TITLE:
A Lightweight MobileViT with a Dual-Path Attention Mechanism for MRI Image Classification
AUTHORS:
Youji Xu, Siyu Xiang, Huifang Feng
KEYWORDS:
MRI Image Classification, MobileViT, Attention Mechanism, Data Enhancement, Transfer Learning, Lightweight
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.3,
March
26,
2026
ABSTRACT: Deep learning has been successfully applied in the field of medical diagnosis, and improving the accurate classification of MRI images through deep learning is important for early treatment and patient prognosis. Aiming at the current deep learning-based MRI image classification algorithms with large parameter counts and high computational complexity, a lightweight MobileViT with a dual-path attention mechanism for MRI image classification is proposed in this paper. Embedding the Convolutional Block Attention Module (CBAM) in the original MobileViT network enhances the extraction of key feature information by attending to both the channel and spatial dimensions of the feature map. A Dual-Path Attention Module (DPAM) is constructed by integrating CSPNet with the CBAM mechanism to further enhance the potential of feature extraction of the proposed model while maintaining a minimal parameter count. The proposed model also employs a transfer learning method to accelerate the learning speed of the network model on the MRI image datasets, and uses a cosine annealing algorithm to optimize the learning rate of the model during the model training process to help the model converge better. The state-of-the-art performance of the proposed model is validated on the Alzheimer’s disease and brain tumor MRI datasets, respectively. We evaluate the performance of our proposed model with the latest deep learning models. The experimental results show that the model not only substantially enhances the accuracy of MRI image classification but also exhibits reduced computational complexity, making it highly suitable for mobile devices with constrained computing resources.