S. Kevin Zhou

Chenyu You, Weicheng Dai, Yifei Min et al. · oxford

For medical image segmentation, contrastive learning is the dominant practice to improve the quality of visual representations by contrasting semantically similar and dissimilar pairs of samples. This is enabled by the observation that without accessing ground truth labels, negative examples with truly dissimilar anatomical features, if sampled, can significantly improve the performance. In reality, however, these samples may come from similar anatomical regions and the models may struggle to distinguish the minority tail-class samples, making the tail classes more prone to misclassification, both of which typically lead to model collapse. In this paper, we propose ARCO, a semi-supervised contrastive learning (CL) framework with stratified group theory for medical image segmentation. In particular, we first propose building ARCO through the concept of variance-reduced estimation and show that certain variance-reduction techniques are particularly beneficial in pixel/voxel-level segmentation tasks with extremely limited labels. Furthermore, we theoretically prove these sampling techniques are universal in variance reduction. Finally, we experimentally validate our approaches on eight benchmarks, i.e., five 2D/3D medical and three semantic segmentation datasets, with different label settings, and our methods consistently outperform state-of-the-art semi-supervised methods. Additionally, we augment the CL frameworks with these sampling techniques and demonstrate significant gains over previous methods. We believe our work is an important step towards semi-supervised medical image segmentation by quantifying the limitation of current self-supervision objectives for accomplishing such challenging safety-critical tasks.

Ziyuan Zhao, Fangcheng Zhou, Kaixin Xu et al.

While deep learning methods hitherto have achieved considerable success in medical image segmentation, they are still hampered by two limitations: (i) reliance on large-scale well-labeled datasets, which are difficult to curate due to the expert-driven and time-consuming nature of pixel-level annotations in clinical practices, and (ii) failure to generalize from one domain to another, especially when the target domain is a different modality with severe domain shifts. Recent unsupervised domain adaptation~(UDA) techniques leverage abundant labeled source data together with unlabeled target data to reduce the domain gap, but these methods degrade significantly with limited source annotations. In this study, we address this underexplored UDA problem, investigating a challenging but valuable realistic scenario, where the source domain not only exhibits domain shift~w.r.t. the target domain but also suffers from label scarcity. In this regard, we propose a novel and generic framework called ``Label-Efficient Unsupervised Domain Adaptation"~(LE-UDA). In LE-UDA, we construct self-ensembling consistency for knowledge transfer between both domains, as well as a self-ensembling adversarial learning module to achieve better feature alignment for UDA. To assess the effectiveness of our method, we conduct extensive experiments on two different tasks for cross-modality segmentation between MRI and CT images. Experimental results demonstrate that the proposed LE-UDA can efficiently leverage limited source labels to improve cross-domain segmentation performance, outperforming state-of-the-art UDA approaches in the literature. Code is available at: https://github.com/jacobzhaoziyuan/LE-UDA.

Qing Wu, Lixuan Chen, Ce Wang et al.

Emerging neural reconstruction techniques based on tomography (e.g., NeRF, NeAT, and NeRP) have started showing unique capabilities in medical imaging. In this work, we present a novel Polychromatic neural representation (Polyner) to tackle the challenging problem of CT imaging when metallic implants exist within the human body. CT metal artifacts arise from the drastic variation of metal's attenuation coefficients at various energy levels of the X-ray spectrum, leading to a nonlinear metal effect in CT measurements. Recovering CT images from metal-affected measurements hence poses a complicated nonlinear inverse problem where empirical models adopted in previous metal artifact reduction (MAR) approaches lead to signal loss and strongly aliased reconstructions. Polyner instead models the MAR problem from a nonlinear inverse problem perspective. Specifically, we first derive a polychromatic forward model to accurately simulate the nonlinear CT acquisition process. Then, we incorporate our forward model into the implicit neural representation to accomplish reconstruction. Lastly, we adopt a regularizer to preserve the physical properties of the CT images across different energy levels while effectively constraining the solution space. Our Polyner is an unsupervised method and does not require any external training data. Experimenting with multiple datasets shows that our Polyner achieves comparable or better performance than supervised methods on in-domain datasets while demonstrating significant performance improvements on out-of-domain datasets. To the best of our knowledge, our Polyner is the first unsupervised MAR method that outperforms its supervised counterparts. The code for this work is available at: https://github.com/iwuqing/Polyner.

Ruinan Jin, Zikang Xu, Yuan Zhong et al.

The advent of foundation models (FMs) in healthcare offers unprecedented opportunities to enhance medical diagnostics through automated classification and segmentation tasks. However, these models also raise significant concerns about their fairness, especially when applied to diverse and underrepresented populations in healthcare applications. Currently, there is a lack of comprehensive benchmarks, standardized pipelines, and easily adaptable libraries to evaluate and understand the fairness performance of FMs in medical imaging, leading to considerable challenges in formulating and implementing solutions that ensure equitable outcomes across diverse patient populations. To fill this gap, we introduce FairMedFM, a fairness benchmark for FM research in medical imaging.FairMedFM integrates with 17 popular medical imaging datasets, encompassing different modalities, dimensionalities, and sensitive attributes. It explores 20 widely used FMs, with various usages such as zero-shot learning, linear probing, parameter-efficient fine-tuning, and prompting in various downstream tasks -- classification and segmentation. Our exhaustive analysis evaluates the fairness performance over different evaluation metrics from multiple perspectives, revealing the existence of bias, varied utility-fairness trade-offs on different FMs, consistent disparities on the same datasets regardless FMs, and limited effectiveness of existing unfairness mitigation methods. Checkout FairMedFM's project page and open-sourced codebase, which supports extendible functionalities and applications as well as inclusive for studies on FMs in medical imaging over the long term.

Fenghe Tang, Ronghao Xu, Qingsong Yao et al.

The generative self-supervised learning strategy exhibits remarkable learning representational capabilities. However, there is limited attention to end-to-end pre-training methods based on a hybrid architecture of CNN and Transformer, which can learn strong local and global representations simultaneously. To address this issue, we propose a generative pre-training strategy called Hybrid Sparse masKing (HySparK) based on masked image modeling and apply it to large-scale pre-training on medical images. First, we perform a bottom-up 3D hybrid masking strategy on the encoder to keep consistency masking. Then we utilize sparse convolution for the top CNNs and encode unmasked patches for the bottom vision Transformers. Second, we employ a simple hierarchical decoder with skip-connections to achieve dense multi-scale feature reconstruction. Third, we implement our pre-training method on a collection of multiple large-scale 3D medical imaging datasets. Extensive experiments indicate that our proposed pre-training strategy demonstrates robust transfer-ability in supervised downstream tasks and sheds light on HySparK's promising prospects. The code is available at https://github.com/FengheTan9/HySparK

Yue Zhang, Chengtao Peng, Qiuli Wang et al.

Multi-modal medical images provide complementary soft-tissue characteristics that aid in the screening and diagnosis of diseases. However, limited scanning time, image corruption and various imaging protocols often result in incomplete multi-modal images, thus limiting the usage of multi-modal data for clinical purposes. To address this issue, in this paper, we propose a novel unified multi-modal image synthesis method for missing modality imputation. Our method overall takes a generative adversarial architecture, which aims to synthesize missing modalities from any combination of available ones with a single model. To this end, we specifically design a Commonality- and Discrepancy-Sensitive Encoder for the generator to exploit both modality-invariant and specific information contained in input modalities. The incorporation of both types of information facilitates the generation of images with consistent anatomy and realistic details of the desired distribution. Besides, we propose a Dynamic Feature Unification Module to integrate information from a varying number of available modalities, which enables the network to be robust to random missing modalities. The module performs both hard integration and soft integration, ensuring the effectiveness of feature combination while avoiding information loss. Verified on two public multi-modal magnetic resonance datasets, the proposed method is effective in handling various synthesis tasks and shows superior performance compared to previous methods.

Fenghe Tang, Bingkun Nian, Yingtai Li et al.

