Pujin Cheng

Pujin Cheng, Li Lin, Junyan Lyu et al.

Contrastive learning based vision-language joint pre-training has emerged as a successful representation learning strategy. In this paper, we present a prototype representation learning framework incorporating both global and local alignment between medical images and reports. In contrast to standard global multi-modality alignment methods, we employ a local alignment module for fine-grained representation. Furthermore, a cross-modality conditional reconstruction module is designed to interchange information across modalities in the training phase by reconstructing masked images and reports. For reconstructing long reports, a sentence-wise prototype memory bank is constructed, enabling the network to focus on low-level localized visual and high-level clinical linguistic features. Additionally, a non-auto-regressive generation paradigm is proposed for reconstructing non-sequential reports. Experimental results on five downstream tasks, including supervised classification, zero-shot classification, image-to-text retrieval, semantic segmentation, and object detection, show the proposed method outperforms other state-of-the-art methods across multiple datasets and under different dataset size settings. The code is available at https://github.com/QtacierP/PRIOR.

Junyan Lyu, Yiqi Zhang, Yijin Huang et al.

Convolutional neural networks have been widely applied to medical image segmentation and have achieved considerable performance. However, the performance may be significantly affected by the domain gap between training data (source domain) and testing data (target domain). To address this issue, we propose a data manipulation based domain generalization method, called Automated Augmentation for Domain Generalization (AADG). Our AADG framework can effectively sample data augmentation policies that generate novel domains and diversify the training set from an appropriate search space. Specifically, we introduce a novel proxy task maximizing the diversity among multiple augmented novel domains as measured by the Sinkhorn distance in a unit sphere space, making automated augmentation tractable. Adversarial training and deep reinforcement learning are employed to efficiently search the objectives. Quantitative and qualitative experiments on 11 publicly-accessible fundus image datasets (four for retinal vessel segmentation, four for optic disc and cup (OD/OC) segmentation and three for retinal lesion segmentation) are comprehensively performed. Two OCTA datasets for retinal vasculature segmentation are further involved to validate cross-modality generalization. Our proposed AADG exhibits state-of-the-art generalization performance and outperforms existing approaches by considerable margins on retinal vessel, OD/OC and lesion segmentation tasks. The learned policies are empirically validated to be model-agnostic and can transfer well to other models. The source code is available at https://github.com/CRazorback/AADG.

Yijin Huang, Junyan Lyu, Pujin Cheng et al.

Self-supervised Learning (SSL) has been widely applied to learn image representations through exploiting unlabeled images. However, it has not been fully explored in the medical image analysis field. In this work, Saliency-guided Self-Supervised image Transformer (SSiT) is proposed for Diabetic Retinopathy (DR) grading from fundus images. We novelly introduce saliency maps into SSL, with a goal of guiding self-supervised pre-training with domain-specific prior knowledge. Specifically, two saliency-guided learning tasks are employed in SSiT: (1) Saliency-guided contrastive learning is conducted based on the momentum contrast, wherein fundus images' saliency maps are utilized to remove trivial patches from the input sequences of the momentum-updated key encoder. Thus, the key encoder is constrained to provide target representations focusing on salient regions, guiding the query encoder to capture salient features. (2) The query encoder is trained to predict the saliency segmentation, encouraging the preservation of fine-grained information in the learned representations. To assess our proposed method, four publicly-accessible fundus image datasets are adopted. One dataset is employed for pre-training, while the three others are used to evaluate the pre-trained models' performance on downstream DR grading. The proposed SSiT significantly outperforms other representative state-of-the-art SSL methods on all downstream datasets and under various evaluation settings. For example, SSiT achieves a Kappa score of 81.88% on the DDR dataset under fine-tuning evaluation, outperforming all other ViT-based SSL methods by at least 9.48%.

Yijin Huang, Pujin Cheng, Roger Tam et al.

