Computer based technique to detect depression in Alzheimer patients

Abstract

Depression is common among Alzheimer's patients in the early and middle stages of Alzheimer, it is estimated that 30% of Alzheimer patients are suffering from depression. Depression in Alzheimer patients can lead to more cognitive decline. There is no official test that diagnoses depression in Alzheimer patients which is a challenge for the medical communities. This research aimed to help the medical communities in this challenge, by presenting machine learning models that classify AD and MCI patients into two classes (depressive and non-depressive). 276 participants (mean age 70.9) are selected from the Pitt Corpus of Dementia Bank. The interviewer’s voice and the silences are removed from the audio records as a preprocessing task. Different features are extracted from the patient's speech to achieve this task. For instance, MFCC’s, Spectral Centroid, Spectral Roll-Off Point, and others. We used three traditional classification techniques (SVM, Random Tree, and Random Forest). The Artifical Neural Networks (ANN) is used also to provide an idea of how this project can be extended. Bootstrapping method is used to solve the sampling bias – 70% of the patients are not suffering from depression. From the results, MFCC’s features set is more appropriate with tree-based classifiers than SVM, where the Random Tree classifier achieved the highest classification accuracy (91.3%). While the other features set are more appropriate with SVM than tree-based classifiers, where SVM achieved 89.1% accuracy, 91.1% recall, and 89.1% precision. As a conclusion of the research results, the Random Tree classifier performed the highest classification performance rates – with 91.3% for accuracy, recall, and precision - using MFCC’s features (Standard Deviation and Kurtosis).

Date	16 Mar 2020
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Sylvie Ratté (Supervisor)