Recently, the state space model Mamba has demonstrated efficient long-sequence modeling capabilities, particularly for addressing long-sequence visual tasks in 3D medical imaging. However, existing generative self-supervised learning methods have not yet fully unleashed Mamba's potential for handling long-range dependencies because they overlook the inherent causal properties of state space sequences in masked modeling. To address this challenge, we propose a general-purpose pre-training framework called MambaMIM, a masked image modeling method based on a novel TOKen-Interpolation strategy (TOKI) for the selective structure state space sequence, which learns causal relationships of state space within the masked sequence. Further, MambaMIM introduces a bottom-up 3D hybrid masking strategy to maintain a masking consistency across different architectures and can be used on any single or hybrid Mamba architecture to enhance its multi-scale and long-range representation capability. We pre-train MambaMIM on a large-scale dataset of 6.8K CT scans and evaluate its performance across eight public medical segmentation benchmarks. Extensive downstream experiments reveal the feasibility and advancement of using Mamba for medical image pre-training. In particular, when we apply the MambaMIM to a customized architecture that hybridizes MedNeXt and Vision Mamba, we consistently obtain the state-of-the-art segmentation performance. The code is available at: https://github.com/FengheTan9/MambaMIM.

Chenxu Wu, Qingpeng Kong, Peiang Zhao et al.

Recent advances in generative models, especially diffusion models, have significantly improved image restoration (IR) performance. However, existing problem-agnostic diffusion model-based image restoration (DMIR) methods face challenges in fully leveraging diffusion priors, resulting in suboptimal performance. In this paper, we address the limitations of current problem-agnostic DMIR methods by analyzing their sampling process and providing effective solutions. We introduce EquS, a DMIR method that imposes equivariant information through dual sampling trajectories. To further boost EquS, we propose the Timestep-Aware Schedule (TAS) and introduce EquS$^+$. TAS prioritizes deterministic steps to enhance certainty and sampling efficiency. Extensive experiments on benchmarks demonstrate that our method is compatible with previous problem-agnostic DMIR methods and significantly boosts their performance without increasing computational costs. Our code is available at https://github.com/FouierL/EquS.

Jiahua Dong, Yue Zhang, Qiuli Wang et al.

Medical image segmentation is crucial in the field of medical imaging, aiding in disease diagnosis and surgical planning. Most established segmentation methods rely on supervised deep learning, in which clean and precise labels are essential for supervision and significantly impact the performance of models. However, manually delineated labels often contain noise, such as missing labels and inaccurate boundary delineation, which can hinder networks from correctly modeling target characteristics. In this paper, we propose a deep self-cleansing segmentation framework that can preserve clean labels while cleansing noisy ones in the training phase. To achieve this, we devise a gaussian mixture model-based label filtering module that distinguishes noisy labels from clean labels. Additionally, we develop a label cleansing module to generate pseudo low-noise labels for identified noisy samples. The preserved clean labels and pseudo-labels are then used jointly to supervise the network. Validated on a clinical liver tumor dataset and a public cardiac diagnosis dataset, our method can effectively suppress the interference from noisy labels and achieve prominent segmentation performance.

Quan Quan, Shang Zhao, Qingsong Yao et al.

The augmentation parameters matter to few-shot semantic segmentation since they directly affect the training outcome by feeding the networks with varying perturbated samples. However, searching optimal augmentation parameters for few-shot segmentation models without annotations is a challenge that current methods fail to address. In this paper, we first propose a framework to determine the ``optimal'' parameters without human annotations by solving a distribution-matching problem between the intra-instance and intra-class similarity distribution, with the intra-instance similarity describing the similarity between the original sample of a particular anatomy and its augmented ones and the intra-class similarity representing the similarity between the selected sample and the others in the same class. Extensive experiments demonstrate the superiority of our optimized augmentation in boosting few-shot segmentation models. We greatly improve the top competing method by 1.27\% and 1.11\% on Abd-MRI and Abd-CT datasets, respectively, and even achieve a significant improvement for SSL-ALP on the left kidney by 3.39\% on the Abd-CT dataset.

Yukun Cao, Shuo Han, Zengyi Gao et al.

Although Large Language Models (LLMs) have demonstrated potential in processing graphs, they struggle with comprehending graphical structure information through prompts of graph description sequences, especially as the graph size increases. We attribute this challenge to the uneven memory performance of LLMs across different positions in graph description sequences, known as ''positional biases''. To address this, we propose GraphInsight, a novel framework aimed at improving LLMs' comprehension of both macro- and micro-level graphical information. GraphInsight is grounded in two key strategies: 1) placing critical graphical information in positions where LLMs exhibit stronger memory performance, and 2) investigating a lightweight external knowledge base for regions with weaker memory performance, inspired by retrieval-augmented generation (RAG). Moreover, GraphInsight explores integrating these two strategies into LLM agent processes for composite graph tasks that require multi-step reasoning. Extensive empirical studies on benchmarks with a wide range of evaluation tasks show that GraphInsight significantly outperforms all other graph description methods (e.g., prompting techniques and reordering strategies) in understanding graph structures of varying sizes.

Haoran Lai, Qingsong Yao, Zihang Jiang et al.

The advancement of Zero-Shot Learning in the medical domain has been driven forward by using pre-trained models on large-scale image-text pairs, focusing on image-text alignment. However, existing methods primarily rely on cosine similarity for alignment, which may not fully capture the complex relationship between medical images and reports. To address this gap, we introduce a novel approach called Cross-Attention Alignment for Radiology Zero-Shot Classification (CARZero). Our approach innovatively leverages cross-attention mechanisms to process image and report features, creating a Similarity Representation that more accurately reflects the intricate relationships in medical semantics. This representation is then linearly projected to form an image-text similarity matrix for cross-modality alignment. Additionally, recognizing the pivotal role of prompt selection in zero-shot learning, CARZero incorporates a Large Language Model-based prompt alignment strategy. This strategy standardizes diverse diagnostic expressions into a unified format for both training and inference phases, overcoming the challenges of manual prompt design. Our approach is simple yet effective, demonstrating state-of-the-art performance in zero-shot classification on five official chest radiograph diagnostic test sets, including remarkable results on datasets with long-tail distributions of rare diseases. This achievement is attributed to our new image-text alignment strategy, which effectively addresses the complex relationship between medical images and reports. Code and models are available at https://github.com/laihaoran/CARZero.

Sizhuo Zhou, Xiaosong Jia, Fanrui Zhang et al.

Generative world models for autonomous driving (AD) have become a trending topic. Unlike the widely studied image modality, in this work we explore generative world models for LiDAR data. Existing generation methods for LiDAR data only support single frame generation, while existing prediction approaches require multiple frames of historical input and can only deterministically predict multiple frames at once, lacking interactivity. Both paradigms fail to support long-horizon interactive generation. To this end, we introduce LaGen, which to the best of our knowledge is the first framework capable of frame-by-frame autoregressive generation of long-horizon LiDAR scenes. LaGen is able to take a single-frame LiDAR input as a starting point and effectively utilize bounding box information as conditions to generate high-fidelity 4D scene point clouds. In addition, we introduce a scene decoupling estimation module to enhance the model's interactive generation capability for object-level content, as well as a noise modulation module to mitigate error accumulation during long-horizon generation. We construct a protocol based on nuScenes for evaluating long-horizon LiDAR scene generation. Experimental results comprehensively demonstrate LaGen outperforms state-of-the-art LiDAR generation and prediction models, especially on the later frames.

Jiangyong Yu, Changyong Shu, Sifan Zhou et al.

