Developing Advanced Deep Learning Layers for Enhanced Automatic Feature Extraction and Higher Accuracy in Alzheimer's Disease Classification

Abstract

Alzheimer's Disease (AD) classification through medical imaging is important for early detection and planning treatment, but it's still hard to do because neuropathological changes are so minor and complicated. This work presents new layers of advanced deep learning (DL), intended to enhance automated feature extraction, hence improving the accuracy of categorisation. Conventional approaches for AD diagnosis may rely on manual identification of brain biomarkers, which can take a lot of time and vary for every observer. Our work developed a novel set of convolutional neural network (CNN) layers able to automatically identify and rank structural MRI features connected to AD, hence addressing these issues. Three well-known deep learning architectures—density network, inception network, and ResNet—were employed in this work. Every model was trained and tested using a set of MRI images from AD patients and healthy controls made accessible for public use. The upgraded layers were meant to highlight early AD symptoms like hippocampus atrophy and loss of cortical thickness. The accuracy of categorisation is shown to be much greater than with standard DL models. Showing a 12% increase above the average, our enhanced layers included into the DenseNet architecture made it the most accurate. Our models were also robust in feature extraction; they could routinely identify significant AD-related changes in photos with different sizes and brightness levels. This approach not only accelerates the classification of AD but also provides the foundation for future research on accurate and automated neurodegenerative disease testing, which can significantly influence patient outcomes and therapeutic practices.