Research on the Construction of an Artificial Intelligence Precision Diagnosis Model for Pulmonary Tuberculosis Based on Multi-modal Data Fusion

Abstract

Objective: To evaluate the value of a deep learning-based CT-assisted diagnosis model for pulmonary tuberculosis in clinical applications. Methods: The case data of 2000 patients who underwent chest CT plain scans at the Chongqing Public Health Medical Treatment Center from March 2021 to July 2024 were retrospectively collected. The patients were divided into the normal lung group (1300 cases), the common pulmonary infection group (100 cases), and the pulmonary tuberculosis group (600 cases). The data set was divided into the training set (1400 cases, 70.0%) and the test set (600 cases, 30.0%) by random grouping (achieved through the sample function of R language for complete random grouping of the training set and test set). All images used the lung field automatic segmentation algorithm to obtain the lung field area. Further, the DenseNet algorithm based on the mixed-domain attention was used for classification research. The classification performance of the model was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy. Finally, in the test data, the optimal model was compared with the diagnostic results of three radiologists of different seniority. Results: The diagnostic performance of the model in the training set and test set was AUC = 0.899, 0.831; ACC = 89.4, 84.1%. Among them, the diagnostic performance of the improved DenseNet model was superior to that of junior doctors (accuracy rates were 90.9% and 89.4%, P = 1.000, Kappa = 0.679), and was highly consistent with the diagnostic levels of middle-aged and senior doctors (accuracy rates were 90.7%, 92.4%, and 95.7%, Kappa values were 0.749 and 0.821). Conclusion: The DenseNet model can accurately identify secondary pulmonary tuberculosis and is comparable to the diagnostic level of middle-aged radiologists, and can be used as an auxiliary diagnostic tool for secondary pulmonary tuberculosis.