DETECTION OF DEMENTIA VIA HYBRID CONVOLUTIONAL NEURAL NETWORKS: A COMPREHENSIVE LITERATURE REVIEW OF OASIS AND ADNI DATASET APPLICATIONS

Abstract

Background: The application of hybrid convolutional neural networks (CNNs) for dementia detection has emerged as a promising approach in medical neuroimaging, particularly when large-scale datasets such as the Open Access Series of Imaging Studies (OASIS) and the Alzheimer’s Disease Neuroimaging Initiative (ADNI) are utilized.

Objective: This comprehensive literature review examines recent developments (2020--2025) in hybrid CNN architectures for multitype dementia classification, with an emphasis on model performance metrics, architectural specifications, and preprocessing methodologies.

Methods: We conducted a systematic search of multiple databases, including PubMed, IEEE Xplore, SciSpace, and ArXiv, with a focus on papers published between 2020 and 2025. Our analysis included 429 initially identified papers, with 156 papers meeting the inclusion criteria for detailed analysis.

Results: Hybrid architectures that combine CNNs with vision transformers, ensemble methods, and multimodal approaches achieve superior performance, with reported accuracies ranging from 85% to 99% across different dementia classification tasks. The best-performing system (DenseNet-ViT fusion) achieved 98.7% accuracy on the OASIS dataset. However, significant variations in preprocessing pipelines, evaluation protocols, and dataset splits highlight challenges in standardization.

Conclusions: Hybrid CNN approaches represent the current state-of-the-art for automated dementia detection, with CNN-transformer hybrids showing particular promise. Future research should focus on standardized evaluation protocols, cross-dataset validation, and clinical translation.