Hongming Shan

Qi Gao, Zilong Li, Junping Zhang et al.

Low-dose computed tomography (CT) images suffer from noise and artifacts due to photon starvation and electronic noise. Recently, some works have attempted to use diffusion models to address the over-smoothness and training instability encountered by previous deep-learning-based denoising models. However, diffusion models suffer from long inference times due to the large number of sampling steps involved. Very recently, cold diffusion model generalizes classical diffusion models and has greater flexibility. Inspired by the cold diffusion, this paper presents a novel COntextual eRror-modulated gEneralized Diffusion model for low-dose CT (LDCT) denoising, termed CoreDiff. First, CoreDiff utilizes LDCT images to displace the random Gaussian noise and employs a novel mean-preserving degradation operator to mimic the physical process of CT degradation, significantly reducing sampling steps thanks to the informative LDCT images as the starting point of the sampling process. Second, to alleviate the error accumulation problem caused by the imperfect restoration operator in the sampling process, we propose a novel ContextuaL Error-modulAted Restoration Network (CLEAR-Net), which can leverage contextual information to constrain the sampling process from structural distortion and modulate time step embedding features for better alignment with the input at the next time step. Third, to rapidly generalize to a new, unseen dose level with as few resources as possible, we devise a one-shot learning framework to make CoreDiff generalize faster and better using only a single LDCT image (un)paired with NDCT. Extensive experimental results on two datasets demonstrate that our CoreDiff outperforms competing methods in denoising and generalization performance, with a clinically acceptable inference time. Source code is made available at https://github.com/qgao21/CoreDiff.

Zhizhong Huang, Junping Zhang, Hongming Shan

Learning from noisy data is a challenging task that significantly degenerates the model performance. In this paper, we present TCL, a novel twin contrastive learning model to learn robust representations and handle noisy labels for classification. Specifically, we construct a Gaussian mixture model (GMM) over the representations by injecting the supervised model predictions into GMM to link label-free latent variables in GMM with label-noisy annotations. Then, TCL detects the examples with wrong labels as the out-of-distribution examples by another two-component GMM, taking into account the data distribution. We further propose a cross-supervision with an entropy regularization loss that bootstraps the true targets from model predictions to handle the noisy labels. As a result, TCL can learn discriminative representations aligned with estimated labels through mixup and contrastive learning. Extensive experimental results on several standard benchmarks and real-world datasets demonstrate the superior performance of TCL. In particular, TCL achieves 7.5\% improvements on CIFAR-10 with 90\% noisy label -- an extremely noisy scenario. The source code is available at \url{https://github.com/Hzzone/TCL}.

Jiaxin Ye, Xin-cheng Wen, Yujie Wei et al.

Speech emotion recognition (SER) plays a vital role in improving the interactions between humans and machines by inferring human emotion and affective states from speech signals. Whereas recent works primarily focus on mining spatiotemporal information from hand-crafted features, we explore how to model the temporal patterns of speech emotions from dynamic temporal scales. Towards that goal, we introduce a novel temporal emotional modeling approach for SER, termed Temporal-aware bI-direction Multi-scale Network (TIM-Net), which learns multi-scale contextual affective representations from various time scales. Specifically, TIM-Net first employs temporal-aware blocks to learn temporal affective representation, then integrates complementary information from the past and the future to enrich contextual representations, and finally, fuses multiple time scale features for better adaptation to the emotional variation. Extensive experimental results on six benchmark SER datasets demonstrate the superior performance of TIM-Net, gaining 2.34% and 2.61% improvements of the average UAR and WAR over the second-best on each corpus. The source code is available at https://github.com/Jiaxin-Ye/TIM-Net_SER.

Tao Chen, Chenhui Wang, Hongming Shan · deepmind

Medical image segmentation is a challenging task with inherent ambiguity and high uncertainty, attributed to factors such as unclear tumor boundaries and multiple plausible annotations. The accuracy and diversity of segmentation masks are both crucial for providing valuable references to radiologists in clinical practice. While existing diffusion models have shown strong capacities in various visual generation tasks, it is still challenging to deal with discrete masks in segmentation. To achieve accurate and diverse medical image segmentation masks, we propose a novel conditional Bernoulli Diffusion model for medical image segmentation (BerDiff). Instead of using the Gaussian noise, we first propose to use the Bernoulli noise as the diffusion kernel to enhance the capacity of the diffusion model for binary segmentation tasks, resulting in more accurate segmentation masks. Second, by leveraging the stochastic nature of the diffusion model, our BerDiff randomly samples the initial Bernoulli noise and intermediate latent variables multiple times to produce a range of diverse segmentation masks, which can highlight salient regions of interest that can serve as valuable references for radiologists. In addition, our BerDiff can efficiently sample sub-sequences from the overall trajectory of the reverse diffusion, thereby speeding up the segmentation process. Extensive experimental results on two medical image segmentation datasets with different modalities demonstrate that our BerDiff outperforms other recently published state-of-the-art methods. Our results suggest diffusion models could serve as a strong backbone for medical image segmentation.

Yujie Wei, Jiaxin Ye, Zhizhong Huang et al.

Online continual learning (CL) studies the problem of learning continuously from a single-pass data stream while adapting to new data and mitigating catastrophic forgetting. Recently, by storing a small subset of old data, replay-based methods have shown promising performance. Unlike previous methods that focus on sample storage or knowledge distillation against catastrophic forgetting, this paper aims to understand why the online learning models fail to generalize well from a new perspective of shortcut learning. We identify shortcut learning as the key limiting factor for online CL, where the learned features may be biased, not generalizable to new tasks, and may have an adverse impact on knowledge distillation. To tackle this issue, we present the online prototype learning (OnPro) framework for online CL. First, we propose online prototype equilibrium to learn representative features against shortcut learning and discriminative features to avoid class confusion, ultimately achieving an equilibrium status that separates all seen classes well while learning new classes. Second, with the feedback of online prototypes, we devise a novel adaptive prototypical feedback mechanism to sense the classes that are easily misclassified and then enhance their boundaries. Extensive experimental results on widely-used benchmark datasets demonstrate the superior performance of OnPro over the state-of-the-art baseline methods. Source code is available at https://github.com/weilllllls/OnPro.

Zhizhong Huang, Mingliang Dai, Yi Zhang et al.

Class-agnostic object counting aims to count all objects in an image with respect to example boxes or class names, \emph{a.k.a} few-shot and zero-shot counting. In this paper, we propose a generalized framework for both few-shot and zero-shot object counting based on detection. Our framework combines the superior advantages of two foundation models without compromising their zero-shot capability: (\textbf{i}) SAM to segment all possible objects as mask proposals, and (\textbf{ii}) CLIP to classify proposals to obtain accurate object counts. However, this strategy meets the obstacles of efficiency overhead and the small crowded objects that cannot be localized and distinguished. To address these issues, our framework, termed PseCo, follows three steps: point, segment, and count. Specifically, we first propose a class-agnostic object localization to provide accurate but least point prompts for SAM, which consequently not only reduces computation costs but also avoids missing small objects. Furthermore, we propose a generalized object classification that leverages CLIP image/text embeddings as the classifier, following a hierarchical knowledge distillation to obtain discriminative classifications among hierarchical mask proposals. Extensive experimental results on FSC-147, COCO, and LVIS demonstrate that PseCo achieves state-of-the-art performance in both few-shot/zero-shot object counting/detection. Code: https://github.com/Hzzone/PseCo

Zhizhong Huang, Junping Zhang, Hongming Shan

To minimize the impact of age variation on face recognition, age-invariant face recognition (AIFR) extracts identity-related discriminative features by minimizing the correlation between identity- and age-related features while face age synthesis (FAS) eliminates age variation by converting the faces in different age groups to the same group. However, AIFR lacks visual results for model interpretation and FAS compromises downstream recognition due to artifacts. Therefore, we propose a unified, multi-task framework to jointly handle these two tasks, termed MTLFace, which can learn the age-invariant identity-related representation for face recognition while achieving pleasing face synthesis for model interpretation. Specifically, we propose an attention-based feature decomposition to decompose the mixed face features into two uncorrelated components -- identity- and age-related features -- in a spatially constrained way. Unlike the conventional one-hot encoding that achieves group-level FAS, we propose a novel identity conditional module to achieve identity-level FAS, which can improve the age smoothness of synthesized faces through a weight-sharing strategy. Benefiting from the proposed multi-task framework, we then leverage those high-quality synthesized faces from FAS to further boost AIFR via a novel selective fine-tuning strategy. Furthermore, to advance both AIFR and FAS, we collect and release a large cross-age face dataset with age and gender annotations, and a new benchmark specifically designed for tracing long-missing children. Extensive experimental results on five benchmark cross-age datasets demonstrate that MTLFace yields superior performance for both AIFR and FAS. We further validate MTLFace on two popular general face recognition datasets, obtaining competitive performance on face recognition in the wild. Code is available at http://hzzone.github.io/MTLFace.