Camera-based multi-view 3D detection is crucial for autonomous driving. PETR and its variants (PETRs) excel in benchmarks but face deployment challenges due to high computational cost and memory footprint. Quantization is an effective technique for compressing deep neural networks by reducing the bit width of weights and activations. However, directly applying existing quantization methods to PETRs leads to severe accuracy degradation. This issue primarily arises from two key challenges: (1) significant magnitude disparity between multi-modal features-specifically, image features and camera-ray positional embeddings (PE), and (2) the inefficiency and approximation error of quantizing non-linear operators, which commonly rely on hardware-unfriendly computations. In this paper, we propose FQ-PETR, a fully quantized framework for PETRs, featuring three key innovations: (1) Quantization-Friendly LiDAR-ray Position Embedding (QFPE): Replacing multi-point sampling with LiDAR-prior-guided single-point sampling and anchor-based embedding eliminates problematic non-linearities (e.g., inverse-sigmoid) and aligns PE scale with image features, preserving accuracy. (2) Dual-Lookup Table (DULUT): This algorithm approximates complex non-linear functions using two cascaded linear LUTs, achieving high fidelity with minimal entries and no specialized hardware. (3) Quantization After Numerical Stabilization (QANS): Performing quantization after softmax numerical stabilization mitigates attention distortion from large inputs. On PETRs (e.g. PETR, StreamPETR, PETRv2, MV2d), FQ-PETR under W8A8 achieves near-floating-point accuracy (1% degradation) while reducing latency by up to 75%, significantly outperforming existing PTQ and QAT baselines.

Fenghe Tang, Qingsong Yao, Wenxin Ma et al.

Medical image segmentation remains a formidable challenge due to the label scarcity. Pre-training Vision Transformer (ViT) through masked image modeling (MIM) on large-scale unlabeled medical datasets presents a promising solution, providing both computational efficiency and model generalization for various downstream tasks. However, current ViT-based MIM pre-training frameworks predominantly emphasize local aggregation representations in output layers and fail to exploit the rich representations across different ViT layers that better capture fine-grained semantic information needed for more precise medical downstream tasks. To fill the above gap, we hereby present Hierarchical Encoder-driven MAE (Hi-End-MAE), a simple yet effective ViT-based pre-training solution, which centers on two key innovations: (1) Encoder-driven reconstruction, which encourages the encoder to learn more informative features to guide the reconstruction of masked patches; and (2) Hierarchical dense decoding, which implements a hierarchical decoding structure to capture rich representations across different layers. We pre-train Hi-End-MAE on a large-scale dataset of 10K CT scans and evaluated its performance across seven public medical image segmentation benchmarks. Extensive experiments demonstrate that Hi-End-MAE achieves superior transfer learning capabilities across various downstream tasks, revealing the potential of ViT in medical imaging applications. The code is available at: https://github.com/FengheTan9/Hi-End-MAE

Peiang Zhao, Han Li, Ruiyang Jin et al.

Recent text-to-image diffusion models have reached an unprecedented level in generating high-quality images. However, their exclusive reliance on textual prompts often falls short in precise control of image compositions. In this paper, we propose LoCo, a training-free approach for layout-to-image Synthesis that excels in producing high-quality images aligned with both textual prompts and layout instructions. Specifically, we introduce a Localized Attention Constraint (LAC), leveraging semantic affinity between pixels in self-attention maps to create precise representations of desired objects and effectively ensure the accurate placement of objects in designated regions. We further propose a Padding Token Constraint (PTC) to leverage the semantic information embedded in previously neglected padding tokens, improving the consistency between object appearance and layout instructions. LoCo seamlessly integrates into existing text-to-image and layout-to-image models, enhancing their performance in spatial control and addressing semantic failures observed in prior methods. Extensive experiments showcase the superiority of our approach, surpassing existing state-of-the-art training-free layout-to-image methods both qualitatively and quantitatively across multiple benchmarks.

Ziqi Gao, S. Kevin Zhou

Implicit visual knowledge in a large latent diffusion model (LLDM) pre-trained on natural images is rich and hypothetically universal to natural and medical images. To test this hypothesis from a practical perspective, we propose a novel framework for undersampled MRI Reconstruction by Prompting a large latent Diffusion model (MRPD). While the existing methods trained on MRI datasets are typically of limited generalizability toward diverse data acquisition scenarios, MRPD supports unsupervised and universally adaptive MRI reconstruction. For unsupervised reconstruction, MRSampler guides LLDM with a random-phase-modulated hard-to-soft control. With any single- or multiple-source MRI dataset, MRPD's performance is boosted universally by a lightweight MRAdapter that only finetunes the LLDM's autoencoder. Experiments on FastMRI and IXI show that MRPD is the only model that supports both MRI database-free and database-available scenarios and attains the best generalizability towards out-of-domain (OOD) samplings, contrasts, and organs among compared unsupervised, supervised, and MRI diffusion methods. To our knowledge, MRPD is the first method that empirically shows the universal prowess of an LLDM pre-trained on vast natural images for MRI. Our official implementation is at https://github.com/Z7Gao/MRPD.

Chenxu Wu, Qingpeng Kong, Zihang Jiang et al.

Magnetic Resonance Imaging (MRI), including diffusion MRI (dMRI), serves as a ``microscope'' for anatomical structures and routinely mitigates the influence of low signal-to-noise ratio scans by compromising temporal or spatial resolution. However, these compromises fail to meet clinical demands for both efficiency and precision. Consequently, denoising is a vital preprocessing step, particularly for dMRI, where clean data is unavailable. In this paper, we introduce Di-Fusion, a fully self-supervised denoising method that leverages the latter diffusion steps and an adaptive sampling process. Unlike previous approaches, our single-stage framework achieves efficient and stable training without extra noise model training and offers adaptive and controllable results in the sampling process. Our thorough experiments on real and simulated data demonstrate that Di-Fusion achieves state-of-the-art performance in microstructure modeling, tractography tracking, and other downstream tasks. Code is available at https://github.com/FouierL/Di-Fusion.

Fanrui Zhang, Qiang Zhang, Sizhuo Zhou et al.

Existing image forgery detection (IFD) methods either exploit low-level, semantics-agnostic artifacts or rely on multimodal large language models (MLLMs) with high-level semantic knowledge. Although naturally complementary, these two information streams are highly heterogeneous in both paradigm and reasoning, making it difficult for existing methods to unify them or effectively model their cross-level interactions. To address this gap, we propose ForenAgent, a multi-round interactive IFD framework that enables MLLMs to autonomously generate, execute, and iteratively refine Python-based low-level tools around the detection objective, thereby achieving more flexible and interpretable forgery analysis. ForenAgent follows a two-stage training pipeline combining Cold Start and Reinforcement Fine-Tuning to enhance its tool interaction capability and reasoning adaptability progressively. Inspired by human reasoning, we design a dynamic reasoning loop comprising global perception, local focusing, iterative probing, and holistic adjudication, and instantiate it as both a data-sampling strategy and a task-aligned process reward. For systematic training and evaluation, we construct FABench, a heterogeneous, high-quality agent-forensics dataset comprising 100k images and approximately 200k agent-interaction question-answer pairs. Experiments show that ForenAgent exhibits emergent tool-use competence and reflective reasoning on challenging IFD tasks when assisted by low-level tools, charting a promising route toward general-purpose IFD. The code will be released after the review process is completed.

Yifan Li, Fenghe Tang, Yingtai Li et al.

General-purpose large Vision-Language Models (VLMs) demonstrate strong capabilities in generating detailed descriptions for natural images. However, their performance in the medical domain remains suboptimal, even for relatively straightforward tasks, primarily due to the lack of large-scale, high-quality, specialized medical imaging datasets and the neglect of the diagnostic process that progresses from coarse to fine-grained. To address the first issue, we construct the CT-RATE-VQA dataset, which has 84K QA pairs. For the second issue, we propose MedReason-R1, a medical VLM with explicit reasoning process for disease diagnosis. MedReason-R1 incorporates a novel strategy that embeds zoom-in disease region-of-interest areas into the image, highlighting the crucial role of both global localization and disease-specific details in enhancing the model's diagnostic performance. Furthermore, we introduce the GRPO reinforcement learning framework to MedReason-R1, which enables effective reasoning without relying on costly manual annotations. Compared to recent general-purpose and medical VLMs, MedReason-R1 achieves state-of-the-art performance in CT disease diagnosis while retaining generalization. The code, checkpoints, and dataset are available at: https://github.com/Leevan001/MedReason-R1

Fenghe Tang, Chengqi Dong, Wenxin Ma et al.

Over the past decade, U-Net has been the dominant architecture in medical image segmentation, leading to the development of thousands of U-shaped variants. Despite its widespread adoption, there is still no comprehensive benchmark to systematically evaluate their performance and utility, largely because of insufficient statistical validation and limited consideration of efficiency and generalization across diverse datasets. To bridge this gap, we present U-Bench, the first large-scale, statistically rigorous benchmark that evaluates 100 U-Net variants across 28 datasets and 10 imaging modalities. Our contributions are threefold: (1) Comprehensive Evaluation: U-Bench evaluates models along three key dimensions: statistical robustness, zero-shot generalization, and computational efficiency. We introduce a novel metric, U-Score, which jointly captures the performance-efficiency trade-off, offering a deployment-oriented perspective on model progress. (2) Systematic Analysis and Model Selection Guidance: We summarize key findings from the large-scale evaluation and systematically analyze the impact of dataset characteristics and architectural paradigms on model performance. Based on these insights, we propose a model advisor agent to guide researchers in selecting the most suitable models for specific datasets and tasks. (3) Public Availability: We provide all code, models, protocols, and weights, enabling the community to reproduce our results and extend the benchmark with future methods. In summary, U-Bench not only exposes gaps in previous evaluations but also establishes a foundation for fair, reproducible, and practically relevant benchmarking in the next decade of U-Net-based segmentation models. The project can be accessed at: https://fenghetan9.github.io/ubench. Code is available at: https://github.com/FengheTan9/U-Bench.

Wenliang Li, Rui Yan, Xu Zhang et al.

Large language models (LLMs) have demonstrated notable potential in medical applications, yet they face substantial challenges in handling complex real-world clinical diagnoses using conventional prompting methods. Current prompt engineering and multi-agent approaches typically optimize isolated inferences, neglecting the accumulation of reusable clinical experience. To address this, this study proposes a novel Multi-Agent Clinical Diagnosis (MACD) framework, which allows LLMs to self-learn clinical knowledge via a multi-agent pipeline that summarizes, refines, and applies diagnostic insights. It mirrors how physicians develop expertise through experience, enabling more focused and accurate diagnosis on key disease-specific cues. We further extend it to a MACD-human collaborative workflow, where multiple LLM-based diagnostician agents engage in iterative consultations, supported by an evaluator agent and human oversight for cases where agreement is not reached. Evaluated on 4,390 real-world patient cases across seven diseases using diverse open-source LLMs (Llama-3.1 8B/70B, DeepSeek-R1-Distill-Llama 70B), MACD significantly improves primary diagnostic accuracy, outperforming established clinical guidelines with gains up to 22.3% (MACD). In direct comparison with physician-only diagnosis under the same evaluation protocol, MACD achieves comparable or superior performance, with improvements up to 16%. Furthermore, the MACD-human workflow yields an 18.6% improvement over physician-only diagnosis, demonstrating the synergistic potential of human-AI collaboration. Notably, the self-learned clinical knowledge exhibits strong cross-model stability, transferability across LLMs, and capacity for model-specific personalization.This work thus presents a scalable self-learning paradigm that bridges the gap between the intrinsic knowledge of LLMs.

Yingtai Li, Haoran Lai, Xiaoqian Zhou et al.

The emergence of Large Language Models (LLMs) presents unprecedented opportunities to revolutionize medical contrastive vision-language pre-training. In this paper, we show how LLMs can facilitate large-scale supervised pre-training, thereby advancing vision-language alignment. We begin by demonstrate that modern LLMs can automatically extract diagnostic labels from radiology reports with remarkable precision (>96\% AUC in our experiments) without complex prompt engineering, enabling the creation of large-scale "silver-standard" datasets at a minimal cost (~\$3 for 50k CT image-report pairs). Further, we find that vision encoder trained on this "silver-standard" dataset achieves performance comparable to those trained on labels extracted by specialized BERT-based models, thereby democratizing the access to large-scale supervised pre-training. Building on this foundation, we proceed to reveal that supervised pre-training fundamentally improves contrastive vision-language alignment. Our approach achieves state-of-the-art performance using only a 3D ResNet-18 with vanilla CLIP training, including 83.8\% AUC for zero-shot diagnosis on CT-RATE, 77.3\% AUC on RAD-ChestCT, and substantial improvements in cross-modal retrieval (MAP@50=53.7\% for image-image, Recall@100=52.2\% for report-image). These results demonstrate the potential of utilizing LLMs to facilitate {\bf more performant and scalable} medical AI systems. Our code is avaiable at https://github.com/SadVoxel/More-performant-and-scalable.

Rongsheng Wang, Fenghe Tang, Qingsong Yao et al.

Medical vision-language pre-training shows great potential in learning representative features from massive paired radiographs and reports. However, in computed tomography (CT) scans, the distribution of lesions which contain intricate structures is characterized by spatial sparsity. Besides, the complex and implicit relationships between different pathological descriptions in each sentence of the report and their corresponding sub-regions in radiographs pose additional challenges. In this paper, we propose a Similarity-Driven Cross-Granularity Pre-training (SimCroP) framework on chest CTs, which combines similarity-driven alignment and cross-granularity fusion to improve radiograph interpretation. We first leverage multi-modal masked modeling to optimize the encoder for understanding precise low-level semantics from radiographs. Then, similarity-driven alignment is designed to pre-train the encoder to adaptively select and align the correct patches corresponding to each sentence in reports. The cross-granularity fusion module integrates multimodal information across instance level and word-patch level, which helps the model better capture key pathology structures in sparse radiographs, resulting in improved performance for multi-scale downstream tasks. SimCroP is pre-trained on a large-scale paired CT-reports dataset and validated on image classification and segmentation tasks across five public datasets. Experimental results demonstrate that SimCroP outperforms both cutting-edge medical self-supervised learning methods and medical vision-language pre-training methods. Codes and models are available at https://github.com/ToniChopp/SimCroP.

Chengqi Dong, Fenghe Tang, Rongge Mao et al.

Medical image segmentation plays a pivotal role in disease diagnosis and treatment planning, particularly in resource-constrained clinical settings where lightweight and generalizable models are urgently needed. However, existing lightweight models often compromise performance for efficiency and rarely adopt computationally expensive attention mechanisms, severely restricting their global contextual perception capabilities. Additionally, current architectures neglect the channel redundancy issue under the same convolutional kernels in medical imaging, which hinders effective feature extraction. To address these challenges, we propose LGMSNet, a novel lightweight framework based on local and global dual multiscale that achieves state-of-the-art performance with minimal computational overhead. LGMSNet employs heterogeneous intra-layer kernels to extract local high-frequency information while mitigating channel redundancy. In addition, the model integrates sparse transformer-convolutional hybrid branches to capture low-frequency global information. Extensive experiments across six public datasets demonstrate LGMSNet's superiority over existing state-of-the-art methods. In particular, LGMSNet maintains exceptional performance in zero-shot generalization tests on four unseen datasets, underscoring its potential for real-world deployment in resource-limited medical scenarios. The whole project code is in https://github.com/cq-dong/LGMSNet.

Yuan Feng, Yukun Cao, Hairu Wang et al.

