Artificial Intelligence based AFB microscopy for Pulmonary Tuberculosis in North India: A pilot study

Abstract

Introduction: Microscopy is the simplest and the most important step in diagnosis of pulmonary tuberculosis. But the microscopy requires considerable experience and is bound to human errors. Artificial intelligence (AI) based microscopy can be an answer to this problem. AI can be used even where expert microscopists are not available. Sevamob provides artificial intelligence enabled healthcare platform to organizations. It uses deep learning for image recognition, machine learning for triaging and computer vision for object counting. AI models of various medical conditions are first trained in the software from anonymized image data procured from various sources.To determine the accuracy of AI based point-of-care screening solution for sputum, following were used. Android Smartphone / tablet with Sevamob app, tripod and a simple microscope. The system was operated by a nurse or a technician with minimal training. Methods: 150 ZN stained smears of sputum samples from clinically suspected pulmonary tuberculosis cases were included in the study. Results:Outof these 150 smears, 118 were found to be positive and 32 as negative by expert microscopist. These smears were also analyzed by AI system. Out of these only 6 smears were found to be false positive for AFB. Thus, the sensitivity and specificity of AI based microscopy was 81.7% and 80% respectively. Conclusion:This shows that our AI based microscopy system can be very useful to find AFB and has potential to replace the requirement of expert microscopist in the coming future. Sensitivity and specificity also depend on the threshold used by our AI system.