Zhihao Chen, Chuang Niu, Qi Gao et al.

This paper studies 3D low-dose computed tomography (CT) imaging. Although various deep learning methods were developed in this context, typically they focus on 2D images and perform denoising due to low-dose and deblurring for super-resolution separately. Up to date, little work was done for simultaneous in-plane denoising and through-plane deblurring, which is important to obtain high-quality 3D CT images with lower radiation and faster imaging speed. For this task, a straightforward method is to directly train an end-to-end 3D network. However, it demands much more training data and expensive computational costs. Here, we propose to link in-plane and through-plane transformers for simultaneous in-plane denoising and through-plane deblurring, termed as LIT-Former, which can efficiently synergize in-plane and through-plane sub-tasks for 3D CT imaging and enjoy the advantages of both convolution and transformer networks. LIT-Former has two novel designs: efficient multi-head self-attention modules (eMSM) and efficient convolutional feedforward networks (eCFN). First, eMSM integrates in-plane 2D self-attention and through-plane 1D self-attention to efficiently capture global interactions of 3D self-attention, the core unit of transformer networks. Second, eCFN integrates 2D convolution and 1D convolution to extract local information of 3D convolution in the same fashion. As a result, the proposed LIT-Former synergize these two subtasks, significantly reducing the computational complexity as compared to 3D counterparts and enabling rapid convergence. Extensive experimental results on simulated and clinical datasets demonstrate superior performance over state-of-the-art models. The source code is made available at https://github.com/hao1635/LIT-Former.

Mingliang Dai, Zhizhong Huang, Jiaqi Gao et al.

Noisy annotations such as missing annotations and location shifts often exist in crowd counting datasets due to multi-scale head sizes, high occlusion, etc. These noisy annotations severely affect the model training, especially for density map-based methods. To alleviate the negative impact of noisy annotations, we propose a novel crowd counting model with one convolution head and one transformer head, in which these two heads can supervise each other in noisy areas, called Cross-Head Supervision. The resultant model, CHS-Net, can synergize different types of inductive biases for better counting. In addition, we develop a progressive cross-head supervision learning strategy to stabilize the training process and provide more reliable supervision. Extensive experimental results on ShanghaiTech and QNRF datasets demonstrate superior performance over state-of-the-art methods. Code is available at https://github.com/RaccoonDML/CHSNet.

Zhihao Chen, Qi Gao, Yi Zhang et al.

While various deep learning methods have been proposed for low-dose computed tomography (CT) denoising, most of them leverage the normal-dose CT images as the ground-truth to supervise the denoising process. These methods typically ignore the inherent correlation within a single CT image, especially the anatomical semantics of human tissues, and lack the interpretability on the denoising process. In this paper, we propose a novel Anatomy-aware Supervised CONtrastive learning framework, termed ASCON, which can explore the anatomical semantics for low-dose CT denoising while providing anatomical interpretability. The proposed ASCON consists of two novel designs: an efficient self-attention-based U-Net (ESAU-Net) and a multi-scale anatomical contrastive network (MAC-Net). First, to better capture global-local interactions and adapt to the high-resolution input, an efficient ESAU-Net is introduced by using a channel-wise self-attention mechanism. Second, MAC-Net incorporates a patch-wise non-contrastive module to capture inherent anatomical information and a pixel-wise contrastive module to maintain intrinsic anatomical consistency. Extensive experimental results on two public low-dose CT denoising datasets demonstrate superior performance of ASCON over state-of-the-art models. Remarkably, our ASCON provides anatomical interpretability for low-dose CT denoising for the first time. Source code is available at https://github.com/hao1635/ASCON.

Chenglong Ma, Zilong Li, Junping Zhang et al.

Sparse-view computed tomography (CT) is a promising solution for expediting the scanning process and mitigating radiation exposure to patients, the reconstructed images, however, contain severe streak artifacts, compromising subsequent screening and diagnosis. Recently, deep learning-based image post-processing methods along with their dual-domain counterparts have shown promising results. However, existing methods usually produce over-smoothed images with loss of details due to (1) the difficulty in accurately modeling the artifact patterns in the image domain, and (2) the equal treatment of each pixel in the loss function. To address these issues, we concentrate on the image post-processing and propose a simple yet effective FREquency-band-awarE and SElf-guidED network, termed FreeSeed, which can effectively remove artifact and recover missing detail from the contaminated sparse-view CT images. Specifically, we first propose a frequency-band-aware artifact modeling network (FreeNet), which learns artifact-related frequency-band attention in Fourier domain for better modeling the globally distributed streak artifact on the sparse-view CT images. We then introduce a self-guided artifact refinement network (SeedNet), which leverages the predicted artifact to assist FreeNet in continuing to refine the severely corrupted details. Extensive experiments demonstrate the superior performance of FreeSeed and its dual-domain counterpart over the state-of-the-art sparse-view CT reconstruction methods. Source code is made available at https://github.com/Masaaki-75/freeseed.

Jiaxin Ye, Junping Zhang, Hongming Shan

Depression is a common mental disorder that affects millions of people worldwide. Although promising, current multimodal methods hinge on aligned or aggregated multimodal fusion, suffering two significant limitations: (i) inefficient long-range temporal modeling, and (ii) sub-optimal multimodal fusion between intermodal fusion and intramodal processing. In this paper, we propose an audio-visual progressive fusion Mamba for multimodal depression detection, termed DepMamba. DepMamba features two core designs: hierarchical contextual modeling and progressive multimodal fusion. On the one hand, hierarchical modeling introduces convolution neural networks and Mamba to extract the local-to-global features within long-range sequences. On the other hand, the progressive fusion first presents a multimodal collaborative State Space Model (SSM) extracting intermodal and intramodal information for each modality, and then utilizes a multimodal enhanced SSM for modality cohesion. Extensive experimental results on two large-scale depression datasets demonstrate the superior performance of our DepMamba over existing state-of-the-art methods. Code is available at https://github.com/Jiaxin-Ye/DepMamba.

Zilong Li, Qi Gao, Yaping Wu et al.

Metal implants and other high-density objects in patients introduce severe streaking artifacts in CT images, compromising image quality and diagnostic performance. Although various methods were developed for CT metal artifact reduction over the past decades, including the latest dual-domain deep networks, remaining metal artifacts are still clinically challenging in many cases. Here we extend the state-of-the-art dual-domain deep network approach into a quad-domain counterpart so that all the features in the sinogram, image, and their corresponding Fourier domains are synergized to eliminate metal artifacts optimally without compromising structural subtleties. Our proposed quad-domain network for MAR, referred to as Quad-Net, takes little additional computational cost since the Fourier transform is highly efficient, and works across the four receptive fields to learn both global and local features as well as their relations. Specifically, we first design a Sinogram-Fourier Restoration Network (SFR-Net) in the sinogram domain and its Fourier space to faithfully inpaint metal-corrupted traces. Then, we couple SFR-Net with an Image-Fourier Refinement Network (IFR-Net) which takes both an image and its Fourier spectrum to improve a CT image reconstructed from the SFR-Net output using cross-domain contextual information. Quad-Net is trained on clinical datasets to minimize a composite loss function. Quad-Net does not require precise metal masks, which is of great importance in clinical practice. Our experimental results demonstrate the superiority of Quad-Net over the state-of-the-art MAR methods quantitatively, visually, and statistically. The Quad-Net code is publicly available at https://github.com/longzilicart/Quad-Net.

