Identification of biomarkers from Radiomics of brain scans for prediction of major depression using Repeated Elastic Net Technique

Abstract

The application of machine learning in the field of medicine is expanding on an almost daily basis. Data from the healthcare industry typically have high dimensionality but a limited sample size. The learning process can be sped up, system performance can be improved, complexity can be minimised, and the risk of overfitting may be decreased by selecting a smaller subset of relevant features from the high-dimensional data set. The primary objective of this investigation was to diagnose patients with major depressive disorder (MDD) using radiomics features extracted from MR images. In addition, the thesis tries to accomplish the objective by locating a collection of biomarkers that can assist in developing individualised treatment plans. Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care (EMBARC) was the source of the data. Before beginning the classification process, the data set’s dimensionality was decreased by applying a technique known as RENT, which stands for Repeated Elastic Net Technique for Feature Selection. Logistic Regression, Support Vector Machines and Random Forest are three common classifiers utilised in computing the performance of all features and the RENT chosen features predictions. A technique known as principal component analysis (PCA) was used for the analysis of the data. Throughout the splits and the dataset, RENT chose eleven characteristics in all. According to RENT, the rostral middle frontal cortex may be a significant biomarker that can predict people who have MDD.