UC San Diego

Cardiac dyssynchrony can lead to heart failure. Cardiac Resynchronization Therapy (CRT) is a procedure which restores synchrony to the heart, allowing the heart to function as intended. Previous research has shown that patient-specific computational modeling of dyssynchrony and CRT can be useful in predicting the outcome of the clinical procedure. Machine learning may be another technique to predict outcomes of medical procedures, including CRT. The application of machine learning to CRT patient outcomes has not been thoroughly investigated, though it may be able to categorize dyssynchrony with similar accuracy as previous complex models, in less time. Patient models were analyzed to determine the efficacy of machine learning as a means to predict outcomes of CRT. Vectorcardiogram data for virtual patients was used to train and test the capabilities of machine learning for determining a dyssynchrony index output, the same output found with the patient-specific computational models. The accuracies of the machine learning outputs were compared to determine the optimal amount of input data needed. Patient physiological parameters were compared to the dyssynchrony index prediction. It was determined that when looking at the simulated patients for whom there was data, varying the amount of input data per patient had no effect. Increasing the number of patients increased the accuracy of the prediction, while most physiological parameters had no clear correlation to the accuracy of the prediction. Further tests with additional data can be carried out to look deeper into using machine learning to predict outcomes of CRT.