Zilong Li, Chenglong Ma, Jie Chen et al.

Sparse-view computed tomography (CT) -- using a small number of projections for tomographic reconstruction -- enables much lower radiation dose to patients and accelerated data acquisition. The reconstructed images, however, suffer from strong artifacts, greatly limiting their diagnostic value. Current trends for sparse-view CT turn to the raw data for better information recovery. The resultant dual-domain methods, nonetheless, suffer from secondary artifacts, especially in ultra-sparse view scenarios, and their generalization to other scanners/protocols is greatly limited. A crucial question arises: have the image post-processing methods reached the limit? Our answer is not yet. In this paper, we stick to image post-processing methods due to great flexibility and propose global representation (GloRe) distillation framework for sparse-view CT, termed GloReDi. First, we propose to learn GloRe with Fourier convolution, so each element in GloRe has an image-wide receptive field. Second, unlike methods that only use the full-view images for supervision, we propose to distill GloRe from intermediate-view reconstructed images that are readily available but not explored in previous literature. The success of GloRe distillation is attributed to two key components: representation directional distillation to align the GloRe directions, and band-pass-specific contrastive distillation to gain clinically important details. Extensive experiments demonstrate the superiority of the proposed GloReDi over the state-of-the-art methods, including dual-domain ones. The source code is available at https://github.com/longzilicart/GloReDi.

Zhizhong Huang, Siteng Ma, Junping Zhang et al.

Recent works for face editing usually manipulate the latent space of StyleGAN via the linear semantic directions. However, they usually suffer from the entanglement of facial attributes, need to tune the optimal editing strength, and are limited to binary attributes with strong supervision signals. This paper proposes a novel adaptive nonlinear latent transformation for disentangled and conditional face editing, termed AdaTrans. Specifically, our AdaTrans divides the manipulation process into several finer steps; i.e., the direction and size at each step are conditioned on both the facial attributes and the latent codes. In this way, AdaTrans describes an adaptive nonlinear transformation trajectory to manipulate the faces into target attributes while keeping other attributes unchanged. Then, AdaTrans leverages a predefined density model to constrain the learned trajectory in the distribution of latent codes by maximizing the likelihood of transformed latent code. Moreover, we also propose a disentangled learning strategy under a mutual information framework to eliminate the entanglement among attributes, which can further relax the need for labeled data. Consequently, AdaTrans enables a controllable face editing with the advantages of disentanglement, flexibility with non-binary attributes, and high fidelity. Extensive experimental results on various facial attributes demonstrate the qualitative and quantitative effectiveness of the proposed AdaTrans over existing state-of-the-art methods, especially in the most challenging scenarios with a large age gap and few labeled examples. The source code is available at https://github.com/Hzzone/AdaTrans.

Yiming Lei, Zilong Li, Yangyang Li et al.

Interpreting the decisions of deep learning models has been actively studied since the explosion of deep neural networks. One of the most convincing interpretation approaches is salience-based visual interpretation, such as Grad-CAM, where the generation of attention maps depends merely on categorical labels. Although existing interpretation methods can provide explainable decision clues, they often yield partial correspondence between image and saliency maps due to the limited discriminative information from one-hot labels. This paper develops a Language-Image COnsistency model for explainable image classification, termed LICO, by correlating learnable linguistic prompts with corresponding visual features in a coarse-to-fine manner. Specifically, we first establish a coarse global manifold structure alignment by minimizing the distance between the distributions of image and language features. We then achieve fine-grained saliency maps by applying optimal transport (OT) theory to assign local feature maps with class-specific prompts. Extensive experimental results on eight benchmark datasets demonstrate that the proposed LICO achieves a significant improvement in generating more explainable attention maps in conjunction with existing interpretation methods such as Grad-CAM. Remarkably, LICO improves the classification performance of existing models without introducing any computational overhead during inference. Source code is made available at https://github.com/ymLeiFDU/LICO.

Jiaxin Ye, Yujie Wei, Xin-Cheng Wen et al.

Cross-corpus speech emotion recognition (SER) seeks to generalize the ability of inferring speech emotion from a well-labeled corpus to an unlabeled one, which is a rather challenging task due to the significant discrepancy between two corpora. Existing methods, typically based on unsupervised domain adaptation (UDA), struggle to learn corpus-invariant features by global distribution alignment, but unfortunately, the resulting features are mixed with corpus-specific features or not class-discriminative. To tackle these challenges, we propose a novel Emotion Decoupling aNd Alignment learning framework (EMO-DNA) for cross-corpus SER, a novel UDA method to learn emotion-relevant corpus-invariant features. The novelties of EMO-DNA are two-fold: contrastive emotion decoupling and dual-level emotion alignment. On one hand, our contrastive emotion decoupling achieves decoupling learning via a contrastive decoupling loss to strengthen the separability of emotion-relevant features from corpus-specific ones. On the other hand, our dual-level emotion alignment introduces an adaptive threshold pseudo-labeling to select confident target samples for class-level alignment, and performs corpus-level alignment to jointly guide model for learning class-discriminative corpus-invariant features across corpora. Extensive experimental results demonstrate the superior performance of EMO-DNA over the state-of-the-art methods in several cross-corpus scenarios. Source code is available at https://github.com/Jiaxin-Ye/Emo-DNA.

Tao Chen, Chenhui Wang, Zhihao Chen et al.

Medical image segmentation has been significantly advanced with the rapid development of deep learning (DL) techniques. Existing DL-based segmentation models are typically discriminative; i.e., they aim to learn a mapping from the input image to segmentation masks. However, these discriminative methods neglect the underlying data distribution and intrinsic class characteristics, suffering from unstable feature space. In this work, we propose to complement discriminative segmentation methods with the knowledge of underlying data distribution from generative models. To that end, we propose a novel hybrid diffusion framework for medical image segmentation, termed HiDiff, which can synergize the strengths of existing discriminative segmentation models and new generative diffusion models. HiDiff comprises two key components: discriminative segmentor and diffusion refiner. First, we utilize any conventional trained segmentation models as discriminative segmentor, which can provide a segmentation mask prior for diffusion refiner. Second, we propose a novel binary Bernoulli diffusion model (BBDM) as the diffusion refiner, which can effectively, efficiently, and interactively refine the segmentation mask by modeling the underlying data distribution. Third, we train the segmentor and BBDM in an alternate-collaborative manner to mutually boost each other. Extensive experimental results on abdomen organ, brain tumor, polyps, and retinal vessels segmentation datasets, covering four widely-used modalities, demonstrate the superior performance of HiDiff over existing medical segmentation algorithms, including the state-of-the-art transformer- and diffusion-based ones. In addition, HiDiff excels at segmenting small objects and generalizing to new datasets. Source codes are made available at https://github.com/takimailto/HiDiff.

Binglei Li, Zhizhong Huang, Hongming Shan et al.

Navigating in the latent space of StyleGAN has shown effectiveness for face editing. However, the resulting methods usually encounter challenges in complicated navigation due to the entanglement among different attributes in the latent space. To address this issue, this paper proposes a novel framework, termed SDFlow, with a semantic decomposition in original latent space using continuous conditional normalizing flows. Specifically, SDFlow decomposes the original latent code into different irrelevant variables by jointly optimizing two components: (i) a semantic encoder to estimate semantic variables from input faces and (ii) a flow-based transformation module to map the latent code into a semantic-irrelevant variable in Gaussian distribution, conditioned on the learned semantic variables. To eliminate the entanglement between variables, we employ a disentangled learning strategy under a mutual information framework, thereby providing precise manipulation controls. Experimental results demonstrate that SDFlow outperforms existing state-of-the-art face editing methods both qualitatively and quantitatively. The source code is made available at https://github.com/phil329/SDFlow.

