Development and evaluation of an artificial intelligence (AI) -assisted chest x-ray diagnostic system for detecting, diagnosing, and monitoring tuberculosis

Abstract

Objectives: To develop an artificial intelligence (AI)-assisted chest x-ray diagnostic system for the detection, differential diagnosis, and follow-up of tuberculosis (TB), and prove its usefulness. Methods: This is a retrospective study. In-house developed AI-assisted chest x-ray diagnostic system was used to identify and diagnose lung abnormalities in participants' chest x-rays and to compare imaging findings from two x-rays. First, 100 chest radiographs were reviewed including TB cases (N = 43) with positive sputum test confirmation and non-TB cases (N = 57) for initial diagnosis and differential diagnosis. Next, 45 pairs of TB cases from the identical patients were reviewed for follow-up. The AI system diagnosed TB and graded the comparison images into three categories (improved, stable, or worsening). The performance was evaluated by four expert radiologists or pulmonary medicine specialists. Results: The AI system demonstrated an exceptional sensitivity of 100 %, successfully identifying all 43 TB cases. Nevertheless, it is also susceptible to misclassify other diseases as TB, resulting in low specificity score of 66.7 %. The comparison function determined that expert physicians and AI-assisted chest x-ray diagnostic system were 58 % in exact agreement and 100 % in within one grade agreement. Conclusions: The AI system successfully detected all TB patients identified in this study and demonstrated a reasonable comparison function. Therefore, our AI assisted chest x-ray diagnostic system is feasible and practical for TB screening.