RespFocus identifies timing of breathing sounds to improve diagnosis accuracy
RespFocus: Learning Temporal Evidence for Weakly Supervised Respiratory Sound Classification
Machine LearningArtificial Intelligence
Summary
Listening to breathing sounds can help doctors find lung problems, but such sounds are often noisy and tricky to understand. The authors created RespFocus, a computer method that learns which parts of breathing sounds are important and shows this information clearly. They also collected a new set of children's breathing sounds recorded in real clinics for testing. RespFocus performed well on two different datasets and helped explain why it made its decisions. This work could help build better tools to automatically find lung issues from breathing sounds.
What this means in practice
- •For medical device developers: Create clearer and more explainable automated tools for diagnosing lung conditions from breathing sounds.$Commercial implications: Enables development of new diagnostic devices that use transparent sound-based algorithms to aid clinicians.
- •For clinical informatics teams: Improve integration of respiratory sound analysis into hospital systems by using models that highlight important sound segments.
Authors
Yi Tian, Xiaohuan Chen, Yufei tang, Xinyu Lin, Hongliang Li, Chengbin Hou, Degen Huang, Rongshan Yu
Abstract
Respiratory auscultation provides a low-cost means of assessing pulmonary function, yet transient crackles and sustained wheezes can be difficult to recognize in noisy clinical recordings. Conventional deep learning classifiers often operate as black boxes, offering limited insight into the temporal acoustic evidence supporting their predictions. We present RespFocus, a weakly supervised framework that uses multiple instance learning (MIL) to learn inspectable temporal evidence from cycle-level labels and integrate it into classification. We also introduce PedResp2026, a multicenter pediatric respiratory sound cohort collected in routine clinical environments with adjudicated cycle-level annotations. In classification experiments, RespFocus achieves four-class Scores of 69.55\% on public ICBHI 2017 and 72.91\% on PedResp2026. Evidence diagnostics and matched temporal interventions support the functional contribution of the learned evidence to model predictions. Together, these findings highlight RespFocus's potential as a tool for automated respiratory sound diagnosis.