SUPPLEMENTARY MATERIAL.docx

<b>A concise statement of the major contents</b> The study developed and validated a nomogram combining BMUS radiomics and the ACR TI-RADS. The predictive performance and stability of the nomogram was assessed with respect to discrimination, calibration and clinical application in the validation and entire cohort. The nomogram yielded an AUC of 0.889 and 0.849 in the training and validation cohort, and the calibration curve were better. Higher interobserver agreement and a significantly improved AUC, NRI and IDI were found. Used the nomogram cutoff, radiologists showed significantly decreased unnecessary FNA rate. The nomogram is an intuitive and convenient tool which could considerably improve the predictive performance of radiologists with different levels of experience and reduce the unnecessary FNA rate during the treatment of thyroid nodules. <b>Keywords </b>Thyroid nodule; Radiomics; Nomogram; Ultrasound; Prediction <b> </b> <b>Abstract</b><b></b> <b>Context:</b> Visual interpretation and radiomics of B-mode ultrasound (BMUS) images both show good performance in predicting the malignancy of thyroid nodules.<b> </b> <b>Objective:</b> Develop and validate a nomogram combining BMUS radiomics and the American College of Radiology (ACR) Thyroid Imaging Reporting and Data System (TI-RADS). <b>Design:</b> Retrospective cohort study <b>Methods:</b> A total of 451 pathologically proven thyroid nodules at our hospital (January 2019 to September 2020) were randomly assigned to training and validation cohorts (7:3 ratio). Combining the BMUS radiomics score (Rad-Score) with the average ACR TI-RADS score (ACR score) from a senior and junior radiologist, a nomogram was developed from the training group. The predictive performance and stability of the nomogram was assessed with respect to discrimination, calibration and clinical application in the validation and entire cohort. <b>Results</b> The nomogram yielded an AUC of 0.889 in the training cohort and 0.849 in the validation cohort, which were significantly better than the average ACR score alone (<i>p</i>=0.002 and &lt;0.001 respectively). The goodness of fit and calibration curve of the nomogram were both better than the ACR score. Higher interobserver agreement and a significantly improved AUC, net reclassification index (NRI) and integrated discriminatory improvement (IDI) of the nomogram were found compared with the corresponding ACR score, as independently scored by a senior and junior radiologist (all <i>p</i> &lt;0.001). <b>Conclusion</b> The nomogram is an intuitive and convenient tool that combines radiomics with the ACR TI-RADS, which could considerably and stably improve the predictive performance of radiologists with different levels of experience.