The segmentation of 2D vascular structures via deep learning holds significant clinical value but is hindered by the scarcity of annotated data, severely limiting its widespread application. Developing a universal few-shot vascular segmentation model is highly desirable, yet remains challenging due to the need for extensive training and the inherent complexities of vascular imaging. In this work, we propose UniVG (Generative Data-engine Foundation Model for Universal Few-shot 2D Vascular Image Segmentation), a novel approach that learns the compositionality of vascular images and constructing a generative foundation model for robust vascular segmentation. UniVG enables the synthesis and learning of diverse and realistic vascular images through two key innovations: 1) Compositional learning for flexible and diverse vascular synthesis: It decomposes and recombines vascular structures with varying morphological features and diverse foreground-background configurations to generate richly diverse synthetic image-label pairs. 2) Few-shot generative adaptation for transferable segmentation: It fine-tunes pre-trained models with minimal annotated data to bridge the gap between synthetic and real vascular domains, synthesizing authentic and diverse vessel images for downstream few-shot vascular segmentation learning. To support our approach, we develop UniVG-58K, a large dataset comprising 58,689 vascular images across five imaging modalities, facilitating robust large-scale generative pre-training. Extensive experiments on 11 vessel segmentation tasks cross 5 modalties (only with 5 labeled images on each task) demonstrate that UniVG achieves performance comparable to fully supervised models, significantly reducing data collection and annotation costs. All code and datasets will be made publicly available at https://github.com/XinAloha/UniVG.

Segment Anything Model (SAM) has demonstrated state-of-the-art performance in most segmentation tasks. However, due to insufficient training in the medical domain, SAM’s ability to generalize to medical images is limited. Although preliminary efforts have fine-tuned SAM for the medical domain, the fine-tuned model still struggles with variability in medical tasks. Some recent studies have explored weakly supervised learning to mitigate SAM’s performance degradation in the medical domain. However, the effectiveness of weakly supervised learning is heavily dependent on the quality of weakly supervised information, with performance significantly dropping as the quality declines. Doctors’ attention is closely related to the target area during diagnosis. Integrating gaze information into SAM’s adaptation process for medical image segmentation enhances efficiency and significantly improves performance in medical tasks. In this paper, we first propose a Gaze-assisted medical segment Anything Model (GAM), which utilizes gaze information to enable the adaptation of SAM in medical images following doctor’s attention. It has two innovations: 1) Feature-level adaptation: Gaze Alignment (GA) learning makes the feature-level adaptation follow the doctor’s attention which mines the human guidance from gaze heatmaps and guides model to extract general features for downstream tasks. 2) Output-level adaptation: Gaze-Balance (GB) learning makes the output-level adaptation follow the doctor’s attention which utilizes gaze heatmaps to enhance the human-focused area and solve the problem of over/under segmentation from the output-level. Our promising results on 7 tasks with 12 targets have demonstrated the powerful adaptation ability of our GAM in the medical domain. Our GAM demonstrates significant potential for low-cost clinical assistance in medical diagnosis, enabling SAM to adapt to the medical image domain without disrupting clinical workflows. We have released the full source code on https://github.com/Ruiz1026/GAM.

Angiographic enhancement of non-contrast CT (NCCT) using AI techniques is essential for diagnosing patients unable to use contrast agents. However, AI angiography remains a challenging task because of the feature fragility, structural complexity, and spatial continuity. In this paper, we propose an angiographic framework based on a conditional multi-view diffusion model called AEGIS with three innovations: multi-view hybrid learning (MHL), conditional angiographic diffusion estimation (CADE), and multi-view map fusion (MMF). 1) MHL targets Contrast Map (CM), the difference between NCCT and CT angiography, from multiple views to perceive 3D features in 2D space, enhancing the stability of feature representation. 2) CADE is a conditional diffusion model using NCCT as spatial guidance, providing crucial information for CM generation. 3) MMF adopts a lightweight AutoEncoder for filtering and fusing multi-view CMs, maintaining coherence between adjacent slices while modifying slight bias in low-dimensional representations, thus optimizing data quality and accuracy. Experiments demonstrate our superior performance, which achieve state-of-the-art image quality (PSNR+6.69, SSIM+3.17, MSE-46.38), segmentation evaluation (CADIR×10.49, HSDIR×5.57) and feature distance (FID-64.27). Visualizations and positive evaluation scores from clinicians further reveals that AEGIS has significant potential in clinical applications.

