arXiv subjects

Wenjie Yan

Publications and source records attributed to Wenjie Yan.

5 recordsLinked to original sources

arXiv2026

Regulating Anatomy-Aware Rewards via Trajectory-Integral Feedback for Volumetric Computed Tomography Analysis

Medical vision-language models (VLMs) have rapidly advanced as general-purpose multimodal assistants, yet their deployment in 3D Computed Tomography (CT) analysis remains constrained by a persistent mismatch between optimization objectives and clinical rigor. Current Reinforcement Learning (RL) paradigms still rely on lexical proxy signals that induce ``\textit{Evaluation Hallucinations}'', where models optimize linguistic fluency rather than factual clinical correctness, leading to diagnostically critical errors. To bridge this gap, we introduce the \textbf{Clinical Abnormality Benchmarking Substrate (CABS)}, a structured system that decomposes radiology reports into verifiable clinical semantic units. Using CABS, we identify a ``\textit{Mechanistic Divergence}'' in standard RL, where surface-similarity rewards drive policy gradients to bypass medical facts. We therefore propose \textbf{Trajectory-Integral Feedback GRPO (TIF-GRPO)}, a novel framework integrating control-theoretic principles into policy optimization. By formulating clinical reasoning as a pseudo-temporal trajectory for anomaly discovery, TIF-GRPO regulates anatomy-aware rewards via an integral feedback loop that penalizes persistent omissions as cumulative state errors and suppresses hallucinations as excessive control effort. Experiments on 3D CT benchmarks demonstrate that our approach significantly enhances abnormality detection and clinical faithfulness, establishing a new paradigm for fine-grained regulation in medical VLMs. Our project is available at \href{https://github.com/ZJU4HealthCare/TIF-GRPO}{GitHub}.

cs.CV↗

arXiv2026

HeartcareGPT: A Unified Multimodal ECG Suite for Dual Signal-Image Modeling and Understanding

Although electrocardiograms (ECG) play a dominant role in cardiovascular diagnosis and treatment, their intrinsic data forms and representational patterns pose significant challenges for medical multimodal large language models (Med-MLLMs) in achieving cross-modal semantic alignment. To address this gap, we propose Heartcare Suite, a unified ECG suite designed for dual signal-image modeling and understanding: (i) Heartcare-400K. A fine-grained ECG instruction dataset on top of our data pipeline engine--HeartAgent--by integrating high quality clinical ECG reports from top hospitals with open-source data. (ii) Heartcare-Bench. A systematic benchmark assessing performance of models in multi-perspective ECG understanding and cross-modal generalization, providing guidance for optimizing ECG comprehension models. (iii) HeartcareGPT. Built upon a structure-aware discrete tokenizer Beat, we propose Dual Stream Projection Alignment (DSPA) paradigm--a dual encoder projection alignment mechanism enabling joint optimizing and modeling native ECG signal-image within a shared feature space. HeartcareGPT achieves consistent improvements across diverse ECG understanding tasks, validating both the effectiveness of the unified modeling paradigm and the necessity of a high-quality data pipeline, and establishing a methodological foundation for extending Med-MLLMs towards physiological signal domains. Our project is available at https://github.com/ZJU4HealthCare/HeartcareGPT .

cs.LG↗

arXiv2026

CLEAR-Mamba:Towards Accurate, Adaptive and Trustworthy Multi-Sequence Ophthalmic Angiography Classification

Medical image classification is a core task in computer-aided diagnosis (CAD), playing a pivotal role in early disease detection, treatment planning, and patient prognosis assessment. In ophthalmic practice, fluorescein fundus angiography (FFA) and indocyanine green angiography (ICGA) provide hemodynamic and lesion-structural information that conventional fundus photography cannot capture. However, due to the single-modality nature, subtle lesion patterns, and significant inter-device variability, existing methods still face limitations in generalization and high-confidence prediction. To address these challenges, we propose CLEAR-Mamba, an enhanced framework built upon MedMamba with optimizations in both architecture and training strategy. Architecturally, we introduce HaC, a hypernetwork-based adaptive conditioning layer that dynamically generates parameters according to input feature distributions, thereby improving cross-domain adaptability. From a training perspective, we develop RaP, a reliability-aware prediction scheme built upon evidential uncertainty learning, which encourages the model to emphasize low-confidence samples and improves overall stability and reliability. We further construct a large-scale ophthalmic angiography dataset covering both FFA and ICGA modalities, comprising multiple retinal disease categories for model training and evaluation. Experimental results demonstrate that CLEAR-Mamba consistently outperforms multiple baseline models, including the original MedMamba, across various metrics-showing particular advantages in multi-disease classification and reliability-aware prediction. This study provides an effective solution that balances generalizability and reliability for modality-specific medical image classification tasks. Our project can be accessed at https://github.com/ZJU4HealthCare/CLEAR-Mamba.

cs.CV↗

arXiv2025

Reconstructing Pristine Molecular Orbitals from Scanning Tunneling Microscopy Images via Artificial Intelligence Approaches

Molecular orbital (MO) is one of the most fundamental concepts for molecules, relating to all branches of chemistry, while scanning tunneling microscopy (STM) has been widely recognized for its potential to measure the spatial distribution of MOs. However, the precise characterization of MO with high resolution in real space is a long-standing challenge owing to the inevitable interference of high-angular-momentum contributions from functionalized tips in STM. Here, leveraging advances in artificial intelligence for image recognition, we establish a physics-driven deep-learning network, named STM-Net, to reconstruct MOs from high-resolution STM images with a functionalized tip, taking advantage of the separable characteristics of different angular momentum contributions. We demonstrate that STM-Net can be directly applied to a variety of experimental observations, successfully reconstructing pristine MO features for molecules under diverse conditions. Moreover, STM-Net can adapt to various states of the functionalized tip and the substrate, illustrating the broad applicability of our physics-driven framework. These results pave the way for accurate characterization of MO with high resolution, potentially leading to new insights and applications for this fundamental concept in chemistry.

physics.chem-ph↗

arXiv2024

Constructing accurate and efficient general-purpose atomistic machine learning model with transferable accuracy for quantum chemistry

Density Functional Theory (DFT) has been a cornerstone in computational science, providing powerful insights into structure-property relationships for molecules and materials through first-principles quantum-mechanical (QM) calculations. However, the advent of atomistic machine learning (ML) is reshaping the landscape by enabling large-scale dynamics simulations and high-throughput screening at DFT-equivalent accuracy with drastically reduced computational cost. Yet, the development of general-purpose atomistic ML models as surrogates for QM calculations faces several challenges, particularly in terms of model capacity, data efficiency, and transferability across chemically diverse systems. This work introduces a novel extension of the polarizable atom interaction neural network (namely, XPaiNN) to address these challenges. Two distinct training strategies have been employed, one direct-learning and the other $Δ$-ML on top of a semi-empirical QM method. These methodologies have been implemented within the same framework, allowing for a detailed comparison of their results. The XPaiNN models, in particular the one using $Δ$-ML, not only demonstrate competitive performance on standard benchmarks, but also demonstrate the effectiveness against other ML models and QM methods on comprehensive downstream tasks, including non-covalent interactions, reaction energetics, barrier heights, geometry optimization and reaction thermodynamics, etc. This work represents a significant step forward in the pursuit of accurate and efficient atomistic ML models of general-purpose, capable of handling complex chemical systems with transferable accuracy.

physics.chem-ph↗