5 - Machine learning methods for robust parameter estimation

Abstract

After having explored how to build efficient models of cardiac physiology, this chapter introduces methods to estimate their parameters such that the simulated cardiac function matches clinical observations. This step, called model personalization, is traditionally performed using inverse problem techniques like optimization or variational approaches. A cost function, which measures the misfit between simulation and measurements, is minimized. In this chapter, we explore how machine learning concepts can be used to solve the inverse problem. A first approach consists in learning a regression model that maps clinical measurements to model parameters directly. The method is illustrated for the case of cardiac electrophysiology. A second approach builds upon reinforcement learning techniques to learn the optimization process, leading to a potentially more efficient (i.e. less number of iterations) and more robust (i.e. less prone to local minima) algorithm than black box optimization methods.