Objective To explore the clinical value of ultrasound radiomics in differentiating parathyroid adenoma (PA) and parathyroid hyperplasia (PH). Methods Totally 133 patients with hyperparathyroidism (181 lesions in total) in our hospital from January 2019 to June 2023 were selected

including 66 cases of adenoma (67 lesions) and 67 cases of hyperplasia (114 lesions). The clinical data and conventional ultrasound characteristics of the two groups were compared. A 7:3 ratio was used to divide the lesions into two sets: training set (126 lesions) and validation set (55 lesions). The region of interest (ROI) was delineated based on ultrasound images and the ultrasound radiomics features were extracted. The least absolute shrinkage and selection operator (LASSO) regression was used to screen feature and construct the radiomics model. Multivariate Logistic regression was used to screen important conventional ultrasound features and construct the conventional ultrasound model. Multivariate Logistic regression was applied to construct the combined model based on radiomics features and conventional ultrasound features. The receiver operating characteristic (ROC) curve was drawn to evaluate the diagnostic efficiency of each model and to verify them in the validation set. The calibration curve was drawn to analyze the consistency between the effect and the actual result of the radiomics model and the combined model. Results A total of 8 radiomics features were selected for constructing the radiomics model based on LASSO. Multivariate Logistic regression analysis showed that lesion maximum diameter

peripheral high-eco bright line and polar vascular sign were important ultrasound features for differential diagnosis of PA and PH (both P<0.05). In the training set and validation set

the area under the curve (AUC) of the radiomics model AUC was 0.764 and 0.750

the AUC of the conventional ultrasound model was 0.812 and 0.838

and the AUC of the combined model was 0.825 and 0.856. The AUC of the combined model was higher than that of the radiomics model in both the training set and validation set

and the differences were statistically significant (both P<0.05). The comparison of the AUC among other models showed no statistical significance. Conclusion The combined model constructed by combining radiomics features and conventional ultrasound features can accurately discriminate parathyroid adenoma and parathyroid hyperplasia

which can provide a certain reference for clinical surgical planning.