Tuberculosis disease severity assessment using clinical variables and radiology enabled by artificial intelligence

Radiology can define tuberculosis (TB) severity and may guide duration of treatment, however the optimal radiological metric to use and which clinical variables to combine it with in the real-world is unclear. We systematically associated baseline chest X-rays (CXR) metrics with TB treatment outcome using real-world data from diverse TB clinical settings. We used logistic regression to associate 10 radiological metrics including percent of lung involved in disease (PLI), cavitation, and Timika score, alone or with other clinical characteristics, stratifying by drug resistance and HIV (n = 2,809). We fine-tuned convolutional neural nets (CNN) to automate PLI measurement from the CXR DICOM images (n = 5,261). PLI is the only CXR finding associated with unfavorable outcome across drug resistance and HIV subgroups [rifampicin-susceptible disease without HIV, adjusted odds ratio 1·11 (1·01, 1·22), P-value 0·025]. The most informed model of baseline characteristics tested predicts outcome with a validation mean area under the curve (AUC) of 0·769. PLI alone predicts unfavorable outcomes equally or better than Timika or cavitary information (AUC PLI 0·656 vs. Timika 0·655 and cavitation best 0·591). PLI>25% provides a better separation of favorable and unfavorable outcomes compared to PLI>50% currently used in some clinical trials. The best performing ensemble of CNNs has an AUC 0·850 for PLI>25% and mean absolute error of 11·7% for the PLI value. PLI is better than cavitation, is accurately predicted with CNNs, and is optimally combined with age, sex, and smear grade for predicting unfavorable treatment outcome in pulmonary TB in real-world settings.

A systematic evaluation of specific CXR findings in combination with clinical variables and their association with unfavorable outcomes in real-world settings is currently lacking. Stratification by severity of pulmonary TB can support personalized treatment, including the identification of patient groups that can be cured reliably with a shortened treatment regimen. Shorter regimens can minimize drug side effects, improve adherence and reduce costs of care. With the wider use of digital CXR and the increased adoption of AI for computer assisted diagnosis, radiology has the potential to be leveraged for multiple uses in the treatment and monitoring of TB disease, including contributing to a more individualized approach to TB treatment.