied in open-world scenarios. In this paper, we challenge this stationary-distribution assumption by introducing the Open-World Continual Category Discovery (OW-CCD) setting. We address this challenge with PRISM (\underline{P}rogressive \underline{R}obust d\underline{I}scovery under \underline{S}trea\underline{M}ing data), an adaptive continual discovery framework consisting of three key components. First, inspired by spectral properties, we develop a high-frequency-driven category separation technique that exploits high-frequency components—preserving more global information—to distinguish known from unknown categories. Second, for known categories, we design a sparse assignment matching strategy, which performs proximal sparse sample-to-label matching to assign reliable cluster labels to known-class samples. Finally, to better recognize novel categories, we propose an invariant knowledge transfer module that enforces domain-invariant category relation consistency, thereby facilitating robust knowledge transfer from known to unknown classes under domain shifts. Extensive experiments on the SSB-C and DomainNet benchmarks demonstrate that our method significantly outperforms state-of-the-art CCD approaches, highlighting its effectiveness and superiority.

Clinical Relevance:The retina is a unique and accessible window to the central nervous system and its vasculature. Ocular fundus examination is crucial for identifying diagnostic red flags indicative of life- and sight-threatening conditions, such as papilledema and central retinal artery occlusion. In the emergency department (ED), prompt recognition of these conditions is essential to reduce the risk of vision loss and serious complications. Methods:This review summarizes advances in ocular fundus examination, imaging, and artificial intelligence (AI) for emergency care. Results:The direct ophthalmoscope, a traditional tool for ocular fundus examination by nonophthalmology providers in the ED, is rarely utilized due to technical limitations and lack of user skill and confidence. Recent advancements in retinal imaging technologies have introduced ocular fundus cameras with OCT as valuable alternatives. The advent of nonmydriatic and portable imaging devices has significantly expanded accessibility, enabling remote image interpretation and facilitating teleophthalmology consultations. Moreover, AI, especially deep-learning technology, has demonstrated considerable potential for automated ocular image interpretation. The latest developments in large language models show promise in providing aids in diagnosis and management. Conclusions:Further research is needed to validate the reliability of AI-powered ocular fundus assessment and to explore how these emerging technologies can be effectively integrated into ED practice to enhance ocular fundus examination and improve patient care.

Large Language Models (LLMs) are increasingly used in advice seeking and decision making that may affect social judgements. Despite stigma's profound effects on people and communities, benchmarks remain scarce. Existing general-domain evaluations typically rely on static prompts and fixed-format tasks, overlooking conversational contexts and audience effects in everyday communication. To address these gaps, we introduce SDARE-Bench, the first scenario-based benchmark evaluating both stigma detection and open-ended response generation in LLMs, comprising 1,138 dyadic queries and 1,388 group dialogue. Empirical results across 8 LLMs consistently demonstrate poor identification of stigma components, especially in group dialogues. In open-ended response generation, stigma expression was substantially higher in group settings than in dyadic, with weaker resistance to stigma and more unrealistic advice. Responses were evaluated using a classifier trained on 1,392 human annotated responses. In constructed group pressure settings, stigma expression rates further increased to a striking average of 97.5