Scholarlink

<p>Parkinson’s disease is a prevalent global neurodegenerative condition, often accompanied by dysarthria, a speech disorder resulting from neurological damage. Dysarthria manifests as weakness, slowness, incoordination, or altered tone in speech muscles, leading to reduced or impaired intelligibility. Conventional assessment of dysarthria severity depends on subjective auditory evaluations, which are time-consuming and require trained specialists. This paper proposes an automated and objective framework for classifying the severity of speech disorders in PD patients using acoustic features and machine learning. Speech recordings from 221 PD patients at the University Hospital of Kiel, Germany, were analyzed. The classification process comprises two phases: frequency-domain feature extraction utilizing linear predictive coding (LPC), mel-frequency cepstral coefficients (MFCCs), and discrete wavelet packet decomposition (DWPD), followed by feature matching using <i>k</i>-nearest neighbours (KNN), support vector machines (SVM), and neural networks (NN). A novel segment-to-subject voting strategy was introduced, with which voice records are segmented into short of 2 s segments that are individually analyzed and classified, and final subject-level predictions are derived by majority voting across segments. This ensemble-style approach significantly improved accuracy, using DWPD with KNN yielded the best results, achieving an accuracy of 89.7% accuracy and a macro-F1 of 89.3%in a three-class (mild, moderate, severe) system. Successful implementation of this methodology in real clinical settings promises to improve the treatment of speech disorders in PD patients and enhance the effectiveness of speech therapies by providing an objective, efficient, reliable, and easy-to-understand assessment tool. It can help clinicians monitor patients more effectively and support faster, more consistent evaluations, particularly during therapy follow-ups and in telemedicine applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Classification of the Severity of Speech Disorders in Patients with Parkinson’s Disease Using Machine Learning Techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月24日

查看全文
  • Muna Alqam,
  • Abdulnasir Hossen,
  • Salima Al-Abri,
  • Rami Al-Hmouz,
  • Muhammet Deveci

摘要

Parkinson’s disease is a prevalent global neurodegenerative condition, often accompanied by dysarthria, a speech disorder resulting from neurological damage. Dysarthria manifests as weakness, slowness, incoordination, or altered tone in speech muscles, leading to reduced or impaired intelligibility. Conventional assessment of dysarthria severity depends on subjective auditory evaluations, which are time-consuming and require trained specialists. This paper proposes an automated and objective framework for classifying the severity of speech disorders in PD patients using acoustic features and machine learning. Speech recordings from 221 PD patients at the University Hospital of Kiel, Germany, were analyzed. The classification process comprises two phases: frequency-domain feature extraction utilizing linear predictive coding (LPC), mel-frequency cepstral coefficients (MFCCs), and discrete wavelet packet decomposition (DWPD), followed by feature matching using k-nearest neighbours (KNN), support vector machines (SVM), and neural networks (NN). A novel segment-to-subject voting strategy was introduced, with which voice records are segmented into short of 2 s segments that are individually analyzed and classified, and final subject-level predictions are derived by majority voting across segments. This ensemble-style approach significantly improved accuracy, using DWPD with KNN yielded the best results, achieving an accuracy of 89.7% accuracy and a macro-F1 of 89.3%in a three-class (mild, moderate, severe) system. Successful implementation of this methodology in real clinical settings promises to improve the treatment of speech disorders in PD patients and enhance the effectiveness of speech therapies by providing an objective, efficient, reliable, and easy-to-understand assessment tool. It can help clinicians monitor patients more effectively and support faster, more consistent evaluations, particularly during therapy follow-ups and in telemedicine applications.