ML cough models fail to generalize for TB screening, study finds

A new study evaluating machine learning models for tuberculosis screening using cough acoustics found that despite strong within-dataset performance, these models fail to generalize to new datasets. The research indicates that audio representations are organized by recording device and dataset rather than disease status, and that device-specific variability is a significant factor in poor generalizability. A clinical-variable baseline model demonstrated more consistent generalization, highlighting the critical need for external validation before cough-based TB models can be considered clinically ready. AI

IMPACT Highlights critical need for external validation in ML models for healthcare applications, impacting deployment readiness.

RANK_REASON Academic paper detailing research findings on ML model generalizability. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• CatalyzeX
• CODA
• DagsHub
• deep learning
• Gotit.pub
• Hugging Face
• machine learning
• ROC-AUC
• ScienceCast
• tuberculosis

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →