Diagnostic Performance of Different TI-RADS in the Evaluation of Thyroid Nodules in Comparison to and in Combination with a Computer-aided Diagnostic System

Abstract

Abstract Background This retrospective study aimed to compare the diagnostic performance of CAD and various TI-RADS applied by sonographers, and to assess if CAD can improve the diagnostic performance of sonographers. Methods 95 patients with a total of 150 thyroid nodules (TNs) were enrolled in the study. Three sonographers and the S-thyroid CAD system separately evaluated the thyroid images according to the TIRADS classification. Four mixed models were created using doctors’ classification and S-thyroid system evaluation (model1: Eu-TR4 + Eu-TR5 and risk, model2: Eu-TR4 + Eu-TR5 or risk, model3: Eu-TR5 and risk, model4: Eu-TR5 and risk) to investigate whether S-thyroid system could assist sonographers in feasibility diagnosis. Results The ICC values for S-thyroid CAD-based ACR-TIRADS and the sonographers-based ACR-TIRADS was 0.76. The ICCs for sonographers and CAD ultrasound features of TNs were 0.60 (shape), 0.77 (edge), 0 .61 (echo), 0.48 (calcification), and 0.46 (composition). Sensitivity, specificity, and AUC of Eu-TIRADS were higher, 90.47%, 92.42%, and 0.94, respectively. However, the differences in ROC curves between these three TI-RADS and the S-thyroid system ACR-TIRADS were not statistically significant (p > 0.05). In the CAD, the AUC, sensitivity, and specificity of risk assessment were higher than those of S-thyroid ACR-TIRADS (0.90, 90.48%, 82.82% vs 79.76%, 90.90%, 0.89). Among the mixed models 1–4, model 1 and model 3 had the higher AUC and PPV, respectively, 0.91 and 94% (model 1) and 0.90 and 96% (model 3). When two sonographers with different qualifications classified TNs using Eu-TRIADS, the AUC was 0.94 for both. Additionally, the AUC was higher in model 1 and model 3, 0.90 and 0.93, respectively. Conclusions C-TIRADS, ACR-TIRADS, Eu-TIRADS, and S-thyroid ACR-TIRADS are all effective in classifying TNs. S-thyroid CAD-based ACR-TIRADS and the sonographers-based ACR-TIRADS showed excellent interobserver agreement. CAD system may become an additional diagnostic method for the diagnosis of TNs and improve the diagnostic performance of less experienced sonographers.