Medical Image Representation Learning

Large-scale, volumetric medical imaging datasets typically aggregate scans from different vendors and devices, resulting in highly variable resolution, slice thicknesses, and numbers of slices per study. Consequently, training representation models usually requires cropping or interpolating along the z-axis to obtain fixed-size blocks, which inevitably causes information loss. We propose a new training approach to overcome this limitation. Instead of absolute position embeddings, we interpret volumes as sequences of 3D chunks and adopt Rotary Position Embeddings, allowing us to treat the z-axis as an unconstrained temporal dimensions. Building on this idea, we introduce a new vision-language model: SigVLP. In SigVLP, we implement Rotary Position Embedding as the positional encoding method, which is applied directly within the attention operation, generating input-conditioned sine and cosine weights on the fly. This design ensures consistent alignment between query and key projections and adapts to any input sizes. To allow for variable input size during training, we sample Computed Tomography volumes in chunks and pair them with localized organ-wise textual observations. Compared to using entire reports for conditioning, chunkwise alignment provides finer-grained supervision, enabling the model to establish stronger correlations between the text and volume representations, thereby improving the precision of text-to-volume alignment. Our models are trained with the Muon optimizer and evaluated on a diverse set of downstream tasks, including zero-shot abnormality and organ classification, segmentation, and retrieval tasks.

Abridged: Clinicians commonly interpret 3D medical images by examining multiple anatomical planes rather than relying on volumetric views. In clinical CT workflows, the axial plane often serves as the primary diagnostic reference, while the auxiliary planes provide complementary spatial context. However, many existing 3D deep learning approaches either process volumetric data holistically or assign equal importance to all planes, failing to reflect this asymmetric, axial-centric interpretation strategy. To address this, we propose an axial-centric cross-plane attention architecture for 3D medical image classification that models asymmetric dependencies between anatomical planes. The architecture employs large-scale axial CT images pretrained MedDINOv3 as a frozen feature extractor for axial, coronal, and sagittal planes. RICA blocks and intra-plane transformer encoders capture plane-specific positional and contextual information, while axial-centric cross-plane transformer encoders selectively condition axial representations on complementary auxiliary representations. Experiments on six datasets from the MedMNIST3D benchmark show that the proposed method consistently outperforms existing 3D and multi-plane models in ACC and AUC. A lightweight variant, AC-Tiny, achieves competitive performance with substantially fewer trainable parameters, suggesting that architectural design contributes more to performance gains than increased model scale. Ablation studies further validate the importance of axial-centric querying, QKV allocation, directional cross-plane fusion, residual-free cross-attention, and classification head design. Slice-level Grad-CAM visualizations demonstrate that the model identifies diagnostically relevant regions across all planes. These findings highlight the value of aligning architectural design with clinical interpretation workflows for robust 3D medical image analysis.

Self-supervised vision models have achieved notable success in digital pathology. However, their domain-agnostic transformer architectures are not originally designed to account for fundamental biological elements of histopathology images, namely cells and their complex interactions. In this work, we hypothesize that a biologically-informed modeling of tissues as cell graphs offers a more efficient representation learning. Thus, we introduce GrapHist, a novel graph-based self-supervised learning framework for histopathology, which learns generalizable and structurally-informed embeddings that enable diverse downstream tasks. GrapHist integrates masked autoencoders and heterophilic graph neural networks that are explicitly designed to capture the heterogeneity of tumor microenvironments. We pre-train GrapHist on a large collection of 11 million cell graphs derived from breast tissues and evaluate its transferability across in- and out-of-domain benchmarks. Our results show that GrapHist achieves competitive performance compared to its vision-based counterparts in slide-, region-, and cell-level tasks, while requiring four times fewer parameters. It also drastically outperforms fully-supervised graph models on cancer subtyping tasks. Finally, we also release five graph-based digital pathology datasets used in our study at https://huggingface.co/ogutsevda/datasets , establishing the first large-scale graph benchmark in this field. Our code is available at https://github.com/ogutsevda/graphist .

