Study on artificial intelligence to judge the activity of tuberculomas

Abstract

Objective

This study aims to explore the auxiliary diagnostic value of artificial intelligence (AI) in determining the activity status of various pulmonary tuberculosis lesions, including but not limited to tuberculomas. By utilizing AI technology to automatically segment tuberculoma lesions in CT images and combining manual adjustment of the region of interest (ROI) to ensure the accuracy of analysis, the study ultimately aims to quantitatively evaluate the activity of tuberculomas.

Methods

A total of 112 patients with pulmonary tuberculomas were retrospectively analyzed. Among them, 60 patients had active tuberculomas and 52 patients had inactive tuberculomas, with a total of 172 tuberculomas (108 active and 64 inactive) studied on chest CT images. AI technology was employed to automatically segment various pulmonary tuberculosis lesions, including tuberculomas and other relevant types, and manual ROI adjustment was performed on some lesions. Statistical analyses, including the T-test and ROC curve analysis, were subsequently carried out to determine differences, thresholds, and calculate the accuracy, sensitivity, and specificity of the diagnosis.

Results

The study revealed significant differences in volumetric CT values between active and inactive tuberculomas. The AUC value of the ROC curve analysis was AUC = 0.997, with an optimal threshold of 45.5 HU. The sensitivity, specificity, and accuracy of the method achieved high levels.

Conclusion

This study demonstrates that utilizing AI technology to measure volumetric CT values of various pulmonary tuberculosis lesions, including tuberculomas, can accurately determine their activity status, enhancing the diagnostic accuracy and applicability across different manifestations of pulmonary tuberculosis.