AI software can accurately detect tuberculosis from chest X-rays, study finds.
- AI software was at least as good as radiologists at detecting tuberculosis from chest X-rays.
- Tuberculosis is a major cause of death and disease worldwide.
- AI could help diagnose more TB cases and save lives.
This is a real breakthrough: an AI has managed to detect tuberculosis at least as effectively as specialists. Using this software to analyze X-rays for abnormalities could thus facilitate the diagnosis of tuberculosis in regions where there is a shortage of radiologists, according to Danish researchers.
Tuberculosis is a life-threatening disease
Tuberculosis is caused by a bacterium (Mycobacterium tuberculosis) which most often affects the lungs. It is a disease that can be prevented and cured, says theWHO. It is transmitted from person to person through the air: “When a person with pulmonary tuberculosis coughs, sneezes, or spits, they project tubercle bacilli into the air. It is enough to inhale only a few of them to become infected”.
Globally, tuberculosis is the 13th cause of death and the second due to an infectious disease, behind Covid-19 (and before AIDS).
The researchers presented their study At European Congress of Clinical Microbiology and Infectious Diseases (ECCMID) in Copenhagen (Denmark), which took place from April 15 to 18. To carry out their work, they compared the performance of this artificial intelligence software in evaluating chest X-rays to that of two Ethiopian radiologists with different levels of experience.
Tuberculosis: AI is effective in diagnosing the disease
And the conclusion is clear: the AI software was as good, or even better, than a radiologist.
Chest radiographs of 498 patients were analyzed retrospectively. Fifty-seven (11%) of these patients had been diagnosed with tuberculosis, including 41 clinically and 16 by PCR tests (Xpert MTB/Rif). This test simultaneously detects Mycobacterium tuberculosis complex (MTBC) and rifampicin resistance (RIF), specific for tuberculosis, in less than 2 hours.
In the experience, the AI software correctly identified 75% of all PCR-confirmed cases and 85.7% of non-TB cases. The less experienced radiologist correctly identified 62.5% of PCR-confirmed cases and correctly identified 91.7% of non-TB cases. The experienced radiologist’s assessments were 75% for PCR-confirmed cases and 82% for non-TB cases.
Tuberculosis diagnosis: AI could replace radiologists if needed
With an estimated 3 million undiagnosed patients in 2021, there is an urgent need to develop new strategies and technologies to improve TB detection in low-resource, high-incidence settings, said Dr Frauke Rudolf, Department of infectious diseases of the university hospital of Aarhus (Denmark) author of the study.
“We have shown that AI software is at least as good at detecting TB as a trained radiologist and that a simple photo taken using a mobile phone is enough for the analysis. In low-resource settings where the incidence of TB is high but there is a shortage of radiologists, chest X-rays could be photographed using a mobile phone and the image sent would be analyzed remotely by the ‘AI’did he declare.
This would allow more chest X-rays to be read correctly and, importantly, more cases of tuberculosis to be diagnosed.