AI Promising in Detecting TB From Chest X-rays

Artificial intelligence (AI) software can detect tuberculosis (TB) from chest x-rays at an accuracy level comparable to, or better than, that of the radiologists tested, new research indicates.

With the AI method, a simple image of a chest x-ray developed on film and captured with a mobile phone could be sent to a central location for analysis, which could be particularly helpful in low-resource areas, where TB is often concentrated.

TB causes 1.6 million deaths a year globally, making it the 13th leading cause of death globally and the second biggest killer from infections after COVID-19, the study authors note. There were an estimated 3 million undiagnosed patients in 2021, they add.

Data on the accuracy of AI-detected TB were presented at this year’s European Congress of Clinical Microbiology & Infectious Diseases in Copenhagen, Denmark.

Frauke Rudolf, PhD, Department of Infectious Diseases, Aarhus University Hospital, Aarhus, Denmark, and colleagues compared the performance of the AI software (qXR, in spotting TB with chest x-rays with that of two Ethiopian radiologists with different levels of experience. Rudolf said both had at least 10 years of experience.

Chest x-rays are particularly important in diagnosing patients who can’t produce good-quality sputum samples for analysis by a microbiologist.

In this study, the AI software was trained on mobile phone photos of nondigital chest x-rays to test how the software might assist in low-resource settings.

Researchers analyzed chest x-rays from 498 patients. Fifty-seven (11%) of patients had been diagnosed with TB, 41 clinically and 16 through polymerase chain reaction (PCR) tests. 

Human vs AI

Neither of the radiologists nor the AI platform was given clinical or laboratory information to accompany the x-rays.

The AI software correctly identified 75% of all PCR-confirmed cases (sensitivity of 75%) and 85.7% of non-TB cases (specificity of 85.7%).

The less experienced radiologist’s assessments had a sensitivity of 62.5% and a specificity of 91.7%. The more experienced radiologist’s assessments had sensitivity of 75% and specificity of 82%.

On some components of TB indication, the AI performed better than the radiologists combined (such as consolidation and nodule findings) and on some indications it performed less well (such as Hilar lymphadenopathy and cavitation).

“It’s useful where you don’t have radiologists, but it also might be useful in settings where you don’t have experienced radiologists to support their decision-making,” said Rudolf. “And now we also know that it’s applicable in settings where you have this physical analog x-ray, which we didn’t know before.”

Patients in this study were adults with TB-indicative symptoms presenting at three public centers in Bissau, Guinea-Bissau in west Africa, and three public health centers in Gondar, Ethiopia.

The patients either had no sputum assessment done or their smear had been negative. If they still had symptoms at a 2-week follow-up, they were sent for x-rays, Rudolf said.

WHO Recommends Computer-Aided Detection

Dr Marieke van der Werf

Marieke van der Werf, MD, PhD, MPH, head of section for STI, Blood-Borne Viruses and TB, with the European Centre for Disease Prevention and Control (ECDC), told Medscape Medical News that the World Health Organization (WHO) has recommended using computer-aided detection software programs for interpreting digital chest X-rays for screening and triage for TB disease among people aged 15 years and older since 2021.

“The recommendation is that software may be used in place of human readers,” she said.

However, she noted that the conditional recommendation is based on low certainty of evidence. “Thus, new well-conducted studies that provide additional evidence are welcome.”

In low-resource countries there often is no trained health staff available to interpret chest x-rays, and even if experts are available there is variability among readers, she added.

“Whether it is feasible to implement computer-aided software programs for reading chest x-rays will depend on whether equipment for conducting digital radiography is available and whether the setting has a stable internet connection,” said van der Werf.

The study authors and Van der Werf declared no relevant financial relationships.

European Congress of Clinical Microbiology & Infectious Diseases (ECCMID 2023):   Abstract 3073. Presented April 18, 2023.

Marcia Frellick is a freelance journalist based in Chicago. She has previously written for the Chicago Tribune, Science News, and, and was an editor at the Chicago Sun-Times, the Cincinnati Enquirer, and the St. Cloud (Minnesota) Times. Follow her on Twitter at @mfrellick

For more news, follow Medscape on Facebook, Twitter, Instagram, YouTube, and LinkedIn

Source: Read Full Article