The success of large-scale pre-trained models has established fine-tuning as a standard method for achieving significant improvements in downstream tasks. However, fine-tuning the entire parameter set of a pre-trained model is costly. Parameter-efficient transfer learning (PETL) has recently emerged as a cost-effective alternative for adapting pre-trained models to downstream tasks. Despite its advantages, the increasing model size and input resolution present challenges for PETL, as the training memory consumption is not reduced as effectively as the parameter usage. In this paper, we introduce Fine-grained Prompt Tuning plus (FPT+), a PETL method designed for high-resolution medical image classification, which significantly reduces the training memory consumption compared to other PETL methods. FPT+ performs transfer learning by training a lightweight side network and accessing pre-trained knowledge from a large pre-trained model (LPM) through fine-grained prompts and fusion modules. Specifically, we freeze the LPM of interest and construct a learnable lightweight side network. The frozen LPM processes high-resolution images to extract fine-grained features, while the side network employs corresponding down-sampled low-resolution images to minimize the memory usage. To enable the side network to leverage pre-trained knowledge, we propose fine-grained prompts and fusion modules, which collaborate to summarize information through the LPM's intermediate activations. We evaluate FPT+ on eight medical image datasets of varying sizes, modalities, and complexities. Experimental results demonstrate that FPT+ outperforms other PETL methods, using only 1.03% of the learnable parameters and 3.18% of the memory required for fine-tuning an entire ViT-B model. Our code is available on https://github.com/YijinHuang/FPT.

Ziqi Huang, Li Lin, Pujin Cheng et al.

Multi-modal magnetic resonance (MR) imaging provides great potential for diagnosing and analyzing brain gliomas. In clinical scenarios, common MR sequences such as T1, T2 and FLAIR can be obtained simultaneously in a single scanning process. However, acquiring contrast enhanced modalities such as T1ce requires additional time, cost, and injection of contrast agent. As such, it is clinically meaningful to develop a method to synthesize unavailable modalities which can also be used as additional inputs to downstream tasks (e.g., brain tumor segmentation) for performance enhancing. In this work, we propose an end-to-end framework named Modality-Level Attention Fusion Network (MAF-Net), wherein we innovatively conduct patchwise contrastive learning for extracting multi-modal latent features and dynamically assigning attention weights to fuse different modalities. Through extensive experiments on BraTS2020, our proposed MAF-Net is found to yield superior T1ce synthesis performance (SSIM of 0.8879 and PSNR of 22.78) and accurate brain tumor segmentation (mean Dice scores of 67.9%, 41.8% and 88.0% on segmenting the tumor core, enhancing tumor and whole tumor).

Ziqi Huang, Li Lin, Pujin Cheng et al.

Contrast-enhanced T1 (T1ce) is one of the most essential magnetic resonance imaging (MRI) modalities for diagnosing and analyzing brain tumors, especially gliomas. In clinical practice, common MRI modalities such as T1, T2, and fluid attenuation inversion recovery are relatively easy to access while T1ce is more challenging considering the additional cost and potential risk of allergies to the contrast agent. Therefore, it is of great clinical necessity to develop a method to synthesize T1ce from other common modalities. Current paired image translation methods typically have the issue of requiring a large amount of paired data and do not focus on specific regions of interest, e.g., the tumor region, in the synthesization process. To address these issues, we propose a Difficulty-perceived common-to-T1ce Semi-Supervised multimodal MRI Synthesis network (DS3-Net), involving both paired and unpaired data together with dual-level knowledge distillation. DS3-Net predicts a difficulty map to progressively promote the synthesis task. Specifically, a pixelwise constraint and a patchwise contrastive constraint are guided by the predicted difficulty map. Through extensive experiments on the publiclyavailable BraTS2020 dataset, DS3-Net outperforms its supervised counterpart in each respect. Furthermore, with only 5% paired data, the proposed DS3-Net achieves competitive performance with state-of-theart image translation methods utilizing 100% paired data, delivering an average SSIM of 0.8947 and an average PSNR of 23.60.

Linkai Peng, Li Lin, Pujin Cheng et al.

Unsupervised domain adaptation has been proposed recently to tackle the so-called domain shift between training data and test data with different distributions. However, most of them only focus on single-target domain adaptation and cannot be applied to the scenario with multiple target domains. In this paper, we propose RVms, a novel unsupervised multi-target domain adaptation approach to segment retinal vessels (RVs) from multimodal and multicenter retinal images. RVms mainly consists of a style augmentation and transfer (SAT) module and a dual-teacher knowledge distillation (DTKD) module. SAT augments and clusters images into source-similar domains and source-dissimilar domains via Bezier and Fourier transformations. DTKD utilizes the augmented and transformed data to train two teachers, one for source-similar domains and the other for source-dissimilar domains. Afterwards, knowledge distillation is performed to iteratively distill different domain knowledge from teachers to a generic student. The local relative intensity transformation is employed to characterize RVs in a domain invariant manner and promote the generalizability of teachers and student models. Moreover, we construct a new multimodal and multicenter vascular segmentation dataset from existing publicly-available datasets, which can be used to benchmark various domain adaptation and domain generalization methods. Through extensive experiments, RVms is found to be very close to the target-trained Oracle in terms of segmenting the RVs, largely outperforming other state-of-the-art methods.

