Active learning of biosounds for determining presence of medical condition
By applying an active learning (AL) system in asthma monitoring to collect and annotate audio data, the problem of high scope and cost of data collection in the prior art is solved, and efficient and scalable data management and model performance improvement is achieved.
Patent Information
- Application Number
- CN202411350011.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-09-26
- Filing Date
- 2024-09-26
- Publication Date
- 2025-03-28
AI Technical Summary
Existing audio-based asthma monitoring solutions rely on contact stethoscopes, limiting the scope and scalability of data collection, and making it difficult to train robust and generalizable models due to high cost of medical participation.
Active learning (AL) system is used to collect and annotate audio data, and automatically extract audio clips and efficiently annotate, reducing annotation costs and improving model performance. The system includes extracting audio blocks from unlabeled data streams, automatically segmenting and annotating using pre-trained models to enable large-scale audio data management.
Under the restricted annotation budget, the annotation cost is reduced, the model performance is improved, and the larger-scale data management of the childhood asthma audio corpus is effectively promoted, and more scalable data collection and annotation are achieved.