基于MLP的帕金森震颤评级系统

Parkinson’s Tremor Rating System Based on MLP

摘要: 目前测量帕金森严重程度的主要方法是医生对照UPDRS量表衡量，存在一定的主观性，为此本文研究出了一种用于衡量帕金森严重程度的评级系统。该系统由可穿戴设备、APP、云平台以及评估模型四部分构成。其中可穿戴设备通过三轴加速度传感器以及蓝牙模块采集震颤加速度传输至APP中进行存储、可视化与上传至云平台中，在平台获取数据后进行预处理并训练模型。通过中值滤波、小波去噪以及信号修正等算法进行数据预处理之后，通过多层感知机对处理后的数据进行学习与分类，最后得到系统的准确率为98%，召回率为96%，F1值为96%。经过实验验证，该系统能较好地反应出患者帕金森震颤的严重程度，为医生的后续判断与治疗提供了重要依据。

Abstract: At present, the main method of measuring the severity of Parkinson’s is that doctors measure it against the UPDRS scale. There is a certain degree of subjectivity. For this reason, this article has developed a rating system to measure the severity of Parkinson’s. The system consists of four parts: wearable device, APP, cloud platform and evaluation model. Among them, the wearable device col-lects the tremor acceleration through the three-axis acceleration sensor and the Bluetooth module and transmits it to the APP for storage, visualization and upload to the cloud platform. After the platform obtains the data, it preprocesses and trains the model. After data preprocessing through median filtering, wavelet denoising and signal correction algorithms, the processed data are learned and classified by a multi-layer perceptron. Finally, the accuracy rate of the system is 98%, and the recall rate is 96%. The F1 value is 96%. After experimental verification, the system can better reflect the severity of the patient’s Parkinson’s tremor, providing an important basis for the doctor’s follow-up judgment and treatment.