HealthDay News — An artificial intelligence system can detect pulmonary tuberculosis (TB) on chest X-rays with a higher sensitivity than radiologists, according to a study published online Sept. 6 in Radiology.
Sahar Kazemzadeh, M.D., from Google Health in Mountain View, California, and colleagues developed a deep learning system (DLS) to detect active pulmonary TB on chest radiographs and compared its performance with that of radiologists. The DLS was trained and tested using 165,754 images acquired between 1996 and 2020 in 22,284 individuals from 10 countries. The DLS was assessed in a four-country test set (1,236 individuals from China, India, the United States, and Zambia; 17 percent with active TB) and in a mining population in South Africa, with confirmation of positive TB with microbiological tests or nucleic acid amplification testing. DLS performance was compared to that of 14 radiologists, and its efficacy was compared to that of nine radiologists.
The researchers found that the receiver operating characteristic (ROC) curve of the DLS was higher than those for all nine India-based radiologists in the four-country test set (area under ROC curve, 0.89). At the prespecified operating point, the sensitivity of DLS was higher than that of radiologists (88 versus 75 percent), while specificity was noninferior (79 versus 84 percent). Similar trends were seen within other patient subgroups, in the data set from South Africa, and across various TB-specific chest radiograph findings.
“The DLS may be able to facilitate TB screening in areas with limited radiologist resources and merits further prospective clinical validation,” the authors write.
Several authors disclosed financial ties to Google and Alphabet; the study was funded by Google.