Population based Deep Cardiac Atlas Phenotypes and Application in Biological Age Prediction

Motivation: Cardiac biological age serves as a crucial indicator of cardiovascular disease risk. It can be assessed through non-invasive imaging. Goal(s): This study aims to construct the morphological atlas using CMR imaging, extract deep phenotypes, and validate their potential value in age prediction. Approach: End-diastolic (ED) and end-systolic (ES) atlases from 1000 healthy volunteers were constructed to extract momenta as deep phenotypes, with a random forest model evaluating their predictive power against conventional indicators. Results: 880 cardiac phenotypes based on ED and ES atlases were extracted. Integrating these with conventional biomarkers enhances age prediction accuracy, reflected by reduced MAE and increased R2 score. Impact: Based on cardiac atlases of two key phases, momenta extracted as deep phenotypes could control deformation and encode age-related anatomical variations. Combining these new phenotypes with conventional biomarkers enables the development of more accurate models for predicting cardiac biological age.