Zhongying Deng, Cheng Tang, Ziyan Huang et al. · pku

Foundation models have demonstrated remarkable success across diverse domains and tasks, primarily due to the thrive of large-scale, diverse, and high-quality datasets. However, in the field of medical imaging, the curation and assembling of such medical datasets are highly challenging due to the reliance on clinical expertise and strict ethical and privacy constraints, resulting in a scarcity of large-scale unified medical datasets and hindering the development of powerful medical foundation models. In this work, we present the largest survey to date of medical image datasets, covering over 1,000 open-access datasets with a systematic catalog of their modalities, tasks, anatomies, annotations, limitations, and potential for integration. Our analysis exposes a landscape that is modest in scale, fragmented across narrowly scoped tasks, and unevenly distributed across organs and modalities, which in turn limits the utility of existing medical image datasets for developing versatile and robust medical foundation models. To turn fragmentation into scale, we propose a metadata-driven fusion paradigm (MDFP) that integrates public datasets with shared modalities or tasks, thereby transforming multiple small data silos into larger, more coherent resources. Building on MDFP, we release an interactive discovery portal that enables end-to-end, automated medical image dataset integration, and compile all surveyed datasets into a unified, structured table that clearly summarizes their key characteristics and provides reference links, offering the community an accessible and comprehensive repository. By charting the current terrain and offering a principled path to dataset consolidation, our survey provides a practical roadmap for scaling medical imaging corpora, supporting faster data discovery, more principled dataset creation, and more capable medical foundation models.

Wenjun Xia, Hongming Shan, Ge Wang et al.

Since 2016, deep learning (DL) has advanced tomographic imaging with remarkable successes, especially in low-dose computed tomography (LDCT) imaging. Despite being driven by big data, the LDCT denoising and pure end-to-end reconstruction networks often suffer from the black box nature and major issues such as instabilities, which is a major barrier to apply deep learning methods in low-dose CT applications. An emerging trend is to integrate imaging physics and model into deep networks, enabling a hybridization of physics/model-based and data-driven elements. %This type of hybrid methods has become increasingly influential. In this paper, we systematically review the physics/model-based data-driven methods for LDCT, summarize the loss functions and training strategies, evaluate the performance of different methods, and discuss relevant issues and future directions.

Weiyi Yu, Zhizhong Huang, Junping Zhang et al.

There are considerable interests in automatic stroke lesion segmentation on magnetic resonance (MR) images in the medical imaging field, as stroke is an important cerebrovascular disease. Although deep learning-based models have been proposed for this task, generalizing these models to unseen sites is difficult due to not only the large inter-site discrepancy among different scanners, imaging protocols, and populations, but also the variations in stroke lesion shape, size, and location. To tackle this issue, we introduce a self-adaptive normalization network, termed SAN-Net, to achieve adaptive generalization on unseen sites for stroke lesion segmentation. Motivated by traditional z-score normalization and dynamic network, we devise a masked adaptive instance normalization (MAIN) to minimize inter-site discrepancies, which standardizes input MR images from different sites into a site-unrelated style by dynamically learning affine parameters from the input; \ie, MAIN can affinely transform the intensity values. Then, we leverage a gradient reversal layer to force the U-net encoder to learn site-invariant representation with a site classifier, which further improves the model generalization in conjunction with MAIN. Finally, inspired by the ``pseudosymmetry'' of the human brain, we introduce a simple yet effective data augmentation technique, termed symmetry-inspired data augmentation (SIDA), that can be embedded within SAN-Net to double the sample size while halving memory consumption. Experimental results on the benchmark Anatomical Tracings of Lesions After Stroke (ATLAS) v1.2 dataset, which includes MR images from 9 different sites, demonstrate that under the ``leave-one-site-out'' setting, the proposed SAN-Net outperforms recently published methods in terms of quantitative metrics and qualitative comparisons.

Jiaqi Gao, Jingqi Li, Hongming Shan et al.

Crowd counting has important applications in public safety and pandemic control. A robust and practical crowd counting system has to be capable of continuously learning with the new incoming domain data in real-world scenarios instead of fitting one domain only. Off-the-shelf methods have some drawbacks when handling multiple domains: (1) the models will achieve limited performance (even drop dramatically) among old domains after training images from new domains due to the discrepancies of intrinsic data distributions from various domains, which is called catastrophic forgetting; (2) the well-trained model in a specific domain achieves imperfect performance among other unseen domains because of the domain shift; and (3) it leads to linearly increasing storage overhead, either mixing all the data for training or simply training dozens of separate models for different domains when new ones are available. To overcome these issues, we investigated a new crowd counting task in the incremental domains training setting called Lifelong Crowd Counting. Its goal is to alleviate the catastrophic forgetting and improve the generalization ability using a single model updated by the incremental domains. Specifically, we propose a self-distillation learning framework as a benchmark (Forget Less, Count Better, or FLCB) for lifelong crowd counting, which helps the model sustainably leverage previous meaningful knowledge for better crowd counting to mitigate the forgetting when the new data arrive. In addition, a new quantitative metric, normalized backward transfer (nBwT), is developed to evaluate the forgetting degree of the model in the lifelong learning process. Extensive experimental results demonstrate the superiority of our proposed benchmark in achieving a low catastrophic forgetting degree and strong generalization ability.

Yiming Lei, Zilong Li, Yan Shen et al.

Lung nodule malignancy prediction has been enhanced by advanced deep-learning techniques and effective tricks. Nevertheless, current methods are mainly trained with cross-entropy loss using one-hot categorical labels, which results in difficulty in distinguishing those nodules with closer progression labels. Interestingly, we observe that clinical text information annotated by radiologists provides us with discriminative knowledge to identify challenging samples. Drawing on the capability of the contrastive language-image pre-training (CLIP) model to learn generalized visual representations from text annotations, in this paper, we propose CLIP-Lung, a textual knowledge-guided framework for lung nodule malignancy prediction. First, CLIP-Lung introduces both class and attribute annotations into the training of the lung nodule classifier without any additional overheads in inference. Second, we designed a channel-wise conditional prompt (CCP) module to establish consistent relationships between learnable context prompts and specific feature maps. Third, we align image features with both class and attribute features via contrastive learning, rectifying false positives and false negatives in latent space. The experimental results on the benchmark LIDC-IDRI dataset have demonstrated the superiority of CLIP-Lung, both in classification performance and interpretability of attention maps.

Yiming Lei, Haiping Zhu, Junping Zhang et al.

The performance of medical image classification has been enhanced by deep convolutional neural networks (CNNs), which are typically trained with cross-entropy (CE) loss. However, when the label presents an intrinsic ordinal property in nature, e.g., the development from benign to malignant tumor, CE loss cannot take into account such ordinal information to allow for better generalization. To improve model generalization with ordinal information, we propose a novel meta ordinal regression forest (MORF) method for medical image classification with ordinal labels, which learns the ordinal relationship through the combination of convolutional neural network and differential forest in a meta-learning framework. The merits of the proposed MORF come from the following two components: a tree-wise weighting net (TWW-Net) and a grouped feature selection (GFS) module. First, the TWW-Net assigns each tree in the forest with a specific weight that is mapped from the classification loss of the corresponding tree. Hence, all the trees possess varying weights, which is helpful for alleviating the tree-wise prediction variance. Second, the GFS module enables a dynamic forest rather than a fixed one that was previously used, allowing for random feature perturbation. During training, we alternatively optimize the parameters of the CNN backbone and TWW-Net in the meta-learning framework through calculating the Hessian matrix. Experimental results on two medical image classification datasets with ordinal labels, i.e., LIDC-IDRI and Breast Ultrasound Dataset, demonstrate the superior performances of our MORF method over existing state-of-the-art methods.

Yujie Wei, Yujin Han, Zhekai Chen et al.

