🏥 Medical Imaging¶

🧠 NeurIPS2025 · 138 paper notes

3D-RAD: A Comprehensive 3D Radiology Med-VQA Dataset with Multi-Temporal Analysis and Diverse Diagnostic Tasks: This paper introduces 3D-RAD — the first large-scale 3D medical VQA benchmark, comprising 170K CT-based question-answer pairs across six clinical task categories (including a novel multi-temporal diagnosis task), accompanied by a 136K training set. The benchmark reveals critical deficiencies of existing VLMs in 3D temporal reasoning.
A Novel Approach to Classification of ECG Arrhythmia Types with Latent ODEs: This work combines a path-minimized Latent ODE encoder with a gradient-boosted decision tree (GBDT) into a two-stage ECG arrhythmia classification pipeline. On the MIT-BIH dataset, the macro AUC-ROC degrades only marginally from 0.984 at 360 Hz to 0.976 at 45 Hz, demonstrating strong robustness to sampling frequency variation.
A Unified Solution to Video Fusion: From Multi-Frame Learning to Benchmarking: This paper proposes UniVF, the first unified video fusion framework based on multi-frame learning, optical flow feature warping, and temporal consistency loss, along with VF-Bench, the first video fusion benchmark covering four major fusion tasks (multi-exposure, multi-focus, infrared-visible, and medical), achieving state-of-the-art performance across all sub-tasks.
A Variational Manifold Embedding Framework for Nonlinear Dimensionality Reduction: This paper proposes a variational manifold embedding framework that formalizes dimensionality reduction as an optimization problem over smooth embedding maps (minimizing the KL divergence between a prior distribution and the pullback of the data distribution), theoretically unifying PCA and nonlinear dimensionality reduction methods, and leverages the calculus of variations (Euler-Lagrange equations) and Noether's theorem to derive interpretable constraints on optimal embeddings.
AANet: Virtual Screening under Structural Uncertainty via Alignment and Aggregation: To address the unavailability of holo protein structures in real-world drug discovery, this paper proposes AANet—a framework that aligns representations via tri-modal contrastive learning (ligand–holo pocket–detected cavity) and aggregates multiple candidate binding sites through cross-attention. AANet substantially outperforms SOTA methods in blind screening on apo/predicted protein structures (EF1% on DUD-E: 11.75 → 37.19).
Active Target Discovery under Uninformative Prior: The Power of Permanent and Transient Memory: This paper proposes EM-PTDM, a framework inspired by the dual-memory system in neuroscience. It leverages a pretrained diffusion model as "permanent memory" and incorporates a lightweight "transient memory" module based on Doob's h-transform to achieve efficient active target discovery without any domain-specific prior data, with theoretical guarantees of monotonic prior improvement.
Amortized Active Generation of Pareto Sets: This paper proposes the A-GPS framework, which learns a conditional generative model over the Pareto set to perform online discrete black-box multi-objective optimization. It employs a non-dominance class probability estimator (CPE) as an implicit substitute for explicit hypervolume computation in PHVI, and achieves amortized posterior preference conditioning via preference direction vectors (without retraining). The approach demonstrates superior sample efficiency on synthetic benchmarks and protein design tasks.
Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra: This paper proposes ChefNMR, the first end-to-end framework based on 3D atomic diffusion models that directly predicts the molecular structure of unknown small molecules (especially complex natural products) from 1D NMR spectra and molecular formulae alone, achieving state-of-the-art performance on both synthetic and experimental datasets.
GraphFLA: Augmenting Biological Fitness Prediction Benchmarks with Landscape Features: GraphFLA is an efficient fitness landscape analysis framework that computes 20 biologically meaningful landscape features (ruggedness / epistasis / navigability / neutrality) across 5,300+ real-world landscapes (ProteinGym / RNAGym / CIS-BP), revealing that model performance is highly dependent on landscape topology—e.g., VenusREM outperforms ProSST on highly navigable landscapes but underperforms it on highly epistatic ones—while processing one million mutants in just 20 seconds (vs. 5 hours for MAGELLAN).
Autoencoding Random Forests: RFAE is the first principled encode-decode framework for random forests. It exploits the positive-definiteness and universality of the RF kernel to derive low-dimensional encodings via diffusion-map spectral decomposition, and decodes back to the original feature space through k-NN regression in leaf-node space. Across 20 tabular datasets, RFAE achieves an average reconstruction rank of 1.80, substantially outperforming TVAE (3.38) and AE (3.27), and is successfully applied to MNIST reconstruction and scRNA-seq batch-effect removal.
BarcodeMamba+: Advancing State-Space Models for Fungal Biodiversity Research: BarcodeMamba+ is an SSM-based foundation model for fungal ITS DNA barcode classification. By adopting a pretrain-then-finetune paradigm to leverage large-scale unlabeled sequences, and incorporating three enhancements—hierarchical label smoothing, inverse square-root weighted loss, and multi-head outputs—it substantially outperforms BLAST, CNN, and Transformer baselines across all taxonomic ranks on three test sets, achieving a top species-level accuracy of 88.9%.
CrossNovo: Bidirectional Representations Augmented Autoregressive Biological Sequence Generation: CrossNovo integrates autoregressive (AR) and non-autoregressive (NAR) decoders through a shared spectrum encoder, importance annealing, and gradient-blocked knowledge distillation, enabling the bidirectional global understanding of NAR to augment AR sequence generation. On the 9-Species benchmark, it achieves amino acid accuracy of 0.811 (+2.6%) and peptide recall of 0.654 (+5.3%).
Brain Harmony: A Multimodal Foundation Model Unifying Morphology and Function into 1D Tokens: The first multimodal brain foundation model that unifies structural morphology (T1 sMRI) and functional dynamics (fMRI), compressing high-dimensional neuroimaging data into compact 1D token representations via Geometric Harmonics Pre-alignment and Temporally Adaptive Patch Embedding (TAPE). The model consistently outperforms prior methods on neurodevelopmental/neurodegenerative disease diagnosis and cognitive prediction tasks.
Bridging Graph and State-Space Modeling for Intensive Care Unit Length of Stay Prediction: This paper proposes S2G-Net, a dual-branch architecture that integrates Mamba state-space temporal encoding with a multi-view graph neural network (GraphGPS) for ICU length-of-stay (LOS) prediction, achieving comprehensive improvements over sequential, graph-based, and hybrid baselines on MIMIC-IV.
Care-PD: A Multi-Site Anonymized Clinical Dataset for Parkinson's Disease Gait Assessment: This work introduces Care-PD — the largest multi-site anonymized 3D mesh dataset for Parkinson's disease (PD) gait analysis to date, comprising 9 cohorts, 8 clinical centers, 362 subjects, and 8,477 walking bouts. It provides a systematic benchmark for UPDRS gait scoring and motion pre-training tasks, demonstrating that fine-tuning on Care-PD reduces MPJPE from 60.8 mm to 7.5 mm and improves F1 by 17 percentage points.
CGBench: Benchmarking Language Model Scientific Reasoning for Clinical Genetics Research: This paper introduces CGBench, a clinical genetics benchmark grounded in ClinGen expert annotations, designed to evaluate the scientific literature reasoning capabilities of LLMs from both variant and gene curation perspectives. The benchmark encompasses three tasks—evidence scoring, evidence verification, and experimental evidence extraction—and finds that reasoning models perform best on fine-grained tasks but underperform non-reasoning models on high-level judgments.