Sirui Li, Li Lin, Yijin Huang et al.

In medical contexts, the imbalanced data distribution in long-tailed datasets, due to scarce labels for rare diseases, greatly impairs the diagnostic accuracy of deep learning models. Recent multimodal text-image supervised foundation models offer new solutions to data scarcity through effective representation learning. However, their limited medical-specific pretraining hinders their performance in medical image classification relative to natural images. To address this issue, we propose a novel Text-guided Foundation model Adaptation for Long-Tailed medical image classification (TFA-LT). We adopt a two-stage training strategy, integrating representations from the foundation model using just two linear adapters and a single ensembler for balanced outcomes. Experimental results on two long-tailed medical image datasets validate the simplicity, lightweight and efficiency of our approach: requiring only 6.1% GPU memory usage of the current best-performing algorithm, our method achieves an accuracy improvement of up to 27.1%, highlighting the substantial potential of foundation model adaptation in this area.

Juntao Chen, Li Lin, Pujin Cheng et al.

Medical image quality assessment (MIQA) is a vital prerequisite in various medical image analysis applications. Most existing MIQA algorithms are fully supervised that request a large amount of annotated data. However, annotating medical images is time-consuming and labor-intensive. In this paper, we propose an unsupervised anomaly-aware framework with test-time clustering for optical coherence tomography angiography (OCTA) image quality assessment in a setting wherein only a set of high-quality samples are accessible in the training phase. Specifically, a feature-embedding-based low-quality representation module is proposed to quantify the quality of OCTA images and then to discriminate between outstanding quality and non-outstanding quality. Within the non-outstanding quality class, to further distinguish gradable images from ungradable ones, we perform dimension reduction and clustering of multi-scale image features extracted by the trained OCTA quality representation network. Extensive experiments are conducted on one publicly accessible dataset sOCTA-3*3-10k, with superiority of our proposed framework being successfully established.

Yizhou Fang, Pujin Cheng, Yixiang Liu et al.

Distribution shift in medical imaging remains a central bottleneck for the clinical translation of medical AI. Failure to address it can lead to severe performance degradation in unseen environments and exacerbate health inequities. Existing methods for domain adaptation are inherently limited by exhausting predefined possibilities through simulated shifts or pseudo-supervision. Such strategies struggle in the open-ended and unpredictable real world, where distribution shifts are effectively infinite. To address this challenge, we introduce an empirical law called ``Rank Stability of Positive Regions'', which states that the relative rank of predicted probabilities for positive voxels remains stable under distribution shift. Guided by this principle, we propose CRISP, a parameter-free and model-agnostic framework requiring no target-domain information. CRISP is the first framework to make segmentation based on rank rather than probabilities. CRISP simulates model behavior under distribution shift via latent feature perturbation, where voxel probability rankings exhibit two stable patterns: regions that consistently retain high probabilities (destined positives according to the principle) and those that remain low-probability (can be safely classified as negatives). Based on these patterns, we construct high-precision (HP) and high-recall (HR) priors and recursively refine them under perturbation. We then design an iterative training framework, making HP and HR progressively ``squeeze'' to the final segmentation. Extensive evaluations on multi-center cardiac MRI and CT-based lung vessel segmentation demonstrate CRISP's superior robustness, significantly outperforming state-of-the-art methods with striking HD95 reductions of up to 0.14 (7.0\% improvement), 1.90 (13.1\% improvement), and 8.39 (38.9\% improvement) pixels across multi-center, demographic, and modality shifts, respectively.

Jin Lyu, Liang An, Pujin Cheng et al.