In multimodal learning, CLIP has emerged as the de-facto approach for mapping different modalities into a shared latent space by bringing semantically similar representations closer while pushing apart dissimilar ones. However, CLIP-based contrastive losses exhibit unintended behaviors that negatively impact true semantic alignment, leading to sparse and fragmented latent spaces. This phenomenon, known as the modality gap, has been partially mitigated for standard text and image pairs but remains unknown and unresolved in more complex multimodal settings, such as the medical domain. In this work, we study this phenomenon in the latter case, revealing that the modality gap is present also in medical alignment, and we propose a modality-agnostic framework that closes this gap, ensuring that semantically related representations are more aligned, regardless of their source modality. Our method enhances alignment between radiology images and clinical text, improving cross-modal retrieval and image captioning.

Ultrasound (US) imaging poses unique challenges for representation learning due to its inherently noisy acquisition process. The low signal-to-noise ratio and stochastic speckle patterns hinder standard self-supervised learning methods relying on a pixel-level reconstruction objective. Joint-Embedding Predictive Architectures (JEPAs) address this drawback by predicting masked latent representations rather than raw pixels. However, standard approaches depend on hyperparameter-brittle and computationally expensive online teachers updated via exponential moving average. We propose US-JEPA, a self-supervised framework that adopts the Static-teacher Asymmetric Latent Training (SALT) objective. By using a frozen, domain-specific teacher to provide stable latent targets, US-JEPA decouples student-teacher optimization and pushes the student to expand upon the semantic priors of the teacher. In addition, we provide the first rigorous comparison of all publicly available state-of-the-art ultrasound foundation models on UltraBench, a public dataset benchmark spanning multiple organs and pathological conditions. Under linear probing for diverse classification tasks, US-JEPA achieves performance competitive with or superior to domain-specific and universal vision foundation model baselines. Our results demonstrate that masked latent prediction provides a stable and efficient path toward robust ultrasound representations.

Self-supervised learning (SSL) and diffusion models have respectively advanced representation learning and generative modeling for high-dimensional 3D visual data, yet they are often developed as separate paradigms. Their unification remains challenging under multi-source heterogeneity, as anatomical content must be preserved for analysis while acquisition-related style varies across centers and affects synthesis. In this paper, we propose MeDUET, a 3D Medical image Disentangled UnifiEd PreTraining framework in the variational autoencoder latent space. MeDUET formulates unified pretraining as an empirical factor identifiability problem, aiming to learn domain-invariant content factors for anatomy and domain-specific style factors for appearance. To improve factor separation, MeDUET first uses token demixing with a standard adversarial domain regularizer to establish basic content-style specialization, and further introduces Mixed Factor Token Distillation and Swap-invariance Quadruplet Contrast to reduce mixed-region factor leakage and organize factor spaces with factor-wise invariance and discriminability. With these learned factors, MeDUET transfers effectively to both synthesis and analysis, yielding higher fidelity, faster convergence, and better controllability for synthesis, while achieving competitive or superior domain generalization and label efficiency on diverse datasets, tasks, and modalities. Overall, MeDUET shows that multi-source heterogeneity can serve as useful supervision, with disentanglement providing an effective interface for unifying 3D medical image synthesis and analysis. Our code is available at https://github.com/JK-Liu7/MeDUET.

Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, yet early risk detection is often limited by available diagnostics. Carotid ultrasound, a non-invasive and widely accessible modality, encodes rich structural and hemodynamic information that is largely untapped. Here, we present a machine learning (ML) framework that extracts clinically meaningful representations of vascular damage (VD) from carotid ultrasound videos, using hypertension as a weak proxy label. The model learns robust features that are biologically plausible, interpretable, and strongly associated with established cardiovascular risk factors, comorbidities, and laboratory measures. High VD stratifies individuals for myocardial infarction, cardiac death, and all-cause mortality, matching or outperforming conventional risk models such as SCORE2. Explainable AI analyses reveal that the model relies on vessel morphology and perivascular tissue characteristics, uncovering novel functional and anatomical signatures of vascular damage. This work demonstrates that routine carotid ultrasound contains far more prognostic information than previously recognized. Our approach provides a scalable, non-invasive, and cost-effective tool for population-wide cardiovascular risk assessment, enabling earlier and more personalized prevention strategies without reliance on laboratory tests or complex clinical inputs.

