ENHANCING ALZHEIMER'S DISEASE DIAGNOSIS: A NOVEL LOW-HRNET APPROACH FOR IMPROVED CLASSIFICATION

Alzheimer's disease is a progressive neurodegenerative disease, primarily characterized by the gradual loss of cognitive and memory functions, which places a significant burden on patients and their families. Therefore, early diagnosis and effective treatment are particularly important. The aim of this paper is to explore the improvement in classification accuracy for AD (Alzheimer's Disease) patients by enhancing the artificial neural network structure and introducing a new attention mechanism without increasing the depth of the CNN network. A novel network structure called Low-HRNet (high-resolution network) is proposed, which combines a HRNet with an inception module for more efficient information extraction. By cross-fertilizing feature information at different resolutions, the computational cost is reduced while preserving high-resolution features. Additionally, the introduction of the Dual Attention mechanism further enhances the feature extraction capability. The results of the experiments demonstrate that the enhanced Low-HRNet reached an accuracy of 94.53% in classifying Alzheimer's disease (AD) versus NC (normal control), representing a significant improvement compared to the 88.14% accuracy achieved by the conventional CNN network. The experimental results show that the accuracy of automatic classification of Alzheimer's disease can be effectively improved by optimising the network structure and introducing an advanced attention mechanism.