Enhanced CBAMWDNet: A deep learning approach for accurate dementia multiclassification using MRI scans

Abstract

The rise in dementia cases emphasizes the critical need for accurate and early diagnosis. While numerous studies have focused on precise classification systems for singular dementia types, a gap exists in comprehensive classification encompassing various dementia subtypes. This research addresses this gap by curating a diverse MRI dataset containing multiple forms of dementia, aiming to develop a robust classification model. The research focuses on enhancing the CBAMWDNet, an advanced deep learning model, to precisely categorize different types of dementia like Alzheimer's, Lewy body, Frontotemporal and Vascular dementia. Originally developed for detecting tuberculosis in chest X-ray images, this model incorporates the architecture of Convolutional Block Attention Module (CBAM), Wide ResNet, and Dense blocks (WDnet). By leveraging a well-balanced and varied MRI dataset, the model's training will encompass a spectrum of dementia presentations, enhancing its capacity for nuanced classification. The proposed research aims not only to advance the capabilities of CBAMWDNet but also to contribute significantly to personalized medical diagnostics. Achieving accurate classification across diverse dementia subtypes holds the potential to revolutionize patient care, enabling tailored interventions and treatments based on precise subtype identification. This research thus underscores its relevance in the broader context of improving healthcare outcomes for individuals affected by dementia.