Detection of MCI from MRI using Gradient Boosting Classifier

Abstract

This work presents a non-invasive approach for detection of Mild Cognitive Impairment (MCI) using Magnetic Resonance Imaging (MRI). The gray matter features of MRI along with the personal characteristics data are used as features for the Gradient Boosting classifier. The MRI and personal characteristics data of Cognitively Normal (CN) and MCI subjects are obtained from Alzheimer's Diseases Neuroimaging Initiative database. First, the MRI scans are subjected to segmentation from which the gray matter images are obtained. Then the resulting images are pre-processed using 2D Dual-Tree Complex Wavelet Transforms. The wavelets obtained are then combined with the personal characteristics data and is fed to the Gradient Boosting classifier. An accuracy of 97.25% is obtained for classifying CN and MCI subjects and the results are compared with other traditional machine learning approaches such as Logistic Regression, Naive Bayes, Support Vector Machine and Random Forest.