The project aims to lay the foundation for developing computational methods for improved diagnosis and prediction of heart disease from noisy data gathered from patients. The data comprises both time-series signals and images gathered from patients suffering from Hypertrophic Cardiomyopathy (HCM), the most common cardiac genetic disease. Current diagnosis limitations lead to undetected cases that may result in sudden cardiac death, often in active young individuals. The project focuses on advancing the computational means for utilizing data that is relatively fast and low-cost to obtain but is hard to interpret, namely sonogram images (Echo) and electrocardiogram time-series (ECG). Developing tools that improve diagnosis and prediction while making the basis for such tools simpler and less-expensive stands to benefit large populations, especially in under-privileged areas where state-of-the-art imaging technology is scarce.
This project develops probabilistic, supervised and unsupervised machine learning methods and tools. Specifically, it extends the framework of hidden Markov model to support effective use of 12-lead ECG as well as integration of Echo image data. These information sources are relatively inexpensive and nonintrusive but have hitherto been under-utilized in computer-assisted HCM-detection. Developing the proposed models and methods provides a unique training opportunity for young scientists, by conducting inter-disciplinary research while utilizing diverse types of data. The research also paves the road toward introducing probabilistic integrative methods in other disease-domains, as well as in other data-intensive applications outside medicine.
