Active Learning Audio Model for Pediatric Asthma Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio-based asthma monitoring solutions face challenges due to reliance on feature engineering designs paired with contact-based auscultation, which are brittle and do not scale beyond point-of-care setups, and the high cost of data curation requiring physicians-in-the-loop.
Innovation Solution
An active learning (AL) system is proposed to facilitate audio data collection and annotation, automatically extracting interesting audio segments and efficiently annotating and training anomaly detector models, reducing annotation costs while increasing model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional machine learning approaches are used with smaller datasets, then data collection difficulty and cost are reduced, but model robustness and generalizability deteriorate
Solution Approach 1:
The system uses automated feature extraction and machine learning models to perform tasks that would otherwise require manual physician analysis, enabling the model to serve itself in identifying and labeling relevant audio segments for training
Solution Approach 2:
The system performs preliminary feature extraction and automated labeling of audio data before formal model training, preparing high-quality training data in advance to improve model robustness without requiring extensive manual annotation
2Measurement precision
If physicians-in-the-loop are used for data curation, then annotation accuracy is improved, but cost and time consumption increase
Solution Approach 1:
The system applies partial human annotation by having physicians review only the automatically selected and pre-labeled audio segments rather than annotating all data, achieving high accuracy while significantly reducing annotation time and cost
Solution Approach 2:
The system introduces an intermediary automated feature extraction and preliminary labeling layer between raw audio data and human annotators, filtering and preparing data to reduce the burden on physicians while maintaining annotation quality
3Measurement precision
If contact-based auscultation methods are used, then diagnostic accuracy is improved, but scalability and ease of operation deteriorate
Solution Approach 1:
The system replaces the mechanical contact-based auscultation process with acoustic sensor arrays and automated signal processing algorithms, maintaining diagnostic accuracy while enabling remote and scalable monitoring capabilities
Solution Approach 2:
The system creates a multi-functional platform that can perform both traditional diagnostic functions and scalable remote monitoring, allowing the same technology to serve both point-of-care and population health management needs
Data Source
AI summary
Methods and systems for training an audio-based machine learning model to predict a health condition based on biological sounds emitted by a person. Audio data corresponding to biological sounds produced by the person is generated from a microphone. The audio data is segmented into a plurality of segments, each segment associated with a respective sound event. An audio-based machine learning model is executed on the plurality of segments. The audio-based machine learning model is configured to output, for each segment, a label of a medical condition and an associated a confidence score. The model is trained via active learning, in which a subset of the plurality of segments are selected based on their confidence score being below a threshold, and provided to a human for annotation.