ALZHEIMER’S DISEASE DETECTION USING DEEP LEARNING ON MRI AND CLASSICAL ML ON CLINICAL RECORDS

Abstract

      For prompt intervention and better patient outcomes, Alzheimer's disease (AD) must be identified early. Despite recent advancements, key questions remain: how do different data modalities influence diagnostic accuracy, and can a single model consistently outperform traditional approaches across diverse datasets? This study investigates a dual-modality architecture that combines structured clinical data and medical imaging in order to overcome these issues.  A baseline Convolutional Neural Network (CNN) trained on MRI scans achieved 98.67% accuracy by effectively capturing spatial features associated with dementia stages. Simultaneously, eight traditional machine learning models—Logistic Regression, Decision Tree,K-Nearest Neighbors, Random Forest, Gaussian Naïve Bayes, AdaBoost, XGBoost, and Gradient Boosting—were evaluated on structured clinical data. To enhance diagnostic performance, we propose a novel hybrid framework that integrates an enhanced CNN for image data and for tabular data. The proposed model outperformed all baselines, achieving more than 97% classification accuracy on MRI images, 97.44% classification accuracy and an AUC of 0.98 on clinical text data. These results highlight the superior predictive capability and generalizability of the proposed method, demonstrating its potential as a robust and clinically applicable tool for early Alzheimer’s disease diagnosis.