AI-driven software enhances TB detection in Chennai

AI-driven software enhances TB detection in Chennai

Pushpa.Narayan@timesofindia.com 15.01.2024

Chennai : A computer-aided detection software can detect patients with tuberculosis from their X-ray reports using algorithms with high accuracy, a city-based study by doctors from two medical colleges has shown. AI could speed up and streamline the TB screening process in India, which exhibits 27% of all TB cases worldwide with an estimated 2.82 million cases and a 12% fatality rate in 2022, they concluded. 

For the study, the software -called Genki -was connected to an X-ray machine on four mobile diagnostic vans in low-resource areas of Chennai. These vans were used to screen patients who were suggestive of TB. The patients visited the van, completed registration and chest X-ray. The reports were triaged for suggestive findings within one minute -as TB suggestive or TB not suggestive. Scans and AI outputs were then reviewed by an expert radiologist for further interpretation. Patients with suggestive TB were referred for sputum or swab collection to confirm the diagnosis, said the study’s first author, Dr Prabakaran Jayaraman from the department of community medicine at Madha Medical College. Out of 25,598 X-rays screened between January and December 2022, Genki demonstrated an aggregated sensitivity (true positive rate) of 98%, specificity (true negative rate) of 96.9%, and accuracy of 96.9% in detec ting TB from chest X-ray scans, based on expert radiologist annotations. The study was published in the Japanese Association of Rural Medicine Journal of Rural Medicine. Of the 25,598 chest radiographs, radiologists reported TB in 1,029 (4.02%) scans, whereas Genki suggested TB in 1,802 (7.03%) scans. The software predicted 1,029 true positives, 23,775 true negatives, 773 false positives and 21 false negatives, Dr Prabhakaran said. The study included a significant number of participants from remote areas with high TB prevalence. Participants included both asymptomatic and symptomatic individuals. "Both these reflect real-world con ditions.

The software was not trained on any of the study participants' data and hence it minimized bias and provided a more objective assessment of its performance. We found Genki demonstrated consistent performance across different age groups and sexes," said Dr S Sangeetha from the department of community medicine at Vinayaka Mission’s Kirupananda Variyar Medical College and Hospitals. However, there were limitations as the study relied on human reader annotations for performance evaluation. While expert opinions are valuable, biological confirmation through sputum smears or NAATs would provide a more definitive reference standard.