Sketches, probabilistic structures for estimating item frequencies in infinite data streams with limited space, are widely used across various domains. Recent studies have shifted the focus from handcrafted sketches to neural sketches, leveraging memory-augmented neural networks (MANNs) to enhance the streaming compression capabilities and achieve better space-accuracy trade-offs.However, existing neural sketches struggle to scale across different data domains and space budgets due to inflexible MANN configurations. In this paper, we introduce a scalable MANN architecture that brings to life the {\it Lego sketch}, a novel sketch with superior scalability and accuracy. Much like assembling creations with modular Lego bricks, the Lego sketch dynamically coordinates multiple memory bricks to adapt to various space budgets and diverse data domains. Our theoretical analysis guarantees its high scalability and provides the first error bound for neural sketch. Furthermore, extensive experimental evaluations demonstrate that the Lego sketch exhibits superior space-accuracy trade-offs, outperforming existing handcrafted and neural sketches. Our code is available at https://github.com/FFY0/LegoSketch_ICML.

Yuetong Fang, Deming Zhou, Ziqing Wang et al.

Spiking Neural Networks promise brain-inspired and energy-efficient computation by transmitting information through binary (0/1) spikes. Yet, their performance still lags behind that of artificial neural networks, often assumed to result from information loss caused by sparse and binary activations. In this work, we challenge this long-standing assumption and reveal a previously overlooked frequency bias: spiking neurons inherently suppress high-frequency components and preferentially propagate low-frequency information. This frequency-domain imbalance, we argue, is the root cause of degraded feature representation in SNNs. Empirically, on Spiking Transformers, adopting Avg-Pooling (low-pass) for token mixing lowers performance to 76.73% on Cifar-100, whereas replacing it with Max-Pool (high-pass) pushes the top-1 accuracy to 79.12%. Accordingly, we introduce Max-Former that restores high-frequency signals through two frequency-enhancing operators: (1) extra Max-Pool in patch embedding, and (2) Depth-Wise Convolution in place of self-attention. Notably, Max-Former attains 82.39% top-1 accuracy on ImageNet using only 63.99M parameters, surpassing Spikformer (74.81%, 66.34M) by +7.58%. Extending our insight beyond transformers, our Max-ResNet-18 achieves state-of-the-art performance on convolution-based benchmarks: 97.17% on CIFAR-10 and 83.06% on CIFAR-100. We hope this simple yet effective solution inspires future research to explore the distinctive nature of spiking neural networks. Code is available: https://github.com/bic-L/MaxFormer.

Baizhi Wang, Rui Yan, Wenxin Ma et al.

Histomorphology is crucial in breast cancer diagnosis. However, existing whole slide image (WSI) classification methods struggle to effectively incorporate histomorphology information, limiting their ability to capture key and fine-grained pathological features. To address this limitation, we propose a novel framework that explicitly incorporates histomorphology (tumor cellularity, cellular morphology, and tissue architecture) into WSI classification. Specifically, our approach consists of three key components: (1) estimating the importance of tumor-related histomorphology information at the patch level based on medical prior knowledge; (2) generating representative cluster-level features through histomorphology-driven cluster pooling; and (3) enabling WSI-level classification through histomorphology-driven multi-instance aggregation. With the incorporation of histomorphological information, our framework strengthens the model's ability to capture key and fine-grained pathological patterns, thereby enhancing WSI classification performance. Experimental results demonstrate its effectiveness, achieving high diagnostic accuracy for molecular subtyping and cancer subtyping. The code will be made available at https://github.com/Badgewho/HMDMIL.

Jianwen Sun, Yukang Feng, Chuanhao Li et al.

Unified multimodal understanding and generation have recently received much attention in the area of vision and language. Existing UniMs are designed to simultaneously learn both multimodal understanding and generation capabilities, demanding substantial computational resources, and often struggle to generate interleaved text-image. We present ARMOR, a resource-efficient and pure autoregressive framework that achieves both understanding and generation by fine-tuning existing multimodal large language models (MLLMs). Specifically, ARMOR extends existing MLLMs from three perspectives: (1) For model architecture, an asymmetric encoder-decoder architecture with a forward-switching mechanism is introduced to unify embedding space integrating textual and visual modalities for enabling natural text-image interleaved generation with minimal computational overhead. (2) For training data, a meticulously curated, high-quality interleaved dataset is collected for fine-tuning MLLMs. (3) For the training algorithm, we propose a ``what or how to generate'' algorithm to empower existing MLLMs with multimodal generation capabilities while preserving their multimodal understanding capabilities, through three progressive training stages based on the collected dataset. Experimental results demonstrate that ARMOR upgrades existing MLLMs to UniMs with promising image generation capabilities, using limited training resources. Our code will be released soon at https://github.com/finyorko/armor.

Ziqi Gao, Wendi Yang, Yujia Li et al.

Non-semantic context information is crucial for visual recognition, as the human visual perception system first uses global statistics to process scenes rapidly before identifying specific objects. However, while semantic information is increasingly incorporated into computer vision tasks such as image reconstruction, non-semantic information, such as global spatial structures, is often overlooked. To bridge the gap, we propose a biologically informed non-semantic context descriptor, \textbf{MS-Glance}, along with the Glance Index Measure for comparing two images. A Global Glance vector is formulated by randomly retrieving pixels based on a perception-driven rule from an image to form a vector representing non-semantic global context, while a local Glance vector is a flattened local image window, mimicking a zoom-in observation. The Glance Index is defined as the inner product of two standardized sets of Glance vectors. We evaluate the effectiveness of incorporating Glance supervision in two reconstruction tasks: image fitting with implicit neural representation (INR) and undersampled MRI reconstruction. Extensive experimental results show that MS-Glance outperforms existing image restoration losses across both natural and medical images. The code is available at \url{https://github.com/Z7Gao/MSGlance}.

Ce Wang, Kun Shang, Haimiao Zhang et al.

While Computed Tomography (CT) reconstruction from X-ray sinograms is necessary for clinical diagnosis, iodine radiation in the imaging process induces irreversible injury, thereby driving researchers to study sparse-view CT reconstruction, that is, recovering a high-quality CT image from a sparse set of sinogram views. Iterative models are proposed to alleviate the appeared artifacts in sparse-view CT images, but the computation cost is too expensive. Then deep-learning-based methods have gained prevalence due to the excellent performances and lower computation. However, these methods ignore the mismatch between the CNN's \textbf{local} feature extraction capability and the sinogram's \textbf{global} characteristics. To overcome the problem, we propose \textbf{Du}al-\textbf{Do}main \textbf{Trans}former (\textbf{DuDoTrans}) to simultaneously restore informative sinograms via the long-range dependency modeling capability of Transformer and reconstruct CT image with both the enhanced and raw sinograms. With such a novel design, reconstruction performance on the NIH-AAPM dataset and COVID-19 dataset experimentally confirms the effectiveness and generalizability of DuDoTrans with fewer involved parameters. Extensive experiments also demonstrate its robustness with different noise-level scenarios for sparse-view CT reconstruction. The code and models are publicly available at https://github.com/DuDoTrans/CODE

Bo Zhou, Zachary Augenfeld, Julius Chapiro et al.

Multimodal image registration has many applications in diagnostic medical imaging and image-guided interventions, such as Transcatheter Arterial Chemoembolization (TACE) of liver cancer guided by intraprocedural CBCT and pre-operative MR. The ability to register peri-procedurally acquired diagnostic images into the intraprocedural environment can potentially improve the intra-procedural tumor targeting, which will significantly improve therapeutic outcomes. However, the intra-procedural CBCT often suffers from suboptimal image quality due to lack of signal calibration for Hounsfield unit, limited FOV, and motion/metal artifacts. These non-ideal conditions make standard intensity-based multimodal registration methods infeasible to generate correct transformation across modalities. While registration based on anatomic structures, such as segmentation or landmarks, provides an efficient alternative, such anatomic structure information is not always available. One can train a deep learning-based anatomy extractor, but it requires large-scale manual annotations on specific modalities, which are often extremely time-consuming to obtain and require expert radiological readers. To tackle these issues, we leverage annotated datasets already existing in a source modality and propose an anatomy-preserving domain adaptation to segmentation network (APA2Seg-Net) for learning segmentation without target modality ground truth. The segmenters are then integrated into our anatomy-guided multimodal registration based on the robust point matching machine. Our experimental results on in-house TACE patient data demonstrated that our APA2Seg-Net can generate robust CBCT and MR liver segmentation, and the anatomy-guided registration framework with these segmenters can provide high-quality multimodal registrations. Our code is available at https://github.com/bbbbbbzhou/APA2Seg-Net.