4D reconstruction of equine family (e.g. horses) from monocular video is important for animal welfare. Previous mainstream 4D animal reconstruction methods require joint optimization of motion and appearance over a whole video, which is time-consuming and sensitive to incomplete observation. In this work, we propose a novel framework called 4DEquine by disentangling the 4D reconstruction problem into two sub-problems: dynamic motion reconstruction and static appearance reconstruction. For motion, we introduce a simple yet effective spatio-temporal transformer with a post-optimization stage to regress smooth and pixel-aligned pose and shape sequences from video. For appearance, we design a novel feed-forward network that reconstructs a high-fidelity, animatable 3D Gaussian avatar from as few as a single image. To assist training, we create a large-scale synthetic motion dataset, VarenPoser, which features high-quality surface motions and diverse camera trajectories, as well as a synthetic appearance dataset, VarenTex, comprising realistic multi-view images generated through multi-view diffusion. While training only on synthetic datasets, 4DEquine achieves state-of-the-art performance on real-world APT36K and AiM datasets, demonstrating the superiority of 4DEquine and our new datasets for both geometry and appearance reconstruction. Comprehensive ablation studies validate the effectiveness of both the motion and appearance reconstruction network. Project page: https://luoxue-star.github.io/4DEquine_Project_Page/.

Li Lin, Yixiang Liu, Jiewei Wu et al.

Federated learning (FL) effectively mitigates the data silo challenge brought about by policies and privacy concerns, implicitly harnessing more data for deep model training. However, traditional centralized FL models grapple with diverse multi-center data, especially in the face of significant data heterogeneity, notably in medical contexts. In the realm of medical image segmentation, the growing imperative to curtail annotation costs has amplified the importance of weakly-supervised techniques which utilize sparse annotations such as points, scribbles, etc. A pragmatic FL paradigm shall accommodate diverse annotation formats across different sites, which research topic remains under-investigated. In such context, we propose a novel personalized FL framework with learnable prompt and aggregation (FedLPPA) to uniformly leverage heterogeneous weak supervision for medical image segmentation. In FedLPPA, a learnable universal knowledge prompt is maintained, complemented by multiple learnable personalized data distribution prompts and prompts representing the supervision sparsity. Integrated with sample features through a dual-attention mechanism, those prompts empower each local task decoder to adeptly adjust to both the local distribution and the supervision form. Concurrently, a dual-decoder strategy, predicated on prompt similarity, is introduced for enhancing the generation of pseudo-labels in weakly-supervised learning, alleviating overfitting and noise accumulation inherent to local data, while an adaptable aggregation method is employed to customize the task decoder on a parameter-wise basis. Extensive experiments on four distinct medical image segmentation tasks involving different modalities underscore the superiority of FedLPPA, with its efficacy closely parallels that of fully supervised centralized training. Our code and data will be available.

Jin Lyu, Tianyi Zhu, Yi Gu et al.

Quantitative analysis of animal behavior and biomechanics requires accurate animal pose and shape estimation across species, and is important for animal welfare and biological research. However, the small network capacity of previous methods and limited multi-species dataset leave this problem underexplored. To this end, this paper presents AniMer to estimate animal pose and shape using family aware Transformer, enhancing the reconstruction accuracy of diverse quadrupedal families. A key insight of AniMer is its integration of a high-capacity Transformer-based backbone and an animal family supervised contrastive learning scheme, unifying the discriminative understanding of various quadrupedal shapes within a single framework. For effective training, we aggregate most available open-sourced quadrupedal datasets, either with 3D or 2D labels. To improve the diversity of 3D labeled data, we introduce CtrlAni3D, a novel large-scale synthetic dataset created through a new diffusion-based conditional image generation pipeline. CtrlAni3D consists of about 10k images with pixel-aligned SMAL labels. In total, we obtain 41.3k annotated images for training and validation. Consequently, the combination of a family aware Transformer network and an expansive dataset enables AniMer to outperform existing methods not only on 3D datasets like Animal3D and CtrlAni3D, but also on out-of-distribution Animal Kingdom dataset. Ablation studies further demonstrate the effectiveness of our network design and CtrlAni3D in enhancing the performance of AniMer for in-the-wild applications. The project page of AniMer is https://luoxue-star.github.io/AniMer_project_page/.

Kai Pan, Linyang Li, Li Lin et al.

