Sifan Song

Jinfeng Wang, Qiming Huang, Feilong Tang et al.

Colonoscopy, currently the most efficient and recognized colon polyp detection technology, is necessary for early screening and prevention of colorectal cancer. However, due to the varying size and complex morphological features of colonic polyps as well as the indistinct boundary between polyps and mucosa, accurate segmentation of polyps is still challenging. Deep learning has become popular for accurate polyp segmentation tasks with excellent results. However, due to the structure of polyps image and the varying shapes of polyps, it easy for existing deep learning models to overfitting the current dataset. As a result, the model may not process unseen colonoscopy data. To address this, we propose a new State-Of-The-Art model for medical image segmentation, the SSFormer, which uses a pyramid Transformer encoder to improve the generalization ability of models. Specifically, our proposed Progressive Locality Decoder can be adapted to the pyramid Transformer backbone to emphasize local features and restrict attention dispersion. The SSFormer achieves statet-of-the-art performance in both learning and generalization assessment.

Sifan Song, Jinfeng Wang, Fengrui Cheng et al.

Chromosomes carry the genetic information of humans. They exhibit non-rigid and non-articulated nature with varying degrees of curvature. Chromosome straightening is an important step for subsequent karyotype construction, pathological diagnosis and cytogenetic map development. However, robust chromosome straightening remains challenging, due to the unavailability of training images, distorted chromosome details and shapes after straightening, as well as poor generalization capability. In this paper, we propose a novel architecture, ViT-Patch GAN, consisting of a self-learned motion transformation generator and a Vision Transformer-based patch (ViT-Patch) discriminator. The generator learns the motion representation of chromosomes for straightening. With the help of the ViT-Patch discriminator, the straightened chromosomes retain more shape and banding pattern details. The experimental results show that the proposed method achieves better performance on Fréchet Inception Distance (FID), Learned Perceptual Image Patch Similarity (LPIPS) and downstream chromosome classification accuracy, and shows excellent generalization capability on a large dataset.

Jinfeng Wang, Sifan Song, Jionglong Su et al.

Self-supervised learning is well known for its remarkable performance in representation learning and various downstream computer vision tasks. Recently, Positive-pair-Only Contrastive Learning (POCL) has achieved reliable performance without the need to construct positive-negative training sets. It reduces memory requirements by lessening the dependency on the batch size. The POCL method typically uses a single loss function to extract the distortion invariant representation (DIR) which describes the proximity of positive-pair representations affected by different distortions. This loss function implicitly enables the model to filter out or ignore the distortion variant representation (DVR) affected by different distortions. However, existing POCL methods do not explicitly enforce the disentanglement and exploitation of the actually valuable DVR. In addition, these POCL methods have been observed to be sensitive to augmentation strategies. To address these limitations, we propose a novel POCL framework named Distortion-Disentangled Contrastive Learning (DDCL) and a Distortion-Disentangled Loss (DDL). Our approach is the first to explicitly disentangle and exploit the DVR inside the model and feature stream to improve the overall representation utilization efficiency, robustness and representation ability. Experiments carried out demonstrate the superiority of our framework to Barlow Twins and Simsiam in terms of convergence, representation quality, and robustness on several benchmark datasets.

Sifan Song, Jinfeng Wang, Zilong Wang et al.

Accurate fovea localization is essential for analyzing retinal diseases to prevent irreversible vision loss. While current deep learning-based methods outperform traditional ones, they still face challenges such as the lack of local anatomical landmarks around the fovea, the inability to robustly handle diseased retinal images, and the variations in image conditions. In this paper, we propose a novel transformer-based architecture called DualStreamFoveaNet (DSFN) for multi-cue fusion. This architecture explicitly incorporates long-range connections and global features using retina and vessel distributions for robust fovea localization. We introduce a spatial attention mechanism in the dual-stream encoder to extract and fuse self-learned anatomical information, focusing more on features distributed along blood vessels and significantly reducing computational costs by decreasing token numbers. Our extensive experiments show that the proposed architecture achieves state-of-the-art performance on two public datasets and one large-scale private dataset. Furthermore, we demonstrate that the DSFN is more robust on both normal and diseased retina images and has better generalization capacity in cross-dataset experiments.

