Image-based Disease-wide Association Study via Self-supervised Learning links Abdominal MRI Features to 158 Diseases

Medical images contain rich phenotypic information that is often not fully captured by manual clinical assessment. Here, we present a systematic framework for extracting such information and conducting image-based disease-wide association studies (iDWAS) using self-supervised learning (SSL). We applied this framework to abdominal MRI data from the UK Biobank and evaluated associations between MRI-derived features and 562 diseases. Features were learned using a VICReg-based SSL model and tested for disease associations using logistic regression models. We identified 158 diseases that were significantly associated with the MRI-derived features, including the ones which are not directly linked to abdominal anatomy. Focusing on Metabolic dysfunction–associated steatohepatitis (MASH), we showed that the MRI-derived features captured disease-relevant information and separated MASH cases from controls better than established biomarkers such as PDFF and Iron-cT1. These findings highlight the ability of SSL to uncover clinically meaningful signals from routine imaging data. The proposed framework is broadly applicable to other imaging datasets and modalities, enabling more systematic approaches to incidental and early disease detection. The trained model and analysis pipeline are publicly available at https://github.com/srm2022/iDWAS .

Background

Medical images, often analyzed manually by clinicians, contain valuable information that may not be fully captured by the human eye. Machine Learning (ML) has demonstrated significant potential in extracting this information, allowing us to enhance our understanding of underlying pathologies. We present a systematic framework for extracting such information and performing image-based disease-wide association studies (iDWAS) using self-supervised learning. We then applied the framework to abdominal Magnetic Resonance Imaging (MRI) from UK Biobank (UKB) and investigated the associations between MRI-derived features and 562 diseases. Finally, we delved deeper into the observed associations for Metabolic dysfunction–associated steatohepatitis (MASH), a prevalent yet underdiagnosed condition with a high unmet need for non-invasive diagnostic tools.

Methods

We extracted features from MRI images using VICReg-based self-supervised learning and modeled the associations of the extracted features with 562 diseases using logistic regression. Using the likelihood ratio (LR) test, we then assessed how much additional variance can be explained by the MRI-derived features relative to the variance explained by a null model including only common confounders such as age, gender, and BMI.

Results

Of these 562 diseases, we identified 158 diseases to be significantly associated with the MRI-derived features. While many of them (e.g. MASH) are known to primarily emerge in the abdomen, others (e.g. neuropsychiatric diseases) are not. Taking MASH as a use-case and adjusting for MASH-specific confounding, we then showed that our MRI-derived features captured additional MASH-relevant information compared to standard MRI-derived biomarkers of liver health such as proton density fat fraction (PDFF) and iron-corrected T1 (Iron-cT1). The trained SSL model and analysis pipeline are available at https://github.com/srm2022/iDWAS .

Conclusions

Our results illustrate (1) how MRIs from one organ of the body can inform about the diseases primarily rooted in other organs of the body, reflecting patients’ overall health status, and (2) how SSL can extract more information relevant for specific diseases compared to current state-of-the-art biomarkers. Our framework can be applied to other MRI datasets as well as other imaging modalities with minimal adjustment, providing opportunities for the development of more systematic approaches to incidental diagnosing, i.e. (early) detection of diseases falling outside the scope of the original imaging procedure.