Enhanced Computational Neural Structures for Accurate Identification of Depression and Alzheimer’s via fMRI

Abstract

This study presents a inventive approach for classing of depression and Alzheimer’s disease patients from normative group using functional Magnetic Resonance Imaging (fMRI) data and deep learning techniques. Leveraging data from the All India Institute of Medical Sciences (AIIMS) and the Alzheimer’s disease Neuroimaging Initiative (ADNI), we developed a customized Convolutional Neural Network (CNN) model to accurately categorising between participants with neurological disorders and normal group. Our approach involves preprocessing fMRI data then all fMRI images were converted into 2D PNG format to facilitate analysis. The custom CNN achieved the highest accuracy of 98.06% demonstrating optimal results over two state-of-theart pre-trained networks, including MobileNetV3 and Inception-ResNetV2 assess its efficacy and generalizability. This comparison underscores the advantages of task-specific network architectures in accurately classifying neurological conditions. The findings contribute to the development of reliable, AI-powered diagnostic tools, offering improved accuracy and clinical relevance for timely intervention of depression and Alzheimer’s disease.