A prior knowledge-informed traceableNeutral Network modeling only using regular laboratory results to assist early diagnosis for tuberculosis: a multiple-center study in China

Abstract

AbstractBackground:To construct a knowledge-informed traceable artificial intelligence (AI)-based model to assist early diagnosis for tuberculosis (TB).Methods:60729 cases were extracted from January 1, 2014, to December 31, 2021, in Beijing Hepingli Hospital. Beijng Jishuitan Hospital was used as an independently external testing set. Only using routine laboratory results, six models based on Neutral Network (NN) algorithm combined with clinical prior knowledge were designed for TB screening and differentials were set up. Our TB model was not only quantitatively evaluated by means of metrology, but also validated by an independently external testing set from Beijing Jishuitan Hospital, and by on-site clinical validation in 37 hospitals.Results:For disease screening, our NN algorithm overall performed better than the other algorithms for diseases & healthy control (HC), and TB & non-TB models. Taking an example for the TB& non-TB model, the AUC, ACC, SPE and SEN were 0.9240, 0.7703, 0.7664 and 0.8958 respectively. For disease differentials, The AUC was 0.8035 for pulmonary tuberculosis (PTB) & other pulmonary diseases (OPD) model; the AUC was 0.7761 for tuberculosis（TB）& extrapulmonary tuberculosis（EPTB）model. For an on-site clinical validation in Baoding No.2 Central Hospital, the average accuracy was stable, achieving 93% for TB& non-TB model.Conclusions:A knowledge-informed AI-based model only based on regular laboratory results offers a more convenient, effective, and highly accurate early diagnosis tool for TB.