Automatic Early Diagnosis of Alzheimer's Disease Using 3D Deep Ensemble Approach

Abstract

Alzheimer's disease (AD) is considered the 6 th leading cause of death worldwide. Early diagnosis of AD is not an easy task, and no preventive cures have been discovered yet. Having an accurate computer-aided system for the early detection of AD is important to help patients with AD. This study proposes a new approach for classifying disease stages. First, we worked on the MRI images and split them into an appropriate format to avoid data leakage. Subsequently, a simple and fast registration-free preprocessing pipeline was applied to the dataset. Numerous experiments were conducted to analyze the performances of different 3D classification architectures. Finally, an ensemble learning approach is applied to the top-performing models. The outstanding performance of the proposed method was demonstrated using augmentation of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Our proposed ensemble approach outperforms studies in literature for distinguishing between people with AD and mild cognitive impairment (MCI), and MCI and cognitive normal (CN) with an AUC score of 91.28% and 88.42%, respectively. We also targeted the multiclass task, which was marginalized in previous work, by differentiating between the three stages of the disease.