Development Of A Serverless, Interactive Application For Alzheimer’s Disease Detection And Visualization Using Mri Images

Abstract

This study presents the development of a serverless, interactive application for Alzheimer's Disease detection and visualization using MRI images. The project combines advanced deep learning techniques with efficient cloud deployment strategies to address the challenges of early AD diagnosis. A multi-view MRI analysis approach was implemented using neural architecture search to develop AD detection models. Two architectures were explored: chain-based and cellbased. The chain-based model, despite incorporating advanced components like Vision Transformers, underperformed with an accuracy of 46.0%. In contrast, the cell-based architecture emerged as superior, achieving an accuracy of 86.7% and an AUC of 0.900 in distinguishing between cognitively normal, mild cognitive impairment, and AD cases, with only 1.7 million parameters. A web application was developed using AWS serverless services (Lambda, S3, DynamoDB) for model deployment, offering features for patient management, image visualization, and AD prediction. The application demonstrated high performance in a clinical setting, with an accuracy of 93.3% on a 30-sample test. Cost analysis revealed extremely low operational expenses. For a hospital managing 10,000 patients with 10 MRI scans each, the monthly cost is estimated at approximately 27,000 VND per 100 scans beyond the free tier, with negligible database costs. While challenges persist, particularly in MCI classification, this integrated approach shows promise in enhancing early AD diagnosis and management. The combination of an accurate, lightweight model with a cost-effective, scalable deployment solution addresses critical needs in healthcare technology adoption, especially in developing countries like Vietnam.