Video generation is rapidly evolving from single-shot synthesis to complex multi-shot audio-video (MSAV) narratives to meet real-world demands. However, evaluating such frontier models remains a fundamental challenge. Existing benchmarks are limited in scope and data diversity, and rely on rigid evaluation pipelines, preventing systematic and reliable assessment of modern MSAV models. To bridge these gaps, we introduce MSAVBench, the first comprehensive benchmark and adaptive hybrid evaluation framework for multi-shot audio-video generation. Our benchmark spans four key dimensions, video, audio, shot, and reference, covering diverse task settings, varying shot counts of up to 15, and challenging non-realistic scenarios. Our evaluation framework improves robustness through an adaptive self-correction mechanism for shot segmentation, instance-wise rubrics for subjective metrics, and tool-grounded evidence extraction for complex judgments. Furthermore, MSAVBench achieves high alignment with human judgments, reaching a Spearman rank correlation of 91.5%. Our systematic evaluation of 19 state-of-the-art closed- and open-source models shows that current systems still struggle with director-level control and fine-grained audio-visual synchronization, while modular or agentic generation pipelines offer a promising path toward narrowing the gap between open- and closed-source models. We will release the benchmark data and evaluation code to facilitate future research.

Jiaxin Ye, Gaoxiang Cong, Chenhui Wang et al.

Video-to-Speech (VTS) generation aims to synthesize speech from a silent video without auditory signals. However, existing VTS methods disregard the hierarchical nature of speech, which spans coarse speaker-aware semantics to fine-grained prosodic details. This oversight hinders direct alignment between visual and speech features at specific hierarchical levels during property matching. In this paper, leveraging the hierarchical structure of Residual Vector Quantization (RVQ)-based codec, we propose HiCoDiT, a novel Hierarchical Codec Diffusion Transformer that exploits the inherent hierarchy of discrete speech tokens to achieve strong audio-visual alignment. Specifically, since lower-level tokens encode coarse speaker-aware semantics and higher-level tokens capture fine-grained prosody, HiCoDiT employs low-level and high-level blocks to generate tokens at different levels. The low-level blocks condition on lip-synchronized motion and facial identity to capture speaker-aware content, while the high-level blocks use facial expression to modulate prosodic dynamics. Finally, to enable more effective coarse-to-fine conditioning, we propose a dual-scale adaptive instance layer normalization that jointly captures global vocal style through channel-wise normalization and local prosody dynamics through temporal-wise normalization. Extensive experiments demonstrate that HiCoDiT outperforms baselines in fidelity and expressiveness, highlighting the potential of discrete modelling for VTS. The code and speech demo are both available at https://github.com/Jiaxin-Ye/HiCoDiT.

Weiyi Yu, Yiming Lei, Hongming Shan

Since stroke is the main cause of various cerebrovascular diseases, deep learning-based stroke lesion segmentation on magnetic resonance (MR) images has attracted considerable attention. However, the existing methods often neglect the domain shift among MR images collected from different sites, which has limited performance improvement. To address this problem, we intend to change style information without affecting high-level semantics via adaptively changing the low-frequency amplitude components of the Fourier transform so as to enhance model robustness to varying domains. Thus, we propose a novel FAN-Net, a U-Net--based segmentation network incorporated with a Fourier-based adaptive normalization (FAN) and a domain classifier with a gradient reversal layer. The FAN module is tailored for learning adaptive affine parameters for the amplitude components of different domains, which can dynamically normalize the style information of source images. Then, the domain classifier provides domain-agnostic knowledge to endow FAN with strong domain generalizability. The experimental results on the ATLAS dataset, which consists of MR images from 9 sites, show the superior performance of the proposed FAN-Net compared with baseline methods.

Yiming Lei, Zilong Li, Yangyang Li et al.

The goal of image ordinal estimation is to estimate the ordinal label of a given image with a convolutional neural network. Existing methods are mainly based on ordinal regression and particularly focus on modeling the ordinal mapping from the feature representation of the input to the ordinal label space. However, the manifold of the resultant feature representations does not maintain the intrinsic ordinal relations of interest, which hinders the effectiveness of the image ordinal estimation. Therefore, this paper proposes learning intrinsic Consistent Ordinal REpresentations (CORE) from ordinal relations residing in groundtruth labels while encouraging the feature representations to embody the ordinal low-dimensional manifold. First, we develop an ordinal totally ordered set (toset) distribution (OTD), which can (i) model the label embeddings to inherit ordinal information and measure distances between ordered labels of samples in a neighborhood, and (ii) model the feature embeddings to infer numerical magnitude with unknown ordinal information among the features of different samples. Second, through OTD, we convert the feature representations and labels into the same embedding space for better alignment, and then compute the Kullback Leibler (KL) divergence between the ordinal labels and feature representations to endow the latent space with consistent ordinal relations. Third, we optimize the KL divergence through ordinal prototype-constrained convex programming with dual decomposition; our theoretical analysis shows that we can obtain the optimal solutions via gradient backpropagation. Extensive experimental results demonstrate that the proposed CORE can accurately construct an ordinal latent space and significantly enhance existing deep ordinal regression methods to achieve better results.

Jingqi Li, Jiaqi Gao, Yuzhen Zhang et al.

As a unique biometric that can be perceived at a distance, gait has broad applications in person authentication, social security, and so on. Existing gait recognition methods suffer from changes in viewpoint and clothing and barely consider extracting diverse motion features, a fundamental characteristic in gaits, from gait sequences. This paper proposes a novel motion modeling method to extract the discriminative and robust representation. Specifically, we first extract the motion features from the encoded motion sequences in the shallow layer. Then we continuously enhance the motion feature in deep layers. This motion modeling approach is independent of mainstream work in building network architectures. As a result, one can apply this motion modeling method to any backbone to improve gait recognition performance. In this paper, we combine motion modeling with one commonly used backbone~(GaitGL) as GaitGL-M to illustrate motion modeling. Extensive experimental results on two commonly-used cross-view gait datasets demonstrate the superior performance of GaitGL-M over existing state-of-the-art methods.

Rodrigo de Barros Vimieiro, Chuang Niu, Hongming Shan et al.

Digital breast tomosynthesis (DBT) exams should utilize the lowest possible radiation dose while maintaining sufficiently good image quality for accurate medical diagnosis. In this work, we propose a convolution neural network (CNN) to restore low-dose (LD) DBT projections to achieve an image quality equivalent to a standard full-dose (FD) acquisition. The proposed network architecture benefits from priors in terms of layers that were inspired by traditional model-based (MB) restoration methods, considering a model-based deep learning approach, where the network is trained to operate in the variance stabilization transformation (VST) domain. To accurately control the network operation point, in terms of noise and blur of the restored image, we propose a loss function that minimizes the bias and matches residual noise between the input and the output. The training dataset was composed of clinical data acquired at the standard FD and low-dose pairs obtained by the injection of quantum noise. The network was tested using real DBT projections acquired with a physical anthropomorphic breast phantom. The proposed network achieved superior results in terms of the mean normalized squared error (MNSE), training time and noise spatial correlation compared with networks trained with traditional data-driven methods. The proposed approach can be extended for other medical imaging application that requires LD acquisitions.

Zhihao Chen, Jiahui Wang, Yizhou Chen et al.

