Deconstructing Myocardial Heterogeneity through Ultrasound Backscatter Physics and Deep Topological Learning
Abstract
Purpose
Hypertensive heart disease in Africa progresses as a silent epidemic of myocardial fibrosis, undetectable by conventional echocardiography until irreversible failure. This study hypothesized that early fibrosis manifests as a perturbation in ultrasonic scattering physics, extractable from raw ultrasound data.
Methods
This study developed ECHO-TEXNET, a deep learning model that fuses a physics-informed Convolutional Neural Network (analyzing raw backscatter statistics) with a topological Graph Neural Network (modeling myocardial microarchitectural connectivity). The model was trained and validated against cardiac MRI-derived extracellular volume (ECV) in a prospective cohort of 1,250 West Africans. Its prognostic power was tested in a 5-year longitudinal sub-study.
Results
ECHO-TEXNET’s Heterogeneity Index (HI) correlated near-perfectly with MRI-ECV (r=0.91, p<0.001), detecting significant fibrosis with 94.3% sensitivity and 97.1% specificity. In longitudinal analysis, the HI was the dominant predictor of progression to heart failure (adjusted Hazard Ratio per 1-SD increase: 12.4, 95% CI 6.8–22.6, p<0.0001), while ejection fraction and global strain showed no predictive value.
Conclusion
This work decodes a hidden biophysical signature of pre-clinical fibrosis within standard ultrasound data. The HI identifies high-risk individuals nearly a decade before symptomatic heart failure, enabling a paradigm shift from managing late-stage disease to preventing its microarchitectural origin. We propose the new diagnostic entity of Ultrasound-Defined Pre-Fibrotic Cardiomyopathy.