Automatic interpretation algorithms of chest radiography in tuberculosis

Abstract

Mass screening and rapid diagnosis of tuberculosis (TB) is a difficult task in developing countries like India. To address this difficulty, many artificial intelligence tools and deep learning methods are used as an automated TB detection system to simplify the diagnosis of tuberculosis and assist clinicians in confirming the severity of disease and treatment execution. Upon reviewing the recent evidences for using automated interpretation in tuberculosis screening and diagnosis by radiological methods, here are the few evidences of its usage suggesting it as a valuable tool for clinicians in assessing patients with tuberculosis. Soares at al demonstrated that chest X-ray automated interpretation algorithms were significantly correlated with sputum bacillary load in mass screening of tuberculosis [1]. Yan et al demonstrated a moderate to strong correlation in diagnostic performance between clinician and AI model based on chest computed tomography [2]. Ma et al. developed an AI tool for automatic detection of patients with active tuberculosis (ATB) and distinguish between ATB and non-ATB cases with chest computed tomography (CT) [3]. Currently more than 55 commercial (CE-marked) artificial intelligence algorithm-based devices are available for chest radiography [4-6]. Recently Kazemzadeh et al demonstrated the efficiency of deep learning system in the detection of active pulmonary TB in 1,65,754 images of 22,284 subjects spanning five countries (India, China, United States, Zambia and South Africa) [7]. Further multicentre prospective studies were needed to validate and generalize these algorithms for clinical implementation in India. Hence, it is important for us to lead this new era radiology powered by artificial intelligence.