PET theranostics is transforming precision oncology, yet treatment response varies substantially; many patients receiving 177Lu-PSMA radioligand therapy (RLT) for metastatic castration-resistant prostate cancer (mCRPC) fail to respond, demanding reliable pre-therapy prediction. While LLM-based agents have shown remarkable potential in complex medical diagnosis, their application to PET theranostic outcome prediction remains unexplored, which faces three key challenges: (1) data and knowledge scarcity: RLT was only FDA-approved in 2022, yielding few training cases and insufficient domain knowledge in general LLMs; (2) heterogeneous information integration: robust prediction hinges on structured knowledge extraction from PET/CT, laboratory tests, and free-text clinical documentation; (3) evidence-grounded reasoning: clinical decisions must be anchored in trial evidence rather than LLM hallucinations. In this paper, we present TheraAgent, to our knowledge, the first agentic framework for PET theranostics, with three core innovations: (1) Multi-Expert Feature Extraction with Confidence-Weighted Consensus, where three specialized experts process heterogeneous inputs with uncertainty quantification; (2) Self-Evolving Agentic Memory (SEA-Mem), which learns prognostic patterns from accumulated cases, enabling case-based reasoning from limited data; (3) Evidence-Calibrated Reasoning, integrating a curated theranostics knowledge base to ground predictions in VISION/TheraP trial evidence. Evaluated on 35 real patients and 400 synthetic cases, TheraAgent achieves 75.7% overall accuracy on real patients and 87.0% on synthetic cases, outperforming MDAgents and MedAgent-Pro by over 20%. These results highlight a promising blueprint for trustworthy AI agents in PET theranostics, enabling trial-calibrated, multi-source decision support. Code will be released upon acceptance.

Xin Zheng, Ziang Peng, Yuan Cao et al.

Video prediction, predicting future frames from the previous ones, has broad applications such as autonomous driving and weather forecasting. Existing state-of-the-art methods typically focus on extracting either spatial, temporal, or spatiotemporal features from videos. Different feature focuses, resulting from different network architectures, may make the resultant models excel at some video prediction tasks but perform poorly on others. Towards a more generic video prediction solution, we explicitly model these features in a unified encoder-decoder framework and propose a novel simple alternating Mixer (SIAM). The novelty of SIAM lies in the design of dimension alternating mixing (DaMi) blocks, which can model spatial, temporal, and spatiotemporal features through alternating the dimensions of the feature maps. Extensive experimental results demonstrate the superior performance of the proposed SIAM on four benchmark video datasets covering both synthetic and real-world scenarios.

Yujie Wei, Xinyu Liu, Shiwei Zhang et al.

While large-scale diffusion models have revolutionized video synthesis, achieving precise control over both multi-subject identity and multi-granularity motion remains a significant challenge. Recent attempts to bridge this gap often suffer from limited motion granularity, control ambiguity, and identity degradation, leading to suboptimal performance on identity preservation and motion control. In this work, we present DreamVideo-Omni, a unified framework enabling harmonious multi-subject customization with omni-motion control via a progressive two-stage training paradigm. In the first stage, we integrate comprehensive control signals for joint training, encompassing subject appearances, global motion, local dynamics, and camera movements. To ensure robust and precise controllability, we introduce a condition-aware 3D rotary positional embedding to coordinate heterogeneous inputs and a hierarchical motion injection strategy to enhance global motion guidance. Furthermore, to resolve multi-subject ambiguity, we introduce group and role embeddings to explicitly anchor motion signals to specific identities, effectively disentangling complex scenes into independent controllable instances. In the second stage, to mitigate identity degradation, we design a latent identity reward feedback learning paradigm by training a latent identity reward model upon a pretrained video diffusion backbone. This provides motion-aware identity rewards in the latent space, prioritizing identity preservation aligned with human preferences. Supported by our curated large-scale dataset and the comprehensive DreamOmni Bench for multi-subject and omni-motion control evaluation, DreamVideo-Omni demonstrates superior performance in generating high-quality videos with precise controllability.

Ziyuan Yang, Zerui Shao, Huijie Huangfu et al.

Federated learning (FL) is a promising distributed paradigm, eliminating the need for data sharing but facing challenges from data heterogeneity. Personalized parameter generation through a hypernetwork proves effective, yet existing methods fail to personalize local model structures. This leads to redundant parameters struggling to adapt to diverse data distributions. To address these limitations, we propose FedOFA, utilizing personalized orthogonal filter attention for parameter recalibration. The core is the Two-stream Filter-aware Attention (TFA) module, meticulously designed to extract personalized filter-aware attention maps, incorporating Intra-Filter Attention (IntraFa) and Inter-Filter Attention (InterFA) streams. These streams enhance representation capability and explore optimal implicit structures for local models. Orthogonal regularization minimizes redundancy by averting inter-correlation between filters. Furthermore, we introduce an Attention-Guided Pruning Strategy (AGPS) for communication efficiency. AGPS selectively retains crucial neurons while masking redundant ones, reducing communication costs without performance sacrifice. Importantly, FedOFA operates on the server side, incurring no additional computational cost on the client, making it advantageous in communication-constrained scenarios. Extensive experiments validate superior performance over state-of-the-art approaches, with code availability upon paper acceptance.

Zilong Li, Yiming Lei, Chenglong Ma et al.

Image restoration, which aims to retrieve and enhance degraded images, is fundamental across a wide range of applications. While conventional deep learning approaches have notably improved the image quality across various tasks, they still suffer from (i) the high storage cost needed for various task-specific models and (ii) the lack of interactivity and flexibility, hindering their wider application. Drawing inspiration from the pronounced success of prompts in both linguistic and visual domains, we propose novel Prompt-In-Prompt learning for universal image restoration, named PIP. First, we present two novel prompts, a degradation-aware prompt to encode high-level degradation knowledge and a basic restoration prompt to provide essential low-level information. Second, we devise a novel prompt-to-prompt interaction module to fuse these two prompts into a universal restoration prompt. Third, we introduce a selective prompt-to-feature interaction module to modulate the degradation-related feature. By doing so, the resultant PIP works as a plug-and-play module to enhance existing restoration models for universal image restoration. Extensive experimental results demonstrate the superior performance of PIP on multiple restoration tasks, including image denoising, deraining, dehazing, deblurring, and low-light enhancement. Remarkably, PIP is interpretable, flexible, efficient, and easy-to-use, showing promising potential for real-world applications. The code is available at https://github.com/longzilicart/pip_universal.

Zhihao Chen, Tao Chen, Chenhui Wang et al.

While various deep learning methods were proposed for low-dose computed tomography (CT) denoising, they often suffer from over-smoothing, blurring, and lack of explainability. To alleviate these issues, we propose a plug-and-play Language-Engaged Dual-space Alignment loss (LEDA) to optimize low-dose CT denoising models. Our idea is to leverage large language models (LLMs) to align denoised CT and normal dose CT images in both the continuous perceptual space and discrete semantic space, which is the first LLM-based scheme for low-dose CT denoising. LEDA involves two steps: the first is to pretrain an LLM-guided CT autoencoder, which can encode a CT image into continuous high-level features and quantize them into a token space to produce semantic tokens derived from the LLM's vocabulary; and the second is to minimize the discrepancy between the denoised CT images and normal dose CT in terms of both encoded high-level features and quantized token embeddings derived by the LLM-guided CT autoencoder. Extensive experimental results on two public LDCT denoising datasets demonstrate that our LEDA can enhance existing denoising models in terms of quantitative metrics and qualitative evaluation, and also provide explainability through language-level image understanding. Source code is available at https://github.com/hao1635/LEDA.

Zilong Li, Chenglong Ma, Yiming Lei et al.

Computed tomography (CT) metal artifact reduction (MAR) aims to reduce the severe streaking artifacts induced by metallic implants and other high-density objects. Effective MAR generally requires both accurate artifact localization and artifact removal. Sinogram-domain methods can exploit explicit geometric cues, such as metal traces, to identify metal-corrupted measurements, while requiring raw projection data, which is often unavailable in clinical and practical scenarios. Image-domain methods are more flexible and widely applicable, yet they usually lack comparable geometric guidance, limiting their ability to localize artifacts and leading to suboptimal results. To address this limitation, we propose GraphMAR, a geometry-aware learning framework for explicit artifact identification and spatially adaptive MAR in the image domain. The key idea is to introduce graph-based geometric modeling as an image-domain analogue of sinogram metal traces. Specifically, we first construct a geometric graph from the metal mask and derive a geometric density graph that coarsely localizes artifact-prone regions according to inter-implant geometry. We then design GraphMoE, a graph-routed mixture-of-experts module that builds a polar-coordinate artifact graph in feature space and adaptively routes different experts to different spatial regions for MAR. By aligning the learned routing maps with the geometric density graph, GraphMAR provides explicit and interpretable artifact localization while enabling region-adaptive artifact reduction. Experiments on both simulated and real-world datasets demonstrate that GraphMAR achieves superior MAR performance compared with existing methods. To the best of our knowledge, this is the first work to introduce graph-based modeling for CT MAR and to enable explicit artifact identification in the image domain, improving both restoration quality and interpretability.