Heqin Zhu, Qingsong Yao, Li Xiao et al.

Detecting anatomical landmarks in medical images plays an essential role in understanding the anatomy and planning automated processing. In recent years, a variety of deep neural network methods have been developed to detect landmarks automatically. However, all of those methods are unary in the sense that a highly specialized network is trained for a single task say associated with a particular anatomical region. In this work, for the first time, we investigate the idea of "You Only Learn Once (YOLO)" and develop a universal anatomical landmark detection model to realize multiple landmark detection tasks with end-to-end training based on mixed datasets. The model consists of a local network and a global network: The local network is built upon the idea of universal U-Net to learn multi-domain local features and the global network is a parallelly-duplicated sequential of dilated convolutions that extract global features to further disambiguate the landmark locations. It is worth mentioning that the new model design requires much fewer parameters than models with standard convolutions to train. We evaluate our YOLO model on three X-ray datasets of 1,588 images on the head, hand, and chest, collectively contributing 62 landmarks. The experimental results show that our proposed universal model behaves largely better than any previous models trained on multiple datasets. It even beats the performance of the model that is trained separately for every single dataset. The code is available at https://github.com/MIRACLE-Center/YOLO_Universal_Anatomical_Landmark_Detection

Pengbo Liu, Hu Han, Yuanqi Du et al.

Purpose: Pelvic bone segmentation in CT has always been an essential step in clinical diagnosis and surgery planning of pelvic bone diseases. Existing methods for pelvic bone segmentation are either hand-crafted or semi-automatic and achieve limited accuracy when dealing with image appearance variations due to the multi-site domain shift, the presence of contrasted vessels, coprolith and chyme, bone fractures, low dose, metal artifacts, etc. Due to the lack of a large-scale pelvic CT dataset with annotations, deep learning methods are not fully explored. Methods: In this paper, we aim to bridge the data gap by curating a large pelvic CT dataset pooled from multiple sources and different manufacturers, including 1, 184 CT volumes and over 320, 000 slices with different resolutions and a variety of the above-mentioned appearance variations. Then we propose for the first time, to the best of our knowledge, to learn a deep multi-class network for segmenting lumbar spine, sacrum, left hip, and right hip, from multiple-domain images simultaneously to obtain more effective and robust feature representations. Finally, we introduce a post-processing tool based on the signed distance function (SDF) to eliminate false predictions while retaining correctly predicted bone fragments. Results: Extensive experiments on our dataset demonstrate the effectiveness of our automatic method, achieving an average Dice of 0.987 for a metal-free volume. SDF post-processor yields a decrease of 10.5% in hausdorff distance by maintaining important bone fragments in post-processing phase. Conclusion: We believe this large-scale dataset will promote the development of the whole community and plan to open source the images, annotations, codes, and trained baseline models at https://github.com/ICT-MIRACLE-lab/CTPelvic1K.

Qingsong Yao, Zecheng He, Hu Han et al.

Recent methods in multiple landmark detection based on deep convolutional neural networks (CNNs) reach high accuracy and improve traditional clinical workflow. However, the vulnerability of CNNs to adversarial-example attacks can be easily exploited to break classification and segmentation tasks. This paper is the first to study how fragile a CNN-based model on multiple landmark detection to adversarial perturbations. Specifically, we propose a novel Adaptive Targeted Iterative FGSM (ATI-FGSM) attack against the state-of-the-art models in multiple landmark detection. The attacker can use ATI-FGSM to precisely control the model predictions of arbitrarily selected landmarks, while keeping other stationary landmarks still, by adding imperceptible perturbations to the original image. A comprehensive evaluation on a public dataset for cephalometric landmark detection demonstrates that the adversarial examples generated by ATI-FGSM break the CNN-based network more effectively and efficiently, compared with the original Iterative FGSM attack. Our work reveals serious threats to patients' health. Furthermore, we discuss the limitations of our method and provide potential defense directions, by investigating the coupling effect of nearby landmarks, i.e., a major source of divergence in our experiments. Our source code is available at https://github.com/qsyao/attack_landmark_detection.

Chao Huang, Hu Han, Qingsong Yao et al.

Fully convolutional neural networks like U-Net have been the state-of-the-art methods in medical image segmentation. Practically, a network is highly specialized and trained separately for each segmentation task. Instead of a collection of multiple models, it is highly desirable to learn a universal data representation for different tasks, ideally a single model with the addition of a minimal number of parameters steered to each task. Inspired by the recent success of multi-domain learning in image classification, for the first time we explore a promising universal architecture that handles multiple medical segmentation tasks and is extendable for new tasks, regardless of different organs and imaging modalities. Our 3D Universal U-Net (3D U$^2$-Net) is built upon separable convolution, assuming that {\it images from different domains have domain-specific spatial correlations which can be probed with channel-wise convolution while also share cross-channel correlations which can be modeled with pointwise convolution}. We evaluate the 3D U$^2$-Net on five organ segmentation datasets. Experimental results show that this universal network is capable of competing with traditional models in terms of segmentation accuracy, while requiring only about $1\%$ of the parameters. Additionally, we observe that the architecture can be easily and effectively adapted to a new domain without sacrificing performance in the domains used to learn the shared parameterization of the universal network. We put the code of 3D U$^2$-Net into public domain. \url{https://github.com/huangmozhilv/u2net_torch/}

Zhidong Yang, Xiuhui Shi, Wei Ba et al.

Whole slide image (WSI) analysis has emerged as an increasingly essential technique in computational pathology. Recent advances in the pathology foundation models (FMs) have demonstrated significant advantages in deriving meaningful patch-level or slide-level multi-scale features from WSIs. However, current pathology FMs have exhibited substantial heterogeneity caused by diverse private training datasets and different network architectures. This heterogeneity introduces performance variability when we utilize the features from different FMs in the downstream tasks. To fully explore the advantages of multiple FMs effectively, in this work, we propose a novel framework for the fusion of multi-scale heterogeneous pathology FMs, called FuseCPath, yielding a model with a superior ensemble performance. The main contributions of our framework can be summarized as follows: (i) To guarantee the representativeness of the training patches, we propose a multi-view clustering-based method to filter out the discriminative patches via multiple FMs' embeddings. (ii) To effectively fuse the patch-level FMs, we devise a cluster-level re-embedding strategy to online capture patch-level local features. (iii) To effectively fuse the slide-level FMs, we devise a collaborative distillation strategy to explore the connections between slide-level FMs. Extensive experiments demonstrate that the proposed FuseCPath achieves state-of-the-art performance across multiple tasks on diverse datasets.

Yingtai Li, Xueming Fu, Han Li et al.

Sparse-view computed tomography (CT) reduces radiation exposure by acquiring fewer projections, making it a valuable tool in clinical scenarios where low-dose radiation is essential. However, this often results in increased noise and artifacts due to limited data. In this paper we propose a novel 3D Gaussian representation (3DGR) based method for sparse-view CT reconstruction. Inspired by recent success in novel view synthesis driven by 3D Gaussian splatting, we leverage the efficiency and expressiveness of 3D Gaussian representation as an alternative to implicit neural representation. To unleash the potential of 3DGR for CT imaging scenario, we propose two key innovations: (i) FBP-image-guided Guassian initialization and (ii) efficient integration with a differentiable CT projector. Extensive experiments and ablations on diverse datasets demonstrate the proposed 3DGR-CT consistently outperforms state-of-the-art counterpart methods, achieving higher reconstruction accuracy with faster convergence. Furthermore, we showcase the potential of 3DGR-CT for real-time physical simulation, which holds important clinical applications while challenging for implicit neural representations.

Haoran Lai, Zihang Jiang, Qingsong Yao et al.

