The College of Information Science and Engineering, Northeastern University, Wenhua Road 3-11, Shenyang, 110819, Liaoning, PR China; The Department of Electrical Engineering and Computer Science, Vanderbilt University, 2301 Vanderbilt Place, Nashville, 37235, TN, USA.
Vanderbilt University Medical Center, 1211 Medical Center Drive, Nashville, 37232, TN, USA.
Comput Biol Med. 2023 Nov;166:107534. doi: 10.1016/j.compbiomed.2023.107534. Epub 2023 Sep 29.
It remains hard to directly apply deep learning-based methods to assist diagnosing essential tremor of voice (ETV) and abductor and adductor spasmodic dysphonia (ABSD and ADSD). One of the main challenges is that, as a class of rare laryngeal movement disorders (LMDs), there are limited available databases to be investigated. Another worthy explored research question is which above sub-disorder benefits most from diagnosis based on sustained phonations. The question is from the fact that sustained phonations can help detect pathological voice from healthy voice.
A transfer learning strategy is developed for LMD diagnosis with limited data, which consists of three fundamental parts. (1) An extra vocally healthy database from the International Dialects of English Archive (IDEA) is employed to pre-train a convolutional autoencoder. (2) The transferred proportion of the pre-trained encoder is explored. And its impact on LMD diagnosis is also evaluated, yielding a two-stage transfer model. (3) A third stage is designed following the initial two stages to embed information of pathological sustained phonation into the model. This stage verifies the different effects of applying sustained phonation on diagnosing the three sub-disorders, and helps boost the final diagnostic performance.
The analysis in this study is based on clinician-labeled LMD data obtained from the Vanderbilt University Medical Center (VUMC). We find that diagnosing ETV shows sensitivity to sustained phonation within the current database. Meanwhile, the results show that the proposed multi-stage transfer learning strategy can produce (1) accuracy of 65.3% on classifying normal and other three sub-disorders all at once, (2) accuracy of 85.3% in differentiating normal, ABSD, and ETV, and (3) accuracy of 77.7% for normal, ADSD and ETV. These findings demonstrate the effectiveness of the proposed approach.
将基于深度学习的方法直接应用于辅助诊断原发性震颤性嗓音（ETV）和外展、内收痉挛性发音障碍（ABSD 和 ADSD）仍然具有挑战性。主要挑战之一是，作为一类罕见的喉部运动障碍（LMD），可供研究的可用数据库有限。另一个值得探索的研究问题是，在基于持续发声的诊断中，上述哪种亚障碍获益最大。这个问题源于持续发声可以帮助从健康声音中检测到病理性声音。
提出了一种基于有限数据的 LMD 诊断的迁移学习策略，该策略由三个基本部分组成。（1）从国际英语方言档案（IDEA）中使用额外的语音健康数据库来预训练卷积自动编码器。（2）探索预训练编码器的转移比例。并评估其对 LMD 诊断的影响，得到两阶段转移模型。（3）在初始的两个阶段之后设计第三个阶段，将病理持续发声的信息嵌入到模型中。这一阶段验证了应用持续发声对诊断三种亚障碍的不同效果，有助于提高最终诊断性能。
本研究的分析基于从范德比尔特大学医学中心（VUMC）获得的临床医生标记的 LMD 数据。我们发现，在当前数据库中，诊断 ETV 对持续发声敏感。同时，结果表明，所提出的多阶段迁移学习策略可以产生（1）一次性对正常和其他三种亚障碍进行分类的准确率为 65.3%，（2）正常、ABSD 和 ETV 之间的准确率为 85.3%，以及（3）正常、ADSD 和 ETV 的准确率为 77.7%。这些发现证明了所提出方法的有效性。
