Lead study author Dr. Véronique Suttels highlighted that the ULTR-AI suite's deep learning algorithms can interpret lung ultrasound in real-time, making it particularly beneficial for minimally trained healthcare workers in rural settings.

From 2020 to 2023, global TB rates rose by 4.6%, underscoring the urgent need for improved diagnostic tools in high-burden countries, where many patients drop out during diagnosis due to costs and a lack of trained personnel.

A study conducted in Benin, West Africa, involved 504 patients, with 192 confirmed cases of pulmonary TB, including a notable 15% who were HIV-positive.

Overall, the results from the ULTR-AI suite represent a significant advancement for accessible TB triage, promising to enhance early detection and improve patient outcomes.

A groundbreaking study presented at ESCMID Global 2025 demonstrated that AI-guided lung ultrasound outperforms human experts by 9% in diagnosing pulmonary tuberculosis.

The ULTR-AI suite comprises three deep-learning models: ULTR-AI for direct TB prediction from ultrasound images, ULTR-AI (signs) for detecting ultrasound patterns, and ULTR-AI (max) which optimizes accuracy by utilizing the highest risk score from both previous models.

The integration of AI models into an app allows for immediate diagnostic results at the point of care, enhancing patient diagnosis and potentially reducing follow-up losses.

High costs of chest x-ray equipment and a shortage of trained radiologists in many high-burden countries contribute significantly to patient dropouts at the diagnostic stage.

Dr. Suttels emphasized the necessity for accessible diagnostic solutions, particularly for healthcare workers in rural areas, to improve early TB detection capabilities.

The ULTR-AI suite introduces portable, smartphone-connected ultrasound devices that offer a rapid, sputum-free alternative for tuberculosis (TB) detection, exceeding World Health Organization (WHO) benchmarks.

Specifically, the ULTR-AI (max) model achieved 93% sensitivity and 81% specificity, surpassing WHO's targets of 90% sensitivity and 70% specificity for non-sputum-based TB triage tests.

The research utilized a standardized 14-point lung ultrasound sliding scan protocol, with human experts interpreting the images against a sputum molecular test (MTB Xpert Ultra) as the reference standard.