AI-Powered Lung Ultrasound Surpasses Human Experts in Tuberculosis Diagnosis

ULTR-AI achieves 93% sensitivity and 81% specificity, exceeding WHO benchmarks, and offers real-time, app-based diagnostics for improved TB triage.

The ULTR-AI suite, presented at ESCMID Global 2025, uses deep learning to analyze portable lung ultrasound images, providing a rapid, sputum-free diagnostic alternative.
Achieving 93% sensitivity and 81% specificity, ULTR-AI outperforms human experts by 9% and surpasses WHO thresholds for non-sputum-based TB triage tests.
The system includes three AI models, with ULTR-AI (max) optimizing accuracy by integrating risk scores from direct image predictions and expert-pattern detection.
Conducted in Benin with 504 patients, the study highlights the tool's applicability in high-burden, resource-limited settings where TB rates have risen by 4.6% from 2020 to 2023.
Integration into a smartphone app enables immediate point-of-care results, reducing patient dropouts and facilitating faster linkage to care.