Assessment of Clonal Hematopoiesis of Indeterminate Potential and Future Cardiomyopathy from Cardiac Magnetic Resonance Imaging using Deep Learning in a Cardio-oncology Population

TL;DR

This study develops a deep learning model that uses routine cardiac MRI scans to identify clonal hematopoiesis of indeterminate potential (CHIP) and predict future cardiomyopathy risk, aiding non-invasive cardiovascular risk assessment.

Contribution

The paper introduces a novel deep learning framework that detects CHIP and stratifies cardiomyopathy risk from cardiac MRI, demonstrating high predictive accuracy and robustness against confounders.

Findings

AUC of 0.71 for CHIP detection

AUC of 0.87 for cardiomyopathy prediction in CHIP-positive patients

Model outperforms demographic-only baselines

Abstract

We propose a novel deep learning framework to identify clonal hematopoiesis of indeterminate potential (CHIP), a somatic mutation condition associated with adverse cardiovascular outcomes, using routine cardiac magnetic resonance (CMR) imaging. Utilizing 152 multi-view late gadolinium enhancement (LGE) scans from 136 cardio-oncology patients, we developed a convolutional neural network to (1) detect CHIP status and (2) stratify the risk of future cardiomyopathy specifically within the CHIP-positive cohort. To ensure robustness, we performed rigorous feature importance analysis to rule out reliance on demographic confounders such as age and immune checkpoint inhibitor usage. The model achieved an AUC of 0.71 for CHIP detection and, notably, an AUC of 0.87 for predicting future cardiomyopathy in CHIP-positive patients, significantly outperforming demographic-only baselines. These results demonstrate the feasibility of using LGE-CMR signatures as a non-invasive "radiogenomic" screening tool, potentially enabling accessible risk stratification and precision medicine for high-risk cardiovascular populations.