Deep Learning-Based Diagnostic Models for Early Detection of Alzheimer's Disease Using MRI and Genetic Data

Abstract

Early detection of Alzheimer's disease (AD) is crucial for timely intervention and management. This research investigates the efficacy of deep learning-based diagnostic models using MRI imaging and genetic data for the early identification of AD. We developed three models: a Convolutional Neural Network (CNN) focused solely on MRI data, a Genomic CNN utilizing genetic information, and a Hybrid CNN integrating both modalities. Our comprehensive analysis included performance evaluations across several metrics, including accuracy, sensitivity, specificity, precision, F1 score, and AUC-ROC.

The CNN on MRI data achieved an accuracy of 89.6%, demonstrating strong capabilities in recognizing structural brain changes indicative of Alzheimer's. The Genomic CNN reached a maximum accuracy of 82.6%, highlighting the potential of genetic markers in AD detection but revealing limitations in sensitivity (80.2%). The Hybrid CNN model outperformed both standalone approaches, achieving an impressive accuracy of 91.2% and an AUC-ROC of 93.7%. These results suggest that integrating MRI and genetic data significantly enhances diagnostic performance.

Hyperparameter optimization studies revealed the importance of tuning learning rates and batch sizes, with optimal configurations leading to substantial improvements in accuracy and sensitivity across all models. Specifically, the CNN on MRI data peaked in performance at a learning rate of 0.006 and a batch size of 64.

This research underscores the potential of deep learning techniques, particularly multimodal approaches, in improving early AD diagnosis. The findings advocate for future exploration of larger datasets, additional imaging modalities, and interpretability methods to enhance clinical applicability, ultimately aiming to facilitate timely interventions for individuals at risk of Alzheimer's disease.