A nomogram for predicting the nature of thyroid adenomatoid nodules on ultrasound: a dual-center study.

PURPOSE

Thyroid Imaging Reporting and Data System (TIRADS) does not perform well in thyroid adenomatoid nodules on ultrasound (TANU). Therefore, we aimed to generate and validate a nomogram based on radiomics features and clinical information to predict the nature of TANU.

METHODS

A total of 200 TANU in 200 patients were enrolled. Firstly, radiomics nomograms (R_Nomogram) and clinical nomograms (C_Nomogram) were constructed using eight machine-learning algorithms. The best R_Nomogram and C_Nomogram generated the Radiomics-clinical nomogram (R-C_nomogram). We compared the Area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA) of different nomograms. The unnecessary intervention rates were compared between nomograms and the 2017 ACR TI-RADS recommendations.

RESULTS

The R-C_Nomogram had a higher AUC than other nomograms [training cohort: R-C_Nomogram (AUC: 0.922) Vs. C_Nomogram (AUC: 0.825): <0.001, R-C_Nomogram Vs. R_ Nomogram (AUC:0.836), =0.007); validation cohort: R-C_Nomogram (AUC: 0.868) Vs. C_Nomogram (AUC: 0.850): =0.778, R-C_Nomogram Vs. R_Nomogram (AUC:0.684), =0.005). The R-C_Nomogram has the lowest unnecessary intervention rate among all approaches.

CONCLUSION

The R-C_Nomogram exhibited excellent diagnostic performances for predicting the nature of TANU. By incorporating clinical and radiomics features, the R-C Nomogram can reduce unnecessary biopsies and guide treatment decisions such as ultrasound-guided thermal ablation, improving patient management and reducing healthcare resource burden.