In the field of medical image processing, vascular image segmentation plays a crucial role in clinical diagnosis, treatment planning, prognosis, and medical decision-making. Accurate and automated segmentation of vascular images can assist clinicians in understanding the vascular network structure, leading to more informed medical decisions. However, manual annotation of vascular images is time-consuming and challenging due to the fine and low-contrast vascular branches, especially in the medical imaging domain where annotation requires specialized knowledge and clinical expertise. Data-driven deep learning models struggle to achieve good performance when only a small number of annotated vascular images are available. To address this issue, this paper proposes a novel Conditional Virtual Imaging (CVI) framework for few-shot vascular image segmentation learning. The framework combines limited annotated data with extensive unlabeled data to generate high-quality images, effectively improving the accuracy and robustness of segmentation learning. Our approach primarily includes two innovations: First, aligned image-mask pair generation, which leverages the powerful image generation capabilities of large pre-trained models to produce high-quality vascular images with complex structures using only a few training images; Second, the Dual-Consistency Learning (DCL) strategy, which simultaneously trains the generator and segmentation model, allowing them to learn from each other and maximize the utilization of limited data. Experimental results demonstrate that our CVI framework can generate high-quality medical images and effectively enhance the performance of segmentation models in few-shot scenarios. Our code will be made publicly available online.

Domain adaptation (DA) for cardiac ultrasound image segmentation is clinically significant and valuable. However, previous domain adaptation methods are prone to be affected by the incomplete pseudo label and low-quality target to source images. Human-centric domain adaptation has great advantages of human cognitive guidance to help model adapt to target domain and reduce reliance on labels. Doctor gaze trajectories contains a large amount of cross-domain human guidance. To leverage gaze information and human cognition for guiding domain adaptation, we propose gaze-assisted human-centric domain adaptation (GAHCDA), which reliably guides the domain adaptation of cardiac ultrasound images. GAHCDA includes following modules: (1) Gaze Augment Alignment (GAA): GAA enables the model to obtain human cognition general features to recognize segmentation target in different domain of cardiac ultrasound images like humans. (2) Gaze Balance Loss (GBL): GBL fused gaze heatmap with outputs which makes the segmentation result structurally closer to the target domain. The experimental results show that our proposed framework is able to segment cardiac ultrasound images more effectively in the target domain than GAN-based methods and other self-train based methods and shown great potential in clinical application.

Contrastive learning (CL) has become a cornerstone of self-supervised pretraining (SSP) in foundation models, however, extending CL to pixel-wise representation, crucial for medical vision, remains an open problem. Standard CL formulates SSP as a binary optimization problem (binary CL) where the excessive pursuit of feature dispersion leads to an over-dispersion problem, breaking pixel-wise feature correlation thus disrupting the intra-class distribution. Our vector CL reformulates CL as a vector regression problem, enabling dispersion quantification in pixel-wise pretraining via modeling feature distances in regressing displacement vectors. To implement this novel paradigm, we propose the COntrast in VEctor Regression (COVER) framework. COVER establishes an extendable vector-based self-learning, enforces a consistent optimization flow from vector regression to distance modeling, and leverages a vector pyramid architecture for granularity adaptation, thus preserving pixel-wise feature correlations in SSP. Extensive experiments across 8 tasks, spanning 2 dimensions and 4 modalities, show that COVER significantly improves pixel-wise SSP, advancing generalizable medical visual foundation models.

Multi-view data involves various data forms, such as multi-feature, multi-sequence and multimodal data, providing rich semantic information for downstream tasks. The inherent challenge of incomplete multi-view missing multi-label learning lies in how to effectively utilize limited supervision and insufficient data to learn discriminative representation. Starting from the sufficiency of multi-view shared information for downstream tasks, we argue that the existing contrastive learning paradigms on missing multi-view data show limited consistency representation learning ability, leading to the bottleneck in extracting multi-view shared information. In response, we propose to minimize task-independent redundant information by pursuing the maximization of cross-view mutual information. Additionally, to alleviate the hindrance caused by missing labels, we develop a dual-branch soft pseudo-label cross-imputation strategy to improve classification performance. Extensive experiments on multiple benchmarks validate our advantages and demonstrate strong compatibility with both missing and complete data.

