CT-Based Radiomics Analysis of Consolidation Characteristics in Differentiating Pulmonary Disease of Nontuberculous Mycobacterium from Tuberculosis

Abstract Background: Nontuberculous mycobacteria (NTM) grows slowly, the course of disease is longer than that of tuberculosis (TB), and the resistance rate to first-line anti tuberculosis drugs is high, so the overall cure rate is low. Radiomics is a new image processing technology developed in recent years. In this study, CT-based radiomics features are evaluated to differentiatenon NTM pulmonary disease with consolidation characteristics from PTB with similar consolidation characteristics. Methods: A total of 156 patients (75 NTM pulmonary disease and 81 PTB) with the Consolidation were evaluated. 305 regions of interest of CT consolidation were outlined. 80% of consolidations were allocated to the training set and 20% to the validation set using a random number generated by a computer. Three supervised learning classifiers (KNN, SVM and LR models) were used to analyze the features. Results: 63 optimal features were selected by these three methods. The AUC (sensitivity, specificity) for the training and validation cohorts were 0.98 (0.90, 0.94) and 0.97 (0.87, 0.97) for KNN, respectively; 0.99 (0.94, 0.93) and 0.96 (0.80, 0.97) for SVM, respectively; and 0.98 (0.96, 0.97) and 0.95 (0.88, 0.87) for LR, respectively. Precision, Recall and F1-scores determined that KNN performed better at diagnosing early NTM pulmonary disease, with the values of the above three indexes being 0.89 and 0.97 and 0.93, respectively. Conclusion: CT-based radiomics analysis consolidation features can provide effective proof in distinguishing the NTM pulmonary disease from PTB.Among the three classifiers, KNN classifier has the best performance in identifying two diseases.