Department of Otolaryngology Head and Neck Surgery, Far Eastern Memorial Hospital, Taipei, Taiwan.
Center of Artificial Intelligence, Far Eastern Memorial Hospital, Taipei, Taiwan.
Laryngoscope. 2024 Nov;134(11):4585-4592. doi: 10.1002/lary.31563. Epub 2024 Jun 12.
This study investigated whether artificial intelligence (AI) models combining voice signals, demographics, and structured medical records can detect glottic neoplasm from benign voice disorders.
We used a primary dataset containing 2-3 s of vowel "ah", demographics, and 26 items of structured medical records (e.g., symptoms, comorbidity, smoking and alcohol consumption, vocal demand) from 60 patients with pathology-proved glottic neoplasm (i.e., squamous cell carcinoma, carcinoma in situ, and dysplasia) and 1940 patients with benign voice disorders. The validation dataset comprised data from 23 patients with glottic neoplasm and 1331 patients with benign disorders. The AI model combined convolutional neural networks, gated recurrent units, and attention layers. We used 10-fold cross-validation (training-validation-testing: 8-1-1) and preserved the percentage between neoplasm and benign disorders in each fold.
Results from the AI model using voice signals reached an area under the ROC curve (AUC) value of 0.631, and additional demographics increased this to 0.807. The highest AUC of 0.878 was achieved when combining voice, demographics, and medical records (sensitivity: 0.783, specificity: 0.816, accuracy: 0.815). External validation yielded an AUC value of 0.785 (voice plus demographics; sensitivity: 0.739, specificity: 0.745, accuracy: 0.745). Subanalysis showed that AI had higher sensitivity but lower specificity than human assessment (p < 0.01). The accuracy of AI detection with additional medical records was comparable with human assessment (82% vs. 83%, p = 0.78).
Voice signal alone was insufficient for AI differentiation between glottic neoplasm and benign voice disorders, but additional demographics and medical records notably improved AI performance and approximated the prediction accuracy of humans.
NA Laryngoscope, 134:4585-4592, 2024.
本研究旨在探讨结合语音信号、人口统计学资料和结构化病历的人工智能（AI）模型是否能从良性嗓音障碍中检测出声带肿瘤。
我们使用了一个包含 60 例经病理证实的声带肿瘤（即鳞状细胞癌、原位癌和发育不良）患者和 1940 例良性嗓音障碍患者的 2-3 秒元音“啊”、人口统计学资料和 26 项结构化病历（如症状、合并症、吸烟和饮酒、嗓音需求）的主要数据集。验证数据集包括 23 例声带肿瘤患者和 1331 例良性疾病患者的数据。AI 模型结合了卷积神经网络、门控循环单元和注意力层。我们使用 10 折交叉验证（训练-验证-测试：8-1-1），并在每个折层中保留肿瘤和良性疾病之间的百分比。
使用语音信号的 AI 模型的 ROC 曲线下面积（AUC）值为 0.631，额外的人口统计学资料将其提高到 0.807。当结合语音、人口统计学资料和病历时，AUC 达到了最高值 0.878（敏感性：0.783，特异性：0.816，准确性：0.815）。外部验证的 AUC 值为 0.785（语音加人口统计学资料；敏感性：0.739，特异性：0.745，准确性：0.745）。亚分析表明，AI 的敏感性高于人类评估，但特异性较低（p<0.01）。添加病历后，AI 的检测准确性与人类评估相当（82%比 83%，p=0.78）。
仅语音信号不足以通过 AI 区分声带肿瘤和良性嗓音障碍，但额外的人口统计学资料和病历显著提高了 AI 的性能，并且与人类的预测准确性相当。
无。喉镜，134：4585-4592，2024。
