Diagnosis and classification of Alzheimer's disease from MRI images using parallel deep convolutional neural networks

Abstract

Alzheimer's disease, an incurable neurological condition affecting memory, predominantly in the elderly, necessitates precise early diagnosis. While Convolutional Neural Networks (CNNs) are effective in classifying Alzheimer's, they may suffer from overfitting due to random feature collection. To address this, a novel parallel deep Convolutional neural network (PDCNN) architecture is proposed. This architecture extracts global and local features through two parallel paths, mitigating overfitting using dropout regularization and batch normalization. Initial steps involve resizing input images and grayscale transformation for complexity reduction, followed by data augmentation to enhance dataset size. The parallel paths leverage two deep CNNs with identical window sizes, enabling the model to learn both local and global information. Evaluation on a multi-class Kaggle dataset showcases the method's efficacy, achieving 100% accuracy on training data and 98.927% on validation data. The proposed approach not only ensures accuracy but also efficiency by extracting low-level and high-level features, outperforming existing techniques.