The development of 3D medical vision-language models holds significant potential for disease diagnosis and patient treatment. However, compared to 2D medical images, 3D medical images, such as CT scans, face challenges related to limited training data and high dimension, which severely restrict the progress of 3D medical vision-language models. To address these issues, we collect a large amount of unlabeled 3D CT data and utilize self-supervised learning to construct a 3D visual foundation model for extracting 3D visual features. Then, we apply 3D spatial convolutions to aggregate and project high-level image features, reducing computational complexity while preserving spatial information. We also construct two instruction-tuning datasets based on BIMCV-R and CT-RATE to fine-tune the 3D vision-language model. Our model demonstrates superior performance compared to existing methods in report generation, visual question answering, and disease diagnosis. Code and data will be made publicly available soon.

Chenhe Du, Xiyue Lin, Qing Wu et al.

Limited-angle and sparse-view computed tomography (LACT and SVCT) are crucial for expanding the scope of X-ray CT applications. However, they face challenges due to incomplete data acquisition, resulting in diverse artifacts in the reconstructed CT images. Emerging implicit neural representation (INR) techniques, such as NeRF, NeAT, and NeRP, have shown promise in under-determined CT imaging reconstruction tasks. However, the unsupervised nature of INR architecture imposes limited constraints on the solution space, particularly for the highly ill-posed reconstruction task posed by LACT and ultra-SVCT. In this study, we introduce the Diffusion Prior Driven Neural Representation (DPER), an advanced unsupervised framework designed to address the exceptionally ill-posed CT reconstruction inverse problems. DPER adopts the Half Quadratic Splitting (HQS) algorithm to decompose the inverse problem into data fidelity and distribution prior sub-problems. The two sub-problems are respectively addressed by INR reconstruction scheme and pre-trained score-based diffusion model. This combination first injects the implicit image local consistency prior from INR. Additionally, it effectively augments the feasibility of the solution space for the inverse problem through the generative diffusion model, resulting in increased stability and precision in the solutions. We conduct comprehensive experiments to evaluate the performance of DPER on LACT and ultra-SVCT reconstruction with two public datasets (AAPM and LIDC), an in-house clinical COVID-19 dataset and a public raw projection dataset created by Mayo Clinic. The results show that our method outperforms the state-of-the-art reconstruction methods on in-domain datasets, while achieving significant performance improvements on out-of-domain (OOD) datasets.

Yuan Feng, Junlin Lv, Yukun Cao et al.

Large language models have revolutionized natural language processing but face significant challenges of high storage and runtime costs, due to the transformer architecture's reliance on self-attention, particularly the large Key-Value (KV) cache for long-sequence inference. Recent efforts to reduce KV cache size by pruning less critical entries based on attention weights remain empirical and lack formal grounding. This paper presents a formal study on identifying critical KV cache entries by analyzing attention output perturbation. Our analysis reveals that, beyond attention weights, the value states within KV entries and pretrained parameter matrices are also crucial. Based on this, we propose a perturbation-constrained selection algorithm that optimizes the worst-case output perturbation to identify critical entries. Evaluations on the Needle-in-a-Haystack test and Longbench benchmark show our algorithm enhances state-of-the-art cache eviction methods. Further empirical analysis confirms that our algorithm achieves lower output perturbations in over 92% attention heads in Llama model, thereby providing a significant improvement over existing methods.

Qingsong Yao, Zecheng He, Yuexiang Li et al. · tencent-ai

Deep learning based methods for medical images can be easily compromised by adversarial examples (AEs), posing a great security flaw in clinical decision-making. It has been discovered that conventional adversarial attacks like PGD which optimize the classification logits, are easy to distinguish in the feature space, resulting in accurate reactive defenses. To better understand this phenomenon and reassess the reliability of the reactive defenses for medical AEs, we thoroughly investigate the characteristic of conventional medical AEs. Specifically, we first theoretically prove that conventional adversarial attacks change the outputs by continuously optimizing vulnerable features in a fixed direction, thereby leading to outlier representations in the feature space. Then, a stress test is conducted to reveal the vulnerability of medical images, by comparing with natural images. Interestingly, this vulnerability is a double-edged sword, which can be exploited to hide AEs. We then propose a simple-yet-effective hierarchical feature constraint (HFC), a novel add-on to conventional white-box attacks, which assists to hide the adversarial feature in the target feature distribution. The proposed method is evaluated on three medical datasets, both 2D and 3D, with different modalities. The experimental results demonstrate the superiority of HFC, \emph{i.e.,} it bypasses an array of state-of-the-art adversarial medical AE detectors more efficiently than competing adaptive attacks, which reveals the deficiencies of medical reactive defense and allows to develop more robust defenses in future.

Zengyi Gao, Yukun Cao, Hairu Wang et al.

To mitigate the hallucination and knowledge deficiency in large language models (LLMs), Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) has shown promising potential by utilizing KGs as external resource to enhance LLMs reasoning. However, existing KG-RAG approaches struggle with a trade-off between flexibility and retrieval quality. Modular methods prioritize flexibility by avoiding the use of KG-fine-tuned models during retrieval, leading to fixed retrieval strategies and suboptimal retrieval quality. Conversely, coupled methods embed KG information within models to improve retrieval quality, but at the expense of flexibility. In this paper, we propose a novel flexible modular KG-RAG framework, termed FRAG, which synergizes the advantages of both approaches. FRAG estimates the hop range of reasoning paths based solely on the query and classify it as either simple or complex. To match the complexity of the query, tailored pipelines are applied to ensure efficient and accurate reasoning path retrieval, thus fostering the final reasoning process. By using the query text instead of the KG to infer the structural information of reasoning paths and employing adaptable retrieval strategies, FRAG improves retrieval quality while maintaining flexibility. Moreover, FRAG does not require extra LLMs fine-tuning or calls, significantly boosting efficiency and conserving resources. Extensive experiments show that FRAG achieves state-of-the-art performance with high efficiency and low resource consumption.

Haoran Lai, Zihang Jiang, Qingsong Yao et al.

3D medical images such as computed tomography are widely used in clinical practice, offering a great potential for automatic diagnosis. Supervised learning-based approaches have achieved significant progress but rely heavily on extensive manual annotations, limited by the availability of training data and the diversity of abnormality types. Vision-language alignment (VLA) offers a promising alternative by enabling zero-shot learning without additional annotations. However, we empirically discover that the visual and textural embeddings after alignment endeavors from existing VLA methods form two well-separated clusters, presenting a wide gap to be bridged. To bridge this gap, we propose a Bridged Semantic Alignment (BrgSA) framework. First, we utilize a large language model to perform semantic summarization of reports, extracting high-level semantic information. Second, we design a Cross-Modal Knowledge Interaction module that leverages a cross-modal knowledge bank as a semantic bridge, facilitating interaction between the two modalities, narrowing the gap, and improving their alignment. To comprehensively evaluate our method, we construct a benchmark dataset that includes 15 underrepresented abnormalities as well as utilize two existing benchmark datasets. Experimental results demonstrate that BrgSA achieves state-of-the-art performances on both public benchmark datasets and our custom-labeled dataset, with significant improvements in zero-shot diagnosis of underrepresented abnormalities.

Junzhe Li, Sifan Zhou, Liya Guo et al.

