Using Echocardiography and AI to Predict Cardiac Biological Age

Biological age is a personalized measure of the health status of an organism, organ, or system, as opposed to simply accounting for chronological age. To date, there have been known attempts to create estimators of biological age based on various biomedical data. In this work, we focused on developing an approach for assessing heart biological age using echocardiographic data. The current study included echocardiographic data from more than 5,000 different cases. As a result, indicators such as EA (E/A ratio of maximum flow rates in the first and second phases), IVS (thickness of the interventricular septum), LVPW (thickness of the posterior left ventricular wall), LVCO (cardiac output), and RWT (relative wall thickness) showed the greatest predictive potential. Based on these parameters, we created and trained neural network models to determine heart biological age, with a Mean Absolute Error (MAE) of approximately 3.5 years, an R-squared (R2) value of around 0.87, and a Spearman's rank correlation coefficient (rho) greater than 0.9 in men. In women, the MAE was approximately 3.4 years, with an R2 value of around 0.89 and a rho greater than 0.9. In addition, we have applied an AI explanation algorithm to improve understanding of how the model performs an assessment. Furthermore, the EchoAGE model was tested on echo cardiographic data from patients with age-related diseases, patients with multimorbidity, children with progeria syndrome, and diachronic data series.