Multi-class Classification of Alzheimer's Disease using 3DCNN Features and Multilayer Perceptron
Manu Raju, Varun P. Gopi, Anitha Venugopal
Abstract
An Alzheimer's individual cognitively goes through three stages: Cognitive Normal, Mild Cognitive Impairment (MCI), and complete cognitive impairment and finally succumb to death. It is very important to understand all these stages to diagnose the disease at different levels properly. Past studies mainly center on issues of binary classification such as Alzheimer's Disease (AD) vs. Normal Control (NC), MCI vs. AD, NC vs. MCI, or MCI converter (MCIc) vs. MCI non-converter (MCInc). There are less efficient works in Multi-level classification. Multiâlevel diagnosis enables one to use the same algorithm to identify the two classes of diseases, namely MCI and AD and normal, which is impossible with a binary level classification. This helps determine the severity levels of the disease and helps identify the disease at an earlier stage, i.e., at the MCI stage. The training of architecture with AD, MCI, and NC data also improves discriminating non-diseased from diseased ones. The transition from one stage to another is indicated by anatomical changes in the Structural MRI (SMRI). For CNN to learn discriminatory features underlying the disease and identify the brain affected regions that correspond to each level from MRI scans, the algorithm should have the three stages in a single algorithm. So we propose a multi-level classification of AD using cascaded 3D CNN features and Multilayer Perceptron classifier to achieve a ternary classification accuracy of 96.66%, which outperforms state of the art.