Classification of diabetic cardiomyopathy-related cells using machine learning

Abstract

The patch-clamp technique is a significant tool in current electrophysiology research, especially in cardiovascular diseases, because it can capture electrical activity of the heart from cardiomyocytes. It is challenging to classify action potential waveforms in cardiological data from these recordings because it relies largely on professional assistance. We discovered that supervised classification may be used to predict the impact of electrophysiological perturbations on cardiomyopathic action potential groups in rat ventricular cells. At the cellular level, action potential classifications are utilized to discern between pathological and control waveforms in recorded cardiac action potentials. The four groups are as follows: (1) control, (2) diabetes, (3) diabetes with angiotensin, and (4) angiotensin. The signal’s biologically relevant features for the treatment of cardiomyopathy have been discovered. After they have been trained with different sets of features, the results of the seven machine learning models are compared. The knearest neighbor approach, along with the decision tree and random forest algorithms, is the best classifier for diagnosing aberrant action potential waveforms, with an accuracy of above 99% when compared to other models. The high classification accuracy demonstrates that the gathered individual cardiac AP features provide useful information regarding the pathological status of cardiomyocytes.