Zhixiang Lu, Mengqi Han, Peixin Guo et al.

Constructing high-resolution 3D geological models from sparse 1D borehole and 2D surface data is a highly ill-posed inverse problem. Traditional heuristic and implicit modeling methods fundamentally fail to capture non-linear topological discontinuities under extreme sparsity, often yielding unrealistic artifacts. Furthermore, while deep generative architectures like Diffusion Models have revolutionized continuous domains, they suffer from severe representation collapse when conditioned on sparse categorical grids. To bridge this gap, we propose 3D-GeoFlow, the first Attention-Guided Continuous Flow Matching framework tailored for sparse multimodal geological modeling. By reformulating discrete categorical generation as a simulation-free, continuous vector field regression optimized via Mean Squared Error, our model establishes stable, deterministic optimal transport paths. Crucially, we integrate 3D Attention Gates to dynamically propagate localized borehole features across the volumetric latent space, ensuring macroscopic structural coherence. To validate our framework, we curated a large-scale multimodal dataset comprising 2,200 procedurally generated 3D geological cases. Extensive out-of-distribution (OOD) evaluations demonstrate that 3D-GeoFlow achieves a paradigm shift, significantly outperforming heuristic interpolations and standard diffusion baselines.

Zhiling Yan, Sifan Song, Dingjie Song et al.

Recent "segment anything" efforts show promise by learning from large-scale data, but adapting such models directly to medical images remains challenging due to the complexity of medical data, noisy annotations, and continual learning requirements across diverse modalities and anatomical structures. In this work, we propose SAMed-2, a new foundation model for medical image segmentation built upon the SAM-2 architecture. Specifically, we introduce a temporal adapter into the image encoder to capture image correlations and a confidence-driven memory mechanism to store high-certainty features for later retrieval. This memory-based strategy counters the pervasive noise in large-scale medical datasets and mitigates catastrophic forgetting when encountering new tasks or modalities. To train and evaluate SAMed-2, we curate MedBank-100k, a comprehensive dataset spanning seven imaging modalities and 21 medical segmentation tasks. Our experiments on both internal benchmarks and 10 external datasets demonstrate superior performance over state-of-the-art baselines in multi-task scenarios. The code is available at: https://github.com/ZhilingYan/Medical-SAM-Bench.

Jinfeng Wang, Sifan Song, Xinkun Wang et al.

With the proposal of the Segment Anything Model (SAM), fine-tuning SAM for medical image segmentation (MIS) has become popular. However, due to the large size of the SAM model and the significant domain gap between natural and medical images, fine-tuning-based strategies are costly with potential risk of instability, feature damage and catastrophic forgetting. Furthermore, some methods of transferring SAM to a domain-specific MIS through fine-tuning strategies disable the model's prompting capability, severely limiting its utilization scenarios. In this paper, we propose an Auto-Prompting Module (APM), which provides SAM-based foundation model with Euclidean adaptive prompts in the target domain. Our experiments demonstrate that such adaptive prompts significantly improve SAM's non-fine-tuned performance in MIS. In addition, we propose a novel non-invasive method called Incremental Pattern Shifting (IPS) to adapt SAM to specific medical domains. Experimental results show that the IPS enables SAM to achieve state-of-the-art or competitive performance in MIS without the need for fine-tuning. By coupling these two methods, we propose ProMISe, an end-to-end non-fine-tuned framework for Promptable Medical Image Segmentation. Our experiments demonstrate that both using our methods individually or in combination achieves satisfactory performance in low-cost pattern shifting, with all of SAM's parameters frozen.

Sekeun Kim, Pengfei Jin, Sifan Song et al.

