基于绕口令的帕金森病计算机辅助诊断系统的构建  

Construction of computer-aided diagnostic system for Parkinson’s disease based on tongue twisters

作　　者：马宇红 王小小 薛生倩 路金叶 MA Yuhong;WANG Xiaoxiao;XUE Shengqian;LU Jinye(College of Mathematics and Statistics,Northwest Normal University,Lanzhou 730070,China;Editorial Department of the University Journal,Northwest Normal University,Lanzhou 730070,China)

机构地区：[1]西北师范大学数学与统计学院,甘肃兰州730070 [2]西北师范大学学报编辑部,甘肃兰州730070

出　　处：《武汉科技大学学报》2024年第4期312-320,共9页Journal of Wuhan University of Science and Technology

基　　金：国家自然科学基金项目(51368055).

摘　　要：早期诊断对于帕金森病的治疗至关重要,语言障碍是帕金森病早期患者的一个主要症状,为此本文结合中文绕口令的语音特征和经典机器学习模型构建帕金森病计算机辅助诊断系统。通过某医院老年病门诊,以中文绕口令“八百标兵奔北坡,炮兵并排北边跑”为语音素材,采集包括67位帕金森病患者和30位非帕金森病患者的语音样本;提取每个样本14个音节的峰值、峰宽、峰面积、峰距和峰跨的平均值以及语音长度6个标志性指标,并结合患者的类别标签,建立帕金森病分类的语音特征集PPFTT;通过K-means聚类、量子聚类、谱聚类、主成分分析4种无监督学习算法验证了PPFTT在识别帕金森病患者上的有效性;应用PPFTT对BP、RF、SVM和LSTM这4个经典机器学习模型进行训练与评估,结果显示,4个模型对帕金森病患者的识别准确率分别为88.42%、91.58%、91.05%和92.64%,其中LSTM性能最佳;选取LSTM模型构建基于绕口令的帕金森病计算机辅助诊断系统,测试表明,该系统可以辅助临床医生快速、准确地识别帕金森病早期患者。Early diagnosis is crucial for the treatment of Parkinson’s disease,and language impairment is a major symptom of patients in the early stage of Parkinson’s disease.Therefore,a computer-aided diagnostic system for Parkinson’s disease was constructed on the basis of speech features of Chinese tongue twisters and classical machine learning models.Firstly,phonetic signals of 67 patients with Parkinson’s disease and 30 patients without Parkinson’s disease were collected from the geriatric cli-nic of a hospital by using Chinese tongue twister“八百标兵奔北坡,炮兵并排北边跑”as speech material.Secondly,six iconic indicators,namely the averages of peak value,peak width,peak area,peak separation and peak gap of 14 syllables in each sample as well as the speech duration,were extracted,and the speech feature set PPFTT for Parkinson’s disease classification was established by attaching the patient’s category label.Thirdly,four unsupervised learning algorithms including K-means clustering,quantum clustering,spectral clustering and principal component analysis were used to verify the effectiveness of PPFTT in identifying Parkinsonian patients.Furthermore,four classical machine learning models(BP,RF,SVM and LSTM)were trained and evaluated by PPFTT,and their recognition accuracies were 88.42%,91.58%,91.05%and 92.64%,respectively,among which LSTM had the best performance.So LSTM model was selected to construct a computer-aided diagnostic system for Parkinson’s disease based on tongue twisters.The test results demonstrate that this system can assist clinicians to diagnose patients in the early stage of Parkinson’s disease quickly and accurately.

关 键 词：帕金森病 计算机辅助诊断系统 语音信号 绕口令 特征提取 机器学习 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术] R742.5[自动化与计算机技术—计算机科学与技术]