DEAM-XCN: Alzheimer’s disease detection using distributed ensemble activation function based explainable convolutional neural network

Abstract

Alzheimer’s disease (AD) is the most common disabling neurological disorder that damages brain tissue and results in loss of memory, disorientation, and neural death. This incurable disease claims the lives of numerous individuals each year. However, early detection plays a crucial role in mitigating its progression. Recent research faced challenges such as low accuracy, issues with overfitting, and the time-consuming nature of manual diagnoses. To address this issue, a distributed ensemble activation function module-based explainable convolutional neural network (DEAM-XCN) is developed for efficient Alzheimer’s disease detection. The developed distributed ensemble attention module is a hybrid of channel attention module (CAM) and position attention module (PAM), which fine-tunes the ensemble convolutional neural network classifier. Channel attention module (CAM) emphasizes diverse channels or features, enabling the model to concentrate on pertinent aspects of the data. Position attention module (PAM) emphasizes spatial relationships within the feature maps, capturing local patterns and contextual information. By combining channel and position attention, the model can integrate comprehensive information about both the importance of different features and the spatial context of those features in Alzheimer’s disease-related patterns within the brain. The combination allows the model to identify spatial abnormalities and structural variations that may be indicative of Alzheimer’s disease, contributing to more accurate detection. The purpose of an explainable convolutional neural network is to automatically detect Alzheimer’s disease effectively and enhance performance. The proposed DEAM-XCN model is evaluated with the ADNI and OASIS datasets to enhance the generalization. Moreover, the experimental results show that the proposed model achieved a high accuracy of 97.63%, MSE of 4.03, precision of 98.19%, and recall of 97.07% based on a Training percentage of 90% based on the ADNI dataset. Simultaneously, the model gained an accuracy of 97.37%, MSE of 4, precision of 97.65%, and recall of 97.09% based on a K-fold value of 10. The proposed DEAM-XCN model demonstrated more efficiency in Alzheimer’s disease detection with the ADNI dataset.