Dense contrastive representation learning (DCRL) has greatly improved the learning efficiency for image dense prediction tasks, showing its great potential to reduce the large costs of medical image collection and dense annotation. However, the properties of medical images make unreliable correspondence discovery, bringing an open problem of large-scale false positive and negative (FP&N) pairs in DCRL. In this paper, we propose GEoMetric vIsual deNse sImilarity (GEMINI) learning which embeds the homeomorphism prior to DCRL and enables a reliable correspondence discovery for effective dense contrast. We proposes a deformable homeomorphism learning (DHL) which models the homeomorphism of medical images and learns to estimate a deformable mapping to predict the pixels' correspondence under the condition of topological preservation. It effectively reduces the searching space of pairing and drives an implicit and soft learning of negative pairs via gradient. We also proposes a geometric semantic similarity (GSS) which extracts semantic information in features to measure the alignment degree for the correspondence learning. It will promote the learning efficiency and performance of deformation, constructing positive pairs reliably. We implement two practical variants on two typical representation learning tasks in our experiments. Our promising results on seven datasets which outperform the existing methods show our great superiority. We will release our code at a companion website.

The segmentation of cardiac structure in magnetic resonance images (CMR) is paramount in diagnosing and managing cardiovascular illnesses, given its 3D+Time (3D+T) sequence. The existing deep learning methods are constrained in their ability to 3D+T CMR segmentation, due to: (1) Limited motion perception. The complexity of heart beating renders the motion perception in 3D+T CMR, including the long-range and cross-slice motions. The existing methods' local perception and slice-fixed perception directly limit the performance of 3D+T CMR perception. (2) Lack of labels. Due to the expensive labeling cost of the 3D+T CMR sequence, the labels of 3D+T CMR only contain the end-diastolic and end-systolic frames. The incomplete labeling scheme causes inefficient supervision. Hence, we propose a novel spatio-temporal adaptation network with clinical prior embedding learning (STANet) to ensure efficient spatio-temporal perception and optimization on 3D+T CMR segmentation. (1) A spatio-temporal adaptive convolution (STAC) treats the 3D+T CMR sequence as a whole for perception. The long-distance motion correlation is embedded into the structural perception by learnable weight regularization to balance long-range motion perception. The structural similarity is measured by cross-attention to adaptively correlate the cross-slice motion. (2) A clinical prior embedding learning strategy (CPE) is proposed to optimize the partially labeled 3D+T CMR segmentation dynamically by embedding clinical priors into optimization. STANet achieves outstanding performance with Dice of 0.917 and 0.94 on two public datasets (ACDC and STACOM), which indicates STANet has the potential to be incorporated into computer-aided diagnosis tools for clinical application.

In this work, we address the task of few-shot medical image segmentation (MIS) with a novel proposed framework based on the learning registration to learn segmentation (LRLS) paradigm. To cope with the limitations of lack of authenticity, diversity, and robustness in the existing LRLS frameworks, we propose the better registration better segmentation (BRBS) framework with three main contributions that are experimentally shown to have substantial practical merit. First, we improve the authenticity in the registration-based generation program and propose the knowledge consistency constraint strategy that constrains the registration network to learn according to the domain knowledge. It brings the semantic-aligned and topology-preserved registration, thus allowing the generation program to output new data with great space and style authenticity. Second, we deeply studied the diversity of the generation process and propose the space-style sampling program, which introduces the modeling of the transformation path of style and space change between few atlases and numerous unlabeled images into the generation program. Therefore, the sampling on the transformation paths provides much more diverse space and style features to the generated data effectively improving the diversity. Third, we first highlight the robustness in the learning of segmentation in the LRLS paradigm and propose the mix misalignment regularization, which simulates the misalignment distortion and constrains the network to reduce the fitting degree of misaligned regions. Therefore, it builds regularization for these regions improving the robustness of segmentation learning. Without any bells and whistles, our approach achieves a new state-of-the-art performance in few-shot MIS on two challenging tasks that outperform the existing LRLS-based few-shot methods. We believe that this novel and effective framework will provide a powerful few-shot benchmark for the field of medical image and efficiently reduce the costs of medical image research. All of our code will be made publicly available online.

Learning inter-image similarity is crucial for 3D medical images self-supervised pre-training, due to their sharing of numerous same semantic regions. However, the lack of the semantic prior in metrics and the semantic-independent variation in 3D medical images make it challenging to get a reliable measurement for the inter-image similarity, hindering the learning of consistent representation for same semantics. We investigate the challenging problem of this task, i.e., learning a consistent representation between images for a clustering effect of same semantic features. We propose a novel visual similarity learning paradigm, Geometric Visual Similarity Learning, which embeds the prior of topological invariance into the measurement of the inter-image similarity for consistent representation of semantic regions. To drive this paradigm, we further construct a novel geometric matching head, the Z-matching head, to collaboratively learn the global and local similarity of semantic regions, guiding the efficient representation learning for different scale-level inter-image semantic features. Our experiments demonstrate that the pre-training with our learning of inter-image similarity yields more powerful inner-scene, inter-scene, and global-local transferring ability on four challenging 3D medical image tasks. Our codes and pre-trained models will be publicly available11https://github.com/YutingHe-list/GVSL.

