The abstract is read at the publisher; the summary is JClub's.
In a Kampala, Uganda prospective study, attention-based deep learning on point-of-care lung ultrasound—alone or with clinical data—achieved AUC up to 0.95 for pulmonary tuberculosis triage.
- Why it matters: Portable, radiation-free ultrasound could expand tuberculosis screening where chest radiography is unavailable, but standardized interpretation criteria and validated decision rules are lacking.
- What they did: Using one still frame from each of 12 lung ultrasound zones, the team trained and internally evaluated multiple-instance attention deep learning models, comparing ultrasound-only, clinical-only, and combined approaches, and deriving a 6-zone protocol from attention weights.
- The result: In the held-out test set, AUC was 0.88 for clinical-only and 0.93–0.95 for 12-zone ultrasound-only and combined models, with specificity at 90% sensitivity of 82.2% (12-zone combined) and 91.1% (6-zone combined), meeting 2025 WHO high-sensitivity screening targets.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.