Role of Artificial Intelligence Enabled Chest X-ray Interpretation to Detect Pulmonary Tuberculosis in a Low-Income Country

Abstract Introduction: Tuberculosis (TB) is a global public health problem causing significant morbidity and mortality, particularly in low and middle income countries (LMIC). Globally, one-third of TB cases are missed, making the fight against TB more challenging. This study aims to determine the role of Artificial Intelligence enabled CXR interpretation in the detection of pulmonary TB in Ethiopia. Methods: This is a prospective single-arm cohort study conducted in Tikur Anbessa Specialized Hospital, Addis Ababa, Ethiopia from January to September 2024. Ethical approval was obtained from the Institutional Review Board of the College of Health Sciences, Addis Ababa University. Consecutive CXRs considered technically acceptable from adults 18 years and above who consented to follow-up based on AI-based CXR reading were included in the study. Confirmed TB cases and patients not willing to come were excluded. Data on socio-demographics, indications for CXR, presenting symptoms, risk factors, and AI-based CXR findings were collected. The AI-based CXR software (Qure.ai, India) processed all chest X-rays that met the eligibility requirements throughout the study period. Once AI-based CXR software flagged an image as high risk for pulmonary TB, the image was reevaluated by a senior radiologist; if the radiologist agreed, a sputum sample was sent for Xpert MTB/RIF. All TB cases identified clinically or bacteriologically were linked to the TB clinic for treatment. Result: A total of 3351 CXRs were pushed to the AI-based software and 552 (16.5%) scans were flagged as suggestive of pulmonary TB. Senior Radiologists evaluated a total of 521 (94.4%) flagged CXR, of which 47 (9.0%) aligned with their interpretation and were eligible for GeneXpert. GeneXpert was performed for 30 participants (71.1%). Of the remaining 17 patients, 10 were lost to follow-up, 6 were unable to give samples but had chest CT to rule out TB, and the remaining 1 patient was unable to give a sample but evaluated by a pulmonologist and TB ruled out. Ten patients with bacteriologically confirmed TB were linked to TB clinic. Out of ten patients with bacteriologically confirmed tuberculosis, CXR was done for possible PTB in seven patients, for CAP in two, and for a medical certificate screening in one. The median age of patients flagged for TB was 48 years and majority were females, 52% Conclusion: This ongoing study showed that AI-based CXR software identified additional active pulmonary cases. AI-based CXR implementation may help to increase case detection and early diagnosis of TB in LMIC settings.