Nomogram based on multi-modal ultrasound radiomics for the differentiation of benign and malignant ACR TI-RADS 4 and 5 categories thyroid nodules

Abstract

Abstract Objective To develop a multi-modal ultrasound including B-mode ultrasound (BMUS), shear wave elastography (SWE), contrast-enhanced ultrasound (CEUS) radiomics nomogram and evaluate the diagnostic performance of radiomics nomogram for the differentiation of benign and malignant American College of Radiology (ACR) thyroid imaging reporting and data system (TI-RADS) 4 and 5 categories thyroid nodules. Materials and Methods From March 2019 to April 2022, a total of 133 patients with 139 ACR TI-RADS 4 and 5 categories thyroid nodules were enrolled. These nodules were divided randomly into a training and validation cohort. Radiomics features were extracted from BMUS, SWE and CEUS images. The least absolute shrinkage and selection operator (LASSO) regression and five cross-validations were used to select features and build radiomics score. A multi-modal ultrasound radiomics nomogram was established based on multivariate logistics regression analysis with multi-modal ultrasound including BMUS, SWE and CEUS radiomics score (Rad-score) and clinical characteristics, including BMUS features in the 2017 edition of ACR TI-RADS. The performance of the nomogram was assessed via discrimination, calibration and clinical usage in the training and validation cohorts. Results Multi-modal ultrasound radiomics had better diagnostic performance compared with using BMUS radiomics, SWE radiomics or CEUS radiomics alone. Based on multivariate logistics regression analysis, calcification and multi-modal ultrasound Rad-score were the independent risk factors for the discrimination of benign and malignant ACR TI-RADS 4 and 5 categories thyroid nodules. The nomogram incorporated with these independent risk factors had better diagnostic performance compared with clinical model. The AUC of nomogram in training and validation cohorts were 0.911 (95% CI, 0.850–0.972) and 0.883 (95% CI, 0.776–0.990), respectively. Decision curve analysis (DCA) indicated the nomogram had a high clinical value. Conclusion In conclusion, multi-modal ultrasound radiomics had better diagnostic performance, compared with using BMUS radiomics, SWE radiomics or CEUS radiomics alone. The multi-modal ultrasound radiomics nomogram showed satisfying accuracy for the differentiation of benign and malignant ACR TI-RADS 4 and 5 categories thyroid nodules, which might be a potentially noninvasive and useful method to discriminate benign and malignant nodules, it could assist surgeons in clinical decision making and thyroid nodules management, improve the prognosis of patients.