Project Details
Early Career
Status: Funded - Open
Performance of multi-modal AI digital stethoscope for pediatric pneumonia diagnosis in Bangladesh
Summary
BACKGROUND: The burden and mortality of pediatric lower respiratory infections (LRIs) like pneumonia are high in low-and middle-income countries (LMICs), but accessible accurate diagnostics are lacking. Artificial intelligence (AI) enabled digital stethoscopes may address this gap. This project will leverage the Bangladesh Lung Auscultation with Artificial Intelligence for Antibiotic STewardship (BLAAAST) trial to assess the performance of an AI-digital stethoscope in safely reducing antibiotic use among children. GAP: The performance of pediatric AI stethoscope RR algorithms has not been assessed in LMICs and the AI stethoscope performance for identifying abnormal lung sounds has not been evaluated in LMICs using chest radiography (CXR) as the reference. HYPOTHESIS: Hypothesis 1: An AI stethoscope RR algorithm will be accurate within 4 breaths/minute of the physician reference RR. Hypothesis 2: AI stethoscopes will achieve >80% accuracy in diagnosing pneumonia compared to CXR, and incorporating RR stratification will further improve accuracy. METHODS: We will conduct a nested prospective cohort study in a subset of 400 enrolled participants within the double-blinded randomized placebo-controlled BLAAAST trial enrolling 2,500 children 2-59 months old with clinically diagnosed pneumonia. Participants will undergo auscultation with StethoMe® at 4 chest positions. AI-identified abnormal lung sounds (i.e., crackles and/or wheeze) constitute a positive test. Aim 1: A StethoMe®-generated AI RR will be compared to reference RR from a physician listening panel. Aim 2a/b: A subset of trial participants will receive enrollment CXRs. A physician panel will determine radiographic pneumonia and compare to the AI-stethoscope test result. Secondary analysis will determine if RR stratification improves accuracy. RESULTS: Pending. IMPACT: This project will generate novel data to support the adoption of pneumonia guidelines enhanced by AI-stethoscopes for improving LRI diagnosis. This work complements a larger body of childhood pneumonia research by our group in Africa and the United States.