A dual-purpose deep learning model for auscultated lung and tracheal sound analysis based on mixed set training

Many deep learning–based computerized respiratory sound analysis methods have previously been developed. However, these studies focus on either lung only or tracheal only. The effectiveness of using a algorithm and vice versa is rarely reported. Furthermore, no one knows whether sounds together in training learning-based model beneficial. In this study, we first constructed database, HF_Tracheal_V1, containing 10,448 15-s recordings, 21,741 inhalation labels, 15,858 exhalation 6414 continuous adventitious (CAS) labels collected from 227 participants undergoing diagnostic/surgical procedure under monitored anesthesia care. HF_Tracheal_V1 our built HF_Lung_V2, were combined (mixed set), used after the other (domain adaptation), alone to train convolutional neural network bidirectional gate recurrent unit models for inhalation, exhalation, CAS detection sounds. results revealed that trained performed poorly versa. mixed set domain adaptation improved performance 1) 2) compared positive controls (the versa). particular, had great flexibility serve two purposes, analyses, at same time.