CodeCrash: Exposing LLM Fragility to Misleading Natural Language in Code Reasoning: This paper proposes CodeCrash, a stress-testing framework that systematically evaluates the code reasoning robustness of 17 LLMs through functionally equivalent structural perturbations and misleading natural language injections (comments, print statements, and hints). The framework reveals an average performance drop of 23.2% across models, with CoT recovering only to 13.8%, and is the first to identify the "Reasoning Collapse" phenomenon in large reasoning models (LRMs).
Compressing Biology: Evaluating the Stable Diffusion VAE for Phenotypic Drug Discovery: This work presents the first systematic evaluation of the Stable Diffusion VAE (SD-VAE) for reconstructing Cell Painting fluorescence microscopy images. Results show that SD-VAE preserves phenotypic information well at both the pixel level and the biological signal level (with negligible drop in Fraction Retrieved), and that the general-purpose feature extractor InceptionV3 matches or outperforms the domain-specific model OpenPhenom on retrieval tasks.
ConfRover: Simultaneous Modeling of Protein Conformation and Dynamics via Autoregression: ConfRover proposes an autoregressive framework that factorizes protein MD trajectories into frame-wise conditional generation $p(\mathbf{x}^{1:L}) = \prod_l p(\mathbf{x}^l | \mathbf{x}^{<l})$, and through a modular architecture consisting of an encoder, a causal Transformer, and an SE(3) diffusion decoder, unifies three tasks—trajectory simulation, time-independent conformational sampling, and conformational interpolation—within a single model for the first time, achieving comprehensive improvements over MDGen on the ATLAS benchmark.
Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models: This paper identifies an inconsistency between sampling and simulation in diffusion models (particularly at small diffusion timesteps), proposes a Fokker-Planck-based regularization term to enforce consistency, and combines it with a time-partitioned Mixture-of-Experts (MoE) strategy to achieve consistent and efficient sampling and molecular dynamics simulation across multiple biomolecular systems.
Convolutional Monge Mapping between EEG Datasets to Support Independent Component Labeling: This paper extends CMMN (Convolutional Monge Mapping Normalization) by proposing two strategies — channel-averaged PSD with $\ell_1$-normalized barycenter and subject-to-subject matching — to generate a single time-domain filter for domain adaptation across EEG datasets with differing channel counts. On independent component (IC) brain/non-brain classification, the F1 score improves from 0.77 to 0.84, surpassing ICLabel (0.88→0.91).
CureAgent: A Training-Free Executor-Analyst Framework for Clinical Reasoning: CureAgent proposes an Executor-Analyst collaborative framework that decouples precise tool invocation (TxAgent/Llama-8B as Executor) from high-level clinical reasoning (Gemini 2.5 as Analyst). Combined with a Stratified Ensemble Late Fusion topology that preserves evidence diversity, the system achieves 83.8% accuracy on CURE-Bench without end-to-end fine-tuning, and reveals two critical scaling findings: the context–performance paradox and the curse of dimensionality in action space.
CXReasonBench: A Benchmark for Evaluating Structured Diagnostic Reasoning in Chest X-rays: This paper proposes CheXStruct and CXReasonBench — a structured diagnostic reasoning evaluation framework for chest X-rays that employs multi-path, multi-stage assessment to reveal critical deficiencies in existing LVLMs at intermediate reasoning steps.
DCA: Graph-Guided Deep Embedding Clustering for Brain Atlases: DCA (Deep Cluster Atlas) proposes a graph-guided deep embedding clustering framework that combines voxel-level spatiotemporal embeddings from a pretrained Swin-UNETR with KNN graph spatial regularization. By aligning soft assignments with atlas clustering auxiliary labels via KL divergence, the framework generates functionally homogeneous and spatially contiguous individualized brain atlases. On the HCP dataset, DCA achieves 98.8% improvement in homogeneity and 29% improvement in silhouette coefficient, and outperforms existing atlases on downstream tasks including autism diagnosis and cognitive decoding.
De novo generation of functional terpene synthases using TpsGPT: TpsGPT fine-tunes a distilled ProtGPT2 Tiny (38.9M parameters) on 79K terpene synthase (TPS) sequences to generate 28K candidate sequences, which are subsequently filtered through a multi-stage pipeline (perplexity / pLDDT / EnzymeExplorer / CLEAN / InterPro / Foldseek) to yield 7 de novo TPS sequences that are evolutionarily distant (<60% sequence identity) yet structurally conserved. Wet-lab experiments confirm that 2 of the 7 candidates possess TPS enzymatic activity—achieving functional enzyme de novo design at a GPU cost below $200.
Demo: Generative AI helps Radiotherapy Planning with User Preference: This paper proposes the Flexible Dose Proposer (FDP), a two-stage training framework (VQ-VAE pretraining + multi-condition encoding) that enables slider-based interactive 3D dose distribution prediction incorporating user preferences. The system is integrated into the Eclipse clinical treatment planning system and outperforms Varian RapidPlan in head-and-neck cancer radiotherapy scenarios.
Demo: Guide-RAG: Evidence-Driven Corpus Curation for Retrieval-Augmented Generation in Long COVID: This paper systematically evaluates six RAG corpus configurations for Long COVID clinical QA. The GS-4 configuration—combining clinical guidelines with high-quality systematic reviews—consistently outperforms both single-guideline and large-scale literature retrieval baselines across faithfulness, relevance, and comprehensiveness. The authors further introduce the Guide-RAG framework and the LongCOVID-CQ evaluation dataset.
DermaCon-IN: A Multi-concept Annotated Dermatological Image Dataset of Indian Skin Disorders: This work introduces DermaCon-IN—the first densely annotated dermatological image dataset predominantly featuring Indian skin tones (5,450 images / 3,002 patients / 245 diagnoses)—providing three-level hierarchical diagnostic labels, 47 lesion descriptors, and 49 anatomical site annotations, with benchmark evaluations using CNN, ViT, and concept bottleneck model architectures.
DesignX: Human-Competitive Algorithm Designer for Black-Box Optimization: This paper proposes DesignX, the first automated algorithm design framework that jointly learns two sub-tasks—optimizer workflow generation and dynamic hyperparameter control—through dual Transformer agents pre-trained at scale on 10k synthetic problems. DesignX surpasses human-designed optimizers on both synthetic benchmarks and real-world tasks including protein docking, AutoML, and UAV path planning.
DIsoN: Decentralized Isolation Networks for Out-of-Distribution Detection in Medical Imaging: This paper proposes Decentralized Isolation Networks (DIsoN), which detects OOD samples by training a binary classifier to "isolate" a test sample from training data, and leverages training data information without sharing it through decentralized parameter exchange. The method achieves state-of-the-art performance across 12 OOD detection tasks on 4 medical imaging datasets.
Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum: A fully self-supervised noise-robust representation learning framework is proposed, leveraging a "denoised→noisy" data curriculum strategy combined with denoised-teacher regularization. This enables SSL models such as DINOv2 to directly process noisy inputs at inference time without any denoiser, achieving a 4.8% improvement in linear probing accuracy under extreme Gaussian noise on ImageNet-1k.