Qi Gao, Zhihao Chen, Dong Zeng et al.

The generalization of deep learning-based low-dose computed tomography (CT) reconstruction models to doses unseen in the training data is important and remains challenging. Previous efforts heavily rely on paired data to improve the generalization performance and robustness through collecting either diverse CT data for re-training or a few test data for fine-tuning. Recently, diffusion models have shown promising and generalizable performance in low-dose CT (LDCT) reconstruction, however, they may produce unrealistic structures due to the CT image noise deviating from Gaussian distribution and imprecise prior information from the guidance of noisy LDCT images. In this paper, we propose a noise-inspired diffusion model for generalizable LDCT reconstruction, termed NEED, which tailors diffusion models for noise characteristics of each domain. First, we propose a novel shifted Poisson diffusion model to denoise projection data, which aligns the diffusion process with the noise model in pre-log LDCT projections. Second, we devise a doubly guided diffusion model to refine reconstructed images, which leverages LDCT images and initial reconstructions to more accurately locate prior information and enhance reconstruction fidelity. By cascading these two diffusion models for dual-domain reconstruction, our NEED requires only normal-dose data for training and can be effectively extended to various unseen dose levels during testing via a time step matching strategy. Extensive qualitative, quantitative, and segmentation-based evaluations on two datasets demonstrate that our NEED consistently outperforms state-of-the-art methods in reconstruction and generalization performance. Source code is made available at https://github.com/qgao21/NEED.

Gaoxiang Cong, Liang Li, Jiaxin Ye et al.

Movie dubbing aims to synthesize speech that preserves the vocal identity of a reference audio while synchronizing with the lip movements in a target video. Existing methods fail to achieve precise lip-sync and lack naturalness due to explicit alignment at the duration level. While implicit alignment solutions have emerged, they remain susceptible to interference from the reference audio, triggering timbre and pronunciation degradation in in-the-wild scenarios. In this paper, we propose a novel flow matching-based movie dubbing framework driven by the Cognitive Synchronous Diffusion Transformer (CoSync-DiT), inspired by the cognitive process of professional actors. This architecture progressively guides the noise-to-speech generative trajectory by executing acoustic style adapting, fine-grained visual calibrating, and time-aware context aligning. Furthermore, we design the Joint Semantic and Alignment Regularization (JSAR) mechanism to simultaneously constrain frame-level temporal consistency on the contextual outputs and semantic consistency on the flow hidden states, ensuring robust alignment. Extensive experiments on both standard benchmarks and challenging in-the-wild dubbing benchmarks demonstrate that our method achieves the state-of-the-art performance across multiple metrics.

Fengzhi Xu, Ziyuan Yang, Zexin Lu et al.

Non-ideal measurement computed tomography (NICT), which lowers radiation at the cost of image quality, is expanding the clinical use of CT. Although unified models have shown promise in NICT enhancement, most methods require paired data, which is an impractical demand due to inevitable organ motion. Unsupervised approaches attempt to overcome this limitation, but their assumption of homogeneous noise neglects the variability of scanning protocols, leading to poor generalization and potential model collapse. We further observe that distinct scanning protocols, which correspond to different physical imaging processes, produce discrete sub-manifolds in the feature space, contradicting these assumptions and limiting their effectiveness. To address this, we propose an Uncertainty-Guided Manifold Smoothing (UMS) framework to bridge the gaps between sub-manifolds. A classifier in UMS identifies sub-manifolds and predicts uncertainty scores, which guide the generation of diverse samples across the entire manifold. By leveraging the classifier's capability, UMS effectively fills the gaps between discrete sub-manifolds, and promotes a continuous and dense feature space. Due to the complexity of the global manifold, it's hard to directly model it. Therefore, we propose to dynamically incorporate the global- and sub-manifold-specific features. Specifically, we design a global- and sub-manifold-driven architecture guided by the classifier, which enables dynamic adaptation to subdomain variations. This dynamic mechanism improves the network's capacity to capture both shared and domain-specific features, thereby improving reconstruction performance. Extensive experiments on public datasets are conducted to validate the effectiveness of our method across different generation paradigms.

Yujie Wei, Chenglong Ma, Jianxiong Gao et al.

Reconstructing dynamic visual experiences as videos from functional magnetic resonance imaging (fMRI) is pivotal for advancing the understanding of neural processes. However, current fMRI-to-video reconstruction methods are hindered by a semantic gap between noisy fMRI signals and the rich content of videos, stemming from a reliance on incomplete semantic embeddings that neither capture video-specific cues (e.g., actions) nor integrate prior knowledge. To this end, we draw inspiration from the dual-pathway processing mechanism in human brain and introduce CineNeuron, a novel hierarchical framework for semantically enhanced video reconstruction from fMRI signals with two synergistic stages. First, a bottom-up semantic enrichment stage maps fMRI signals to a rich embedding space that comprehensively captures textual semantics, image contents, action concepts, and object categories. Second, a top-down memory integration stage utilizes the proposed Mixture-of-Memories method to dynamically select relevant "memories" from previously seen data and fuse them with the fMRI embedding to refine the video reconstruction. Extensive experimental results on two fMRI-to-video benchmarks demonstrate that CineNeuron surpasses state-of-the-art methods across various metrics.

Chenglong Ma, Yuanfeng Ji, Jin Ye et al.

Advanced autoregressive models have reshaped multimodal AI. However, their transformative potential in medical imaging remains largely untapped due to the absence of a unified visual tokenizer -- one capable of capturing fine-grained visual structures for faithful image reconstruction and realistic image synthesis, as well as rich semantics for accurate diagnosis and image interpretation. To this end, we present MedITok, the first unified tokenizer tailored for medical images, encoding both low-level structural details and high-level clinical semantics within a unified latent space. To balance these competing objectives, we introduce a novel two-stage training framework: a visual representation alignment stage that cold-starts the tokenizer reconstruction learning with a visual semantic constraint, followed by a textual semantic representation alignment stage that infuses detailed clinical semantics into the latent space. Trained on the meticulously collected large-scale dataset with over 30 million medical images and 2 million image-caption pairs, MedITok achieves state-of-the-art performance on more than 30 datasets across 9 imaging modalities and 4 different tasks. By providing a unified token space for autoregressive modeling, MedITok supports a wide range of tasks in clinical diagnostics and generative healthcare applications. Model and code will be made publicly available at: https://github.com/Masaaki-75/meditok.

Zhihao Chen, Tao Chen, Chenhui Wang et al.

Low-dose computed tomography (LDCT) reduces radiation exposure but often degrades image quality, potentially compromising diagnostic accuracy. Existing deep learning-based denoising methods focus primarily on pixel-level mappings, overlooking the potential benefits of high-level semantic guidance. Recent advances in vision-language models (VLMs) suggest that language can serve as a powerful tool for capturing structured semantic information, offering new opportunities to improve LDCT reconstruction. In this paper, we introduce LangMamba, a Language-driven Mamba framework for LDCT denoising that leverages VLM-derived representations to enhance supervision from normal-dose CT (NDCT). LangMamba follows a two-stage learning strategy. First, we pre-train a Language-guided AutoEncoder (LangAE) that leverages frozen VLMs to map NDCT images into a semantic space enriched with anatomical information. Second, we synergize LangAE with two key components to guide LDCT denoising: Semantic-Enhanced Efficient Denoiser (SEED), which enhances NDCT-relevant local semantic while capturing global features with efficient Mamba mechanism, and Language-engaged Dual-space Alignment (LangDA) Loss, which ensures that denoised images align with NDCT in both perceptual and semantic spaces. Extensive experiments on two public datasets demonstrate that LangMamba outperforms conventional state-of-the-art methods, significantly improving detail preservation and visual fidelity. Remarkably, LangAE exhibits strong generalizability to unseen datasets, thereby reducing training costs. Furthermore, LangDA loss improves explainability by integrating language-guided insights into image reconstruction and offers a plug-and-play fashion. Our findings shed new light on the potential of language as a supervisory signal to advance LDCT denoising. The code is publicly available on https://github.com/hao1635/LangMamba.

