📄 中文摘要
帕金森病（PD）是一种慢性神经退行性疾病，早期诊断对于减缓患者生活质量的进行性恶化至关重要。该疾病最显著的运动症状在早期阶段非常轻微，导致诊断困难。近期研究表明，利用患者语音记录有助于早期诊断。尽管从临床角度来看，分析这类记录成本较高，但技术进步使得通过分析语音特征来识别帕金森病成为可能。研究通过提取语音记录中的多种特征，如梅尔频率倒谱系数（MFCCs）、音高、抖动和颤音等，来构建分类模型。这些特征能够捕捉到帕金森病患者常见的语音障碍，例如发音模糊、音量减弱、语速变慢以及音调和响度的不稳定性。通过对这些语音特征进行深入分析，可以识别出与疾病相关的独特模式。例如，MFCCs可以表征语音的频谱包络，反映发音器官的运动协调性；音高和其变异性则能揭示声带功能和神经肌肉控制的异常。研究通常采用机器学习算法，如支持向量机（SVM）、随机森林或深度学习模型，对这些特征进行训练和分类。通过交叉验证等方法评估模型的性能，旨在实现高准确率、灵敏度和特异性，从而为帕金森病的早期筛查和辅助诊断提供非侵入性、客观的工具。这种方法有望降低诊断门槛，提高诊断效率，并为患者提供更及时的干预。
📄 English Summary
Analysis of voice recordings features for Classification of Parkinson's Disease
Parkinson's disease (PD) is a chronic neurodegenerative disorder where early diagnosis is crucial to mitigate progressive deterioration in patients' quality of life. The most characteristic motor symptoms are very subtle in the early stages, making diagnosis challenging. Recent studies have demonstrated that utilizing patient voice recordings can assist in early diagnosis. While the analysis of such recordings can be clinically costly, advancements in technology enable the identification of Parkinson's disease through the analysis of vocal features. Research involves extracting various features from voice recordings, such as Mel-frequency cepstral coefficients (MFCCs), pitch, jitter, and shimmer, to construct classification models. These features are capable of capturing common speech impairments observed in PD patients, including dysarthria, reduced vocal loudness, slowed speech rate, and instability in pitch and loudness. In-depth analysis of these vocal features allows for the identification of unique patterns associated with the disease. For instance, MFCCs characterize the spectral envelope of speech, reflecting the coordination of articulatory organs; pitch and its variability can reveal abnormalities in vocal cord function and neuromuscular control. Machine learning algorithms, such as Support Vector Machines (SVMs), Random Forests, or deep learning models, are typically employed to train and classify these features. Model performance is evaluated using methods like cross-validation, aiming for high accuracy, sensitivity, and specificity. This approach seeks to provide a non-invasive, objective tool for early screening and auxiliary diagnosis of Parkinson's disease, potentially lowering diagnostic barriers, improving diagnostic efficiency, and enabling more timely interventions for patients.