Super‐Resolution Ultrasound Radiomics Can Predict the Upstaging of Ductal Carcinoma In Situ

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Introduction: Preoperatively distinguishing pure ductal carcinoma in situ (DCIS) from upstaged DCIS is important for deciding optimal surgical strategies. However, it is hard to preoperatively predict the upstaging of biopsy‐proven DCIS. This study aims to develop an effective radiomics model for predicting the upstaging of DCIS based on super‐resolution (SR) ultrasound images. Methods: In this multicentre retrospective study, patients with biopsy‐proven DCIS who underwent ultrasound examination were included. A super‐resolution reconstruction algorithm was used to enhance the resolution of original high resolution (HR) ultrasound images and obtain SR images. Pyradiomics was used for feature extraction. The selected HR radiomics features and SR radiomics features were combined with clinical features to construct the HR fusion model and SR fusion model, respectively. The area under the receiver operating characteristic curve (AUC) of the models and radiologists was analyzed and compared by the Delong test. Results: A total of 681 women (median age, 47 years; interquartile range, 42–54) with 681 biopsy‐proven DCIS lesions were included, with 422 lesions in the training set, 106 lesions in the validation set, and 153 lesions in the external test set. The SR Fusion model achieved an AUC of 0.819 (0.732–0.887) in the validation set and 0.800 (95% CI 0.728–0.860) in the external test set. It outperformed the radiologists (AUC = 0.603–0.627; p < 0.001) in the validation set. Additionally, it surpassed the clinical model (AUC = 0.682, 95% CI 0.602–0.755; p = 0.02) and the HR Fusion model (AUC = 0.724, 95% CI 0.646–0.793; p = 0.03) in the external test set. Conclusion: The SR Fusion model integrating SR features and clinical features can effectively predict the upstaging of DCIS.
Funding Information
  • National Natural Science Foundation of China (82320108011)
  • Natural Science Foundation of Guangdong Province (2023A1515011564)

This publication has 31 references indexed in Scilit: