Comparison of different binary classification models on radiomic features

Abstract

Improved cancer prognosis is an important goal of precision health medicine. Radiomics is the extraction of a high number of features from medical images. Machine Learning (ML) has advanced significantly in the last few years and offers many different approaches on how to detect and model out associations. By applying different machine learning methods to the abundance of data provided by radiomic features, it will assist in carrying out cancer detection, prognosis as well as the prediction of treatment response. In this paper, the goal is to create a pipeline that doctors at SGH would be able to use by just attaching placing the csv file containing the radiomics feature into the work path folder and begin running through the code where different ML techniques will be used to carry out binary classification to classify either outcome 1 which indicates a pathological complete response or outcome 0 which indicates a non-pathological complete response. The workflow of the pipeline will be data preprocessing, feature selection, ML modeling and finally analysis of the results.