New Machine Learning Framework Improves Respiratory Sound Classification Across Different Recording Conditions
Researchers have developed QLung, a machine learning framework that uses quality-adaptive angular margin learning to better classify respiratory sounds from audio recordings. The method adjusts its learning parameters based on recording quality, using spectral entropy and energy measurements to improve performance. The approach shows significant improvements on standard datasets and particularly excels at generalizing to recordings from different sources than those used in training.
A new machine learning framework called QLung has been developed to improve the classification of respiratory sounds from audio recordings. The system uses a quality-adaptive angular margin learning approach that enforces both intra-class compactness (keeping similar sounds grouped together) and inter-class separability (keeping different sound types distinct). A key innovation is a no-reference audio quality margin derived from spectral entropy and root-mean-square energy that automatically scales learning parameters based on recording quality. The framework also employs a log-scaled angular margin to handle severe class imbalance in training data and uses an angular classifier that normalizes features and weights on a unit hypersphere. Testing shows the approach improves performance on the ICBHI dataset by 2.46% over standard cross-entropy baselines, and most notably achieves state-of-the-art out-of-distribution performance on the SPRSound dataset, suggesting strong generalization to recordings from different sources.
What's missing
The study does not discuss potential clinical applications, validation with actual patient populations, or comparison with existing clinical respiratory sound classification systems. Additionally, the computational requirements and real-time applicability of the method are not addressed.
What different sources said
- arXiv cs.AICenter
Quality Adaptive Angular Margin Learning for Respiratory Sound Classification
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.