Doctor Approved: Generating Medically Accurate Skin Disease Images through AI-Expert Feedback: This paper proposes MAGIC, a framework that encodes dermatologist-defined clinical checklists into structured evaluation prompts executable by MLLMs (e.g., GPT-4o), and uses the resulting feedback to fine-tune diffusion models via DPO or reward-based fine-tuning (RFT), generating clinically accurate skin disease images for data augmentation. MAGIC achieves +9.02% improvement on a 20-class skin disease classification task and +13.89% in few-shot settings.
Domain-Adaptive Transformer for Data-Efficient Glioma Segmentation in Sub-Saharan MRI: This paper proposes SegFormer3D+, a domain-adaptive Transformer architecture tailored for heterogeneous MRI data from Sub-Saharan Africa. By integrating histogram matching, radiomics-guided stratified sampling, a frequency-aware dual-path encoder, and a dual attention mechanism, the model achieves a mean Dice of 0.81 for glioma segmentation with only 60 annotated cases for fine-tuning, outperforming nnU-Net by +2.5%.
Dual Mixture-of-Experts Framework for Discrete-Time Survival Analysis: This paper proposes a Dual Mixture-of-Experts (Dual MoE) framework for discrete-time survival analysis, combining a feature encoder MoE (for modeling patient subgroup heterogeneity) with a hazard network MoE (for capturing temporal dynamics). The framework achieves improvements of up to 0.04 in time-dependent C-index on the METABRIC and GBSG breast cancer datasets.
DyG-Mamba: Continuous State Space Modeling on Dynamic Graphs: DyG-Mamba introduces continuous state space models (SSMs) into dynamic graph learning. It proposes a temporal span-aware continuous SSM that models irregular time intervals via an exponential decay function inspired by the Ebbinghaus forgetting curve, combined with input-dependent parameters constrained by spectral norm for Lipschitz robustness. The method achieves an average rank of 2.42 across 12 dynamic graph benchmarks (vs. DyGFormer's 2.92) while maintaining $O(bdL)$ linear complexity.
Dynamic Causal Discovery in Alzheimer's Disease through Latent Pseudotime Modelling: This paper applies BN-LTE (Bayesian Network with Latent Time Embedding) to real-world ADNI data from AD patients to infer dynamic causal graphs that evolve along a disease pseudotime axis. The learned pseudotime achieves a diagnostic AUC of 0.82, substantially outperforming chronological age (AUC 0.59), and reveals dynamic causal relationships between emerging biomarkers NfL/GFAP and established AD markers.
EDBench: Large-Scale Electron Density Data for Molecular Modeling: This work constructs EDBench, the largest electron density (ED) dataset to date (3.3 million molecules, computed via B3LYP/6-31G** DFT), and designs a three-category benchmark evaluation framework covering prediction, retrieval, and generation tasks. It provides the first systematic assessment of deep learning models' ability to understand and exploit electron density.
EndoBench: A Comprehensive Evaluation of Multi-Modal Large Language Models for Endoscopy Analysis: This paper introduces EndoBench, the first comprehensive MLLM evaluation benchmark covering 4 endoscopic scenarios, 12 clinical tasks, and 5 levels of visual prompt granularity, comprising 6,832 clinically validated VQA pairs. Evaluation of 23 MLLMs reveals that commercial models generally outperform open-source and medical-specific counterparts, yet all remain below human expert performance.
Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling: This paper proposes Energy Matching, which unifies flow matching and energy-based models via a single time-independent scalar potential field: far from the data manifold, the model performs efficient transport along optimal transport paths; near the manifold, it transitions to a Boltzmann equilibrium distribution for likelihood modeling. The method achieves FID 3.34 on CIFAR-10, substantially outperforming existing EBMs by more than 50%.
EWC-Guided Diffusion Replay for Exemplar-Free Continual Learning in Medical Imaging: This paper proposes an exemplar-free continual learning framework that combines class-conditional DDPM diffusion replay with Elastic Weight Consolidation (EWC), achieving an AUROC of 0.851 on MedMNIST v2 (8 tasks across 2D/3D) and CheXpert, reducing forgetting by over 30% compared to DER++, approaching the joint training upper bound (0.869), while requiring no storage of original patient data.
FAPEX: Fractional Amplitude-Phase Expressor for Robust Cross-Subject Seizure Prediction: This paper proposes FAPEX, a framework that achieves adaptive time-frequency decomposition via a learnable Fractional Neural Frame Operator (FrNFO), combined with Amplitude-Phase Cross-Encoding (APCE) and Spatial Correlation Aggregation (SCA). FAPEX comprehensively outperforms 33 baseline methods across 12 cross-species, cross-modality seizure prediction benchmarks.
Few-Shot Learning from Gigapixel Images via Hierarchical Vision-Language Alignment and Modeling: This paper proposes HiVE-MIL, a hierarchical vision-language MIL framework that constructs a unified heterogeneous graph to model cross-scale hierarchical relationships (5× and 20×) and intra-scale multimodal alignment. Combined with a text-guided dynamic filtering mechanism and a hierarchical contrastive loss, HiVE-MIL consistently outperforms existing methods under the 16-shot setting on three TCGA datasets (lung, breast, and renal cancer), achieving up to 4.1% improvement in Macro F1.
FGBench: A Dataset and Benchmark for Molecular Property Reasoning at Functional Group-Level in Large Language Models: This paper presents FGBench, a dataset comprising 625K molecular property reasoning questions focused on functional group-level reasoning evaluation. Through three dimensions (single functional group effect, multi-functional group interaction, and molecular comparison), it systematically reveals the severe deficiencies of current LLMs in fine-grained chemical reasoning.
FireGNN: Neuro-Symbolic Graph Neural Networks with Trainable Fuzzy Rules for Interpretable Medical Image Classification: This paper proposes FireGNN, which for the first time embeds trainable fuzzy rules into the GNN forward pass. Using three topological descriptors—node degree, clustering coefficient, and label consistency—FireGNN achieves endogenous interpretability for medical image classification, outperforming standard GCN/GAT/GIN and auxiliary-task baselines on 5 MedMNIST datasets and MorphoMNIST.
Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning: This paper proposes Flow Density Control (FDC), which generalizes the fine-tuning of pretrained flow/diffusion models from KL-regularized expected reward maximization to a unified framework supporting arbitrary distributional utility functions with arbitrary divergence regularization. The approach decomposes nonlinear objectives into a sequence of linear fine-tuning subproblems and provides convergence guarantees.
FOXES: A Framework For Operational X-ray Emission Synthesis: This paper proposes FOXES, a Vision Transformer-based framework that translates multi-channel solar EUV observation images into soft X-ray (SXR) flux, achieving an overall Pearson correlation of 0.982. The framework lays the groundwork for far-side solar flare detection and the construction of more complete flare catalogs.
Fractional Diffusion Bridge Models: This paper proposes Fractional Diffusion Bridge Models (FDBM), which incorporate fractional Brownian motion (fBM) into the generative diffusion bridge framework. The Hurst exponent $H$ controls the roughness and long-range dependence of trajectories, yielding improvements over Brownian motion baselines on protein conformation prediction and image translation tasks.
From Black Box to Biomarker: Sparse Autoencoders for Interpreting Speech Models of Parkinson's Disease: This work adapts sparse autoencoder (SAE) techniques from large language model interpretability research to speech-based Parkinson's disease (PD) detection, proposes a Mask-based SAE to address small-dataset limitations, discovers that model predictions rely primarily on spectral flux and spectral flatness in low-energy regions, and further reveals that these features correlate significantly with MRI putamen volume—establishing a bridge from internal model representations to clinical biomarkers.
Generalizable, Real-Time Neural Decoding with Hybrid State-Space Models: POSSM proposes a hybrid SSM-attention architecture that combines spike-level tokenization with a recurrent state-space model backbone, achieving generalizable real-time neural decoding with inference speeds up to 9× faster than Transformers while maintaining comparable accuracy.
Generating Multi-Table Time Series EHR from Latent Space with Minimal Preprocessing: This paper proposes RawMed — the first framework to synthesize multi-table time series EHR data from raw records with minimal lossy preprocessing: events are textualized → compressed into a discrete latent space via Residual Quantization → temporal dynamics are modeled with an autoregressive Transformer. RawMed comprehensively outperforms existing baselines in fidelity, clinical utility, and privacy protection.
Generative Distribution Embeddings: Lifting Autoencoders to the Space of Distributions for Multiscale Representation Learning: This paper proposes Generative Distribution Embeddings (GDE), which lifts autoencoders to the space of distributions — the encoder operates on sets of samples while the decoder is replaced by a conditional generative model — thereby learning distribution-level representations. The framework is validated on 6 computational biology tasks.
Generative Modeling of Full-Atom Protein Conformations using Latent Diffusion on Graph Embeddings: This work proposes LD-FPG, a framework that encodes full-atom MD trajectories into a low-dimensional latent space via Chebyshev graph neural networks and applies DDPM in that space to generate novel conformational ensembles. To the authors' knowledge, this is the first approach to generate protein conformations that includes all heavy atoms of the side chains.
GeoDynamics: A Geometric State-Space Neural Network for Understanding Brain Dynamics on Riemannian Manifolds: This paper proposes GeoDynamics, which generalizes the classical state-space model (SSM) from Euclidean space to the symmetric positive definite (SPD) manifold. By employing weighted Fréchet mean aggregation and orthogonal group translations, it achieves geometrically consistent state evolution on the manifold, attaining state-of-the-art performance on brain connectome analysis (early diagnosis of AD/PD/ASD) and human action recognition.
GFlowNets for Learning Better Drug-Drug Interaction Representations: To address the severe class imbalance in drug-drug interaction (DDI) prediction, this paper proposes combining GFlowNet with a variational graph autoencoder (VGAE). By reward-guided generative sampling, the framework synthesizes training samples for rare interaction types, thereby enhancing predictive performance on infrequent yet clinically critical interaction categories.
H-DDx: A Hierarchical Evaluation Framework for Differential Diagnosis: H-DDx proposes a differential diagnosis evaluation framework grounded in the ICD-10 classification hierarchy. By expanding both predicted and ground-truth diagnoses to their ancestor nodes and computing a Hierarchical Diagnostic F1 (HDF1), the framework rewards "clinically relevant approximate correctness" rather than exact match only. Evaluating 22 LLMs reveals that the domain-specialized model MediPhi rises from 20th to 2nd place under HDF1, an advantage completely obscured by Top-5 metrics.
ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression: By proposing a systematic feature suppression framework—rather than cue-conflict experiments—this work re-evaluates the feature reliance of CNNs, finding that CNNs are not inherently texture-biased but instead rely primarily on local shape features; moreover, feature reliance patterns differ substantially across domains (CV/MI/RS).
Interpreting GFlowNets for Drug Discovery: Extracting Actionable Insights for Medicinal Chemistry: This work constructs a multi-level interpretability toolkit for SynFlowNet (a GFlowNet grounded in synthetic reaction templates), integrating gradient saliency, counterfactual perturbation, sparse autoencoders (SAE), and motif probes to reveal how internal representations encode physicochemical properties and functional group information relevant to medicinal chemistry.
Is Sequence Information All You Need for Bayesian Optimization of Antibodies?: This paper systematically compares the roles of sequence and structural information in antibody Bayesian optimization, finding that sequence-only methods augmented with protein language model (pLM) soft constraints can match the performance of structure-based methods, thereby questioning the necessity of structural information in antibody Bayesian optimization.
Iterative Foundation Model Fine-Tuning on Multiple Rewards: This paper proposes IterativeRS (Iterative Rewarded Soups), which alternates between independent fine-tuning of per-objective expert policies and policy merging. The method unifies reward combination and expert merging approaches, outperforming MORLHF and Rewarded Soups on small molecule design, DNA sequence generation, and text summarization tasks.
JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensembles: This paper proposes JAMUN, a conformational ensemble generation method built on the Walk-Jump Sampling (WJS) framework. By performing Langevin dynamics on a noise-smoothed manifold and using an SE(3)-equivariant denoiser to jump back to the original distribution, JAMUN achieves peptide conformational sampling an order of magnitude faster than conventional molecular dynamics while retaining transferability to out-of-training systems.
JanusDNA: A Powerful Bi-directional Hybrid DNA Foundation Model: JanusDNA is proposed as the first bidirectional DNA foundation model, combining a Mamba-Attention-MoE hybrid architecture with the Janus Modeling pretraining paradigm to achieve bidirectional understanding at the training efficiency of autoregressive methods, attaining state-of-the-art performance across multiple genomic benchmarks.
Large Language Models as Medical Codes Selectors: A Benchmark Using the International Classification of Primary Care: This work constructs a medical coding benchmark based on an extract-retrieve-select framework, evaluating ICPC-2 code selection capability across 33 LLMs. Results show that 28 models achieve F1 > 0.8, demonstrating that LLMs can effectively automate primary care coding without fine-tuning.
Learning Conformational Ensembles of Proteins Based on Backbone Geometry: This paper proposes BBFlow, a flow matching generative model based on protein backbone geometry for conformational ensemble sampling. BBFlow requires neither evolutionary sequence information nor pretrained folding models, achieves inference speeds more than an order of magnitude faster than AlphaFlow, and generalizes to multi-chain proteins.
Learning Relative Gene Expression Trends from Pathology Images in Spatial Transcriptomics: This paper proposes STRank, a loss function that reformulates gene expression estimation from pathology images as a ranking score estimation task. By modeling the stochastic noise inherent in expression counts via binomial/multinomial distributions, STRank enables models to learn robust relative expression relationships from spatial transcriptomics data subject to batch effects and random fluctuations.
LLM-Assisted Emergency Triage Benchmark: Bridging Hospital-Rich and MCI-Like Field Simulation: This work constructs an open, LLM-assisted emergency triage benchmark based on MIMIC-IV-ED, defining two evaluation scenarios—hospital-rich and mass casualty incident (MCI)-like field simulation—and providing baseline models along with SHAP-based interpretability analysis to promote reproducibility and accessibility in triage prediction research.
LoMix: Learnable Weighted Multi-Scale Logits Mixing for Medical Image Segmentation: LoMix introduces a Combinatorial Mutation Module (CMM) that generates "mutant" logits from multi-scale outputs via four fusion operators (addition / multiplication / concatenation / attention-weighted fusion) across all subset combinations, paired with NAS-style Softplus learnable weights for automatic contribution balancing. On Synapse 8-organ segmentation, Dice improves from 80.9% to 85.1% (+4.2%), and by +9.23% under 5% training data.
Magical: Medical Lay Language Generation via Semantic Invariance and Layperson-tailored Adaptation: This paper proposes Magical, an asymmetric LoRA architecture for medical lay language generation (MLLG) that enforces a semantic invariance constraint on the shared matrix $A$ while employing multiple independent matrices $B$ to enable semantically faithful and stylistically diverse lay language generation. Magical reduces trainable parameters by 31.66% while outperforming all LoRA variants.