Chenglong Ma, Zilong Li, Yuanlin Li et al.

Metal artifacts in computed tomography (CT) images can significantly degrade image quality and impede accurate diagnosis. Supervised metal artifact reduction (MAR) methods, trained using simulated datasets, often struggle to perform well on real clinical CT images due to a substantial domain gap. Although state-of-the-art semi-supervised methods use pseudo ground-truths generated by a prior network to mitigate this issue, their reliance on a fixed prior limits both the quality and quantity of these pseudo ground-truths, introducing confirmation bias and reducing clinical applicability. To address these limitations, we propose a novel Radiologist-In-the-loop SElf-training framework for MAR, termed RISE-MAR, which can integrate radiologists' feedback into the semi-supervised learning process, progressively improving the quality and quantity of pseudo ground-truths for enhanced generalization on real clinical CT images. For quality assurance, we introduce a clinical quality assessor model that emulates radiologist evaluations, effectively selecting high-quality pseudo ground-truths for semi-supervised training. For quantity assurance, our self-training framework iteratively generates additional high-quality pseudo ground-truths, expanding the clinical dataset and further improving model generalization. Extensive experimental results on multiple clinical datasets demonstrate the superior generalization performance of our RISE-MAR over state-of-the-art methods, advancing the development of MAR models for practical application. Code is available at https://github.com/Masaaki-75/rise-mar.

Chuang Niu, Wenjun Xia, Hongming Shan et al.

Self-supervised learning (SSL) has emerged as a crucial technique in image processing, encoding, and understanding, especially for developing today's vision foundation models that utilize large-scale datasets without annotations to enhance various downstream tasks. This study introduces a novel SSL approach, Information-Maximized Soft Variable Discretization (IMSVD), for image representation learning. Specifically, IMSVD softly discretizes each variable in the latent space, enabling the estimation of their probability distributions over training batches and allowing the learning process to be directly guided by information measures. Motivated by the MultiView assumption, we propose an information-theoretic objective function to learn transform-invariant, non-travail, and redundancy-minimized representation features. We then derive a joint-cross entropy loss function for self-supervised image representation learning, which theoretically enjoys superiority over the existing methods in reducing feature redundancy. Notably, our non-contrastive IMSVD method statistically performs contrastive learning. Extensive experimental results demonstrate the effectiveness of IMSVD on various downstream tasks in terms of both accuracy and efficiency. Thanks to our variable discretization, the embedding features optimized by IMSVD offer unique explainability at the variable level. IMSVD has the potential to be adapted to other learning paradigms. Our code is publicly available at https://github.com/niuchuangnn/IMSVD.

Boyuan Cao, Xingbo Yao, Chenhui Wang et al.

Diffusion transformers (DiTs) have emerged as a powerful architecture for high-fidelity image generation, yet the quadratic cost of self-attention poses a major scalability bottleneck. To address this, linear attention mechanisms have been adopted to reduce computational cost; unfortunately, the resulting linear diffusion transformers (LiTs) models often come at the expense of generative performance, frequently producing over-smoothed attention weights that limit expressiveness. In this work, we introduce Dynamic Differential Linear Attention (DyDiLA), a novel linear attention formulation that enhances the effectiveness of LiTs by mitigating the oversmoothing issue and improving generation quality. Specifically, the novelty of DyDiLA lies in three key designs: (i) dynamic projection module, which facilitates the decoupling of token representations by learning with dynamically assigned knowledge; (ii) dynamic measure kernel, which provides a better similarity measurement to capture fine-grained semantic distinctions between tokens by dynamically assigning kernel functions for token processing; and (iii) token differential operator, which enables more robust query-to-key retrieval by calculating the differences between the tokens and their corresponding information redundancy produced by dynamic measure kernel. To capitalize on DyDiLA, we introduce a refined LiT, termed DyDi-LiT, that systematically incorporates our advancements. Extensive experiments show that DyDi-LiT consistently outperforms current state-of-the-art (SOTA) models across multiple metrics, underscoring its strong practical potential.

Yujie Wei, Shiwei Zhang, Zhiwu Qing et al.

Customized generation using diffusion models has made impressive progress in image generation, but remains unsatisfactory in the challenging video generation task, as it requires the controllability of both subjects and motions. To that end, we present DreamVideo, a novel approach to generating personalized videos from a few static images of the desired subject and a few videos of target motion. DreamVideo decouples this task into two stages, subject learning and motion learning, by leveraging a pre-trained video diffusion model. The subject learning aims to accurately capture the fine appearance of the subject from provided images, which is achieved by combining textual inversion and fine-tuning of our carefully designed identity adapter. In motion learning, we architect a motion adapter and fine-tune it on the given videos to effectively model the target motion pattern. Combining these two lightweight and efficient adapters allows for flexible customization of any subject with any motion. Extensive experimental results demonstrate the superior performance of our DreamVideo over the state-of-the-art methods for customized video generation. Our project page is at https://dreamvideo-t2v.github.io.

Chenhui Wang, Boyun Zheng, Liuxin Bao et al.

Precise prognostic modeling of glioblastoma (GBM) under varying treatment interventions is essential for optimizing clinical outcomes. While generative AI has shown promise in simulating GBM evolution, existing methods typically treat interventions as static conditional inputs rather than dynamic decision variables. Consequently, they fail to capture the complex, reciprocal interplay between tumor evolution and treatment response. To bridge this gap, we present Brain-WM, a pioneering brain GBM world model that unifies next-step treatment prediction and future MRI generation, thereby capturing the co-evolutionary dynamics between tumor and treatment. Specifically, Brain-WM encodes spatiotemporal dynamics into a shared latent space for joint autoregressive treatment prediction and flow-based future MRI generation. Then, instead of a conventional monolithic framework, Brain-WM adopts a novel Y-shaped Mixture-of-Transformers (MoT) architecture. This design structurally disentangles heterogeneous objectives, successfully leveraging cross-task synergies while preventing feature collapse. Finally, a synergistic multi-timepoint mask alignment objective explicitly anchors latent representations to anatomically grounded tumor structures and progression-aware semantics. Extensive validation on internal and external multi-institutional cohorts demonstrates the superiority of Brain-WM, achieving 91.5% accuracy in treatment planning and SSIMs of 0.8524, 0.8581, and 0.8404 for FLAIR, T1CE, and T2W sequences, respectively. Ultimately, Brain-WM offers a robust clinical sandbox for optimizing patient healthcare. The source code is made available at https://github.com/thibault-wch/Brain-GBM-world-model.

Xingyu Ai, Shaoyu Wang, Zhiyuan Jia et al.

During raw-data acquisition in CT imaging, diverse factors can degrade the collected sinograms, with undersampling and noise leading to severe artifacts and noise in reconstructed images and compromising diagnostic accuracy. Conventional correction methods rely on manually designed algorithms or fixed empirical parameters, but these approaches often lack generalizability across heterogeneous artifact types. To address these limitations, we propose UniSino, a foundation model for universal CT sinogram standardization. Unlike existing foundational models that operate in image domain, UniSino directly standardizes data in the projection domain, which enables stronger generalization across diverse undersampling scenarios. Its training framework incorporates the physical characteristics of sinograms, enhancing generalization and enabling robust performance across multiple subtasks spanning four benchmark datasets. Experimental results demonstrate thatUniSino achieves superior reconstruction quality both single and mixed undersampling case, demonstrating exceptional robustness and generalization in sinogram enhancement for CT imaging. The code is available at: https://github.com/yqx7150/UniSino.