Department of Paediatrics, Monash University, Melbourne, Australia.
Department of Respiratory Medicine, Monash Children's Hospital, 246 Clayton Road, Clayton, Melbourne, Victoria, 3168, Australia.
Respir Res. 2020 Sep 29;21(1):253. doi: 10.1186/s12931-020-01523-9.
Manual auscultation to detect abnormal breath sounds has poor inter-observer reliability. Digital stethoscopes with artificial intelligence (AI) could improve reliable detection of these sounds. We aimed to independently test the abilities of AI developed for this purpose.
One hundred and ninety two auscultation recordings collected from children using two different digital stethoscopes (Clinicloud™ and Littman™) were each tagged as containing wheezes, crackles or neither by a pediatric respiratory physician, based on audio playback and careful spectrogram and waveform analysis, with a subset validated by a blinded second clinician. These recordings were submitted for analysis by a blinded AI algorithm (StethoMe AI) specifically trained to detect pathologic pediatric breath sounds.
With optimized AI detection thresholds, crackle detection positive percent agreement (PPA) was 0.95 and negative percent agreement (NPA) was 0.99 for Clinicloud recordings; for Littman-collected sounds PPA was 0.82 and NPA was 0.96. Wheeze detection PPA and NPA were 0.90 and 0.97 respectively (Clinicloud auscultation), with PPA 0.80 and NPA 0.95 for Littman recordings.
AI can detect crackles and wheeze with a reasonably high degree of accuracy from breath sounds obtained from different digital stethoscope devices, although some device-dependent differences do exist.
手动听诊检测异常呼吸音的观察者间可靠性较差。具有人工智能（AI）的数字听诊器可以提高这些声音的可靠检测能力。我们旨在独立测试为此目的开发的 AI 的能力。
使用两种不同的数字听诊器（Clinicloud™ 和 Littman™）收集了 192 份来自儿童的听诊录音，由儿科呼吸医师根据音频播放以及仔细的频谱图和波形分析，将每份录音标记为包含喘鸣音、爆裂音或两者都不包含，其中一部分由盲法第二临床医生进行验证。这些录音被提交给一个盲法 AI 算法（StethoMe AI）进行分析，该算法专门用于检测病理性儿科呼吸音。
在优化的 AI 检测阈值下，Clinicloud 记录的爆裂音检测阳性百分比一致率（PPA）为 0.95，阴性百分比一致率（NPA）为 0.99；Littman 采集的声音 PPA 为 0.82，NPA 为 0.96。喘鸣音检测的 PPA 和 NPA 分别为 0.90 和 0.97（Clinicloud 听诊），Littman 记录的 PPA 为 0.80，NPA 为 0.95。
AI 可以从不同数字听诊器设备获得的呼吸音中以相当高的准确度检测爆裂音和喘鸣音，尽管存在一些与设备相关的差异。