Accurate segmentation of topological tubular structures, such as blood vessels and roads, is crucial in various fields, ensuring accuracy and efficiency in downstream tasks. However, many factors complicate the task, including thin local structures and variable global morphologies. In this work, we note the specificity of tubular structures and use this knowledge to guide our DSCNet to simultaneously enhance perception in three stages: feature extraction, feature fusion, and loss constraint. First, we propose a dynamic snake convolution to accurately capture the features of tubular structures by adaptively focusing on slender and tortuous local structures. Subsequently, we propose a multi-view feature fusion strategy to complement the attention to features from multiple perspectives during feature fusion, ensuring the retention of important information from different global morphologies. Finally, a continuity constraint loss function, based on persistent homology, is proposed to constrain the topological continuity of the segmentation better. Experiments on 2D and 3D datasets show that our DSCNet provides better accuracy and continuity on the tubular structure segmentation task compared with several methods. Our codes are publicly available1.

Pulmonary arterial hypertension (PAH) prognosis prediction on 3D non-contrast CT images is one of the most important tasks for PAH treatment. It will help clinicians stratify patients into different groups for early diagnosis and timely intervention via automatically extracting the potential biomarkers of PAH to predict mortality. However, it is still a task of great challenges due to the large volume and low-contrast regions of interest in 3D chest CT images. In this paper, we propose the first multi-task learning-based PAH prognosis prediction framework, P$^{2}$-Net, which effectively optimizes the model and powerfully represents task-dependent features via our Memory Drift (MD) and Prior Prompt Learning (PPL) strategies. 1) Our MD maintains a large memory bank to provide a dense sampling of the deep biomarkers' distribution. Therefore, although the batch size is very small caused by our large volume, a reliable (negative log partial) likelihood loss is still able to be calculated on a representative probability distribution for robust optimization. 2) Our PPL simultaneously learns an additional manual biomarkers prediction task to embed clinical prior knowledge into our deep prognosis prediction task in hidden and explicit ways. Therefore, it will prompt the prediction of deep biomarkers and improve the perception of task-dependent features in our low-contrast regions. Our P$^{2}$-Net achieves a high prognostic correlation of the prediction and great generalization with the highest 70.19% C-index and 2.14 HR. Extensive experiments with promising results on our PAH prognosis prediction reveal powerful prognosis performance and great clinical significance in PAH treatment. All of our code will be made publicly available online.

The nature of thick-slice scanning causes severe inter-slice discontinuities of 3D medical images, and the vanilla 2D/3D convolutional neural networks (CNNs) fail to represent sparse inter-slice information and dense intra-slice information in a balanced way, leading to severe underfitting to inter-slice features (for vanilla 2D CNNs) and overfitting to noise from long-range slices (for vanilla 3D CNNs). In this work, a novel mesh network (MNet) is proposed to balance the spatial representation inter axes via learning. 1) Our MNet latently fuses plenty of representation processes by embedding multi-dimensional convolutions deeply into basic modules, making the selections of representation processes flexible, thus balancing representation for sparse inter-slice information and dense intra-slice information adaptively. 2) Our MNet latently fuses multi-dimensional features inside each basic module, simultaneously taking the advantages of 2D (high segmentation accuracy of the easily recognized regions in 2D view) and 3D (high smoothness of 3D organ contour) representations, thus obtaining more accurate modeling for target regions. Comprehensive experiments are performed on four public datasets (CT\&MR), the results consistently demonstrate the proposed MNet outperforms the other methods. The code and datasets are available at: https://github.com/zfdong-code/MNet

Deformable medical image registration estimates corresponding deformation to align the regions of interest (ROIs) of two images to a same spatial coordinate system. However, recent unsupervised registration models only have correspondence ability without perception, making misalignment on blurred anatomies and distortion on task-unconcerned backgrounds. Label-constrained (LC) registration models embed the perception ability via labels, but the lack of texture constraints in labels and the expensive labeling costs causes distortion internal ROIs and overfitted perception. We propose the first few-shot deformable medical image registration framework, Perception-Correspondence Registration (PC-Reg), which embeds perception ability to registration models only with few labels, thus greatly improving registration accuracy and reducing distortion. 1) We propose the Perception-Correspondence Decoupling which decouples the perception and correspondence actions of registration to two CNNs. Therefore, independent optimizations and feature representations are available avoiding interference of the correspondence due to the lack of texture constraints. 2) For few-shot learning, we propose Reverse Teaching which aligns labeled and unlabeled images to each other to provide supervision information to the structure and style knowledge in unlabeled images, thus generating additional training data. Therefore, these data will reversely teach our perception CNN more style and structure knowledge, improving its generalization ability. Our experiments on three datasets with only five labels demonstrate that our PC-Reg has competitive registration accuracy and effective distortion-reducing ability. Compared with LC-VoxelMorph( λ = 1), we achieve the 12.5%, 6.3% and 1.0% Reg-DSC improvements on three datasets, revealing our framework with great potential in clinical application.

