Ultrasound radimoics nomogram for predicting lymph node metastasis in papillary thyroid carcinoma

Objective To establish an ultrasound (US) radimoics nomogram for predicting central neck lymph node metastasis (LNM) in papillary thyroid carcinoma (PTC). Method A training cohort of 300 patients and validation cohort of 143 patients were established from January to December 2017. US radiomics score was built by using radiomics to analyze ultrasound image of PTC lesion. A nomogram was established by logistic regression from selection of gender, age, thyroglobulin (TG), thyroglobulin antibodies (TGAB), thyroid peroxidase antibody (TPOAB), US radiomics score and US report LNM status. C-index was calculated to evaluate discrimination of this nomogram. Also, calibration curve was drawn to evaluate calibration of it. Results The positive rates of LNM in two cohorts were 29.7% and 34.6%, respectively. In this study, 23 predictive US radiomics features (26:1) were screened out from 609 cases in the training set, which were nonzero coefficients in the LASSO logistic regression model. US radiomics score yield a C-index of 0.802 (95% CI, 0.791 to 0.815) in training cohort and 0.809 (95% CI, 0.801 to 0.816). Through binary logistic regression, indexes were filtered as age, TPOAB, US radiomics score and US report LNM status. The area under curve, accuracy, sensitivity and specificity of this nomogram was 0.801, 0.812, 0.628 and 0861, respectively, from validation cohort. The C-index was 0.811 (95% CI, 0.804 to 0.818). Conclusion US radiomics nomogram can predict LNM in PTC effectively.