OC01.05: Ultrasound radiomics for differentiating borderline from epithelial ovarian cancer, compared with intraoperative frozen section

An accurate preoperative diagnosis method is crucial for patients with borderline ovarian tumours (BOT) who want to preserve their reproductive function, as the accuracy rate of intraoperative frozen section pathology is only 70-80%. This can lead to unnecessary overtreatment or additional surgeries. This study aims to establish an ultrasound radionics model, including colour Doppler features to distinguish between BOT and epithelial ovarian cancer (EOC) and compare the accuracy with intraoperative frozen section (IFS). In this multicentre retrospective study, patients diagnosed with BOT or MOC underwent grayscale ultrasound and colour Doppler from August 2013 to December 2021. Five centres were involved, with centre I divided into training and validation sets and centres II to V for the test set. Radiomics features were extracted, reduced, and selected to construct an optimal model with a feature range of 10 to 60. We built four models based on the colour Doppler model (CD), grayscale model (GS), GSCD model and C-GSCD with clinical features. A total of 1,041 patients were included in the training set (344 BOT and 489 MOC) and in the validation set (85 BOT and 123 MOC) in centre I. In the test set from centre II to V, a total of 203 patients were included (91 BOT and 112 MOC). The highest mean AUC value of 0.9301 was achieved with a feature number of 20 in the training set. The C-GSCD yields the best diagnostic performance to identify BOT, AUC was 0.985, 0.973 and 0.816 in the training, test and validation set. Additionally, the C-GSCD showed good prediction between BOT and Stage I EOC with AUC values of 0.930. the accuracy of IFS was 79.5% (828/1041) for patients from centre I. The radiomics model, utilising grayscale and colour Doppler images, has demonstrated the ability to differentiate between borderline and malignant ovarian tumours. The diagnostic accuracy of radiomics model surpasses IFS diagnosis.