An Efficient Lightweight Network Based on Magnetic Resonance Images for Predicting Alzheimer's Disease

Boan Ji, Huabin Wang, Mengxin Zhang, Borun Mao, Xuejun Li

Source Title: International Journal on Semantic Web and Information Systems (IJSWIS)18(1)

ISSN: 1552-6283|EISSN: 1552-6291|EISBN13: 9781799893967|DOI: 10.4018/IJSWIS.313715

MLA

Ji, Boan, et al. "An Efficient Lightweight Network Based on Magnetic Resonance Images for Predicting Alzheimer's Disease." IJSWIS vol.18, no.1 2022: pp.1-18. http://doi.org/10.4018/IJSWIS.313715

APA

Ji, B., Wang, H., Zhang, M., Mao, B., & Li, X. (2022). An Efficient Lightweight Network Based on Magnetic Resonance Images for Predicting Alzheimer's Disease. International Journal on Semantic Web and Information Systems (IJSWIS), 18(1), 1-18. http://doi.org/10.4018/IJSWIS.313715

Chicago

Ji, Boan, et al. "An Efficient Lightweight Network Based on Magnetic Resonance Images for Predicting Alzheimer's Disease," International Journal on Semantic Web and Information Systems (IJSWIS) 18, no.1: 1-18. http://doi.org/10.4018/IJSWIS.313715

Export Reference

Favorite Full-Issue Download

View Full Text HTML

View Full Text PDF

Abstract

Brain magnetic resonance images (MRI) are widely used for the classification of Alzheimer's disease (AD). The size of 3D images is, however, too large. Some of the sliced image features are lost, which results in conflicting network size and classification performance. This article uses key components in the transformer model to propose a new lightweight method, ensuring the lightness of the network and achieving highly accurate classification. First, the transformer model is imitated by using image patch input to enhance feature perception. Second, the Gaussian error linear unit (GELU), commonly used in transformer models, is used to enhance the generalization ability of the network. Finally, the network uses MRI slices as learning data. The depthwise separable convolution makes the network more lightweight. Experiments are carried out on the ADNI public database. The accuracy rate of AD vs. normal control (NC) experiments reaches 98.54%. The amount of network parameters is 1.3% of existing similar networks.