Classification of Alzheimer's Disease Based on Multi-view Lightweight Neural Network

As a degenerative neurodegenerative disease, early detection and intervention are still the key to prevent the aggravation of Alzheimer's disease (AD). Computer aided diagnosis based on Magnetic Resonance Imaging (MRI) can provide assistance for early detection of AD. In the application of limited MRI data, a multi-view lightweight neural network is proposed to effectively identify AD by reducing the amount of model computation and parameters and making full use of the image information. In the proposed method, we design an optimized lightweight neural network for feature extraction on limited MRI data. Attention mechanism is adopted to fuse multi-slice features and ensure the consistency of results. For full use of MRI data, personalized weights are learned and applied to fuse the multi-view features of transverse plane, sagittal plane and coronal plane. The experimental results indicate that compare with state-of-the-art lightweight methods, the proposed multi-view method achieves good balance on parameters (3.75M) and FLOPs (4.45G). Meanwhile, the classification performances of the proposed multi-view method are better in the three AD classification tasks. Through the construction of the model, it is expected to provide a feasible reference lightweight network for further AD detection.