Machine learning radiomics models based on B-mode and contrast-enhanced ultrasound for assisted diagnosis of benign and malignant thyroid nodules: A prospective study

Abstract

Abstract Objectives To develop a radiomics model for differential diagnosis of thyroid nodules based on B-mode ultrasound and contrast-enhanced ultrasound images. To determine if the diagnostic efficiency of radiologists is improved by the use of this radiomics model. Methods In this prospective single-center study, from October 2021 through January 2023, patients scheduled for thyroidectomy or fine-needle aspiration cytology have been preoperatively examined using a standardized B-mode ultrasound combined with contrast-enhanced ultrasound examination. Radiomics models were developed based on B-mode and contrast-enhanced ultrasound images. Two rounds of reader studies were performed to verify the clinical application value of the model. Results A total of 404 patients were enrolled, and the 412 nodules were split into training and test sets. The AUCs in differential diagnosis of thyroid nodules were 0.799 for the B-US radiomics model, 0.766 for the CEUS radiomics model, and 0.890 for the B-US+CEUS radiomics model. The sensitivity of the B-US+CEUS radiomics model in diagnosis was higher than that of the three radiologists, and the accuracy of the model was higher than the diagnoses of the intermediate and junior radiologists. The diagnostic sensitivity of all radiologists was further improved with the aid of the B-US+CEUS radiomics model. Conclusion The findings of this study suggest that both B-mode ultrasound and contrast-enhanced ultrasound radiomics features offer a high clinical value. Using them in combination leads to improved diagnostic performance. Our B-US+CEUS radiomics model is an effective tool to assist radiologists in differential diagnosis of thyroid nodules.