Differential diagnosis of pulmonary tuberculosis and talsromycosis marneffei by computed tomography-derived radiomics in patients with acquired immunodeficiency syndrome

OBJECTIVE: To investigate the value of computed tomography (CT)-derived radiomics features in the differential diagnosis of pulmonary tuberculosis (PTB) and talaromycosis marneffei (TSM) in patients with acquired immunodeficiency syndrome (AIDS). MATERIALS AND METHODS: The venous phase images for 166 patients with AIDS (PTB, n = 66; TSM, n = 99) were retrospectively analyzed, and the radiomics features of lung lesions and mediastinal lymph nodes were extracted. The samples were divided into a training set and a test set in a ratio of 8:2. The optimal eigenvalues were used to establish four prediction models: radiomics model 1 (PTB group and TSM lung lesions), radiomics model 2 (PTB group and TSM lung lesions), radiomics model 3 (pulmonary lesions without lymph node enhancement), and radiomics model 4 (pulmonary lesions with lymph node enhancement). The working characteristic curve was used to evaluate the predictive performance of the model. RESULTS: The accuracy, sensitivity, specificity, and area under the curve values were 0.67, 0.78, 0.78, and 0.735, respectively, for the radiomics model 1 test set; 0.67, 0.62, 0.67, and 0.654, respectively, for radiomics model 2; 0.89, 0.76, 0.80, and 0.833, respectively, for radiomics model 3; and 0.76, 0.80, 0.88, and 0.886, respectively, for radiomics model 4. CONCLUSION: The prediction model based on CT-derived radiomics features has value for the identification of PTB and TSM. The radiomics model based on the optimal eigenvalues of lung lesions combined with lymph node plain scan images is compared with the establishment of a single lung. The focal omics feature model has better predictive power.