Bi-CRCL: Bidirectional Conservative-Radical Complementary Learning with Pre-trained Foundation Models for Class-incremental Medical Image Analysis

Class-incremental learning (CIL) in medical image-guided diagnosis requires retaining prior diagnostic knowledge while adapting to newly emerging disease categories, which is critical for scalable clinical deployment. This problem is particularly challenging due to heterogeneous data and privacy constraints that prevent memory replay. Although pretrained foundation models (PFMs) have advanced general-domain CIL, their potential in medical imaging remains underexplored, where domain-specific adaptation is essential yet difficult due to anatomical complexity and inter-institutional heterogeneity. To address this gap, we conduct a systematic benchmark of recent PFM-based CIL methods and propose Bidirectional Conservative-Radical Complementary Learning (Bi-CRCL), a dual-learner framework inspired by complementary learning systems. Bi-CRCL integrates a conservative learner that preserves prior knowledge through stability-oriented updates and a radical learner that rapidly adapts to new categories via plasticity-oriented learning. A bidirectional interaction mechanism enables forward transfer and backward consolidation, allowing continual integration of new knowledge while mitigating catastrophic forgetting. During inference, outputs from both learners are adaptively fused for robust predictions. Experiments on five medical imaging datasets demonstrate consistent improvements over state-of-the-art methods under diverse settings, including cross-dataset shifts and varying task configurations.

InViC: Intent-aware Visual Cues for Medical Visual Question Answering

Medical visual question answering (Med-VQA) aims to answer clinically relevant questions grounded in medical images. However, existing multimodal large language models (MLLMs) often exhibit shortcut answering, producing plausible responses by exploiting language priors or dataset biases while insufficiently attending to visual evidence. This behavior undermines clinical reliability, especially when subtle imaging findings are decisive. We propose a lightweight plug-in framework, termed Intent-aware Visual Cues (InViC), to explicitly enhance image-based answer generation in medical VQA. InViC introduces a Cue Tokens Extraction (CTE) module that distills dense visual tokens into a compact set of K question-conditioned cue tokens, which serve as structured visual intermediaries injected into the LLM decoder to promote intent-aligned visual evidence. To discourage bypassing of visual information, we further design a two-stage fine-tuning strategy with a cue-bottleneck attention mask. In Stage I, we employ an attention mask to block the LLM's direct view of raw visual features, thereby funneling all visual evidence through the cue pathway. In Stage II, standard causal attention is restored to train the LLM to jointly exploit the visual and cue tokens. We evaluate InViC on three public Med-VQA benchmarks (VQA-RAD, SLAKE, and ImageCLEF VQA-Med 2019) across multiple representative MLLMs. InViC consistently improves over zero-shot inference and standard LoRA fine-tuning, demonstrating that intent-aware visual cues with bottlenecked training is a practical and effective strategy for improving trustworthy Med-VQA.

Topology-Guided Biomechanical Profiling: A White-Box Framework for Opportunistic Screening of Spinal Instability on Routine CT

Routine oncologic computed tomography (CT) presents an ideal opportunity for screening spinal instability, yet prophylactic stabilization windows are frequently missed due to the complex geometric reasoning required by the Spinal Instability Neoplastic Score (SINS). Automating SINS is fundamentally hindered by metastatic osteolysis, which induces topological ambiguity that confounds standard segmentation and black-box AI. We propose Topology-Guided Biomechanical Profiling (TGBP), an auditable white-box framework decoupling anatomical perception from structural reasoning. TGBP anchors SINS assessment on two deterministic geometric innovations: (i) canal-referenced partitioning to resolve posterolateral boundary ambiguity, and (ii) context-aware morphometric normalization via covariance-based oriented bounding boxes (OBB) to quantify vertebral collapse. Integrated with auxiliary radiomic and large language model (LLM) modules, TGBP provides an end-to-end, interpretable SINS evaluation. Validated on a multi-center, multi-cancer cohort ($N=482$), TGBP achieved 90.2\% accuracy in 3-tier stability triage. In a blinded reader study ($N=30$), TGBP significantly outperformed medical oncologists on complex structural features ($κ=0.857$ vs.\ $0.570$) and prevented compounding errors in Total Score estimation ($κ=0.625$ vs.\ $0.207$), democratizing expert-level opportunistic screening.

