Early diagnosis of Parkinson's disease using Continuous Convolution Network: Handwriting recognition based on off-line hand drawing without template.

The examination of patients' handwriting has become an important auxiliary method for the diagnosis and treatment of Parkinson's disease which can be used for early self-diagnosis of patients with Parkinson's disease. However, at present, the recognition of writing disorders based on artificial intelligence technology mainly relies on pattern templates and intelligent dynamic acquisition equipment, which has some design limitations. And professional acquisition equipment is not suitable for ordinary home patients. In order to facilitate the diagnosis of Parkinson's disease and get more accurate diagnostic results, this paper is devoted to studying various features of spiral hand drawing of Parkinson's disease and developing an auxiliary diagnosis scheme based on hand drawing. Firstly, through the ablation experiment with open dataset, it is verified that the visual information of hand drawing can better reflect the characteristics of hand drawing of patients with Parkinson's disease than the original dynamic information. Secondly, an Archimedes spiral hand drawing dataset is established that can accurately reflect the tremor, shape and spacing characteristics of the image, with no limitation of the application scenario. Finally, Continuous Convolution Network (CC-Net) is proposed to reduce the pooling layer. Compared with the traditional classification network, CC-Net can accurately extract diversified features of hand drawings and maximize the retention of image information, and obtain a higher classification accuracy with qualified stability (the classification accuracy on the dataset of this paper is about 89.3%, MCC is about 0.733, and average AUC is about 0.934).