A new AI-driven solution for lung ultrasound shows promising results in the fight against tuberculosis (TB). A study was recently presented showing that the ULTR-AI suite – a deep learning algorithm that analyses lung images – outperforms radiologists and other human experts (+9%) in detecting lung TB. The system offers a fast, sputum-free and scalable alternative for diagnostics, with direct impact on accessibility and efficiency, especially in resource-limited areas.
Between 2020 and 2023, the number of tuberculosis cases worldwide increased by almost 5 percent. A remarkable development, given that in the decades before that, the number of cases had been decreasing. This increase is particularly noticeable in countries with a high tuberculosis burden. These countries suffer from a significant drop-out of patients in the diagnostic phase due to the high costs of X-ray equipment for lung ultrasound and the lack of radiologists.
Accessible triage
The WHO emphasizes the importance of early detection in its ‘End TB Strategy’. The ULTR-AI suite cleverly responds to this with an AI solution that is linked to portable, smartphone-controlled ultrasound devices. This allows lung ultrasound to be used as a true point-of-care test – without the dependence on expensive infrastructure or specialist knowledge.
According to lead researcher Dr. Véronique Suttels, this technology can make a difference, especially in rural or understaffed healthcare environments, thanks to the direct interpretation of images in the app. “The AI makes it possible to standardize lung ultrasound and apply it directly, even by healthcare providers with minimal training,” she says. The research was presented at the ESCMID Global congress in Vienna.
Clinically validated, immediately applicable
The study was conducted in a tertiary center in Benin, West Africa, with 504 participants. 38% of them tested positive for pulmonary TB, with the AI models delivering exceptional performances:
- Sensitivity: 93%
- Specificity: 81%
- AUROC: 0.93
These figures far exceed the WHO target values for sputum-free triage tests. The sensitivity and specificity are at least 90 and at least 70 percent, respectively.
ULTR-AI consists of three models that analyze different aspects of the ultrasound images. By cleverly combining their results, a very robust prediction is created. The model also detects both classic TB characteristics and subtle patterns that human eyes can miss, such as early pleural lesions.
Direct diagnosis, faster care
A major advantage of the technology is the direct feedback: results are available while the patient is still with the healthcare provider. “This reduces the chance of patients dropping out of the diagnostic pathway, and strengthens the link between diagnosis and care,” says Suttels. The AI solution not only offers diagnostic added value, but also supports the broader objectives of the WHO End TB Strategy: faster detection, faster treatment and better access to care.
Better medication
In addition to faster and accessible diagnostics, the development of new, better medication is also an important development for reducing the number of cases of tuberculosis. For example, researchers from four Leiden institutes worked a few years ago in the Leiden interdisciplinary Tuberculosis Consortium (LiTBC) on the development of more effective and efficient therapies against tuberculosis.
They focused on improving the body's self-defense mechanisms and gaining better insight into immune responses in tuberculosis. In addition, they also developed a rapid test for early diagnosis and are investigating new antibiotics, nanomedicines and therapies that strengthen the self-defense of infected cells.