Step-CoT: Stepwise Visual Chain-of-Thought for Medical Visual Question Answering

Chain-of-thought (CoT) reasoning has advanced medical visual question answering (VQA), yet most existing CoT rationales are free-form and fail to capture the structured reasoning process clinicians actually follow. This work asks: Can traceable, multi-step reasoning supervision improve reasoning accuracy and the interpretability of Medical VQA? To this end, we introduce Step-CoT, a large-scale medical reasoning dataset with expert-curated, structured multi-step CoT aligned to clinical diagnostic workflows, implicitly grounding the model's reasoning in radiographic evidence. Step-CoT comprises more than 10K real clinical cases and 70K VQA pairs organized around diagnostic workflows, providing supervised intermediate steps that guide models to follow valid reasoning trajectories. To effectively learn from Step-CoT, we further introduce a teacher-student framework with a dynamic graph-structured focusing mechanism that prioritizes diagnostically informative steps while filtering out less relevant contexts. Our experiments show that using Step-CoT can improve reasoning accuracy and interpretability. Benchmark: github.com/hahaha111111/Step-CoT. Dataset Card: huggingface.co/datasets/fl-15o/Step-CoT

MedPruner: Training-Free Hierarchical Token Pruning for Efficient 3D Medical Image Understanding in Vision-Language Models

While specialized Medical Vision-Language Models (VLMs) have achieved remarkable success in interpreting 2D and 3D medical modalities, their deployment for 3D volumetric data remains constrained by significant computational inefficiencies. Current architectures typically suffer from massive anatomical redundancy due to the direct concatenation of consecutive 2D slices and lack the flexibility to handle heterogeneous information densities across different slices using fixed pruning ratios. To address these challenges, we propose MedPruner, a training-free and model-agnostic hierarchical token pruning framework specifically designed for efficient 3D medical image understanding. MedPruner introduces a two-stage mechanism: an Inter-slice Anchor-based Filtering module to eliminate slice-level temporal redundancy, followed by a Dynamic Information Nucleus Selection strategy that achieves adaptive token-level compression by quantifying cumulative attention weights. Extensive experiments on three 3D medical benchmarks and across three diverse medical VLMs reveal massive token redundancy in existing architectures. Notably, MedPruner enables models such as MedGemma to maintain or even exceed their original performance while retaining fewer than 5% of visual tokens, thereby drastically reducing computational overhead and validating the necessity of dynamic token selection for practical clinical deployment. Our code will be released.

This Looks Distinctly Like That: Grounding Interpretable Recognition in Stiefel Geometry against Neural Collapse

Prototype networks provide an intrinsic case based explanation mechanism, but their interpretability is often undermined by prototype collapse, where multiple prototypes degenerate to highly redundant evidence. We attribute this failure mode to the terminal dynamics of Neural Collapse, where cross entropy optimization suppresses intra class variance and drives class conditional features toward a low dimensional limit. To mitigate this, we propose Adaptive Manifold Prototypes (AMP), a framework that leverages Riemannian optimization on the Stiefel manifold to represent class prototypes as orthonormal bases and make rank one prototype collapse infeasible by construction. AMP further learns class specific effective rank via a proximal gradient update on a nonnegative capacity vector, and introduces spatial regularizers that reduce rotational ambiguity and encourage localized, non overlapping part evidence. Extensive experiments on fine-grained benchmarks demonstrate that AMP achieves state-of-the-art classification accuracy while significantly improving causal faithfulness over prior interpretable models.

Structure Observation Driven Image-Text Contrastive Learning for Computed Tomography Report Generation

Computed Tomography Report Generation (CTRG) aims to automate the clinical radiology reporting process, thereby reducing the workload of report writing and facilitating patient care. While deep learning approaches have achieved remarkable advances in X-ray report generation, their effectiveness may be limited in CTRG due to larger data volumes of CT images and more intricate details required to describe them. This work introduces a novel two-stage (structure- and report-learning) framework tailored for CTRG featuring effective structure-wise image-text contrasting. In the first stage, a set of learnable structure-specific visual queries observe corresponding structures in a CT image. The resulting observation tokens are contrasted with structure-specific textual features extracted from the accompanying radiology report with a structure-wise image-text contrastive loss. In addition, text-text similarity-based soft pseudo targets are proposed to mitigate the impact of false negatives, i.e., semantically identical image structures and texts from non-paired images and reports. Thus, the model learns structure-level semantic correspondences between CT images and reports. Further, a dynamic, diversity-enhanced negative queue is proposed to guide the network in learning to discriminate various abnormalities. In the second stage, the visual structure queries are frozen and used to select the critical image patch embeddings depicting each anatomical structure, minimizing distractions from irrelevant areas while reducing memory consumption. Also, a text decoder is added and trained for report generation.Our extensive experiments on two public datasets demonstrate that our framework establishes new state-of-the-art performance for CTRG in clinical efficiency, and its components are effective.

