Yan Yuyang, van Bemmel Loes, Franssen Frits M E, Simons Sami O, Urovi Visara
Institute of Data Science, Maastricht University, Paul-Henri Spaaklaan 1, Maastricht, 6229 EN, The Netherlands.
Department of Respiratory Medicine, NUTRIM Research Institute of Nutrition and Translational Research in Metabolism, Faculty of Health Medicine and Life Sciences, Maastricht University, P. Debyelaan 25, Maastricht, 6229 HX, The Netherlands.
Comput Methods Programs Biomed. 2025 Aug;268:108796. doi: 10.1016/j.cmpb.2025.108796. Epub 2025 Apr 30.
Deteriorations in respiratory health, also known as exacerbations, are important events in the progression of chronic respiratory diseases such as Chronic Obstructive Pulmonary Disease (COPD) and asthma. Changes in vocal characteristics during episodes of respiratory distress suggest that voice analysis could be a valuable tool for monitoring exacerbations. This study aims to develop a remote monitoring method for automatically detecting exacerbations in COPD and asthma patients using only speech data.
This study proposes a speech-based approach for remote monitoring of asthma and COPD exacerbations, leveraging optimized Mel-Frequency Cepstral Coefficients (MFCC) alongside multi-domain acoustic features. We demonstrate that the optimized MFCC outperforms state-of-the-art feature extraction techniques, while integrating complementary features from the time, frequency, energy, and spectral domains further enhances predictive accuracy. To ensure model transparency and facilitate clinical adoption, we employ SHapley Additive exPlanations (SHAP) to identify key speech biomarkers contributing to exacerbation detection.
Compared with the state-of-the-art methods, our method exhibits excellent classification performance with an accuracy of 0.892 and an AUC of 0.955 on the TACTICAS dataset. Moreover, the most salient features ranked by SHAP values are MFCC-related features and energy features, which explains the reason behind the improvement observed with feature fusion.
Comprehensive experiments and comparisons with existing algorithms highlight the potential of speech-based monitoring for respiratory conditions in real-world settings. The proposed method outperforms state-of-the-art approaches, offering a promising avenue for exacerbation diagnosis and monitoring while potentially reducing the burden on both patients and healthcare providers.
呼吸健康恶化，也称为急性加重，是慢性阻塞性肺疾病（COPD）和哮喘等慢性呼吸道疾病进展过程中的重要事件。呼吸窘迫发作期间声音特征的变化表明，语音分析可能是监测急性加重的一种有价值的工具。本研究旨在开发一种仅使用语音数据自动检测COPD和哮喘患者急性加重的远程监测方法。
本研究提出了一种基于语音的方法，用于远程监测哮喘和COPD急性加重，利用优化的梅尔频率倒谱系数（MFCC）以及多域声学特征。我们证明，优化后的MFCC优于现有最先进的特征提取技术，同时整合来自时间、频率、能量和频谱域的互补特征可进一步提高预测准确性。为确保模型的透明度并促进临床应用，我们采用夏普利值加法解释（SHAP）来识别有助于急性加重检测的关键语音生物标志物。
与现有最先进的方法相比，我们的方法在TACTICAS数据集上表现出优异的分类性能，准确率为0.892，曲线下面积（AUC）为0.955。此外，根据SHAP值排名的最显著特征是与MFCC相关的特征和能量特征，这解释了特征融合后观察到的性能提升背后的原因。
综合实验以及与现有算法的比较突出了基于语音的监测在实际环境中用于呼吸状况的潜力。所提出的方法优于现有最先进的方法，为急性加重的诊断和监测提供了一条有前景的途径，同时可能减轻患者和医疗服务提供者的负担。
