Multiscale Low-Level Feature Fused Multilayer Convolution Neural Network for Alzheimer's Disease Detection

Abstract

Alzheimer's disease (AD) is an acute neurological disorder that can cause heart and respiratory dysfunction. Thus, early detection of such a disease is of the utmost importance. Recently, several researchers have developed AI-based solutions, specifically using deep learning techniques to predict AD using MRI images. Most of these approaches have adopted transfer learning techniques or developed convolutional neural networks, which suffer from limited representation when extracting fine-grained features from the MRI images, which leads to degradation of performance. The proposed model overcomes such limitations by introducing a novel multiscale low-level feature-fused multilayer convolutional neural network (CNN) for AD detection. The proposed model enhances the quality of low-level features by extracting multiscale features using a proposed multiscale low-level feature extraction module (MLLFEM). The multiscale features are known as scale-invariant features, which are inputted to a high-level feature extraction module (HLFEM). The HLFEM contains minutely designed multilayers of CNN to exploit fine-grained object-level features. These features are given to an output layer to predict categories of AD disease. A publicly available MRI image-based AD dataset is used to demonstrate the model's efficacy. The experimental analysis shows that the proposed model achieves an accuracy of 93.43%, which outperforms other state-of-the-art models. © The Institution of Engineering & Technology 2023.