A pipelined, resource-efficient convolutional neural network architecture for detecting and diagnosing Alzheimer's...

Alzheimer's disease (AD) is a progressive neurological disorder that impairs memory and cognitive function in elderly individuals. Early detection is vital to slow disease progression and enable timely therapeutic intervention. Traditional diagnostic approaches for AD, however, often involve high time complexity and significant computational resource utilization, highlighting the need for more efficient automated solutions.

This study introduces a Resource Efficient Convolutional Neural Network (RECNN) framework for AD detection and diagnosis using brain MRI images. The methodology incorporates three main modules: Gabor transformation, data augmentation, and classification with an anomalous pixel segmentation algorithm. Gabor transforms are employed to enhance spatial frequency features and improve detection rates. Data augmentation techniques are applied to increase the diversity of training samples. The RECNN classifier is then used for image classification, and functional morphological segmentation is applied to classify affected pixels into mild or advanced stages of AD. Two benchmark datasets are utilized for training and testing the proposed framework.

The proposed RECNN-based system demonstrated superior detection and classification performance compared with conventional AD detection methods. The model achieved improved accuracy and robustness, with segmentation results enabling the differentiation between mild and advanced AD cases. Comparative evaluation confirmed that RECNN significantly reduces computational complexity while maintaining high diagnostic reliability.

The findings suggest that the RECNN framework offers a resource-efficient and accurate approach for AD detection using MRI data. By combining Gabor-based feature transformation, augmented data diversity, and advanced segmentation, the proposed method provides a scalable and clinically applicable tool for early diagnosis. Future work will extend the model to larger and more diverse datasets and explore hybrid architectures to further enhance diagnostic performance.