Automated classification of Alzheimer's disease based on deep belief neural networks

• Published in: Neural computing & applications Vol. 36; no. 13; pp. 7405 - 7419
• Main Authors: Nanthini, K., Tamilarasi, A., Sivabalaselvamani, D., Suresh, P.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.05.2024; Springer Nature B.V
• Subjects: Alzheimer's disease; Artificial Intelligence; Belief networks; Bioinformatics; Brain; Classification; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Datasets; Feature selection; Image contrast; Image Processing and Computer Vision; Machine learning; Multilayers; Neural networks; Original Article; Prediction models; Principal components analysis; Probability and Statistics in Computer Science; Deep belief network; Alzheimer's disease; Two-tiered feature selection; Classification
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-024-09468-6

When it comes to the causes of dementia, Alzheimer's disease is the most mysterious. There is no central genetic component connected to Alzheimer's disease. Previous approaches and tools for determining Alzheimer's disease genetic risk factors are unreliable. The brain images provided the bulk of the available information. In contrast, large-scale approaches in bioinformatics have seen significant development in recent years. It has encouraged efforts to identify the hereditary risk factors for developing Alzheimer's disease. A large amount of data on the brain's prefrontal cortex as a consequence of recent studies has allowed for the creation of classification and prediction models for Alzheimer's disease. Using the OASIS-4 dataset, which suffers from High Dimension Low Sample Size (HDLSS) problems, a Deep belief network with a Restricted Boltzmann Machine (RBM)-based classification model for processing multimodal data has been constructed. The multi-layer feature selection procedure that took into account both the technical and biological aspects of the characteristics to solve the HDLSS problem has been proposed. In molecular-level information, in the first stage of the two-tiered feature selection method, abnormal places in the dataset are found. Second, combining multiple different feature selection methods is used to refine the set of candidate genes. The principal component analysis is used for dimensionality reduction in MRI, and well pre-processed cognitive assessment scores like MMSE and ADA-cog are considered. Deep belief networks with multiple RBM are used to do unsupervised feature learning. Fivefold cross-validation has been used in all classification studies.