Mamba Goes HoME: Hierarchical Soft Mixture-of-Experts for 3D Medical Image Segmentation: This paper proposes Mamba-HoME, an architecture that integrates a Hierarchical Soft Mixture-of-Experts (HoME) with the Mamba SSM. Through a two-level token routing mechanism, it achieves local-to-global feature modeling and surpasses existing state-of-the-art methods on 3D medical image segmentation across CT, MRI, and ultrasound modalities, while maintaining linear computational complexity.
Manipulating 3D Molecules in a Fixed-Dimensional E(3)-Equivariant Latent Space: This paper proposes MolFLAE, a 3D molecular variational autoencoder that learns a fixed-dimensional, E(3)-equivariant latent space. By introducing learnable virtual nodes and a Bayesian Flow Network (BFN) decoder, MolFLAE enables zero-shot molecular editing — including atom-count editing, structural reconstruction, and property interpolation — and demonstrates practical utility in drug optimization targeting the human glucocorticoid receptor (hGR).
MATCH: Multi-faceted Adaptive Topo-Consistency for Semi-Supervised Histopathology Segmentation: This paper proposes the MATCH framework, which tightly couples topological reasoning with the perturbation-robustness principle of semi-supervised learning. By exploiting dual-level topological consistency across random perturbations and temporal training snapshots, MATCH adaptively identifies reliable topological structures without requiring manually defined thresholds, substantially reducing topological errors in histopathology image segmentation.
MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks: This paper proposes MedAgentBoard, a comprehensive benchmark that systematically evaluates multi-agent collaboration, single-LLM, and conventional methods across diverse medical tasks, revealing that multi-agent collaboration does not consistently outperform strong single models or specialized conventional approaches.
MedMKG: Benchmarking Medical Knowledge Exploitation with Multimodal Knowledge Graph: This paper constructs MedMKG, a medical multimodal knowledge graph that integrates MIMIC-CXR imaging data with UMLS clinical concepts, proposes a Neighbor-aware Filtering (NaF) algorithm for image selection, and conducts comprehensive benchmarking of 24 baseline methods across three tasks: link prediction, text-image retrieval, and VQA.
Mind the (Data) Gap: Evaluating Vision Systems in Small Data Applications: This paper systematically compares MLLMs (e.g., Gemini, Qwen2.5-VL) and vision encoder + SVM pipelines on the NeWT ecological classification benchmark across the "small data regime" (10–1000 labeled samples). MLLMs plateau after 10–30 samples, whereas vision-based methods exhibit near-logarithmic growth throughout, calling on the community to prioritize small-data evaluation.
Mind the Gap: Aligning Knowledge Bases with User Needs to Enhance Mental Health Retrieval: This paper proposes a knowledge base augmentation framework grounded in "demand gap" analysis. By overlaying real user data (forum posts) onto existing mental health resource repositories to identify content voids, the framework applies targeted augmentation strategies to achieve near-full-corpus RAG retrieval quality with minimal document additions.
MIRA: Medical Time Series Foundation Model for Real-World Health Data: This paper presents MIRA, a foundation model specifically designed for irregular medical time series. Through continuous-time rotary position encoding (CT-RoPE), frequency-specific Mixture-of-Experts (MoE), and a Neural ODE-based extrapolation module, MIRA is pretrained on 454 billion observation points and achieves zero-shot forecasting performance that reduces average error by 8% and 6% in OOD and in-distribution (ID) settings, respectively.
Modeling X-ray Photon Pile-up with a Normalizing Flow: This paper proposes a Simulation-Based Inference (SBI) framework based on Normalizing Flows. A CNN extracts spatially resolved X-ray spectral features, which are then passed to a neural spline flow to perform accurate posterior estimation of astrophysical source parameters in the presence of photon pile-up, substantially outperforming the conventional PSF-core excision approach.
Mol-LLaMA: Towards General Understanding of Molecules in Large Molecular Language Models: This paper proposes Mol-LLaMA, a large molecular language model for general molecular understanding. By designing three types of instruction data and a 2D-3D molecular representation fusion module, Mol-LLaMA surpasses GPT-4o in molecular feature understanding while exhibiting interpretability and reasoning capabilities.
MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology: This paper introduces MTBBench—the first clinical benchmark simultaneously covering three dimensions: multimodality, longitudinal temporal sequencing, and interactive agent workflows. It simulates the decision-making process of Molecular Tumor Boards (MTBs) to evaluate and enhance the multimodal longitudinal reasoning capabilities of AI agents in precision oncology.
Multi-Objective Reinforcement Learning with Max-Min Criterion: A Game-Theoretic Approach: This paper reformulates max-min multi-objective reinforcement learning as a two-player zero-sum regularized continuous game, proposes the ERAM/ARAM algorithms, and leverages mirror descent to achieve a concise closed-form weight update. The approach guarantees global last-iterate convergence and significantly outperforms existing methods on tasks such as traffic signal control.
Multimodal 3D Genome Pre-training: This paper proposes MIX-HIC — the first multimodal foundation model for 3D genomics — which integrates Hi-C contact maps and epigenomic signals via cross-modal interaction blocks and cross-modal mapping blocks. Pre-trained on over 1.27 million paired samples, MIX-HIC achieves state-of-the-art performance across three downstream tasks: Hi-C prediction, chromatin loop detection, and CAGE-seq expression prediction.
Multimodal Bayesian Network for Robust Assessment of Casualties in Autonomous Triage: This paper proposes an expert-knowledge-driven Bayesian network decision-support framework that fuses outputs from multiple computer vision models to assess casualty conditions. Requiring no training data and supporting inference under incomplete information, the framework improved triage accuracy from 14% to 53% and diagnostic coverage from 31% to 95% in the DARPA Triage Challenge.
Multimodal Disease Progression Modeling via Spatiotemporal Disentanglement and Multiscale Alignment: This paper proposes DiPro, a framework that addresses redundancy in longitudinal chest X-ray sequences and cross-modal temporal misalignment through region-aware spatiotemporal disentanglement (separating static anatomical from dynamic pathological features) and multiscale alignment (local–global fusion of CXR and EHR), achieving state-of-the-art performance on disease progression recognition and ICU prediction tasks.
Multiscale Guidance of Protein Structure Prediction with Heterogeneous Cryo-EM Data: CryoBoltz leverages cryo-EM density maps to guide the sampling trajectory of a pretrained diffusion-based structure prediction model (Boltz-1) via a multiscale guidance mechanism (global → local), generating multi-conformational atomic models consistent with experimental data without any retraining.
MuSLR: Multimodal Symbolic Logical Reasoning: This paper introduces MuSLR, the first multimodal symbolic logical reasoning task, along with its benchmark MuSLR-Bench (1,093 instances spanning 7 domains, 35 atomic symbolic logic rules, and reasoning depths of 2–9). It further proposes LogiCAM, a modular framework comprising premise selection, reasoning type identification, and symbolic reasoning modules, which improves GPT-4.1's CoT performance by 14.13%.