Vision Transformers (ViTs) have emerged as the state-of-the-art architecture in representation learning, leveraging self-attention mechanisms to excel in various tasks. ViTs split images into fixed-size patches, constraining them to a predefined size and necessitating pre-processing steps like resizing, padding, or cropping. This poses challenges in medical imaging, particularly with irregularly shaped structures like tumors. A fixed bounding box crop size produces input images with highly variable foreground-to-background ratios. Resizing medical images can degrade information and introduce artefacts, impacting diagnosis. Hence, tailoring variable-sized crops to regions of interest can enhance feature representation capabilities. Moreover, large images are computationally expensive, and smaller sizes risk information loss, presenting a computation-accuracy tradeoff. We propose VariViT, an improved ViT model crafted to handle variable image sizes while maintaining a consistent patch size. VariViT employs a novel positional embedding resizing scheme for a variable number of patches. We also implement a new batching strategy within VariViT to reduce computational complexity, resulting in faster training and inference times. In our evaluations on two 3D brain MRI datasets, VariViT surpasses vanilla ViTs and ResNet in glioma genotype prediction and brain tumor classification. It achieves F1-scores of 75.5% and 76.3%, respectively, learning more discriminative features. Our proposed batching strategy reduces computation time by up to 30% compared to conventional architectures. These findings underscore the efficacy of VariViT in image representation learning. Our code can be found here: https://github.com/Aswathi-Varma/varivit

Objective: X-ray angiography is the gold standard imaging modality for cardiovascular diseases. However, current deep learning approaches for X-ray angiogram analysis are severely constrained by the scarcity of annotated data. While large-scale self-supervised learning (SSL) has emerged as a promising solution, its potential in this domain remains largely unexplored, primarily due to the lack of effective SSL frameworks and large-scale datasets. Methods: To bridge this gap, we introduce a vascular anatomy-aware masked image modeling (VasoMIM) framework that explicitly integrates domain-specific anatomical knowledge. Specifically, VasoMIM comprises two key designs: an anatomy-guided masking strategy and an anatomical consistency loss. The former strategically masks vessel-containing patches to compel the model to learn robust vascular semantics, while the latter preserves structural consistency of vessels between the original and reconstructed images, enhancing the discriminability of the learned representations. In conjunction with VasoMIM, we curate XA-170K, the largest X-ray angiogram pre-training dataset to date. Results: We validate VasoMIM on four downstream tasks across six datasets, where it demonstrates superior transferability and achieves state-of-the-art performance compared to existing methods. Conclusion: Incorporating vascular anatomical knowledge into self-supervised pre-training enables more discriminative and transferable representations for X-ray angiogram analysis. Significance: These findings highlight the significant potential of VasoMIM as a foundation model for advancing a wide range of X-ray angiogram analysis tasks. VasoMIM and XA-170K are available on GitHub and Hugging Face, respectively.

Vision-language pretraining has driven progress in medical image representation learning, but it depends on paired image-text data and can inherit reporting bias from clinical narratives. We study whether language-free predictive pretraining can produce an image encoder that transfers effectively to radiology report generation. RadJEPA is a chest-X-ray adaptation of I-JEPA, pretrained on approximately 840K unlabeled radiographs using latent context-to-target prediction. Our primary contribution is an extensive empirical evaluation of this language-free encoder for report generation: the frozen image encoder is coupled to a trainable two-layer projector and language decoder, and is also substituted into four established vision-language backbones. Across MIMIC-CXR and IU-Xray, RadJEPA matches or exceeds the evaluated image-only and image-text baselines on lexical, entity-relation, and clinical-label metrics. Controlled MIMIC-only comparisons provide evidence that the predictive objective contributes beyond domain-specific pretraining, while broader comparisons also reflect differences in pretraining data, model capacity, and input resolution. Complementary classification and segmentation experiments assess transfer beyond report generation.

