The accuracy of CT imaging in differential diagnosis of accidental thyroid nodules based on histopathology findings

Abstract

Introduction: To explore the differential diagnosis effect of CT imaging histology model on incidental thyroid nodules that is otherwise not available using routine techniques. Method: The clinical and imaging data of 139 thyroid nodule patients pathologically confirmed in Western Theater General Hospital of China from October 2019 to December 2020, were analyzed retrospectively. The 41 benign and 98 malignant nodules were proportionally (7:3) divided into training set (n = 98) and verification set (n = 41) based on gender and age of the patients. Result: After screening, optimum efficiency towards diagnosing accidental thyroid nodules was achieved by the model constructed from 9 imaging histology features using SVM method. Specifically, the area under curve (AUC) of ROC was 0.94 in training set (95% confidence interval: 0.89–0.98), the sensitivity, specificity and diagnostic accuracy were 86.9%, 89.6% and 95.2%, respectively. While the AUC of prediction model in verification model was 0.78 (95% confidence interval: 0.60–0.93), the sensitivity, specificity and diagnostic accuracy were 68.9%, 83.3% and 90.9%, respectively. In contrast, the sensitivity, specificity and diagnostic accuracy of routine CT imaging were 86.9%, 89.6% and 95.2%, respectively (OR 95% confidence interval: 2.48–11.38). Conclusion: The imaging histology model based on CT images is characterized by favorable diagnosis effect towards predicting the benign or malignant nature of incidental thyroid nodules, upon which the texture information derived from single-layer image can be effectively used. The combined findings can potentially afford efficient and time-saving strategy via the established models.

Keywords

• Imaging histology
• CT with Histology
• CT imaging
• Thyroid
• Incidental nodule