A groundbreaking study presented at ESCMID Global 2025 in Vienna is driving a reassessment of tuberculosis (TB) diagnoses. The paper explains how an AI-enhanced lung ultrasonography approach not only meets but exceeds accepted diagnostic criteria. By using the capabilities of the ULTR-AI suite, this technology provides an efficient and cost-effective option at a time when traditional screening is being overtaken by developing needs. 

.The unique method employs portable, smartphone-connected ultrasound instruments, significantly lowering dependency on sputum samples. Notably, the system established a new benchmark by surpassing human specialists by 9%, highlighting its potential significance in resource-constrained environments. The ULTR-AI suite has three separate deep-learning models: one predicts tuberculosis from ultrasound images, another finds patterns based on doctors' interpretations, and the third picks the highest risk score from the combined results. The multimodal architecture improves the diagnostic output and reliability of the screening process. 

The relevance of this achievement is heightened by the worrying global trend of a 4.6% increase in tuberculosis incidence between 2020 and 2023. The need for accessible diagnostic techniques has never been more important, with early screening and rapid detection at the heart of the World Health Organisation's 'End TB Strategy'. In many high-burden nations, the prohibitively expensive cost of chest x-ray equipment and a scarcity of skilled radiologists result in the loss of several patients during the diagnostic procedure. The prospect of a quick, point-of-care test is thus a positive development. 

Lead study author Dr. Véronique Suttel emphasised the technology's transformational potential, saying, "These challenges highlight the urgent need for more accessible diagnostic tools." The ULTR-AI suite uses deep learning algorithms to interpret lung ultrasonography in real time, making it more accessible for tuberculosis triage, particularly among minimally educated healthcare professionals in remote locations. This technology can help diagnose patients more quickly and efficiently by decreasing operator dependency and standardising the exam. Her remarks highlight the system's twofold benefit: improved diagnostic accuracy and instant applicability in low-resource areas. 

The study, which took place in a tertiary metropolitan hub in Benin, West Africa, included 504 patients, 192 of whom (38%) were diagnosed with pulmonary tuberculosis. With 15% of the patients HIV-positive and 13% with a history of tuberculosis, the patient group represented a realistic range of clinical circumstances. A defined 14-point lung ultrasound sliding scan methodology was used, with a sputum molecular test as the reference benchmark. The diagnostic performance of ULTR-AI (max) was 93% sensitivity and 81% specificity (AUROC 0.93, 95% CI 0.92-0.95), exceeding WHO's targets of 90% sensitivity and 70% specificity. 

Dr Suttels went on: "Our model clearly detects human-recognisable lung ultrasound findings—like large consolidations and interstitial changes—but an end-to-end deep learning approach captures even subtler features beyond the human eye." This statement exemplifies the inventiveness of the deep learning system, which not only mimics but significantly improves human observation by detecting minute disease indicators such as small sub-centimetre pleural lesions, which are frequently neglected in early tuberculosis. 

Once integrated into a mobile application, lung ultrasonography becomes an efficient triage tool due to its quick response time. Patients receive rapid diagnosis while still interacting with healthcare practitioners, which is likely to improve access to care and reduce follow-up losses. This strategic mix of accessibility, speed, and accuracy represents a substantial improvement in worldwide tuberculosis diagnoses.