Treatment management algorithm for natural frozen embryo transfer cycles using a real-time ovulation prediction machine learning model.

Our aim was to develop an AI-based NC-FET Treatment Management Algorithm (NTMA), decision-support system that predicts ovulation in real time to manage and optimize natural frozen embryo transfer cycle (NC-FET) scheduling. The algorithm was developed using a “teacher-student” machine learning approach and was trained on a total of 3,975 labeled NC-FET, including 3,432 training cycles and 543 test cycles. A second test group included 166 documented ovulation cycles (documented follicular rupture and LH surge in two consecutive days of ultrasound scans). The algorithm showed high ovulation detection accuracy in both tests’ groups particularly one day before and the day of ovulation (95.4% and 94.6% in the labeled test group and 95.5% and 95.3% in the documented test group, respectively). Most influential predictive features of the algorithm included LH levels, the estrogen/progesterone ratio, and leading follicle size during the monitored test days. The NTMA yielded 92.04% correct prediction 7 in identifying ovulation with an average of 3.1 tests per cycle. We propose an AI-based algorithm for the complete management of a NC-FET. The algorithm shows high accuracy in predicting the time of ovulation and therefore can serve as a useful decision support tool for clinicians in their daily practice. Prospective studies are warranted to validate these results.