WSI AFB and AI deliver highest sensitivity for TB detection

Background Individuals infected by Mycobacterium tuberculosis (Mtb) develop tuberculosis (TB) which is a chronic infectious disease with the main transmission route being the respiratory tract. Currently, 24% of TB patients are still not detected in time, which shows the shortcomings of current diagnostic methodology. Methods We developed a novel Whole Slide Imaging (WSI) platform for TB detection, integrating a proprietary Curved Surface Focus Algorithm (CSFA) for high-speed, full-slide digitization under oil immersion, and a two-stage deep learning AI pipeline (YOLOv5 for sensitive candidate detection and ResNet-18 for specific classification) for automated acid-fast bacilli (AFB) identification. We prospectively and retrospectively evaluated its diagnostic performance against conventional smear microscopy, culture, and Xpert MTB/RIF in 1097 patients. Results The results indicate that in the 1097 study population, WSI-TB showed an overall sensitivity of 42.43% and a specificity of 100.00%. Its sensitivity was higher than that of traditional acid-fast staining smear (18.80%) and culture method (30.36%). Compared with other methodologies, the sensitivity was significantly improved. In the sputum smear microscopy group with 600 visual fields, the positive rate of WSI-TB compared with manual microscopy was 42.43% versus 18.8%; in the sputum culture group, it was 43.46% vs 30.36%; in the Xpert group, it was 62.95% versus 44.26% CONCLUSIONS: The WSI-TB technology significantly improves the sensitivity of tuberculosis sputum smear testing while maintaining 100% specificity, providing a new approach to enhance TB detection rates.