Foundation models have recently gained significant attention because of their generalizability and adaptability across multiple tasks and data distributions. Although medical foundation models have emerged, solutions for cardiac imaging, especially echocardiography videos, are still unexplored. In this paper, we introduce EchoFM, a foundation model specifically designed to represent and analyze echocardiography videos. In EchoFM, we propose a self-supervised learning framework that captures both spatial and temporal variability patterns through a spatio-temporal consistent masking strategy and periodic-driven contrastive learning. This framework can effectively capture the spatio-temporal dynamics of echocardiography and learn the representative video features without any labels. We pre-train our model on an extensive dataset comprising over 290,000 echocardiography videos covering 26 scan views across different imaging modes, with up to 20 million frames of images. The pre-trained EchoFM can then be easily adapted and fine-tuned for a variety of downstream tasks, serving as a robust backbone model. Our evaluation was systemically designed for four downstream tasks after the echocardiography examination routine. Experiment results show that EchoFM surpasses state-of-the-art methods, including specialized echocardiography methods, self-supervised pre-training models, and general-purposed pre-trained foundation models, across all downstream tasks.

Yuang Wang, Siyeop Yoon, Pengfei Jin et al.

Diffusion-based models have demonstrated remarkable effectiveness in image restoration tasks; however, their iterative denoising process, which starts from Gaussian noise, often leads to slow inference speeds. The Image-to-Image Schrödinger Bridge (I$^2$SB) offers a promising alternative by initializing the generative process from corrupted images while leveraging training techniques from score-based diffusion models. In this paper, we introduce the Implicit Image-to-Image Schrödinger Bridge (I$^3$SB) to further accelerate the generative process of I$^2$SB. I$^3$SB restructures the generative process into a non-Markovian framework by incorporating the initial corrupted image at each generative step, effectively preserving and utilizing its information. To enable direct use of pretrained I$^2$SB models without additional training, we ensure consistency in marginal distributions. Extensive experiments across many image corruptions, including noise, low resolution, JPEG compression, and sparse sampling, and multiple image modalities, such as natural, human face, and medical images, demonstrate the acceleration benefits of I$^3$SB. Compared to I$^2$SB, I$^3$SB achieves the same perceptual quality with fewer generative steps, while maintaining or improving fidelity to the ground truth.

Jinfeng Wang, Yanhao Huang, Sifan Song et al.

EEG emotion recognition faces significant hurdles due to noise interference, signal nonstationarity, and the inherent complexity of brain activity which make accurately emotion classification. In this study, we present the Fourier Adjacency Transformer, a novel framework that seamlessly integrates Fourier-based periodic analysis with graph-driven structural modeling. Our method first leverages novel Fourier-inspired modules to extract periodic features from embedded EEG signals, effectively decoupling them from aperiodic components. Subsequently, we employ an adjacency attention scheme to reinforce universal inter-channel correlation patterns, coupling these patterns with their sample-based counterparts. Empirical evaluations on SEED and DEAP datasets demonstrate that our method surpasses existing state-of-the-art techniques, achieving an improvement of approximately 6.5% in recognition accuracy. By unifying periodicity and structural insights, this framework offers a promising direction for future research in EEG emotion analysis.

Yujin Oh, Robert Seifert, Yihan Cao et al.

In oncology, Positron Emission Tomography-Computed Tomography (PET/CT) is widely used in cancer diagnosis, staging, and treatment monitoring, as it combines anatomical details from CT with functional metabolic activity and molecular marker expression information from PET. However, existing artificial intelligence-driven PET/CT analyses rely predominantly on task-specific models trained from scratch or on limited datasets, limiting their generalizability and robustness. To address this, we propose a foundation model approach specifically designed for multimodal PET/CT imaging. We introduce the Cross-Fraternal Twin Masked Autoencoder (FratMAE), a novel framework that effectively integrates whole-body anatomical and functional or molecular information. FratMAE employs separate Vision Transformer (ViT) encoders for PET and CT scans, along with cross-attention decoders that enable synergistic interactions between modalities during masked autoencoder training. Additionally, it incorporates textual metadata to enhance PET representation learning. By pre-training on PET/CT datasets, FratMAE captures intricate cross-modal relationships and global uptake patterns, achieving superior performance on downstream tasks and demonstrating its potential as a generalizable foundation model.

Pengfei Jin, Peng Shu, Sifan Song et al.

