September 22, 2020: Pham Minh Tuan, AutoEncoder-Based Feature Ranking for Alzheimer Disease

Alzheimer Disease (AD) is becoming a major type of neurodegenerative brain disease in elderly people. Early detection and diagnosis of AD is of crucial importance for developing treatments. Fluorodeoxyglucose positron emission tomography (FDG-PET) is one of the most effective biomarkers which helps to diagnose AD early. One major challenge of PET-based classification is the very high dimensionality of image features. To address this problem, dimensionality reduction could be used. In this work we propose a method for selecting the most important features from PET images. More specifically, PET images are mapped into region of interest (ROI) using an anatomical atlas. Then multiple AutoEncoder (AE) are trained and fine-tuned with softmax. After that, the connection weights learned from AEs are used to rank ROIs according to the total contribution of ROIs to the networks. To improve the ranking results, we proposed a 2-phases feature ranking method which is able to rank and select the most important ROIs. Lastly, the topranked ROIs are then input into a support vector machine (SVM) classifier. In experiments on Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset, the proposed method significantly improves the accuracy of the classifier when compared to other popular feature ranking methods such as: Fisher score, Tscore, Conditional Mutual Information Maximization (CMIM), Lasso.

Speaker: Mr. Pham Minh Tuan, Aix-Marseille University

Time: 2020-09-22

Venue: Webinar – Microsoft Teams