Photoacoustic Microscopy (PAM) images integrating the advantages of optical contrast and acoustic resolution have been widely used in brain studies. However, there exists a trade-off between scanning speed and image resolution. Compared with traditional raster scanning, rotational scanning provides good opportunities for fast PAM imaging by optimizing the scanning mechanism. Recently, there is a trend to incorporate deep learning into the scanning process to further increase the scanning speed.Yet, most such attempts are performed for raster scanning while those for rotational scanning are relatively rare. In this study, we propose a novel and well-performing super-resolution framework for rotational scanning-based PAM imaging. To eliminate adjacent rows' displacements due to subject motion or high-frequency scanning distortion,we introduce a registration module across odd and even rows in the preprocessing and incorporate displacement degradation in the training. Besides, gradient-based patch selection is proposed to increase the probability of blood vessel patches being selected for training. A Transformer-based network with a global receptive field is applied for better performance. Experimental results on both synthetic and real datasets demonstrate the effectiveness and generalizability of our proposed framework for rotationally scanned PAM images'super-resolution, both quantitatively and qualitatively. Code is available at https://github.com/11710615/PAMSR.git.

Jialin Li, Yiwei Ren, Kai Pan et al.

Recently, Multi-Contrast MR Reconstruction (MCMR) has emerged as a hot research topic that leverages high-quality auxiliary modalities to reconstruct undersampled target modalities of interest. However, existing methods often struggle to generalize across different k-space undersampling patterns, requiring the training of a separate model for each specific pattern, which limits their practical applicability. To address this challenge, we propose UniFS, a Unified Frequency-Spatial Fusion model designed to handle multiple k-space undersampling patterns for MCMR tasks without any need for retraining. UniFS integrates three key modules: a Cross-Modal Frequency Fusion module, an Adaptive Mask-Based Prompt Learning module, and a Dual-Branch Complementary Refinement module. These modules work together to extract domain-invariant features from diverse k-space undersampling patterns while dynamically adapt to their own variations. Another limitation of existing MCMR methods is their tendency to focus solely on spatial information while neglect frequency characteristics, or extract only shallow frequency features, thus failing to fully leverage complementary cross-modal frequency information. To relieve this issue, UniFS introduces an adaptive prompt-guided frequency fusion module for k-space learning, significantly enhancing the model's generalization performance. We evaluate our model on the BraTS and HCP datasets with various k-space undersampling patterns and acceleration factors, including previously unseen patterns, to comprehensively assess UniFS's generalizability. Experimental results across multiple scenarios demonstrate that UniFS achieves state-of-the-art performance. Our code is available at https://github.com/LIKP0/UniFS.

Huaqing He, Li Lin, Zhiyuan Cai et al.

Deep learning-based image segmentation and detection models have largely improved the efficiency of analyzing retinal landmarks such as optic disc (OD), optic cup (OC), and fovea. However, factors including ophthalmic disease-related lesions and low image quality issues may severely complicate automatic OD/OC segmentation and fovea detection. Most existing works treat the identification of each landmark as a single task, and take into account no prior information. To address these issues, we propose a prior guided multi-task transformer framework for joint OD/OC segmentation and fovea detection, named JOINEDTrans. JOINEDTrans effectively combines various spatial features of the fundus images, relieving the structural distortions induced by lesions and other imaging issues. It contains a segmentation branch and a detection branch. To be noted, we employ an encoder pretrained in a vessel segmentation task to effectively exploit the positional relationship among vessel, OD/OC, and fovea, successfully incorporating spatial prior into the proposed JOINEDTrans framework. There are a coarse stage and a fine stage in JOINEDTrans. In the coarse stage, OD/OC coarse segmentation and fovea heatmap localization are obtained through a joint segmentation and detection module. In the fine stage, we crop regions of interest for subsequent refinement and use predictions obtained in the coarse stage to provide additional information for better performance and faster convergence. Experimental results demonstrate that JOINEDTrans outperforms existing state-of-the-art methods on the publicly available GAMMA, REFUGE, and PALM fundus image datasets. We make our code available at https://github.com/HuaqingHe/JOINEDTrans

Li Lin, Linkai Peng, Huaqing He et al.