Recent advances in parameter-efficient transfer learning have demonstrated the utility of composing LoRA adapters from libraries of pretrained modules. However, most existing approaches rely on simple retrieval heuristics or uniform averaging, which overlook the latent structure of task relationships in representation space. We propose a new framework for adapter reuse that moves beyond retrieval, formulating adapter composition as a geometry-aware sparse reconstruction problem. Specifically, we represent each task by a latent prototype vector derived from the base model's encoder and aim to approximate the target task prototype as a sparse linear combination of retrieved reference prototypes, under an $\ell_1$-regularized optimization objective. The resulting combination weights are then used to blend the corresponding LoRA adapters, yielding a composite adapter tailored to the target task. This formulation not only preserves the local geometric structure of the task representation manifold, but also promotes interpretability and efficient reuse by selecting a minimal set of relevant adapters. We demonstrate the effectiveness of our approach across multiple domains-including medical image segmentation, medical report generation and image synthesis. Our results highlight the benefit of coupling retrieval with latent geometry-aware optimization for improved zero-shot generalization.

Siyeop Yoon, Yujin Oh, Pengfei Jin et al.

We present Surf2CT, a novel cascaded flow matching framework that synthesizes full 3D computed tomography (CT) volumes of the human torso from external surface scans and simple demographic data (age, sex, height, weight). This is the first approach capable of generating realistic volumetric internal anatomy images solely based on external body shape and demographics, without any internal imaging. Surf2CT proceeds through three sequential stages: (1) Surface Completion, reconstructing a complete signed distance function (SDF) from partial torso scans using conditional 3D flow matching; (2) Coarse CT Synthesis, generating a low-resolution CT volume from the completed SDF and demographic information; and (3) CT Super-Resolution, refining the coarse volume into a high-resolution CT via a patch-wise conditional flow model. Each stage utilizes a 3D-adapted EDM2 backbone trained via flow matching. We trained our model on a combined dataset of 3,198 torso CT scans (approximately 1.13 million axial slices) sourced from Massachusetts General Hospital (MGH) and the AutoPET challenge. Evaluation on 700 paired torso surface-CT cases demonstrated strong anatomical fidelity: organ volumes exhibited small mean percentage differences (range from -11.1% to 4.4%), and muscle/fat body composition metrics matched ground truth with strong correlation (range from 0.67 to 0.96). Lung localization had minimal bias (mean difference -2.5 mm), and surface completion significantly improved metrics (Chamfer distance: from 521.8 mm to 2.7 mm; Intersection-over-Union: from 0.87 to 0.98). Surf2CT establishes a new paradigm for non-invasive internal anatomical imaging using only external data, opening opportunities for home-based healthcare, preventive medicine, and personalized clinical assessments without the risks associated with conventional imaging techniques.

Siyeop Yoon, Sifan Song, Pengfei Jin et al.

We propose a cascaded 3D diffusion model framework to synthesize high-fidelity 3D PET/CT volumes directly from demographic variables, addressing the growing need for realistic digital twins in oncologic imaging, virtual trials, and AI-driven data augmentation. Unlike deterministic phantoms, which rely on predefined anatomical and metabolic templates, our method employs a two-stage generative process. An initial score-based diffusion model synthesizes low-resolution PET/CT volumes from demographic variables alone, providing global anatomical structures and approximate metabolic activity. This is followed by a super-resolution residual diffusion model that refines spatial resolution. Our framework was trained on 18-F FDG PET/CT scans from the AutoPET dataset and evaluated using organ-wise volume and standardized uptake value (SUV) distributions, comparing synthetic and real data between demographic subgroups. The organ-wise comparison demonstrated strong concordance between synthetic and real images. In particular, most deviations in metabolic uptake values remained within 3-5% of the ground truth in subgroup analysis. These findings highlight the potential of cascaded 3D diffusion models to generate anatomically and metabolically accurate PET/CT images, offering a robust alternative to traditional phantoms and enabling scalable, population-informed synthetic imaging for clinical and research applications.

Sifan Song, Siyeop Yoon, Pengfei Jin et al.

