Modelling Follicular Growth During Ovarian Stimulation Using Agent-based Artificial Intelligence.

Abstract

Context: Ovarian stimulation is a key step in medically assisted reproduction (MAR), whereby supraphysiological doses of FSH extend the 'FSH window' and induce multi-follicular growth. However, only limited data exist examining individual follicular growth rates during fertility treatment.

Objective: To model growth rates of individual ovarian follicles during ovarian stimulation in MAR cycles using an agent-based artificial intelligence (AI) model.

Design: Observational cohort study.

Setting: Eleven assisted conception clinics in Europe.

Patients: 11,572 patients (2005-2023) who underwent ovarian stimulation during MAR.

Intervention: Predictive modelling was conducted using 39,698 scans including 434,082 follicles from 12,950 cycles during ovarian stimulation.

Main outcome measures: Daily growth rates of individual ovarian follicles during stimulation were modelled to enable prediction of follicle sizes at the end of ovarian stimulation.

Results: Mean follicle growth rate of ovarian follicles was 1.35mm per day (95% CI 1.346-1.353), and was significantly associated with antral follicle count and FSH dose changes (both p < 0.001). Using only the first scan, the model enabled prediction of follicles sizes within 2mm at the end of ovarian stimulation with 75.0% accuracy (95% CI 74.6-75.3%), increasing to 80.1% (95% CI 79.8-80.5%) when incorporating the first two scans. Predictive performance was stable across clinics, with a mean accuracy of 78.0% in a random training-test split, and 77.1% using cross-validation by clinic.

Conclusion: We utilized advanced AI techniques to progress our understanding of follicle growth dynamics during ovarian stimulation. This model can reliably predict follicle size profiles at the end of stimulation enabling moderation of the number of scans required.