Explainable AI Framework for Alzheimer’s Diagnosis Using Convolutional Neural Networks

Alzheimer’s disease (AD) stands as a form of dementia characterized by the gradual degeneration of brain cells, resulting in compromised memory, cognitive functions, and the loss of fundamental skills, ultimately leading to fatality. While a definitive cure for AD remains elusive, early detection plays a pivotal role in managing its progression and enhancing the quality of life for patients. This study delves into the realm of Alzheimer’s disease identification through the application of various Neural Network models employing classification techniques. Leveraging a contemporary hybrid dataset, the investigation yielded four distinct classifications. Moreover, the study delved into elucidating the specific brain regions contributing to each classification using the Grad-CAM (Gradient-weighted Class Activation Mappings) based XAI (eXplainable Artificial Intelligence) framework applied to patients’ MRI images. A comprehensive assessment was conducted on pre-trained deep neural networks, particularly focusing on Convolutional Neural Network (CNN) models trained exclusively on authentic MRIs and a combination of authentic and synthetic MRIs. The efficacy of deep learning in disease detection was exemplified, with the CNN model trained on both real and synthetic MRIs outperforming its counterpart trained solely on real MRIs. The former achieved an impressive accuracy of 97.50%, a Balanced Accuracy Score (BA) of $98.58 \%$, and a Matthew’s Correlation Coefficient (MCC) of 95.95%. In contrast, the model trained exclusively on real MRIs exhibited an accuracy of $88.98 \%$, a BA of 94.01%, and an MCC of $83.67 \%$.