Free Lunch in Medical Image Foundation Model Pre-training via Randomized Synthesis and Disentanglement

• URL: http://arxiv.org/abs/2602.12317v1
• Date: Thu, 12 Feb 2026 18:09:22 GMT
• Title: Free Lunch in Medical Image Foundation Model Pre-training via Randomized Synthesis and Disentanglement
• Authors: Yuhan Wei, Yuting He, Linshan Wu, Fuxiang Huang, Junlin Hou, Hao Chen,
• Abstract summary: RaSD (Randomized Synthesis and Disentanglement) is a scalable framework for pre-training medical image foundation models (MIFMs) entirely on synthetic data.<n>We pre-trained RaSD on 1.2 million 3D volumes and 9.6 million 2D images, and extensively evaluated the resulting models across 6 imaging modalities, 48 datasets, and 56 downstream tasks.<n>Across all evaluated downstream tasks, RaSD consistently outperforms training-from-scratch models, achieves the best performance on 17 tasks, and remains comparable to models pre-trained on large real datasets in most others.
• Score: 17.69771768062763
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Medical image foundation models (MIFMs) have demonstrated remarkable potential for a wide range of clinical tasks, yet their development is constrained by the scarcity, heterogeneity, and high cost of large-scale annotated datasets. Here, we propose RaSD (Randomized Synthesis and Disentanglement), a scalable framework for pre-training MIFMs entirely on synthetic data. By modeling anatomical structures and appearance variations with randomized Gaussian distributions, RaSD exposes models to sufficient multi-scale structural and appearance perturbations, forcing them to rely on invariant and task-relevant anatomical cues rather than dataset-specific textures, thereby enabling robust and transferable representation learning. We pre-trained RaSD on 1.2 million 3D volumes and 9.6 million 2D images, and extensively evaluated the resulting models across 6 imaging modalities, 48 datasets, and 56 downstream tasks. Across all evaluated downstream tasks, RaSD consistently outperforms training-from-scratch models, achieves the best performance on 17 tasks, and remains comparable to models pre-trained on large real datasets in most others. These results demonstrate that the capacity of synthetic data alone to drive robust representation learning. Our findings establish a paradigm shift in medical AI, demonstrating that synthetic data can serve as a "free lunch" for scalable, privacy-preserving, and clinically generalizable foundation models.