Weakly-supervised learning (WSL) has been proposed to alleviate the conflict between data annotation cost and model performance through employing sparsely-grained (i.e., point-, box-, scribble-wise) supervision and has shown promising performance, particularly in the image segmentation field. However, it is still a very challenging task due to the limited supervision, especially when only a small number of labeled samples are available. Additionally, almost all existing WSL segmentation methods are designed for star-convex structures which are very different from curvilinear structures such as vessels and nerves. In this paper, we propose a novel sparsely annotated segmentation framework for curvilinear structures, named YoloCurvSeg. A very essential component of YoloCurvSeg is image synthesis. Specifically, a background generator delivers image backgrounds that closely match the real distributions through inpainting dilated skeletons. The extracted backgrounds are then combined with randomly emulated curves generated by a Space Colonization Algorithm-based foreground generator and through a multilayer patch-wise contrastive learning synthesizer. In this way, a synthetic dataset with both images and curve segmentation labels is obtained, at the cost of only one or a few noisy skeleton annotations. Finally, a segmenter is trained with the generated dataset and possibly an unlabeled dataset. The proposed YoloCurvSeg is evaluated on four publicly available datasets (OCTA500, CORN, DRIVE and CHASEDB1) and the results show that YoloCurvSeg outperforms state-of-the-art WSL segmentation methods by large margins. With only one noisy skeleton annotation (respectively 0.14\%, 0.03\%, 1.40\%, and 0.65\% of the full annotation), YoloCurvSeg achieves more than 97\% of the fully-supervised performance on each dataset. Code and datasets will be released at https://github.com/llmir/YoloCurvSeg.

Yijin Huang, Li Lin, Pujin Cheng et al.

Although deep learning based diabetic retinopathy (DR) classification methods typically benefit from well-designed architectures of convolutional neural networks, the training setting also has a non-negligible impact on the prediction performance. The training setting includes various interdependent components, such as objective function, data sampling strategy and data augmentation approach. To identify the key components in a standard deep learning framework (ResNet-50) for DR grading, we systematically analyze the impact of several major components. Extensive experiments are conducted on a publicly-available dataset EyePACS. We demonstrate that (1) the DR grading framework is sensitive to input resolution, objective function, and composition of data augmentation, (2) using mean square error as the loss function can effectively improve the performance with respect to a task-specific evaluation metric, namely the quadratically-weighted Kappa, (3) utilizing eye pairs boosts the performance of DR grading and (4) using data resampling to address the problem of imbalanced data distribution in EyePACS hurts the performance. Based on these observations and an optimal combination of the investigated components, our framework, without any specialized network design, achieves the state-of-the-art result (0.8631 for Kappa) on the EyePACS test set (a total of 42670 fundus images) with only image-level labels. We also examine the proposed training practices on other fundus datasets and other network architectures to evaluate their generalizability. Our codes and pre-trained model are available at https://github.com/YijinHuang/pytorch-classification.

Shiyun Chen, Li Lin, Pujin Cheng et al.

Liver tumor segmentation is essential for computer-aided diagnosis, surgical planning, and prognosis evaluation. However, obtaining and maintaining a large-scale dataset with dense annotations is challenging. Semi-Supervised Learning (SSL) is a common technique to address these challenges. Recently, Segment Anything Model (SAM) has shown promising performance in some medical image segmentation tasks, but it performs poorly for liver tumor segmentation. In this paper, we propose a novel semi-supervised framework, named ASLseg, which can effectively adapt the SAM to the SSL setting and combine both domain-specific and general knowledge of liver tumors. Specifically, the segmentation model trained with a specific SSL paradigm provides the generated pseudo-labels as prompts to the fine-tuned SAM. An adaptation network is then used to refine the SAM-predictions and generate higher-quality pseudo-labels. Finally, the reliable pseudo-labels are selected to expand the labeled set for iterative training. Extensive experiments on the LiTS dataset demonstrate overwhelming performance of our ASLseg.

Yijin Huang, Pujin Cheng, Roger Tam et al.

Parameter-efficient transfer learning (PETL) is proposed as a cost-effective way to transfer pre-trained models to downstream tasks, avoiding the high cost of updating entire large-scale pre-trained models (LPMs). In this work, we present Fine-grained Prompt Tuning (FPT), a novel PETL method for medical image classification. FPT significantly reduces memory consumption compared to other PETL methods, especially in high-resolution input contexts. To achieve this, we first freeze the weights of the LPM and construct a learnable lightweight side network. The frozen LPM takes high-resolution images as input to extract fine-grained features, while the side network is fed low-resolution images to reduce memory usage. To allow the side network to access pre-trained knowledge, we introduce fine-grained prompts that summarize information from the LPM through a fusion module. Important tokens selection and preloading techniques are employed to further reduce training cost and memory requirements. We evaluate FPT on four medical datasets with varying sizes, modalities, and complexities. Experimental results demonstrate that FPT achieves comparable performance to fine-tuning the entire LPM while using only 1.8% of the learnable parameters and 13% of the memory costs of an encoder ViT-B model with a 512 x 512 input resolution.