The accurate 3D left ventricular (LV) myocardium segmentation in short-axis (SAX) view of cardiac magnetic resonance (CMR) is challenged by the sparse spatial structure of CMR. The strategy of multi-view CMR fusion can provide fine-grained spatial structure for accurate segmentation. However, the large information misalignment and lack of dense 3D CMR as fusion target in multi-view CMR fusion, and the different spatial resolution between the fusion result and the ground truth in segmentation limit the strategy. In this study, we propose a multi-view spatial-aware adversarial network (MVSGAN). It studies the perception of fine-grained cardiac structure for accurate segmentation by the spatialaware multi-view CMR fusion. It consists of three modules: (1) A residual adversarial fusion (RAF) module takes inter-slices deep correlation and anatomical prior to refine the spatial structures by residual supplement and adversarial optimization. (2) A structural perception-aggregation (SPA) module establishes the spatial correlation between the dense cardiac model and sparse label for accurate CMR LV myocardium segmentation. (3) A joint training strategy utilizes the dense SAX volume as explicit and implicit goals to jointly optimize the framework. The experiments are applied on a public dataset and a clinical dataset to evaluate the performance of MVSGAN. The average Dice and Jaccard score of LV myocardium segmentation obtained by MVSGAN are highest among seven existing state-of-the-art methods, which are up to 0.92 and 0.75. It is concluded that the spatial-aware multi-view CMR fusion can provide meaningful spatial correlation for accurate LV myocardium segmentation.

Receptive field-based segmentation models represent features in receptive fields having weak perception for thin semantics in thin structures segmentation, due to the challenges in small local size and large global variation. High-frequency (HiFe) components have strong thin perception ability and is stable for global variation, but its weak adaptability limits its direct application. We propose a HiFe priori rule which enables the network to adaptively extract and fuse HiFe components, enhancing the thin semantics and making the network naturally prefer thin structures for their segmentation. We further propose High-Frequency Semantics Enhancement Network (HiFeNet) based on our HiFe priori rule, boosting the SOTA methods in thin structures segmentation: 1) Our Deep High Frequency (DHiFe) block learns to extract task-dependent HiFe components and adds them to feature maps, achieving great perception of thin structures. 2) Our Latent Residual Denoising (LRD) block progressively weakens task-independent features via hierarchical residuals and learns to fuse HiFe components back to feature maps, further enhancing the thin semantics and weakening the interference of global variation. Extensive experiments on the retinal vessel [1], [2], [3] and Massachusetts road [4] segmentation datasets show great superiority of our HiFeNet.

Accurate coronary lumen segmentation on coronary-computed tomography angiography (CCTA) images is crucial for quantification of coronary stenosis and the subsequent computation of fractional flow reserve. Many factors including difficulty in labeling coronary lumens, various morphologies in stenotic lesions, thin structures and small volume ratio with respect to the imaging field complicate the task. In this work, we fused the continuity topological information of centerlines which are easily accessible, and proposed a novel weakly supervised model, Examinee-Examiner Network (EE-Net), to overcome the challenges in automatic coronary lumen segmentation. First, the EE-Net was proposed to address the fracture in segmentation caused by stenoses by combining the semantic features of lumens and the geometric constraints of continuous topology obtained from the centerlines. Then, a Centerline Gaussian Mask Module was proposed to deal with the insensitiveness of the network to the centerlines. Subsequently, a weakly supervised learning strategy, Examinee-Examiner Learning, was proposed to handle the weakly supervised situation with few lumen labels by using our EE-Net to guide and constrain the segmentation with customized prior conditions. Finally, a general network layer, Drop Output Layer, was proposed to adapt to the class imbalance by dropping well-segmented regions and weights the classes dynamically. Extensive experiments on two different data sets demonstrated that our EE-Net has good continuity and generalization ability on coronary lumen segmentation task compared with several widely used CNNs such as 3D-UNet. The results revealed our EE-Net with great potential for achieving accurate coronary lumen segmentation in patients with coronary artery disease. Code at http://github.com/qiyaolei/Examinee-Examiner-Network.