Designing a clinical decision support system for Alzheimer’s diagnosis on OASIS-3 data set

• Published in: Biomedical signal processing and control Vol. 74; p. 103527
• Main Authors: Salami, Farzaneh, Bozorgi-Amiri, Ali, Hassan, Ghulam Mubashar, Tavakkoli-Moghaddam, Reza, Datta, Amitava
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2022
• Subjects: Alzheimer’s disease; Clinical decision support system; Deep learning; OASIS-3; Deep learning; Clinical decision support system; OASIS-3; Alzheimer’s disease
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.103527

•Performance of different 3D CNNs in diagnosing Alzheimer’s on OASIS-3 dataset.•Proposing a deep learning architecture which provides new state-of-the-art results.•Using MMSE and APOE4 as biomarkers along MRI images to improve Alzheimer’s diagnoses. Background and Objective: Alzheimer’s disease (AD) is the most common neurodegenerative disease, and its early detection is crucial for appropriate treatment. To analyse 3D-magnetic resonance imaging (MRI) data, deep learning (DL) methods have become powerful tools. In this paper, we propose a clinical decision support system (CDSS) based on DL methods for diagnosing AD using 3D-MRI images. Methods: We conducted several experiments to choose the best model to use as the backbone of our CDSS. For this purpose, we utilize existing convolutional neural network (CNN), ResNet, DenseNet, and Inception-v3 models. We evaluated the models on the recently released part 3 of the Open Access Series of Imaging Studies (OASIS-3) data set. On analysis of the results, we propose a novel network which performed better than the tested models. Results: Compared to the mentioned networks, the proposed model performed best in AD classification. The 3D image inputs alongside clinical factors in our proposed ensemble architecture increased the performance of the model. The trained version of the proposed model with a graphical interface is proposed as a CDSS to help physicians. We also used person disjoint subsets of the data to avoid reporting biased performance of our research work. Conclusions: Our results show that our proposed model significantly enhances the precision of clinical examinations and makes the process more robust. This CDSS can potentially help to identify AD subjects with high confidence. To the best of our knowledge, this is the first comprehensive work on OASIS-3 data set, with significant results.