Shiyun Chen, Li Lin, Pujin Cheng et al.

Multimodal learning has been demonstrated to enhance performance across various clinical tasks, owing to the diverse perspectives offered by different modalities of data. However, existing multimodal segmentation methods rely on well-registered multimodal data, which is unrealistic for real-world clinical images, particularly for indistinct and diffuse regions such as liver tumors. In this paper, we introduce Diff4MMLiTS, a four-stage multimodal liver tumor segmentation pipeline: pre-registration of the target organs in multimodal CTs; dilation of the annotated modality's mask and followed by its use in inpainting to obtain multimodal normal CTs without tumors; synthesis of strictly aligned multimodal CTs with tumors using the latent diffusion model based on multimodal CT features and randomly generated tumor masks; and finally, training the segmentation model, thus eliminating the need for strictly aligned multimodal data. Extensive experiments on public and internal datasets demonstrate the superiority of Diff4MMLiTS over other state-of-the-art multimodal segmentation methods.

Liang An, Jin Lyu, Li Lin et al.

In the era of foundation models, achieving a unified understanding of different dynamic objects through a single network has the potential to empower stronger spatial intelligence. Moreover, accurate estimation of animal pose and shape across diverse species is essential for quantitative analysis in biological research. However, this topic remains underexplored due to the limited network capacity of previous methods and the scarcity of comprehensive multi-species datasets. To address these limitations, we introduce AniMer+, an extended version of our scalable AniMer framework. In this paper, we focus on a unified approach for reconstructing mammals (mammalia) and birds (aves). A key innovation of AniMer+ is its high-capacity, family-aware Vision Transformer (ViT) incorporating a Mixture-of-Experts (MoE) design. Its architecture partitions network layers into taxa-specific components (for mammalia and aves) and taxa-shared components, enabling efficient learning of both distinct and common anatomical features within a single model. To overcome the critical shortage of 3D training data, especially for birds, we introduce a diffusion-based conditional image generation pipeline. This pipeline produces two large-scale synthetic datasets: CtrlAni3D for quadrupeds and CtrlAVES3D for birds. To note, CtrlAVES3D is the first large-scale, 3D-annotated dataset for birds, which is crucial for resolving single-view depth ambiguities. Trained on an aggregated collection of 41.3k mammalian and 12.4k avian images (combining real and synthetic data), our method demonstrates superior performance over existing approaches across a wide range of benchmarks, including the challenging out-of-domain Animal Kingdom dataset. Ablation studies confirm the effectiveness of both our novel network architecture and the generated synthetic datasets in enhancing real-world application performance.

Tianyunxi Wei, Yijin Huang, Li Lin et al.

Medical image datasets often exhibit long-tailed distributions due to the inherent challenges in medical data collection and annotation. In long-tailed contexts, some common disease categories account for most of the data, while only a few samples are available in the rare disease categories, resulting in poor performance of deep learning methods. To address this issue, previous approaches have employed class re-sampling or re-weighting techniques, which often encounter challenges such as overfitting to tail classes or difficulties in optimization during training. In this work, we propose a novel approach, namely \textbf{S}aliency-guided and \textbf{P}atch-based \textbf{Mix}up (SPMix) for long-tailed skin cancer image classification. Specifically, given a tail-class image and a head-class image, we generate a new tail-class image by mixing them under the guidance of saliency mapping, which allows for preserving and augmenting the discriminative features of the tail classes without any interference of the head-class features. Extensive experiments are conducted on the ISIC2018 dataset, demonstrating the superiority of SPMix over existing state-of-the-art methods.

Linkai Peng, Li Lin, Pujin Cheng et al.