The functions of different regions of the human brain are closely linked to their distinct cytoarchitecture, which is defined by the spatial arrangement and morphology of the cells. Identifying brain regions by their cytoarchitecture enables various scientific analyses of the brain. However, delineating these areas manually in brain histological sections is time-consuming and requires specialized knowledge. An automated approach is necessary to minimize the effort needed from human experts. To address this, we propose CytoCLIP, a suite of vision-language models derived from pre-trained Contrastive Language-Image Pre-Training (CLIP) frameworks to learn joint visual-text representations of brain cytoarchitecture. CytoCLIP comprises two model variants: one is trained using low-resolution whole-region images to understand the overall cytoarchitectural pattern of an area, and the other is trained on high-resolution image tiles for detailed cellular-level representation. The training dataset is created from NISSL-stained histological sections of developing fetal brains of different gestational weeks. It includes 86 distinct regions for low-resolution images and 379 brain regions for high-resolution tiles. We evaluate the model's understanding of the cytoarchitecture and generalization ability using region classification and cross-modal retrieval tasks. Multiple experiments are performed under various data setups, including data from samples of different ages and sectioning planes. Experimental results demonstrate that CytoCLIP outperforms existing methods. It achieves a weighted F1 score of 0.87 for whole-region classification and 0.91 for high-resolution image tile classification.

Temporally aware image representations are crucial for capturing disease progression in 3D volumes of longitudinal medical datasets. However, recent state-of-the-art self-supervised learning approaches like Masked Autoencoding (MAE), despite their strong representation learning capabilities, lack temporal awareness. In this paper, we propose STAMP (Stochastic Temporal Autoencoder with Masked Pretraining), a Siamese MAE framework that encodes temporal information through a stochastic process by conditioning on the time difference between the 2 input volumes. Unlike deterministic Siamese approaches, which compare scans from different time points but fail to account for the inherent uncertainty in disease evolution, STAMP learns temporal dynamics stochastically by reframing the MAE reconstruction loss as a conditional variational inference objective. We evaluated STAMP on two OCT and one MRI datasets with multiple visits per patient. STAMP pretrained ViT models outperformed both existing temporal MAE methods and foundation models on different late stage Age-Related Macular Degeneration and Alzheimer's Disease progression prediction which require models to learn the underlying non-deterministic temporal dynamics of the diseases.

Building robust medical machine learning systems requires pretraining strategies that exploit the intrinsic structure present in clinical data. We introduce Multiview Masked Autoencoder (MVMAE), a self-supervised framework that leverages the natural multi-view organization of radiology studies to learn view-invariant and disease-relevant representations. MVMAE combines masked image reconstruction with cross-view alignment, transforming clinical redundancy across projections into a powerful self-supervisory signal. We further extend this approach with MVMAE-V2T, which incorporates radiology reports as an auxiliary text-based learning signal to enhance semantic grounding while preserving fully vision-based inference. Evaluated on a downstream disease classification task on three large-scale public datasets, MIMIC-CXR, CheXpert, and PadChest, MVMAE consistently outperforms supervised and vision-language baselines. Furthermore, MVMAE-V2T provides additional gains, particularly in low-label regimes where structured textual supervision is most beneficial. Together, these results establish the importance of structural and textual supervision as complementary paths toward scalable, clinically grounded medical foundation models.