Recent advances in representation learning often rely on holistic embeddings that entangle multiple semantic components, limiting interpretability and generalization. These issues are especially critical in medical imaging, where downstream tasks depend on anatomically interpretable features. To address these limitations, we propose an Organ-Wise Tokenization (OWT) framework with a Token Group-based Reconstruction (TGR) training paradigm. Unlike conventional approaches, OWT explicitly disentangles an image into separable token groups, each corresponding to a distinct organ or semantic entity. Our design ensures each token group encapsulates organ-specific information, boosting interpretability, generalization, and efficiency while enabling fine-grained control for targeted clinical applications. Experiments on CT and MRI datasets demonstrate OWT's power: it not only achieves strong performance on standard tasks like image reconstruction and segmentation, but also unlocks novel, high-impact clinical capabilities including organ-specific tumor identification, organ-level retrieval and semantic-level generation, without requiring any additional training. These findings underscore the potential of OWT as a foundational framework for semantically disentangled representation learning, offering broad scalability and a new perspective on how representations can be leveraged.

Runqi Meng, Sifan Song, Pengfei Jin et al.

Accurate tumor segmentation is crucial for cancer diagnosis and treatment. While foundation models have advanced general-purpose segmentation, existing methods still struggle with: (1) limited incorporation of medical priors, (2) imbalance between generic and tumor-specific features, and (3) high computational costs for clinical adaptation. To address these challenges, we propose MAST-Pro (Mixture-of-experts for Adaptive Segmentation of pan-Tumors with knowledge-driven Prompts), a novel framework that integrates dynamic Mixture-of-Experts (D-MoE) and knowledge-driven prompts for pan-tumor segmentation. Specifically, text and anatomical prompts provide domain-specific priors, guiding tumor representation learning, while D-MoE dynamically selects experts to balance generic and tumor-specific feature learning, improving segmentation accuracy across diverse tumor types. To enhance efficiency, we employ Parameter-Efficient Fine-Tuning (PEFT), optimizing MAST-Pro with significantly reduced computational overhead. Experiments on multi-anatomical tumor datasets demonstrate that MAST-Pro outperforms state-of-the-art approaches, achieving up to a 5.20% improvement in average DSC while reducing trainable parameters by 91.04%, without compromising accuracy.

Siyeop Yoon, Yujin Oh, Matthew Tivnan et al.

This study presents a 3D flow-matching model designed to predict the progression of the frozen region (iceball) during kidney cryoablation. Precise intraoperative guidance is critical in cryoablation to ensure complete tumor eradication while preserving adjacent healthy tissue. However, conventional methods, typically based on physics driven or diffusion based simulations, are computationally demanding and often struggle to represent complex anatomical structures accurately. To address these limitations, our approach leverages intraoperative CT imaging to inform the model. The proposed 3D flow matching model is trained to learn a continuous deformation field that maps early-stage CT scans to future predictions. This transformation not only estimates the volumetric expansion of the iceball but also generates corresponding segmentation masks, effectively capturing spatial and morphological changes over time. Quantitative analysis highlights the model robustness, demonstrating strong agreement between predictions and ground-truth segmentations. The model achieves an Intersection over Union (IoU) score of 0.61 and a Dice coefficient of 0.75. By integrating real time CT imaging with advanced deep learning techniques, this approach has the potential to enhance intraoperative guidance in kidney cryoablation, improving procedural outcomes and advancing the field of minimally invasive surgery.

Sifan Song, Jinfeng Wang, Qiaochu Zhao et al.

Invariant Contrastive Learning (ICL) methods have achieved impressive performance across various domains. However, the absence of latent space representation for distortion (augmentation)-related information in the latent space makes ICL sub-optimal regarding training efficiency and robustness in downstream tasks. Recent studies suggest that introducing equivariance into Contrastive Learning (CL) can improve overall performance. In this paper, we revisit the roles of augmentation strategies and equivariance in improving CL's efficacy. We propose CLeVER (Contrastive Learning Via Equivariant Representation), a novel equivariant contrastive learning framework compatible with augmentation strategies of arbitrary complexity for various mainstream CL backbone models. Experimental results demonstrate that CLeVER effectively extracts and incorporates equivariant information from practical natural images, thereby improving the training efficiency and robustness of baseline models in downstream tasks and achieving state-of-the-art (SOTA) performance. Moreover, we find that leveraging equivariant information extracted by CLeVER simultaneously enhances rotational invariance and sensitivity across experimental tasks, and helps stabilize the framework when handling complex augmentations, particularly for models with small-scale backbones.

