Automated Early-Stage Dementia Screening: A Multi-View MRI Analysis Platform with AI-Generated Clinical Insights and Similarity-Based Validation

Jada Sukanya¹, Immidisetty Giresh¹, Ashapu Mohan¹, Chodavarapu Revanth Kumar¹, P. Ramya²

¹,²Department of Information Technology, MVGR College of Engineering (A), Vizianagaram, India

Abstract

Alzheimer’s Disease (AD) stands among the most clinically significant and socially burdensome neurological conditions worldwide, eroding cognitive function, personal identity, and independent living in affected individuals. Despite rising global incidence, early-stage diagnostic capability remains constrained by resource demands, procedural complexity, and the scarcity of specialised clinical expertise. This paper proposes a deep learning framework capable of automatically interpreting MRI brain scans to determine Alzheimer’s Disease progression stage, eliminating the need for per-image expert review.

The system is built around ResNet18, a proven deep neural network architecture that learns complex visual patterns from images using residual connections. It is trained on the publicly available OASIS neuroimaging dataset, which contains labelled brain MRI images spanning four disease stages: non-demented, very mild, mild, and moderate. After training for 15 epochs with weighted sampling to handle class imbalance, the model achieves an overall validation accuracy of 77%, with particularly strong recall for Moderate Dementia (96.7%) and an F1-score of 0.87 for Non-Demented cases. The system is designed not just as a classifier, but as a complete diagnostic tool — generating AI-based insights, heatmap visualisations of key brain regions, and downloadable clinical reports that healthcare professionals can use in real-world settings.

Keywords—Alzheimer’s Disease; Deep Learning; Residual Neural Network (ResNet18); MRI Classification; Transfer Learning; OASIS Dataset; Medical Image Analysis; Convolutional Neural Networks; Early Diagnosis.