Neurological diseases can cause significant cognitive impairment, yet linking brain structure to cognitive function remains challenging due to the limitations of conventional methods and the black-box nature of deep learning. We propose InfoVAE-Med3D, an adaptation of InfoVAE for three-dimensional brain MRI that learns informative latent representations by relaxing the mutual information regularization penalty between inputs and latent representations. We evaluated the method on a healthy control (HC) dataset ($n=6{,}503$) with age and sex information and a multiple sclerosis (MS) dataset ($n=916$) with age, sex, and Symbol Digit Modalities Test (SDMT) scores. Two-dimensional visualizations revealed sex-related clustering and smooth transitions in grey and white matter volumes. Removing the mutual information regularization penalty generally improved MRI reconstruction and brain age and SDMT prediction using support vector regression, outperforming other variational autoencoder variants and achieving competitive performance with deep learning predictors, particularly on the MS dataset. InfoVAE-Med3D provides informative latent representations for investigating brain-related characteristics and supports latent-space analysis and imaging biomarker discovery for cognitive impairment, particularly in MS.

Anomaly detection in medical images is challenging due to limited annotations and the domain gap. Existing reconstruction-based methods often rely on frozen pre-trained encoders, restricting adaptation to domain-specific patterns and degrading localization accuracy. Meanwhile, prototype-based learning offers interpretable representations but commonly suffers from prototype collapse, where a few prototypes dominate training and reduce diversity. To address these issues, we propose DNP-ConFormer, a unified framework that integrates a trainable encoder with prototype-guided reconstruction and a Diversity-Aware Alignment Loss. A momentum encoder enables stable domain-adaptive representation learning, while a lightweight Prototype Extractor discovers informative normal prototypes and injects them into the decoder via attention to guide reconstruction. The proposed alignment objective further encourages balanced feature-to-prototype assignments, effectively mitigating prototype collapse. Extensive experiments on multiple medical imaging benchmarks demonstrate improved representation quality and anomaly localization compared with prior methods. Visualization and prototype assignment analyses further validate the effectiveness and interpretability of our approach. The code is available at https://github.com/liluhu0/DNP-ConFormer.

Melanoma is a highly aggressive skin cancer, making early and accurate diagnosis critical. While deep learning excels in skin lesion classification, standard ``black-box" models struggle to explain diagnostic uncertainty, limiting clinical trust. This work introduces a hybrid framework combining a class-aware adversarial Variational Autoencoder and an XGBoost classifier, transcending simple binary classification by leveraging a generative latent space for interpretable decision support. Guided by adversarial training, the model learns the visual characteristics of skin lesions and projects them into a continuous latent space, ensuring that similar images are grouped closely together. Trained on this latent space, the XGBoost classifier achieves a robust AUC of 0.868, competing closely with state-of-the-art models. For borderline cases, the framework enables clinicians to leverage the latent topology through Content-Based Image Retrieval. This provides a dual benefit: it allows the clinician to visually compare an ambiguous lesion against biopsy-confirmed precedents and acts as an early warning sign since a borderline classification can indicate that a lesion shares features of both nevi and melanomas, potentially requiring close monitoring. Our approach translates algorithmic hesitation into transparent, evidence-based visual support, bridging the gap between predictive performance and clinical trust.

Manually refining radiological segmentation masks is highly resource-intensive. To determine when this expert commitment is truly justified for the training of segmentation models, we investigate the relationship between label quality and model performance. Expanding beyond models trained directly for inference, we conduct the first study isolating the impact of label quality in pre-training datasets. While high-quality labels remain essential for models proceeding directly to deployment, we find no evidence that strict label quality is crucial for pre-training efficacy. These results question the necessity of exhaustive human-in-the-loop refinement for massive corpora intended for pretraining and suggest that expert effort is more effectively invested in well-curated downstream target datasets.

Breast cancer survival prediction in computational pathology presents a remarkable challenge due to tumor heterogeneity. For instance, different regions of the same tumor in the pathology image can show distinct morphological and molecular characteristics. This makes it difficult to extract representative features from whole slide images (WSIs) that truly reflect the tumor's aggressive potential and likely survival outcomes. In this paper, we present PathoHR, a novel pipeline for accurate breast cancer survival prediction that enhances any size of pathological images to enable more effective feature learning. Our approach entails (1) the incorporation of a plug-and-play high-resolution Vision Transformer (ViT) to enhance patch-wise WSI representation, enabling more detailed and comprehensive feature extraction, (2) the systematic evaluation of multiple advanced similarity metrics for comparing WSI-extracted features, optimizing the representation learning process to better capture tumor characteristics, (3) the demonstration that smaller image patches enhanced follow the proposed pipeline can achieve equivalent or superior prediction accuracy compared to raw larger patches, while significantly reducing computational overhead. Experimental findings valid that PathoHR provides the potential way of integrating enhanced image resolution with optimized feature learning to advance computational pathology, offering a promising direction for more accurate and efficient breast cancer survival prediction. Code will be available at https://github.com/AIGeeksGroup/PathoHR.

