AI-powered lung ultrasound outperforms human experts in diagnosing tuberculosis

AI-powered lung ultrasound outperforms human experts in diagnosing tuberculosis

A landmark study presented today at ESCMID Global 2025 has shown that an AI-powered lung ultrasound outperforms experts in diagnosing pulmonary tuberculosis (TB) by 9%.

The Ultr-AI suite analyzes images from portable, smartphone-connected ultrasound devices, providing a sputum-free, fast and scalable alternative to TB detection. The results exceed World Health Organization (WHO) benchmarks for the diagnosis of pulmonary tuberculosis and mark an important opportunity for accessible and efficient TB triage.

Despite previous global declines, TB rates increased by 4.6% from 2020 to 2023.2 and are the critical components of the WHO End TB strategy. However, in many high-burden countries, they suffer significantly due to the high cost of chest X-Ray machines and a lack of trained radiologists.

These challenges highlight the urgent need for more accessible diagnostic tools. The Ultr-AI suite uses deep learning algorithms to interpret lung ultrasound in real-time, making the tool more accessible for TB triage, especially for minimally trained health workers in rural areas. By reducing operator dependency and standardizing the test, this technology can help diagnose patients more quickly and efficiently. “

Dr. Véronique Suttels,Lead study author

The Ultr-AI suite includes three deep learning models: Ultr-ai predicts TB directly from lung ultrasound images; Ultrisch (Sign) captures ultrasound patterns that are interpreted by human experts; and Ultr-AI (MAX) uses the highest risk score of both models to optimize accuracy.

The study was conducted in a tertiary urban center in Benin, West Africa. After exclusions, 504 patients were included, with 192 (38%) confirmed to have pulmonary TB. In the study population, 15% were HIV positive and 13% had a history of TB. A standardized 14-point lung ultrasound slide scan protocol was performed, with human experts interpreting images based on typical lung ultrasound findings. A single sputum molecular test (MTB Xpert Ultra) served as a reference standard.

Ultr-AI (max) demonstrated a sensitivity of 93% and a specificity of 81% (AUROC 0.93, 95% CI 0.92-0.95), exceeding WHO's target thresholds of 90% sensitivity and 70% specificity for non-sputum-based TB triage testing.

"Our model clearly recognizes the human-recognized lung ultrasound findings, which capture similar large consolidations and interstitial changes - an end-to-end deep learning approach captures even more subtle features beyond the human eye," said Dr. Suttels. “We hope this will help identify early pathologic signs such as small, sub-centimeter pleural lesions that are common in TB.”

“A key advantage of our AI models is the immediate turnaround time once they are integrated into an app,” added Dr. Suttels added. "This allows lung ultrasound to act as a true point-of-care test with good diagnostic performance at triage, providing immediate results while still engaging the patient with the health worker. Faster diagnosis could also improve linkage to care and reduce the risk of patients being lost to follow-up."

Sources: