Benchmarking of eight recurrent neural network variants for breath phase and adventitious sound detection on a self-developed open-access lung sound database-HF_Lung_V1

Abstract

A reliable, remote, and continuous real-time respiratory sound monitor with automated respiratory sound analysis ability is urgently required in many clinical scenarios-such as in monitoring disease progression of coronavirus disease 2019-to replace conventional auscultation with a handheld stethoscope. However, a robust computerized respiratory sound analysis algorithm for breath phase detection and adventitious sound detection at the recording level has not yet been validated in practical applications. In this study, we developed a lung sound database (HF_Lung_V1) comprising 9,765 audio files of lung sounds (duration of 15 s each), 34,095 inhalation labels, 18,349 exhalation labels, 13,883 continuous adventitious sound (CAS) labels (comprising 8,457 wheeze labels, 686 stridor labels, and 4,740 rhonchus labels), and 15,606 discontinuous adventitious sound labels (all crackles). We conducted benchmark tests using long short-term memory (LSTM), gated recurrent unit (GRU), bidirectional LSTM (BiLSTM), bidirectional GRU (BiGRU), convolutional neural network (CNN)-LSTM, CNN-GRU, CNN-BiLSTM, and CNN-BiGRU models for breath phase detection and adventitious sound detection. We also conducted a performance comparison between the LSTM-based and GRU-based models, between unidirectional and bidirectional models, and between models with and without a CNN. The results revealed that these models exhibited adequate performance in lung sound analysis. The GRU-based models outperformed, in terms of F1 scores and areas under the receiver operating characteristic curves, the LSTM-based models in most of the defined tasks. Furthermore, all bidirectional models outperformed their unidirectional counterparts. Finally, the addition of a CNN improved the accuracy of lung sound analysis, especially in the CAS detection tasks.

Conflict of interest statement

FSH, SRH, YRC, YCL, BFH, YLW, TLT and CTT are full-time employees and CJH, NJL, WLT and YTC are part-time employees of Heroic Faith Medical Science Co. Ltd. CWH and CHC are with Avalanche Computing Inc., whom Heroic Faith Medical Science Co. Ltd. commissioned to train the deep learning models. This does not alter our adherence to PLOS ONE policies on sharing data and materials.”

Figures

Fig 1. Customized multichannel acoustic recording device (HF-Type-1) connected to a tablet.

Fig 2. Auscultation locations and lung sound recording protocol.

(a) Auscultation locations (L1–L8): L1: second intercostal space (ICS) on the right midclavicular line (MCL); L2: fifth ICS on the right MCL; L3: fourth ICS on the right midaxillary line (MAL); L4: tenth ICS on the right MAL; L5: second ICS on the left MCL; L6: fifth ICS on the left MCL; L7: fourth ICS on the left MAL; and L8: tenth ICS on the left MAL. (b) A standard round of breathing lung sound recording with Littmann 3200 and HF-Type-1 devices. The white arrows represent a continuous recording, and the small red blocks represent 15-s recordings. When the Littmann 3200 device was used, 15.8-s signals were recorded sequentially from L1 to L8. Subsequently, all recordings were truncated to 15 s. When the HF-Type-1 device was used, sounds at L1, L2, L4, L5, L6, and L8 were recorded simultaneously. Subsequently, each 2-min signal was truncated to generate new 15-s audio files.

Fig 3. Pipeline of detection framework.

Fig 4. Model architectures and postprocessing for inhalation, exhalation, CAS, and DAS segment and event detection.

(a) LSTM and GRU models; (b) BiLSTM and BiGRU models; and (c) CNN-LSTM, CNN-GRU, CNN-BiLSTM, and CNN-BiGRU models.

Fig 5. Architectures of simplified bidirectional models.

(a) SIMP BiLSTM and SIMP BiGRU models; and (b) SIMP CNN-BiLSTM, and SIMP CNN-BiGRU models.

Fig 6. Task definition and evaluation metrics.

(a) Ground-truth event labels, (b) ground-truth time segments, (c) AI inference results, (d) segment classification, (e) event detection, and (f) legend. JI: Jaccard index.

Fig 7

ROC curves for (a) inhalation, (b) exhalation, (c) CAS, and (d) DAS segment detection. The corresponding AUC values are presented.

Fig 8

MAPE curves for (a) inhalation, (b) exhalation, (c) CAS, and (d) DAS event detection.

Fig 9. Patterns of normal breathing lung sounds.

(a) General lung sound patterns and (b) general lung sound patterns with unidentifiable exhalations. “I” represents an identifiable inhalation event, “E” represents an identifiable exhalation event, and the black areas represent pause phases.