Discrimination of ischemia and normal sinus rhythm for cardiac signals using a modified k means clustering algorithm.

Over 15 million Americans are affected by coronary heart disease, according to the American Heart Association. Approximately 8 million have suffered a myocardial infarction. The economic and social consequences of this disease are staggering. A plethora of experimental and established therapies exist for this disease, such as stem cell therapy, growth factor injection, engineered cell transfection, etc. The use of these techniques relies on targeted therapeutic delivery. This paper describes mathematical techniques to extract key features from acquired action potential signals from ischemic and normal regions of the same heart. Using a modified means clustering technique on paired data, the best features are evaluated in multidimensional space. The results indicate promising clustering and separation of ischemic and normal beats using frequency domain computations, morphology analyses, and isoelectric point evaluations. Features were tested with data collected from a swine model of localized ischemia implanted with transmural electrodes and evaluated with a cross-validation approach. This research may have clinical significance to aid in the efficacious diagnosis or treatment of myocardial ischemia.