Optimizing Left Ventricular Assist Device Therapy: A Machine Learning Approach for Predicting Cardiac Output

Abstract

Abstract Heart failure (HF) is a significant concern worldwide, with left ventricular assist devices (LVADs) providing effective mechanical circulatory support for end-stage HF patients. However, the static nature of current LVAD pumping rates poses challenges in adapting to patients' physiological needs. To address this limitation, we propose a novel approach utilizing Multi-Layer Perceptron (MLPRegressor), a machine learning algorithm, to predict cardiac output (CO) accurately and adaptively adjust LVAD speed based on non-invasive physiological data. Our approach includes data preprocessing, feature engineering, and model evaluation. Our study demonstrates the superior performance of MLPRegressor over other machine learning models, with a Root mean squared error (RMSE) of 0.652 L/min and an R-squared score of 0.786. Personalized LVAD treatment based on predicted CO has the potential to improve patient outcomes and reduce complications associated with static pumping rates. Future research should explore additional physiological parameters, validation on larger datasets, and real-time monitoring for dynamic LVAD control in clinical settings. The integration of ML in cardiac care holds promise for enhancing heart failure management and patient care.