NeurIPT: Foundation Model for Neural Interfaces: NeurIPT is an EEG foundation model for diverse brain–computer interface (BCI) applications. Through four key innovations—Amplitude-Aware Masking Pre-training (AAMP), Progressive Mixture-of-Experts (PMoE) architecture, 3D electrode spatial encoding, and Intra- and Inter-Lobe Pooling (IILP)—it achieves state-of-the-art performance across eight downstream BCI tasks.
One Small Step with Fingerprints, One Giant Leap for De Novo Molecule Generation from Mass Spectra: By employing MIST as a spectra-to-fingerprint encoder and MolForge as a fingerprint-to-structure decoder, combined with a prior-adjusted thresholding strategy, this work achieves a tenfold performance improvement on the MassSpecGym benchmark for de novo molecular structure generation from mass spectra (top-1 accuracy from 2.3% to 31%).
Online Feedback Efficient Active Target Discovery in Partially Observable Environments: This paper proposes DiffATD, which leverages the reverse process of diffusion models to construct a belief distribution for balancing exploration and exploitation, enabling efficient target region discovery in partially observable environments without any supervised training. The framework is applicable across multiple domains including medical imaging, species discovery, and remote sensing.
Ordinal Label-Distribution Learning with Constrained Asymmetric Priors for Imbalanced Retinal Grading: This paper proposes CAP-WAE (Constrained Asymmetric Prior Wasserstein Autoencoder), which addresses the challenges of long-tailed distribution and ordinal structure in diabetic retinopathy (DR) grading through three innovations: asymmetric priors, a margin-aware orthogonality and compactness loss, and a direction-aware ordinal loss, achieving state-of-the-art performance on multiple DR benchmarks.
Orochi: Versatile Biomedical Image Processor: This paper proposes Orochi—the first general-purpose foundation model for low-level biomedical image processing. Through Task-related Joint-embedding Pre-training (TJP) and a Multi-head Hierarchy Mamba architecture, Orochi matches or surpasses task-specific state-of-the-art models across four tasks—registration, fusion, restoration, and super-resolution—with lightweight fine-tuning of fewer than 5% of parameters.
Pancakes: Consistent Multi-Protocol Image Segmentation Across Biomedical Domains: This paper proposes the Pancakes framework, which, given a collection of biomedical images from an unseen domain, automatically generates label maps for multiple plausible segmentation protocols, ensuring semantic consistency across images within the same protocol—i.e., the same label refers to the same anatomical structure across all images.
PatientSim: A Persona-Driven Simulator for Realistic Doctor-Patient Interactions: This paper introduces PatientSim — an LLM-based patient simulator grounded in real MIMIC clinical data and a four-dimensional persona framework (personality, language proficiency, medical history recall, and cognitive confusion), generating 37 unique persona combinations. The system is evaluated across 8 LLMs for factual accuracy and persona fidelity, and validated by 4 clinical experts with a mean quality score of 3.89/4.
Pharmacophore-Guided Generative Design of Novel Drug-Like Molecules: This paper proposes a pharmacophore-guided molecular generation framework that simultaneously maximizes pharmacophore similarity and minimizes structural similarity within the reward function of a reinforcement learning model (FREED++), generating candidate drug molecules that retain bioactivity features while exhibiting high structural novelty.
PhysioWave: A Multi-Scale Wavelet-Transformer for Physiological Signal Representation: This paper proposes PhysioWave, a multi-scale Transformer architecture based on learnable wavelet decomposition and frequency-guided masking. It establishes, for the first time, large-scale pretrained foundation models for EMG and ECG, and achieves state-of-the-art performance on both unimodal and multimodal physiological signal tasks through a multimodal fusion framework.
PolyPose: Deformable 2D/3D Registration via Polyrigid Transformations: This paper presents PolyPose, a deformable 2D/3D registration method based on polyrigid transformations. Leveraging the anatomical prior that bones are rigid bodies, PolyPose parameterizes complex 3D deformation fields as weighted combinations of multiple rigid transformations in the Lie algebra $\mathfrak{se}(3)$, enabling accurate 3D volumetric registration from as few as two X-ray images without any regularization or hyperparameter tuning.
Position: Thematic Analysis of Unstructured Clinical Transcripts with Large Language Models: This position paper systematically reviews the current state of LLM-assisted thematic analysis (TA) on unstructured clinical transcripts, identifies highly fragmented evaluation practices across the literature, and proposes a standardized evaluation framework centered on three dimensions: Validity, Reliability, and Interpretability.
Posterior Sampling by Combining Diffusion Models with Annealed Langevin Dynamics: This paper proposes an algorithm combining diffusion models with annealed Langevin dynamics that requires only $L^4$-accurate score estimates to achieve polynomial-time posterior sampling under (locally) log-concave distributions, providing the first theoretical guarantees for warm-started inverse problem solving.
Prior-Guided Flow Matching for Target-Aware Molecule Design with Learnable Atom Number: This paper proposes PAFlow, a 3D molecule generation model built on the flow matching framework, which guides the vector field via a protein–ligand interaction predictor and determines atom counts through a learnable atom number predictor. PAFlow achieves a new state-of-the-art Avg. Vina Score of −8.31 on CrossDocked2020, substantially outperforming existing methods.
PROSPERO: Active Learning for Robust Protein Design Beyond Wild-Type Neighborhood: This paper proposes ProSpero, an active learning framework that discovers high-fitness and novel protein sequences even under surrogate model mismatch, via inference-time sampling of a frozen pretrained generative model (EvoDiff) guided by a surrogate, a targeted masking strategy, and biologically-constrained SMC sampling.
Protein Design with Dynamic Protein Vocabulary: ProDVa introduces natural protein fragments as a "dynamic vocabulary" for generative protein design, employing a three-component architecture consisting of a text encoder, a protein language model, and a fragment encoder. Using less than 0.04% of the training data required by prior work, ProDVa designs functionally aligned and structurally foldable protein sequences, surpassing the SOTA model Pinal by 7.38% on the pLDDT>70 ratio.
QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training: QoQ-Med constructs a multimodal clinical foundation model spanning 9 clinical modalities (1D ECG + 6 types of 2D images + 2 types of 3D scans), and proposes Domain-aware Relative Policy Optimization (DRPO)—which employs hierarchical temperature scaling (inter-domain × intra-domain K-means clustering) to address modality/difficulty imbalance. Trained on 2.61 million instruction-tuning pairs, it achieves an average F1 of 0.295 (vs. GRPO 0.193, +52.8%), ranking best in 6 out of 8 modalities.
Quantifying the Role of OpenFold Components in Protein Structure Prediction: This paper proposes a systematic methodology for evaluating the contribution of individual Evoformer components in OpenFold/AlphaFold2 to protein structure prediction accuracy. The study finds that MSA column attention and MLP Transition layers are the most critical components, and that the importance of multiple components is significantly correlated with protein sequence length.
RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis: This paper proposes RAD, a retrieval-augmented diagnostic framework that retrieves disease guidelines from multi-source medical corpora and injects them throughout the full pipeline of a multimodal model — from feature extraction to cross-modal fusion. A dual-axis explainability evaluation protocol is also introduced. RAD achieves state-of-the-art performance on four datasets spanning distinct anatomical regions.
RadZero: Similarity-Based Cross-Attention for Explainable Vision-Language Alignment in Chest X-ray: This paper proposes RadZero, a framework centered on VL-CABS (Vision-Language Cross-Attention Based on Similarity), enabling explainable and fine-grained vision-language alignment on chest X-rays with unified support for zero-shot classification, localization, and segmentation.