Related papers

• Adapting HFMCA to Graph Data: Self-Supervised Learning for Generalizable fMRI Representations [57.054499278843856]
  Functional magnetic resonance imaging (fMRI) analysis faces significant challenges due to limited dataset sizes and domain variability between studies.<n>Traditional self-supervised learning methods inspired by computer vision often rely on positive and negative sample pairs.<n>We propose adapting a recently developed Hierarchical Functional Maximal Correlation Algorithm (HFMCA) to graph-structured fMRI data.
  arXiv  Detail & Related papers  (2025-10-05T12:35:01Z)
• Improving Performance, Robustness, and Fairness of Radiographic AI Models with Finely-Controllable Synthetic Data [14.791557943114737]
  RoentGen-v2 is a text-to-image diffusion model for chest radiographs.<n>It generates clinically plausible images with demographic conditioning.<n>We use this large synthetic dataset to evaluate optimal training pipelines for downstream disease classification models.
  arXiv  Detail & Related papers  (2025-08-22T20:30:58Z)
• impuTMAE: Multi-modal Transformer with Masked Pre-training for Missing Modalities Imputation in Cancer Survival Prediction [75.43342771863837]
  We introduce impuTMAE, a novel transformer-based end-to-end approach with an efficient multimodal pre-training strategy.<n>It learns inter- and intra-modal interactions while simultaneously imputing missing modalities by reconstructing masked patches.<n>Our model is pre-trained on heterogeneous, incomplete data and fine-tuned for glioma survival prediction using TCGA-GBM/LGG and BraTS datasets.
  arXiv  Detail & Related papers  (2025-08-08T10:01:16Z)
• Towards a general-purpose foundation model for fMRI analysis [58.06455456423138]
  We introduce NeuroSTORM, a framework that learns from 4D fMRI volumes and enables efficient knowledge transfer across diverse applications.<n>NeuroSTORM is pre-trained on 28.65 million fMRI frames (>9,000 hours) from over 50,000 subjects across multiple centers and ages 5 to 100.<n>It outperforms existing methods across five tasks: age/gender prediction, phenotype prediction, disease diagnosis, fMRI-to-image retrieval, and task-based fMRI.
  arXiv  Detail & Related papers  (2025-06-11T23:51:01Z)
• SynthFM: Training Modality-agnostic Foundation Models for Medical Image Segmentation without Real Medical Data [0.5242869847419834]
  Foundation models like the Segment Anything Model (SAM) excel in zero-shot segmentation for natural images.<n>But they struggle with medical image segmentation due to differences in texture, contrast, and noise.<n>Annotating medical images is costly and requires domain expertise, limiting large-scale annotated data availability.<n>We propose SynthFM, a synthetic data generation framework that mimics the complexities of medical images.
  arXiv  Detail & Related papers  (2025-04-11T00:14:28Z)
• Latent Drifting in Diffusion Models for Counterfactual Medical Image Synthesis [55.959002385347645]
  Latent Drifting enables diffusion models to be conditioned for medical images fitted for the complex task of counterfactual image generation.<n>We evaluate our method on three public longitudinal benchmark datasets of brain MRI and chest X-rays for counterfactual image generation.
  arXiv  Detail & Related papers  (2024-12-30T01:59:34Z)
• Residual Vision Transformer (ResViT) Based Self-Supervised Learning Model for Brain Tumor Classification [0.08192907805418585]
  Self-supervised learning models provide data-efficient and remarkable solutions to limited dataset problems. This paper introduces a generative SSL model for brain tumor classification in two stages. The proposed model attains the highest accuracy, achieving 90.56% on the BraTs dataset with T1 sequence, 98.53% on the Figshare, and 98.47% on the Kaggle brain tumor datasets.
  arXiv  Detail & Related papers  (2024-11-19T21:42:57Z)
• Evaluating Utility of Memory Efficient Medical Image Generation: A Study on Lung Nodule Segmentation [0.0]
  This work proposes a memory-efficient patch-wise denoising diffusion probabilistic model (DDPM) for generating synthetic medical images. Our approach generates high-utility synthetic images with nodule segmentation while efficiently managing memory constraints. We evaluate the method in two scenarios: training a segmentation model exclusively on synthetic data, and augmenting real-world training data with synthetic images.
  arXiv  Detail & Related papers  (2024-10-16T13:20:57Z)
• Generative models of MRI-derived neuroimaging features and associated dataset of 18,000 samples [17.576301478946775]
  GenMIND is a collection of generative models of normative regional volumetric features derived from structural brain imaging. We offer 18,000 synthetic samples spanning the adult lifespan (ages 22-90 years), alongside the model's capability to generate unlimited data.
  arXiv  Detail & Related papers  (2024-07-17T15:33:10Z)
• Towards a clinically accessible radiology foundation model: open-access and lightweight, with automated evaluation [113.5002649181103]
  Training open-source small multimodal models (SMMs) to bridge competency gaps for unmet clinical needs in radiology. For training, we assemble a large dataset of over 697 thousand radiology image-text pairs. For evaluation, we propose CheXprompt, a GPT-4-based metric for factuality evaluation, and demonstrate its parity with expert evaluation. The inference of LlaVA-Rad is fast and can be performed on a single V100 GPU in private settings, offering a promising state-of-the-art tool for real-world clinical applications.
  arXiv  Detail & Related papers  (2024-03-12T18:12:02Z)
• Modeling Shared Responses in Neuroimaging Studies through MultiView ICA [94.31804763196116]
  Group studies involving large cohorts of subjects are important to draw general conclusions about brain functional organization. We propose a novel MultiView Independent Component Analysis model for group studies, where data from each subject are modeled as a linear combination of shared independent sources plus noise. We demonstrate the usefulness of our approach first on fMRI data, where our model demonstrates improved sensitivity in identifying common sources among subjects.
  arXiv  Detail & Related papers  (2020-06-11T17:29:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.