Convolutional neural network based on T-SPOT.TB assay promoting the discrimination between active tuberculosis and latent tuberculosis infection.

OBJECTIVES: The study aims to investigate the potential of convolutional neural network (CNN) based on spot image of T-SPOT assay for distinguishing active tuberculosis (ATB) from latent tuberculosis infection (LTBI). METHODS: CNN was applied to recognize and classify T-SPOT spot image. Logistic regression was used to establish prediction model based on CNN. RESULTS: Areas under the receiver operating characteristic curve (AUCs) of early secreted antigenic target 6 (ESAT-6) CNN, culture filtrate protein 10 (CFP-10) CNN, and phytohemagglutinin (PHA) CNN were more than 0.7 in differentiating ATB from LTBI, while the performance of these indicators was significantly better than that of spot number. Furthermore, prediction model based on the combination of CNNs yielded an AUC of 0.898. The model presented a sensitivity of 85.76% and a specificity of 90.23%. CONCLUSIONS: The current study identified CNN based on T-SPOT spot image with the potential to serve as a tool for TB diagnostics.