Background:
Ultraportable X-ray devices coupled with artificial intelligence (AI) are beginning to be used widely to screen for TB in remote settings. Few studies have documented their performance.
Methods:
We organized screening camps in Gombe and Adamawa states of Northeast Nigeria using a team of a registration officer, data entry staff, a radiographer, and a coordination officer. All consenting individuals ≥ 15 years were screened for TB symptoms (cough, fever, night sweats, and weight loss) and received a CXR. We used a MinXray Impact system interpreted by AI (qXR V3), which is a fully integrated wireless setup, can be run without electricity, and weighs approximately 30kg. We collected sputum samples from individuals with an abnormality score of .30 or higher or if they reported any TB symptoms. Samples were transported for testing with Xpert MTB/RIF. We documented the TB screening cascade and evaluated the performance of screening with different combinations of CXR read by AI and symptoms.
Results:
We screened 5,298 individuals during 66 camps; 2,685 (51%) were females, and 2,613 (49%) were males. Using ≥ 2 weeks of cough to define presumptive TB, 1,056 people (20%) would be identified. If a cough of any duration was used, the number with presumptive TB increased to 1,889 (36%) and to 3,084 (58%) if any of the four symptoms were used. Overall, 770 (14.5%) had abnormality scores of 0.3 or higher, and 447 (8.4%) had a score of 0.5 or higher. We collected 1,022 samples for Xpert testing and detected 85 (8%) individuals with TB. Traditional symptom screening of prolonged cough only identified 40% of people with TB, while using any symptom detected 90.6% of people with TB, but specificity was 11.4%. Using an AI abnormality score of .50, only used 424 tests to identify 89.4% of people with TB with a specificity of 62.8%.
Conclusion:
Ultra-portable CXR can be used to provide better TB screening in hard-to-reach areas. Symptom screening may miss large proportions of people with bacteriologically confirmed TB. Employing AI to read CXR can improve triaging when human readers are unavailable and can save expensive diagnostic testing costs.