Optimizing Deep-Transfer Learning Model for Early Disease Diagnosis of Alzheimer’s in Clinical Data

Abstract

Early and precise identification of Alzheimer's disease (AD) must be addressed in order to improve patient outcomes and reduce the socioeconomic impact of this degenerative neurological condition. The subjectivity of tested processes, such as magnetic resonance imaging (MRI) decoding and cognitive evaluation, is limited by factors including cost and time constraints. This work proposes a more effective approach to deep transfer learning with the VGG-19 architecture to accurately and robustly distinguish AD using MRI images. The suggested architecture incorporates essential preprocessing techniques as data augmentation, scaling, normalization, denoising, contrast improvement, and skull stripping. These phases improve the quality of images and broaden the variety of data. The model was trained and evaluated using 10-fold cross-validation with an 80:20 split on an image dataset containing both AD and healthy controls (OASIS). The model indicates excellent accuracy, precision, recall, and F1-score of 99.22%. Compared with ML techniques, the VGG-19 is more effective at identifying AD-specific traits. Based on these findings, the proposed technique is a clinically scalable solution, an efficient and automated tool for facilitating the real-time detection of Alzheimer's disease.