3D transfer learning network for classification of Alzheimer's disease with MRI

Background As a kind of dementia, Alzheimer's disease (AD) cannot be cured once diagnosed. Hence, it is very important to diagnose early and delay the deterioration of the disease through drugs. Objective To reduce the computational complexity of conventional 3D convolutional networks, this paper uses machine learning as an auxiliary diagnosis of AD, and proposes three-dimensional (3D) transfer network which is based on two-dimensional (2D) transfer network to classify AD and normal groups with magnetic resonance imaging (MRI). Method First, the method uses a 2D transfer Mobilenet to extract features from 2D slices of MRI, and further perform dimension reduction for the extracted features. Then, all of the 2D slice features of one subject are merged to classify. Results The experiment in this paper uses an open access Alzheimer's disease database to evaluate the method. The experiment result show that the classification accuracy of the proposed 3D network is better than that of the existing 2D transfer network, increased by about 10 percentage points and the classification time is only about 1/4 of the existing one. Conclusion The proposed method is to realize the classification of 3D MRI data through an existing 2D transfer network, and it not only reduces the complexity of conventional 3D networks, but also improves the classification accuracy. Because of the shared weight of the transfer network, besides, the classification time is reduced.

Automated summary

This paper proposes a method utilizing machine learning as an auxiliary diagnosis for Alzheimer's disease (AD), and uses a three-dimensional transfer network to classify magnetic resonance imaging (MRI). The experiment shows significant improvement in classification accuracy and reduced classification time.