Unified multimodal models (UMMs) have emerged as a powerful paradigm in fundamental cross-modality research, demonstrating significant potential in both image understanding and generation. However, existing research in the face domain primarily faces two challenges: $\textbf{(1)}$ $\textbf{fragmentation development}$, with existing methods failing to unify understanding and generation into a single one, hindering the way to artificial general intelligence. $\textbf{(2) lack of fine-grained facial attributes}$, which are crucial for high-fidelity applications. To handle those issues, we propose $\textbf{UniF$^2$ace}$, $\textit{the first UMM specifically tailored for fine-grained face understanding and generation}$. $\textbf{First}$, we introduce a novel theoretical framework with a Dual Discrete Diffusion (D3Diff) loss, unifying masked generative models with discrete score matching diffusion and leading to a more precise approximation of the negative log-likelihood. Moreover, this D3Diff significantly enhances the model's ability to synthesize high-fidelity facial details aligned with text input. $\textbf{Second}$, we propose a multi-level grouped Mixture-of-Experts architecture, adaptively incorporating the semantic and identity facial embeddings to complement the attribute forgotten phenomenon in representation evolvement. $\textbf{Finally}$, to this end, we construct UniF$^2$aceD-1M, a large-scale dataset comprising 130K fine-grained image-caption pairs and 1M visual question-answering pairs, spanning a much wider range of facial attributes than existing datasets. Extensive experiments demonstrate that UniF$^2$ace outperforms existing models with a similar scale in both understanding and generation tasks, with 7.1\% higher Desc-GPT and 6.6\% higher VQA-score, respectively.

Yuncheng Jiang, Chun-Mei Feng, Jinke Ren et al.

Ultrasound imaging is widely used in clinical diagnosis due to its non-invasive nature and real-time capabilities. However, conventional ultrasound diagnostics face several limitations, including high dependence on physician expertise and suboptimal image quality, which complicates interpretation and increases the likelihood of diagnostic errors. Artificial intelligence (AI) has emerged as a promising solution to enhance clinical diagnosis, particularly in detecting abnormalities across various biomedical imaging modalities. Nonetheless, current AI models for ultrasound imaging face critical challenges. First, these models often require large volumes of labeled medical data, raising concerns over patient privacy breaches. Second, most existing models are task-specific, which restricts their broader clinical utility. To overcome these challenges, we present UltraFedFM, an innovative privacy-preserving ultrasound foundation model. UltraFedFM is collaboratively pre-trained using federated learning across 16 distributed medical institutions in 9 countries, leveraging a dataset of over 1 million ultrasound images covering 19 organs and 10 ultrasound modalities. This extensive and diverse data, combined with a secure training framework, enables UltraFedFM to exhibit strong generalization and diagnostic capabilities. It achieves an average area under the receiver operating characteristic curve of 0.927 for disease diagnosis and a dice similarity coefficient of 0.878 for lesion segmentation. Notably, UltraFedFM surpasses the diagnostic accuracy of mid-level ultrasonographers and matches the performance of expert-level sonographers in the joint diagnosis of 8 common systemic diseases. These findings indicate that UltraFedFM can significantly enhance clinical diagnostics while safeguarding patient privacy, marking an advancement in AI-driven ultrasound imaging for future clinical applications.

Jiangyong Yu, Changyong Shu, Sifan Zhou et al.

Camera-based multi-view 3D detection is crucial for autonomous driving. PETR and its variants (PETRs) excel in benchmarks but face deployment challenges due to high computational cost and memory footprint. Quantization is an effective technique for compressing deep neural networks by reducing the bit width of weights and activations. However, directly applying existing quantization methods to PETRs leads to severe accuracy degradation. This issue primarily arises from two key challenges: (1) significant magnitude disparity between multi-modal features-specifically, image features and camera-ray positional embeddings (PE), and (2) the inefficiency and approximation error of quantizing non-linear operators, which commonly rely on hardware-unfriendly computations. In this paper, we propose FQ-PETR, a fully quantized framework for PETRs, featuring three key innovations: (1) Quantization-Friendly LiDAR-ray Position Embedding (QFPE): Replacing multi-point sampling with LiDAR-prior-guided single-point sampling and anchor-based embedding eliminates problematic non-linearities (e.g., inverse-sigmoid) and aligns PE scale with image features, preserving accuracy. (2) Dual-Lookup Table (DULUT): This algorithm approximates complex non-linear functions using two cascaded linear LUTs, achieving high fidelity with minimal entries and no specialized hardware. (3) Quantization After Numerical Stabilization (QANS): Performing quantization after softmax numerical stabilization mitigates attention distortion from large inputs. On PETRs (e.g. PETR, StreamPETR, PETRv2, MV2d), FQ-PETR under W8A8 achieves near-floating-point accuracy (1% degradation) while reducing latency by up to 75%, significantly outperforming existing PTQ and QAT baselines.

Qing Wu, Xu Guo, Lixuan Chen et al.

X-ray CT often suffers from shadowing and streaking artifacts in the presence of metallic materials, which severely degrade imaging quality. Physically, the linear attenuation coefficients (LACs) of metals vary significantly with X-ray energy, causing a nonlinear beam hardening effect (BHE) in CT measurements. Reconstructing CT images from metal-corrupted measurements consequently becomes a challenging nonlinear inverse problem. Existing state-of-the-art (SOTA) metal artifact reduction (MAR) algorithms rely on supervised learning with numerous paired CT samples. While promising, these supervised methods often assume that the unknown LACs are energy-independent, ignoring the energy-induced BHE, which results in limited generalization. Moreover, the requirement for large datasets also limits their applications in real-world scenarios. In this work, we propose Density neural representation (Diner), a novel unsupervised MAR method. Our key innovation lies in formulating MAR as an energy-independent density reconstruction problem that strictly adheres to the photon-tissue absorption physical model. This model is inherently nonlinear and complex, making it a rarely considered approach in inverse imaging problems. By introducing the water-equivalent tissues approximation and a new polychromatic model to characterize the nonlinear CT acquisition process, we directly learn the neural representation of the density map from raw measurements without using external training data. This energy-independent density reconstruction framework fundamentally resolves the nonlinear BHE, enabling superior MAR performance across a wide range of scanning scenarios. Extensive experiments on both simulated and real-world datasets demonstrate the superiority of our unsupervised Diner over popular supervised methods in terms of MAR performance and robustness.

Afifa Khaled, Mohammed Sabir, Rizwan Qureshi et al.

The Medical Information Mart for Intensive Care (MIMIC) datasets have become the Kernel of Digital Health Research by providing freely accessible, deidentified records from tens of thousands of critical care admissions, enabling a broad spectrum of applications in clinical decision support, outcome prediction, and healthcare analytics. Although numerous studies and surveys have explored the predictive power and clinical utility of MIMIC based models, critical challenges in data integration, representation, and interoperability remain underexplored. This paper presents a comprehensive survey that focuses uniquely on open problems. We identify persistent issues such as data granularity, cardinality limitations, heterogeneous coding schemes, and ethical constraints that hinder the generalizability and real-time implementation of machine learning models. We highlight key progress in dimensionality reduction, temporal modelling, causal inference, and privacy preserving analytics, while also outlining promising directions including hybrid modelling, federated learning, and standardized preprocessing pipelines. By critically examining these structural limitations and their implications, this survey offers actionable insights to guide the next generation of MIMIC powered digital health innovations.

Ziqi Gao, Yue Zhang, Xinwen Liu et al.

Multi-contrast (MC) Magnetic Resonance Imaging (MRI) reconstruction aims to incorporate a reference image of auxiliary modality to guide the reconstruction process of the target modality. Known MC reconstruction methods perform well with a fully sampled reference image, but usually exhibit inferior performance, compared to single-contrast (SC) methods, when the reference image is missing or of low quality. To address this issue, we propose DuDoUniNeXt, a unified dual-domain MRI reconstruction network that can accommodate to scenarios involving absent, low-quality, and high-quality reference images. DuDoUniNeXt adopts a hybrid backbone that combines CNN and ViT, enabling specific adjustment of image domain and k-space reconstruction. Specifically, an adaptive coarse-to-fine feature fusion module (AdaC2F) is devised to dynamically process the information from reference images of varying qualities. Besides, a partially shared shallow feature extractor (PaSS) is proposed, which uses shared and distinct parameters to handle consistent and discrepancy information among contrasts. Experimental results demonstrate that the proposed model surpasses state-of-the-art SC and MC models significantly. Ablation studies show the effectiveness of the proposed hybrid backbone, AdaC2F, PaSS, and the dual-domain unified learning scheme.