【作者】 罗正潮； 刘蔚； 余越； 夏晓婷； 杨柳； 陈铖； 朱丹丹； 陈凌； 付艳；

【Author】 Zhengchao Luo;Wei Liu;Yue Yu;Xiaoting Xia;Liu Yang;Cheng Chen;Dandan Zhu;Ling Chen;Yan Fu;Huazhong University of Science and Technology;

【机构】 华中科技大学；

【摘要】 帕金森疾病(PD)是中老年人常见的中枢神经系统变性疾病。传统的PD诊断步骤十分复杂,且患者多在出现明显的运动障碍后就诊,导致PD普遍发现较晚。多数PD患者会在呼吸、发音、调音和韵律方面出现构音障碍,且发生在患者出现明显的运动障碍5年之前。基于此,本文提出基于机器学习算法,通过中文语音识别中国PD患者病情,从而实现帕金森疾病的早期诊断,同时基于此原理开发的器件将为患者提供一个方便的自我诊断手段。本文采用的语音数据为包含元音发音和连续语句的中文数据集,与传统元音发音数据集相比,能更有效地检测处语音损伤。我们通过发声分析、韵律分析和轮替运动分析以提取能反映患者生理状况的语音特征。通过对比KNN、随机森林、LS、CART、感知机、XGBoost六种分类算法,我们验证了XGBoost在中文PD识别中的优势。此工作为后续PD自我诊疗平台的开发工作奠定了基础。更多还原

【Abstract】 Parkinson’s disease(PD) is a common central nervous system degenerative disease in middle-aged and elderly people. The traditional PD diagnosis steps are very complicated, and patients usually go to the doctor after obvious movement disorders, which leads to the late detection of PD. Most patients with PD will have dysarthria in breathing, pronunciation, tuning, and rhythm, and it occurs 5 years before the patient has obvious dyskinesias. Based on this, this paper proposes a machine learning algorithm to recognize the condition of Chinese PD patients through the detection of Chinese speech, so as to realize the early diagnosis of Parkinson’s disease. At the same time, the device developed based on this principle will provide patients with a convenient self-diagnosis method. The speech data used in this paper is a Chinese data set containing vowel pronunciation and continuous sentences. Compared with the traditional vowel pronunciation data set, it can detect local speech damage more effectively. We use vocal analysis, prosody analysis, and rotation motion analysis to extract voice features that can reflect the patient’s physiological condition. By comparing the six classification algorithms of KNN, Random Forest, LS, CART, Perceptron, and XGBoost, we verified the advantages of XGBoost in Chinese PD recognition. This work lays the foundation for the development of the self-diagnosis and treatment platform for people with PD.更多还原