A Deep Learning Multi-Task Approach for the Detection of Alzheimer’s Disease in a Longitudinal Study

Alzheimer's Disease (AD) is one of the most common illnesses in the world, affecting approximately fifty million people. In the United States, AD is the sixth leading cause of death. Currently, AD does not have a cure; therefore, it is crucial to detect this disease early to enhance the quality of life of patients. Several studies have been conducted to detect this disease; however, most of these techniques perform a binary classification to distinguish individuals with AD from healthy ones. Even when such techniques are highly accurate in detecting the disease, they serve merely as a partial AD diagnostic tool since no information regarding its stage is identified. To address this issue, we propose a deep feed-forward neural network based on multi-task learning capable of simultaneously detecting AD and identifying its progression stage. Relying on multiple loss functions simultaneously, this model recognizes four classes of individuals: healthy, with early mild cognitive impairment, with late mild cognitive impairment, and with AD. The proposed technique is evaluated using the Alzheimer's Disease Neuroimaging Initiative dataset in two different periods, five and ten years. The evaluation is performed in terms of accuracy, precision, recall, and F1 score. Experiment results demonstrate the effectiveness of the proposed technique in detecting AD.