RAM-W600: A Multi-Task Wrist Dataset and Benchmark for Rheumatoid Arthritis: RAM-W600 is the first publicly available multi-task wrist conventional radiograph dataset, comprising 1,048 images and supporting two clinically relevant tasks: carpal bone instance segmentation and SvdH bone erosion (BE) scoring, accompanied by comprehensive benchmarking.
Random Search Neural Networks for Efficient and Expressive Graph Learning: This paper proposes Random Search Neural Networks (RSNN), which replace random walks with randomized depth-first search (DFS) for graph structure sampling. On sparse graphs, RSNN achieves complete edge coverage with only $O(\log|V|)$ searches. Paired with a universal sequence model, RSNN attains universal approximation capability, and consistently outperforms RWNN on molecular and protein benchmarks using up to 16× fewer samples.
RAxSS: Retrieval-Augmented Sparse Sampling for Explainable Variable-Length Medical Time Series Classification: This paper proposes RAxSS, a framework that integrates retrieval augmentation into the random sparse sampling (SSS) pipeline. By replacing uniform averaging with intra-window similarity-weighted aggregation, RAxSS maintains competitive performance on variable-length medical time series classification while providing an interpretable evidence chain spanning from "where" to "why."
Revisiting End-to-End Learning with Slide-level Supervision in Computational Pathology: This paper revisits end-to-end (E2E) learning with slide-level supervision in computational pathology, and is the first to identify optimization difficulties induced by sparse-attention MIL under E2E training. It proposes ABMILX, which addresses this issue via multi-head attention and a global attention correction module, enabling E2E-trained ResNets to surpass state-of-the-art foundation models on multiple benchmarks.
Robust or Suggestible? Exploring Non-Clinical Induction in LLM Drug-Safety Decisions: Through a persona-based evaluation framework, this paper finds that ChatGPT-4o and Bio-Medical-Llama-3-8B are systematically influenced by clinically irrelevant sociodemographic attributes (education, insurance, housing, etc.) in adverse drug event prediction, exhibiting both explicit and implicit bias patterns.
Scaling Laws and Pathologies of Single-Layer PINNs: Network Width and PDE Nonlinearity: This work establishes empirical scaling laws for single-layer PINNs on representative nonlinear PDEs, identifying a dual optimization failure: a width-scaling pathology (error does not decrease with width) and a compound pathology (nonlinearity exacerbates this failure), demonstrating that optimization rather than approximation capacity is the primary bottleneck.
Securing the Language of Life: Inheritable Watermarks from DNA Language Models to Proteins: This paper proposes DNAMark and CentralMark, two watermarking schemes for embedding robust watermarks in sequences generated by DNA language models. DNAMark achieves function-preserving watermarks via synonymous codon substitution, while CentralMark realizes inheritable watermarks that propagate from DNA to protein through the central dogma.
Self-supervised Learning of Echocardiographic Video Representations via Online Cluster Distillation: This paper proposes DISCOVR, a self-supervised dual-branch framework that transfers fine-grained spatial semantics from an image encoder to the temporal representations of a video encoder via online semantic cluster distillation, achieving state-of-the-art performance across six cross-population cardiac ultrasound datasets on anomaly detection, classification, and segmentation tasks.
Self-Supervised Learning via Flow-Guided Neural Operator on Time-Series Data: This paper proposes FGNO (Flow-Guided Neural Operator), which combines Flow Matching with operator learning for self-supervised pre-training on time-series data. By leveraging STFT for resolution-invariant function-space learning and treating flow time and network layer depth as adjustable "knobs" for controlling feature granularity, FGNO substantially outperforms baselines such as MAE on biomedical tasks.
Self Iterative Label Refinement via Robust Unlabeled Learning: This paper proposes an iterative pipeline that leverages a robust unlabeled-unlabeled (UU) learning framework to refine LLM-generated pseudo-labels, surpassing the self-refinement approaches of GPT-4o and DeepSeek-R1 on both classification and generative safety alignment tasks with minimal human annotation.
Semantic and Visual Crop-Guided Diffusion Models for Heterogeneous Tissue Synthesis in Histopathology: This paper proposes HeteroTissue-Diffuse (HTD), a dual-conditioned Latent Diffusion Model that generates heterogeneous pathology images by simultaneously conditioning on semantic segmentation maps and real tissue crops (visual crops). On Camelyon16, the method reduces Fréchet Distance from 430 to 72 (a 6× improvement). DeepLabv3+ segmentation IoU trained on synthetic data falls within 1–2% of models trained on real data. The approach is further extended to 11,765 unannotated TCGA whole-slide images via self-supervised clustering.
Sequential Attention-based Sampling for Histopathological Analysis: This paper proposes SASHA, a framework integrating a Hierarchical Attention-based Feature Distillation (HAFED) module with deep reinforcement learning (RL). By sampling only 10–20% of high-resolution patches, SASHA achieves classification performance on par with full-resolution SOTA methods, while yielding a 4–8× inference speedup and a WSI compression ratio exceeding 16×.
Shallow Robustness, Deep Vulnerabilities: Multi-Turn Evaluation of Medical LLMs: This paper proposes the MedQA-Followup framework to systematically evaluate the multi-turn robustness of medical LLMs. It reveals that models exhibit acceptable performance under single-turn perturbations (shallow robustness), yet accuracy can catastrophically drop from 91.2% to 13.5% under multi-turn follow-up challenges (deep vulnerability). Notably, indirect contextual manipulation proves more destructive than direct incorrect suggestions.
SMMILE: An Expert-Driven Benchmark for Multimodal Medical In-Context Learning: This paper presents SMMILE — the first expert-driven benchmark for multimodal medical in-context learning (ICL), comprising 111 questions (517 image-text QA triplets) spanning 6 medical specialties and 13 imaging modalities, constructed by 11 clinical experts. The benchmark systematically exposes critical deficiencies of current MLLMs in medical multimodal ICL and reveals the pivotal impact of in-context example quality and ordering on model performance.
SpecMER: Fast Protein Generation with K-mer Guided Speculative Decoding: SpecMER introduces speculative decoding into protein sequence generation, employing a K-mer-guided batch selection strategy to choose the candidate most consistent with evolutionary conservation from multiple draft model outputs for target model verification. It achieves 24–32% speedup while preserving distributional consistency, and the generated sequences demonstrate significantly improved NLL and pLDDT structural confidence scores compared to unguided baselines.
STAMP: Spatial-Temporal Adapter with Multi-Head Pooling: STAMP introduces a lightweight spatial-temporal adapter with only 750K parameters for Time Series Foundation Models (TSFMs). Through three sets of positional encodings (token/spatial/temporal), cross-gated MLP mixing, and multi-head attention pooling, it enables a frozen TSFM (e.g., MOMENT 385M) to compete with or surpass EEG-specific models with 29M parameters (CBraMod) across 8 EEG datasets, achieving 193% higher Kappa than CBraMod on BCIC-IV-2a.
STARC-9: A Large-scale Dataset for Multi-Class Tissue Classification for CRC Histopathology: This paper introduces STARC-9, a large-scale colorectal cancer (CRC) tissue classification dataset comprising 630K patches across 9 tissue classes, along with its construction framework DeepCluster++. The framework combines domain-specific autoencoder feature extraction, K-means clustering, and equal-frequency binning sampling to ensure morphological diversity. Models trained on STARC-9 significantly outperform those trained on NCT and HMU.
