Binary Classification of Medical Images by Symbolic Regression

This experimental study investigates the application of symbolic regression to classification of medical diagnostic images, and the effectiveness of using alternative fitness functions for symbolic regression in order to account for the class imbalance that is often found in medical datasets. Three fitness functions are tested: one naïve function to act as a control, and two with different approaches to preventing performance bias. The performances of these fitness functions are compared and analysed, and the performance of symbolic regression for classification of medical images is evaluated. Analysis of these results shows such approaches to be insufficient to prevent performance bias for symbolic regression. However, symbolic regression is shown to be a promising machine learning algorithm for applications in medical diagnostic imaging nonetheless.