In a game-changing breakthrough revealed at ESCMID Global 2025, researchers unveiled a revolutionary AI-powered lung ultrasound tool that outperforms human experts by 9% in diagnosing pulmonary tuberculosis (TB). Dubbed ULTR-AI, this deep-learning suite analyzes lung images from portable, smartphone-connected devices, eliminating the need for sputum tests and expensive x-rays. It’s fast, accurate, and tailor-made for frontline care—even in remote settings.
The study, led by Dr. Véronique Suttels in Benin, West Africa, involved 504 patients, 38% of whom were confirmed to have TB. Using a 14-point ultrasound protocol and comparing results against gold-standard molecular testing, ULTR-AI (max) achieved 93% sensitivity and 81% specificity, surpassing the WHO’s benchmarks for TB triage tools.
ULTR-AI’s strength lies in three models: one directly detects TB, another reads lung signs like a human expert, and a third combines both for maximum accuracy. Its real-time results empower minimally trained healthcare workers to diagnose on the spot—no lab needed, no waiting.
Dr. Suttels emphasized the model’s unique ability to spot both obvious and subtle lung abnormalities, including early TB signs that the human eye might miss. With global TB rates climbing, this AI-driven innovation could dramatically improve early detection, reduce dropouts, and boost access to care—especially in resource-limited areas.