Steering Generative Models with Experimental Data for Protein Fitness Optimization: This work systematically evaluates strategies for steering protein generative models (discrete diffusion models and language models) toward fitness optimization, finding that plug-and-play guidance methods using small labeled datasets (~200 samples)—particularly DAPS—outperform RL-based fine-tuning, and proposes a Thompson sampling strategy incorporating predictive uncertainty for adaptive optimization.
Surf2CT: Cascaded 3D Flow Matching Models for Torso 3D CT Synthesis from Skin Surface: This paper proposes Surf2CT, a cascaded 3D Flow Matching framework that, for the first time, synthesizes complete high-resolution 3D CT volumes solely from external body surface scans and demographic data (age, sex, height, weight), without requiring any internal imaging input.
The Biased Oracle: Assessing LLMs' Understandability and Empathy in Medical Diagnoses: This work systematically evaluates GPT-4o and Claude-3.7 on readability and empathy in medical diagnostic communication. Both models produce reading levels well above recommended standards (grades 9–13 vs. the recommended grades 6–8). Affective empathy varies significantly with diagnosis type and patient education level, and LLM-as-Judge exhibits severe self-serving bias (GPT inflates its own empathy scores by ~0.3 points).
The Boundaries of Fair AI in Medical Image Prognosis: A Causal Perspective: FairTTE is the first comprehensive framework to systematically investigate fairness in time-to-event (TTE) prediction for medical imaging. It leverages causal analysis to quantify five sources of bias, and through training over 20,000 models, reveals the limitations of existing fairness methods — particularly the fundamental challenge of maintaining fairness under distribution shift.
THUNDER: Tile-level Histopathology image UNDERstanding benchmark: This paper presents THUNDER, a comprehensive tile-level benchmark for digital pathology foundation models, enabling efficient comparison of 23 foundation models across 16 datasets, covering downstream task performance, feature space analysis, robustness, and uncertainty estimation.
Toward a Vision-Language Foundation Model for Medical Data: Multimodal Dataset and Benchmarks for Vietnamese PET/CT Report Generation: This work introduces ViMed-PET, the first Vietnamese PET/CT image-report dataset comprising 2,757 whole-body PET/CT volumes paired with complete clinical reports. Through a data augmentation strategy and a three-stage fine-tuning pipeline, the approach substantially improves VLM performance on medical report generation and VQA tasks. Novel evaluation metrics based on clinically critical information are also proposed.
Towards Multiscale Graph-based Protein Learning with Geometric Secondary Structural Motifs: This paper proposes SSHG (Secondary Structure-based Hierarchical Graph), a framework that constructs two-level hierarchical graph representations from protein secondary structure motifs — an intra-motif residue-level graph and an inter-motif global graph — and employs a two-stage GNN to learn local and global features respectively. Theoretical guarantees of maximal expressiveness are provided, with empirical improvements in both accuracy and computational efficiency on enzyme classification and ligand affinity prediction tasks.
Towards Self-Supervised Foundation Models for Critical Care Time Series: A self-supervised foundation model for critical care time series is constructed by pre-training a Biaxial Transformer (BAT) architecture on multiple ICU datasets, substantially outperforming supervised baselines in low-data regimes.
Towards Unified and Lossless Latent Space for 3D Molecular Latent Diffusion Modeling: This paper proposes UAE-3D, a multimodal variational autoencoder that compresses atomic types, chemical bonds, and 3D coordinates of molecules into a unified, near-lossless latent space. By eliminating the complexity of handling multimodality and equivariance, a general-purpose Diffusion Transformer achieves state-of-the-art 3D molecular generation.
Uncertainty-Aware Multi-Objective Reinforcement Learning-Guided Diffusion Models for 3D De Novo Molecular Design: This paper proposes an uncertainty-aware multi-objective reinforcement learning framework that guides a 3D molecular diffusion model (EDM) to simultaneously optimize drug-likeness (QED), synthetic accessibility (SAS), and binding affinity. The framework dynamically shapes the reward function using predictive uncertainty from surrogate models, consistently outperforms baselines across three benchmark datasets, and validates candidate molecules through molecular dynamics simulations and ADMET analysis.
Unified All-Atom Molecule Generation with Neural Fields: This paper proposes FuncBind, a framework that represents molecules as continuous atomic density functions via neural fields, constructing a unified conditional generative model capable of target-conditioned generation across three drug modalities: small molecules, macrocyclic peptides, and antibody CDR loops.
UniMRSeg: Unified Modality-Relax Segmentation via Hierarchical Self-Supervised Compensation: This paper proposes UniMRSeg, a unified missing-modality segmentation framework that employs a Hierarchical Self-Supervised Compensation (HSSC) mechanism—spanning input-level modality reconstruction, feature-level contrastive learning, and output-level consistency regularization—to achieve optimal average performance and minimal performance variance across all possible modality combinations using 100% shared parameters.
UniSite: The First Cross-Structure Dataset and Learning Framework for End-to-End Ligand Binding Site Detection: This work introduces UniSite-DS, the first UniProt (unique protein)-centric ligand binding site dataset, and UniSite, the first end-to-end binding site detection framework. UniSite directly predicts multiple potentially overlapping binding sites via set prediction loss and bijective matching, and further proposes IoU-based AP as a more accurate evaluation metric.
Unlearned but Not Forgotten: Data Extraction after Exact Unlearning in LLM: This paper reveals that even exact unlearning (retraining from scratch to remove data influence) is susceptible to privacy leakage. By exploiting the divergence between model checkpoints before and after unlearning, an adversary can apply reversed model guidance with token filtering to substantially improve extraction success rates for deleted data—in some settings doubling the extraction rate.
Unpaired Image-to-Image Translation for Segmentation and Signal Unmixing: This paper proposes Ui2i, a model built upon CycleGAN that achieves high content-fidelity unpaired image-to-image translation through four key innovations: a UNet-based generator, approximate bidirectional spectral normalization (ABSN) as a replacement for feature normalization, channel-spatial attention, and scale augmentation. The model is successfully applied to two biomedical tasks: IHC→H&E domain adaptation for nucleus segmentation and single-channel immunofluorescence signal unmixing.
Variational Autoencoder with Normalizing Flow for X-ray Spectral Fitting: This work embeds a Normalizing Flow (NF) into an autoencoder architecture to enable fast physical parameter inference and full posterior distribution estimation for NICER spectral data of black hole X-ray binaries, achieving approximately 2000× speedup over traditional MCMC methods while maintaining comparable accuracy.
VQ-Seg: Vector-Quantized Token Perturbation for Semi-Supervised Medical Image Segmentation: VQ-Seg is proposed as the first method to introduce vector quantization into semi-supervised medical image segmentation. A Quantization Perturbation Module (QPM) replaces conventional dropout to achieve more controllable feature perturbation, complemented by a dual-branch architecture and foundation-model-guided alignment to compensate for quantization information loss.
Why Masking Diffusion Works: Condition on the Jump Schedule for Improved Discrete Diffusion: This paper reveals the fundamental reason for the superiority of masking diffusion models — they implicitly condition on the known jump-time distribution — and proposes the Schedule-Conditioned Diffusion (SCUD) framework, which generalizes this advantage to arbitrary discrete diffusion models. Combined with structured forward processes, SCUD surpasses masking diffusion on both image and protein generation tasks.