A Novel Semi-supervised Neural Network for Recognizing Parkinson's Disease

Zhehao Zhang, Xiaobo Zhang, Dengmin Wen, Lilan Peng, Yuxin Zhou

Abstract: Generative adversarial networks (GAN) have been mainly applied to tasks such as synthesis, segmentation, and reconstruction in the field of medical images since its appearance, and the research results in the field of classification are relatively rare. In the field of Parkinson’s disease, the development of deep learning in this field has been limited due to the lack of available data sets and the differences between medical images and natural images. This paper proposes a new neural network for recognizing Parkinson’s disease called Triple Progressive Generative Adversarial Networks (TP-GAN). Adding a classifier makes the model change from a two-person game to a three-person game, and introduces a manifold regularization method to guide the direction of classification decisions. The use of progressive networks to replace the traditional convolutional network makes the model perform better than the original network structure when processing large resolution data. The exprimental results demonstrate that our model performs better than the state-of-the-art baselines on the dataset of brain Magnetic Resonance Imaging (MRI).