| dc.description.abstract | This thesis addresses the imperative need for bridging the gap between theoretical
machine learning frameworks and their practical clinical applications in the diagnosis of
Alzheimer's Disease (AD) through the deployment of a trained Convolutional Neural
Network (CNN) model onto a web platform. The demand for accurate and efficient
diagnostic tools for AD is on the rise, necessitating innovative solutions that can
seamlessly integrate machine learning models into clinical workflows. The research
identifies the challenge of deploying machine learning models in real-world settings and
proposes a solution by leveraging web-based platforms for AD diagnosis. The study
meticulously outlines the methodologies employed in training, validating, and deploying
the CNN model, shedding light on the intricacies of model interaction with the interface
and backend architecture. Through a comparative analysis, the performance of two
prominent models, EfficientNetB7 and Vision Transformer (ViT), is evaluated. The
results reveal that EfficientNetB7 achieves remarkable accuracy, with a validation rate of
93.5%, positioning it as a robust and deployable solution for early AD detection in clinical
settings. However, ViT, while exhibiting competitive accuracy of 87.4%, encounters
deployment challenges attributed to incompatibilities in the training environment. These
findings underscore the practical considerations and challenges associated with deploying
machine learning models in real-world healthcare settings. Overall, the thesis validates
the potential of user-friendly web applications in facilitating the deployment of
sophisticated machine learning models for medical diagnostics, setting a precedent for
future innovations in AD diagnosis and beyond. | en_US |
