Multi-agent large language model systems offer a promising paradigm for multidisciplinary medical decision support by integrating heterogeneous expert perspectives. However, existing frameworks often rely on loosely structured interactions, where disagreement is insufficiently diagnosed and discussion may collapse into superficial consensus or parallel opinion aggregation. We propose Structured Adaptive Multi-disciplinary Collaboration (SAMAC), a structured and strategy-aware multi-agent framework for adaptive MDT-style medical reasoning. SAMAC introduces Structured Deliberation to detect opinion disagreement and majority not confident, along with a Consensus Purification mechanism that transforms disagreement into constructive and collaborative refinement. When disagreement arises, SAMAC adaptively selects between Self-refinement and Counter-questioning to enable evidence-grounded negotiation. Experiments on three medical datasets—RADCURE, HANCOCK, and OSCCPI—show that SAMAC consistently improves overall performance and collaborative gains over existing multi-agent baselines. In particular, SAMAC improves the Average Score by 4.29 points on the RADCURE dataset and achieves higher Relative Self-Correction Scores (RSS), indicating more effective resolution of disagreement and stable collective decisions. These results highlight the importance of controlled and diverse interaction dynamics in multi-agent medical reasoning. Code will be released upon publication.

Reliable evaluation of vision-language models (VLMs) and medical vision-language models (Medical-VLMs) requires calibrated confidence, particularly under realistic clinical conditions. However, existing efforts mainly focused on improving accuracy, leaving calibration in the medical domain underexplored. To this end, we propose MVC-Bench, a calibration-centric benchmark for medical image classification with VLMs and Medical-VLMs. MVC-Bench assesses the calibration across three axes: (i) robustness to modality, backbone, and domain shift (ii) effectiveness of calibration strategies and prompt-tuning methods (iii) stability under prompt-template and random-seed variations. The benchmark covers eight different backbones, three medical modalities, including fundus imaging, histopathology, and chest X-ray under in-domain and domain shift settings. It compares post-hoc calibration, train-time calibration, and zero-shot inference methods, together with six prompt-tuning methods. Across more than 1638 controlled experiments, we report accuracy and Expected Calibration Error (ECE) as primary metrics, and further report results with complementary calibration measures, including Maximum Calibration Error (MCE) and Adaptive Calibration Error (ACE). We further investigate the underlying causes of miscalibration in VLMs and Medical-VLMs and propose a simple train-time calibration method, Multi-Class Margin (MCM) regularization, which achieves lowest ECE on 10 out of 12 settings in in-domain and remains competitive under domain shifts. Collectively, MVC-Bench provides a structured evaluation framework and actionable guidance for improving calibration in safety-critical medical workflows.

Recent advances such as Chain-of-Thought prompting have significantly improved large language models (LLMs) in zero-shot medical reasoning. However, prompting-based methods often remain shallow and unstable, while fine-tuned medical LLMs suffer from poor generalization under distribution shifts and limited adaptability to unseen clinical scenarios. To address these limitations, we present TAGS, a test-time framework that combines a broadly capable generalist with a domain-specific specialist to offer complementary perspectives without any model fine-tuning or parameter updates. To support this generalist–specialist reasoning process, we introduce two auxiliary modules: a hierarchical retrieval mechanism that provides multi-scale exemplars by selecting examples based on both semantic and rationale-level similarity, and a reliability scorer that evaluates reasoning consistency to guide final answer aggregation. TAGS achieves strong performance across nine MedQA benchmarks, boosting GPT-4o accuracy by 13.8%, DeepSeek-R1 by 16.8%, and improving a vanilla 7B model from 14.1% to 23.9%. These results surpass several fine-tuned medical LLMs, without any parameter updates.

Vision-Language Pretraining (VLP) for 3D CT scans is severely constrained by the massive GPU memory required for high-resolution volumes and large contrastive batches. We propose ECT-CLIP, a resource-efficient CT VLP framework that achieves negative scaling—scaling up both the quantity and quality of negative samples—under limited computational resources. To scale the quantity, we exploit the memory asymmetry between CT and text encoding to decouple the image and text batch sizes, thereby enlarging the negative pool. To scale the quality, we design two complementary strategies: (1) deterministically separating false negatives from informative hard negatives via UMLS concept normalization, and (2) fusing multiple frozen medical text encoders to enrich semantic diversity without extra training cost. Trained on a single consumer-grade RTX 4090 GPU, ECT-CLIP achieves state-of-the-art zero-shot classification AUCs of 83.4% on CT-RATE and 75.0% on the external RAD-ChestCT benchmark, outperforming prior methods that require 4–8 high-performance GPUs. Extensive ablation studies validate the effectiveness of each negative-scaling component.

Vision-and-language pretraining (VLP) in medicine leverages contrastive learning on image–text pairs, often enhanced with masked modeling. However, existing methods face two challenges: difficulty reconstructing key pathological features due to limited data, and reliance on either paired or image-only datasets without combining both. To address this, we propose MMCLIP (Masked Medical Contrastive Language–Image Pre-training), which introduces two modules: AttMIM: Masks image features highly correlated with text to improve reconstruction of fine medical details. EntMLM: Masks key medical entities in text and reconstructs them using visual cues. Furthermore, MMCLIP incorporates unpaired data through disease-kind prompts, achieving state-of-the-art performance in zero-shot and fine-tuning across five benchmarks.

Medical decision-making often involves integrating knowledge from multiple clinical specialties, typically achieved through multidisciplinary teams. Inspired by this collaborative process, recent work has leveraged large language models (LLMs) in multi-agent collaboration frameworks to emulate expert teamwork. While these approaches improve reasoning through agent interaction, they are limited by static, pre-assigned roles, which hinder adaptability and dynamic knowledge integration. To address these limitations, we propose KAMAC, a Knowledge-driven Adaptive Multi-Agent Collaboration framework that enables LLM agents to dynamically form and expand expert teams based on the evolving diagnostic context. KAMAC begins with one or more expert agents and then conducts a knowledge-driven discussion to identify and fill knowledge gaps by recruiting additional specialists as needed. This supports flexible, scalable collaboration in complex clinical scenarios, with decisions finalized through reviewing updated agent comments. Experiments on two real-world medical benchmarks demonstrate that KAMAC significantly outperforms both single-agent and advanced multi-agent methods, particularly in complex clinical scenarios (i.e., cancer prognosis) requiring dynamic, cross-specialty expertise. Our code is publicly available at: https://github.com/XiaoXiao-Woo/KAMAC.

Recent multimodal large language models (MLLMs) have demonstrated significant potential in open-ended conversation, generating more accurate and personalized responses. However, their abilities to memorize, recall, and reason in sustained interactions within real-world scenarios remain underexplored. This paper introduces MMRC, a Multi-Modal Real-world Conversation benchmark for evaluating six core open-ended abilities of MLLMs: information extraction, multi-turn reasoning, information update, image management, memory recall, and answer refusal. With data collected from real-world scenarios, MMRC comprises 5,120 conversations and 28,720 corresponding manually labeled questions, posing a significant challenge to existing MLLMs. Evaluations on 20 MLLMs in MMRC indicate an accuracy drop during open-ended interactions. We identify four common failure patterns: long-term memory degradation, inadequacies in updating factual knowledge, accumulated assumption of error propagation, and reluctance to “say no.” To mitigate these issues, we propose a simple yet effective NOTE-TAKING strategy, which can record key information from the conversation and remind the model during its responses, enhancing conversational capabilities. Experiments across six MLLMs demonstrate significant performance improvements.