Detecting Nonsynchronous Heart Cells from Video – An Unsupervised Machine Learning Approach

Cardiac myocytes possess the property of automaticity; however, irregularities in the automaticity can lead to cardiac arrhythmias and heart blocks, which can cause life-threatening health concerns. Currently, full-field, high-resolution, anomaly detection methods for non-contact, electro-mechanical dynamics of cardiac myocytes do not exist. This research uses emerging structural dynamic techniques coupled with cutting edge signal processing methods to examine the dynamics of beating heart cells. Applications of this research could offer novel advances in medical diagnostics for heart disorders. A properly functioning heart will have cells that contract together in one global motion. However, in certain pathologic conditions, one cell –or a collection of cells – will beat out of phase with the rest of the heart. Using a combination of blind source separation techniques (non-negative matrix factorization, sparse and low-rank matrix decompositions) and mutual information on videos, these nonsynchronous cardiac myocytes can be identified.