Lunit's AI chest X-ray tool beats global competitors in detecting tuberculosis

Lunit, a medical software company specializing in AI-driven cancer diagnostics and therapeutics, said Tuesday that its chest X-ray analysis software, Lunit INSIGHT CXR, has demonstrated superior performance in detecting tuberculosis (TB). 

Such results are based on a large-scale, independent study published in Lancet Digital Health, which evaluated the effectiveness of various AI products for TB detection, the company added.

TB, an infectious disease predominantly affecting the lungs and often referred to as the “white death,” remains a critical global health issue. Despite being treatable, TB is the world’s second leading cause of death from a single infectious agent, the bacteria Mycobacterium tuberculosis, with about 10.6 million incidents reported in 2022. One of the major obstacles in managing TB is the underdiagnosis of the disease, as an estimated 3.1 million cases went undetected that year.

Chest X-rays are better at detecting abnormalities related to TB than symptom screenings. Still, their effectiveness is frequently hampered by inconsistencies in human interpretation and a shortage of radiologists, particularly in countries heavily burdened by TB, indicating the growing importance of advanced AI technologies, Lunit noted.

The study, led by Dr. Zhi Zhen Qin from Heidelberg University Hospital in Germany and a research team from the Stop TB Partnership, evaluated 12 commercially available AI products for TB detection. The Stop TB Partnership, founded in 2001 to eradicate TB as a public health issue and based in Switzerland, has been managed by the United Nations Office for Project Services (UNOPS) since 2015.

The research team evaluated the AI tools using data from chest X-rays of 774 patients, collected over two years from South Africa's national TB prevalence survey. 

The results showed that Lunit Insight CXR achieved the highest TB detection performance, with an area under the receiver operating characteristic curve (AUC) score of 0.902, surpassing all other AI products tested.

This performance was close to the World Health Organization’s (WHO) targets for computer-aided diagnostic devices (CAD), which recommend 90 percent sensitivity and 70 percent specificity for TB classification in individuals over 15.

When set to a sensitivity threshold of 90 percent, Lunit INSIGHT CXR achieved a specificity of 67.7 percent, the nearest among the 12 products to the WHO’s target. Conversely, at a specificity threshold of 70 percent, its sensitivity reached 89.5 percent, also the closest to the WHO’s recommended metrics.

Lunit said these results suggest that Lunit INSIGHT CXR could be a valuable tool for TB screening in developed and developing countries. 

“Lunit INSIGHT CXR’s capacity to maintain high sensitivity across diverse populations and diagnostic thresholds is particularly crucial in resource-limited settings where each undetected case can have significant repercussions, especially given the ongoing impact of TB in many developing countries,” Lunit CEO Brandon Suh said. “By bridging the gap in TB diagnosis, we are working toward a future where no case goes unnoticed, particularly in regions where it is needed most.”

The findings are available in Lancet Digitial Health (IF: 23.8), The Lancet's sister journal for digital health.