Junde Wu, Huihui Fang, Fei Li et al.

Color fundus photography and Optical Coherence Tomography (OCT) are the two most cost-effective tools for glaucoma screening. Both two modalities of images have prominent biomarkers to indicate glaucoma suspected. Clinically, it is often recommended to take both of the screenings for a more accurate and reliable diagnosis. However, although numerous algorithms are proposed based on fundus images or OCT volumes in computer-aided diagnosis, there are still few methods leveraging both of the modalities for the glaucoma assessment. Inspired by the success of Retinal Fundus Glaucoma Challenge (REFUGE) we held previously, we set up the Glaucoma grAding from Multi-Modality imAges (GAMMA) Challenge to encourage the development of fundus \& OCT-based glaucoma grading. The primary task of the challenge is to grade glaucoma from both the 2D fundus images and 3D OCT scanning volumes. As part of GAMMA, we have publicly released a glaucoma annotated dataset with both 2D fundus color photography and 3D OCT volumes, which is the first multi-modality dataset for glaucoma grading. In addition, an evaluation framework is also established to evaluate the performance of the submitted methods. During the challenge, 1272 results were submitted, and finally, top-10 teams were selected to the final stage. We analysis their results and summarize their methods in the paper. Since all these teams submitted their source code in the challenge, a detailed ablation study is also conducted to verify the effectiveness of the particular modules proposed. We find many of the proposed techniques are practical for the clinical diagnosis of glaucoma. As the first in-depth study of fundus \& OCT multi-modality glaucoma grading, we believe the GAMMA Challenge will be an essential starting point for future research.

Sifan Song, Kang Dang, Qinji Yu et al.

The fovea is an important anatomical landmark of the retina. Detecting the location of the fovea is essential for the analysis of many retinal diseases. However, robust fovea localization remains a challenging problem, as the fovea region often appears fuzzy, and retina diseases may further obscure its appearance. This paper proposes a novel Vision Transformer (ViT) approach that integrates information both inside and outside the fovea region to achieve robust fovea localization. Our proposed network, named Bilateral-Vision-Transformer (Bilateral-ViT), consists of two network branches: a transformer-based main network branch for integrating global context across the entire fundus image and a vessel branch for explicitly incorporating the structure of blood vessels. The encoded features from both network branches are subsequently merged with a customized Multi-scale Feature Fusion (MFF) module. Our comprehensive experiments demonstrate that the proposed approach is significantly more robust for diseased images and establishes the new state of the arts using the Messidor and PALM datasets.

Sifan Song, Daiyun Huang, Yalun Hu et al.

In medical imaging, chromosome straightening plays a significant role in the pathological study of chromosomes and in the development of cytogenetic maps. Whereas different approaches exist for the straightening task, typically geometric algorithms are used whose outputs are characterized by jagged edges or fragments with discontinued banding patterns. To address the flaws in the geometric algorithms, we propose a novel framework based on image-to-image translation to learn a pertinent mapping dependence for synthesizing straightened chromosomes with uninterrupted banding patterns and preserved details. In addition, to avoid the pitfall of deficient input chromosomes, we construct an augmented dataset using only one single curved chromosome image for training models. Based on this framework, we apply two popular image-to-image translation architectures, U-shape networks and conditional generative adversarial networks, to assess its efficacy. Experiments on a dataset comprised of 642 real-world chromosomes demonstrate the superiority of our framework, as compared to the geometric method in straightening performance, by rendering realistic and continued chromosome details. Furthermore, our straightened results improve the chromosome classification by 0.98%-1.39% mean accuracy.