An ultrasound-based deep learning radiomic model combined with clinical data to predict clinical pregnancy after frozen embryo transfer: a pilot cohort study.

RESEARCH QUESTION: Can a multi-modal fusion model based on ultrasound-based deep learning radiomics combined with clinical parameters provide personalized evaluation of endometrial receptivity and predict the occurrence of clinical pregnancy after frozen embryo transfer (FET)? DESIGN: Prospective cohort study of women (n = 326) who underwent FET between August 2019 and December 2021. Input quantitative variables and input image data for radiomic feature extraction were collected to establish a multi-modal fusion prediction model. An additional independent dataset of 453 ultrasound endometrial images was used to establish the segmentation model to determine the endometrial region on ultrasound images for analysis. The performance of different algorithms and different input data for prediction of FET outcome were compared. RESULTS: A total of 240 patients with complete data were included in the final cohort. The proposed multi-modal fusion model performed significantly better than the use of either image or quantitative variables alone to predict the occurrence of clinical pregnancy after FET (P ≤ 0.034). Its area under the curve, accuracy, sensitivity, specificity, positive predictive value and negative predictive value of the proposed model were 0.825, 72.5%, 96.2%, 58.3%, 72.3% and 89.5%, respectively. The Dice coefficient of the multi-task endometrial ultrasound segmentation model was 0.89. Use of endometrial segmentation features significantly improved the prediction performance of the model (P = 0.041). CONCLUSIONS: The multi-modal fusion model based on ultrasound-based deep learning radiomics combined with clinical quantitative variables offers a favourable and rapid non-invasive approach for personalized prediction of FET outcome.