Multimodal Mixture-of-Experts with Retrieval Augmentation for Protein Active Site Identification

Accurate identification of protein active sites at the residue level is crucial for understanding protein function and advancing drug discovery. However, current methods face two critical challenges: vulnerability in single-instance prediction due to sparse training data, and inadequate modality reliability estimation that leads to performance degradation when unreliable modalities dominate fusion processes. To address these challenges, we introduce Multimodal Mixture-of-Experts with Retrieval Augmentation (MERA), the first retrieval-augmented framework for protein active site identification. MERA employs hierarchical multi-expert retrieval that dynamically aggregates contextual information from chain, sequence, and active-site perspectives through residue-level mixture-of-experts gating. To prevent modality degradation, we propose a reliability-aware fusion strategy based on Dempster-Shafer evidence theory that quantifies modality trustworthiness through belief mass functions and learnable discounting coefficients, enabling principled multimodal integration. Extensive experiments on ProTAD-Gen and TS125 datasets demonstrate that MERA achieves state-of-the-art performance, with 90% AUPRC on active site prediction and significant gains on peptide-binding site identification, validating the effectiveness of retrieval-augmented multi-expert modeling and reliability-guided fusion.

RegionReasoner: Region-Grounded Multi-Round Visual Reasoning

Large vision-language models have achieved remarkable progress in visual reasoning, yet most existing systems rely on single-step or text-only reasoning, limiting their ability to iteratively refine understanding across multiple visual contexts. To address this limitation, we introduce a new multi-round visual reasoning benchmark with training and test sets spanning both detection and segmentation tasks, enabling systematic evaluation under iterative reasoning scenarios. We further propose RegionReasoner, a reinforcement learning framework that enforces grounded reasoning by requiring each reasoning trace to explicitly cite the corresponding reference bounding boxes, while maintaining semantic coherence via a global-local consistency reward. This reward extracts key objects and nouns from both global scene captions and region-level captions, aligning them with the reasoning trace to ensure consistency across reasoning steps. RegionReasoner is optimized with structured rewards combining grounding fidelity and global-local semantic alignment. Experiments on detection and segmentation tasks show that RegionReasoner-7B, together with our newly introduced benchmark RegionDial-Bench, considerably improves multi-round reasoning accuracy, spatial grounding precision, and global-local consistency, establishing a strong baseline for this emerging research direction.

Attention Needs to Focus: A Unified Perspective on Attention Allocation

The Transformer architecture, a cornerstone of modern Large Language Models (LLMs), has achieved extraordinary success in sequence modeling, primarily due to its attention mechanism. However, despite its power, the standard attention mechanism is plagued by well-documented issues: representational collapse and attention sink. Although prior work has proposed approaches for these issues, they are often studied in isolation, obscuring their deeper connection. In this paper, we present a unified perspective, arguing that both can be traced to a common root -- improper attention allocation. We identify two failure modes: 1) Attention Overload, where tokens receive comparable high weights, blurring semantic features that lead to representational collapse; 2) Attention Underload, where no token is semantically relevant, yet attention is still forced to distribute, resulting in spurious focus such as attention sink. Building on this insight, we introduce Lazy Attention, a novel mechanism designed for a more focused attention distribution. To mitigate overload, it employs positional discrimination across both heads and dimensions to sharpen token distinctions. To counteract underload, it incorporates Elastic-Softmax, a modified normalization function that relaxes the standard softmax constraint to suppress attention on irrelevant tokens. Experiments on the FineWeb-Edu corpus, evaluated across nine diverse benchmarks, demonstrate that Lazy Attention successfully mitigates attention sink and achieves competitive performance compared to both standard attention and modern architectures, while reaching up to 59.58% attention sparsity.