Various deep learning models have been developed to segment anatomical structures from medical images, but they typically have poor performance when tested on another target domain with different data distribution. Recently, unsupervised domain adaptation methods have been proposed to alleviate this so-called domain shift issue, but most of them are designed for scenarios with relatively small domain shifts and are likely to fail when encountering a large domain gap. In this paper, we propose DCDA, a novel cross-modality unsupervised domain adaptation framework for tasks with large domain shifts, e.g., segmenting retinal vessels from OCTA and OCT images. DCDA mainly consists of a disentangling representation style transfer (DRST) module and a collaborative consistency learning (CCL) module. DRST decomposes images into content components and style codes and performs style transfer and image reconstruction. CCL contains two segmentation models, one for source domain and the other for target domain. The two models use labeled data (together with the corresponding transferred images) for supervised learning and perform collaborative consistency learning on unlabeled data. Each model focuses on the corresponding single domain and aims to yield an expertized domain-specific segmentation model. Through extensive experiments on retinal vessel segmentation, our framework achieves Dice scores close to target-trained oracle both from OCTA to OCT and from OCT to OCTA, significantly outperforming other state-of-the-art methods.

Huilin Yang, Junyan Lyu, Pujin Cheng et al.

We innovatively propose a flexible and consistent face alignment framework, LDDMM-Face, the key contribution of which is a deformation layer that naturally embeds facial geometry in a diffeomorphic way. Instead of predicting facial landmarks via heatmap or coordinate regression, we formulate this task in a diffeomorphic registration manner and predict momenta that uniquely parameterize the deformation between initial boundary and true boundary, and then perform large deformation diffeomorphic metric mapping (LDDMM) simultaneously for curve and landmark to localize the facial landmarks. Due to the embedding of LDDMM into a deep network, LDDMM-Face can consistently annotate facial landmarks without ambiguity and flexibly handle various annotation schemes, and can even predict dense annotations from sparse ones. Our method can be easily integrated into various face alignment networks. We extensively evaluate LDDMM-Face on four benchmark datasets: 300W, WFLW, HELEN and COFW-68. LDDMM-Face is comparable or superior to state-of-the-art methods for traditional within-dataset and same-annotation settings, but truly distinguishes itself with outstanding performance when dealing with weakly-supervised learning (partial-to-full), challenging cases (e.g., occluded faces), and different training and prediction datasets. In addition, LDDMM-Face shows promising results on the most challenging task of predicting across datasets with different annotation schemes.

Yijin Huang, Li Lin, Pujin Cheng et al.

Manually annotating medical images is extremely expensive, especially for large-scale datasets. Self-supervised contrastive learning has been explored to learn feature representations from unlabeled images. However, unlike natural images, the application of contrastive learning to medical images is relatively limited. In this work, we propose a self-supervised framework, namely lesion-based contrastive learning for automated diabetic retinopathy (DR) grading. Instead of taking entire images as the input in the common contrastive learning scheme, lesion patches are employed to encourage the feature extractor to learn representations that are highly discriminative for DR grading. We also investigate different data augmentation operations in defining our contrastive prediction task. Extensive experiments are conducted on the publicly-accessible dataset EyePACS, demonstrating that our proposed framework performs outstandingly on DR grading in terms of both linear evaluation and transfer capacity evaluation.

Li Lin, Zhonghua Wang, Jiewei Wu et al.

Optical coherence tomography angiography (OCTA) is a novel non-invasive imaging technique that allows visualizations of vasculature and foveal avascular zone (FAZ) across retinal layers. Clinical researches suggest that the morphology and contour irregularity of FAZ are important biomarkers of various ocular pathologies. Therefore, precise segmentation of FAZ has great clinical interest. Also, there is no existing research reporting that FAZ features can improve the performance of deep diagnostic classification networks. In this paper, we propose a novel multi-level boundary shape and distance aware joint learning framework, named BSDA-Net, for FAZ segmentation and diagnostic classification from OCTA images. Two auxiliary branches, namely boundary heatmap regression and signed distance map reconstruction branches, are constructed in addition to the segmentation branch to improve the segmentation performance, resulting in more accurate FAZ contours and fewer outliers. Moreover, both low-level and high-level features from the aforementioned three branches, including shape, size, boundary, and signed directional distance map of FAZ, are fused hierarchically with features from the diagnostic classifier. Through extensive experiments, the proposed BSDA-Net is found to yield state-of-the-art segmentation and classification results on the OCTA-500, OCTAGON, and FAZID datasets.