Deep Learning Models for Discrimination of Parkinson's disease and Multiple System Atrophy on Brain Magnetic Resonance Imaging

作者：Yuji Du, Xuanyan Wu, Hao Yang, Dong Liang, Jianing Jin, Dongning Su, Chengzhang He, Yun Ling, Zhonglue Chen, Kang Ren, Tao Feng, Shan Tan · 年份：2024 · DOI：10.1145/3653644.3653648 · 研究领域：Parkinson's Disease Mechanisms and Treatments、Neurological disorders and treatments、Genetic Neurodegenerative Diseases

Parkinson's disease (PD) and Multiple System Atrophy (MSA) are both classic neurodegenerative diseases. However, the significant overlap of PD and MSA in clinical symptoms poses a substantial challenge for their diagnosis. To solve this problem, five deep learning-based 3D models are designed to classify PD and MSA using brain Magnetic Resonance Imaging (MRI) of patients. We conduct a comprehensive comparison and evaluation of these models with traditional clinical approaches. The network architecture with the best performance, i.e., ResNet34, is finally determined, which achieves an impressive accuracy of up to 91% and a sensitivity of 99%. In the experiment, we find that the performance of all five DNN methods is superior to that of both traditional clinical analysis methods in the experiment. Our study introduces an innovative method for computer-assisted diagnosis in the early detection of neurological diseases, which has important clinical significance.