Learning robust representations from functional magnetic resonance imaging (fMRI) is fundamentally challenged by the temporal irregularity and noise inherent in data from heterogeneous sources. Existing self-supervised learning (SSL) methods often discard critical temporal information by discretizing or averaging fMRI signals. To address this, we introduce a novel framework that reframes SSL as a Stochastic Optimal Control (SOC) problem. Our approach models brain activity as continuous-time latent dynamics, learning a robust representation of brain dynamics by optimizing a control policy that is agnostic to the temporal irregularity. This SOC framework naturally unifies masked autoencoding (MAE) and joint-embedding prediction (JEPA) to extract compact, control-derived representations. Furthermore, a simulation-free inference strategy ensures computational efficiency and scalability for large-scale fMRI datasets. Our model demonstrates state-of-the-art performance across diverse downstream applications, highlighting the potential of the SOC-based continuous-time representation learning framework.

Dense prediction is a fundamental requirement for many medical vision tasks such as medical image restoration, registration, and segmentation. The most popular vision model, Convolutional Neural Networks (CNNs), has reached bottlenecks due to the intrinsic locality of convolution operations. Recently, transformers have been widely adopted for dense prediction for their capability to capture long-range visual dependence. However, due to the high computational complexity and large memory consumption of self-attention operations, transformers are usually used at downsampled feature resolutions. Such usage cannot effectively leverage the tissue-level textural information available only at the full image resolution. This textural information is crucial for medical dense prediction as it can differentiate the subtle human anatomy in medical images. In this study, we hypothesize that Multi-layer Perceptrons (MLPs) are superior alternatives to transformers in medical dense prediction where tissue-level details dominate the performance, as MLPs enable long-range dependence at the full image resolution. To validate our hypothesis, we develop a full-resolution hierarchical MLP framework that uses MLPs beginning from the full image resolution. We evaluate this framework with various MLP blocks on a wide range of medical dense prediction tasks including restoration, registration, and segmentation. Extensive experiments on six public well-benchmarked datasets show that, by simply using MLPs at full resolution, our framework outperforms its CNN and transformer counterparts and achieves state-of-the-art performance on various medical dense prediction tasks.

Medical landmark detection usually requires accurate expert annotations, which are laborious and difficult to scale across anatomical regions. In this work, we study an extreme annotation-efficient setting where only a single annotated template image is available. We propose SGB-Match, a structure-guided coarse-to-fine self-supervised matching framework for one-shot medical landmark detection. The framework first learns dense anatomical correspondence from unlabeled augmented image pairs, and then transfers the landmark definition from the annotated template to each target image through feature matching. Different from standard contrastive correspondence learning, where negative candidates are penalized by a structure-agnostic rule, we introduce a structure-guided bias into the contrastive objective. The bias is constructed from relative distance and edge-aware anatomical cues, and explicitly reweights the negative gradients: nearby structure-relevant candidates are weakly repelled, while distant or structure-irrelevant negatives are strongly suppressed. As a result, the learned feature space better preserves local anatomical structures around template landmarks and reduces confusing responses from repeated textures. We further adopt a global-to-local design, where a global encoder provides coarse landmark localization and a local encoder refines the prediction in a cropped region. Extensive experiments on four 2D radiological landmark datasets demonstrate that SGB-Match achieves strong one-shot performance across both public and newly collected datasets and consistently benefits from both structure-guided bias and two-stage refinement.