EP02.06: Ovarian 3D Vaginal Power Doppler Improves the Predictivity of Ovarian Response in Artificial Intelligence Models

The aim of the present study is to estimate the utility of 3D vaginal power Doppler (3DVPD) ovarian vascularity in the prediction of the number of oocytes collected using Artificial intelligence Models during controlled ovarian stimulation (COS). In fact poor ovarian responses to gonadotrophins leads to cycle cancellation during controlled ovarian stimulation while high responses could jeopardise patient life. In this context, gonadotrophin dose is often individualised using patient features that predict ovarian response (such as age, AFC and AMH. A prospective study was conducted on 296 couples undergoing antagonist protocol at Al HADI IVF centre, Lebanon. It was performed between January 2020 and December 2021. On the second and seventh day of stimulation, vascularity index (VI), flow index (FI), and vascularity flow index (VFI) were measured using the 3DVPD. On day 2, the AFC was also evaluated. Ten Artificial intelligence classifiers were tested using Orange data mining version 3.34.0. Couples were categorised into poor responders (3 eggs or less) according to bologna criteria (36.1%), high responders (10 or more egg collected) (6.7%), and normal responders (between 3 and 10 eggs collected) (57,2%). Of particular interest, when removing 3D vaginal power Doppler parameters from the models, the accuracy of Random Forest (RF) method for the prediction of low responders drops from 63.5% to 48%. And the accuracy of the Random Forest and Support Vector Machine (SVM) for prediction of high responders dropped from 73.3% to 63.5% for RF and to 66.9% for SVM. Assessing ovarian vascularity by 3DVPD during ovarian stimulation can help to customise the dosage of gonadotrophins. It will be necessary to perform analysis on a broad sample size to validate these findings.