Is antral follicular recruitment wave-like or continuous? A prospective evaluation of temporal patterns using AI-assisted ultrasonography

Abstract

Study question Does antral follicular recruitment (AFR) in healthy reproductive-age women occur in deterministic or discrete waves or does it follow a continuous and stochastic pattern? Summary answer AFR seems to be a continuous stochastic process, without regular wave patterns, showing only short-term fluctuations and no dominant cyclicity observed within or between individuals. What is known already Follicular recruitment in humans has traditionally been described using single-wave or multi-wave models across the menstrual cycle. However, increasing histological, endocrine and clinical evidence suggests that early antral follicles may grow continuously and independently of menstrual cycle phase. Human studies supporting wave-like behavior have been limited by short observation periods and lower than current imaging quality. Recent advances in high-resolution ultrasonography and artificial intelligence–assisted follicle tracking can enable more precise assessment of antral follicle dynamics over time. Study design, size, duration This single-center prospective observational study was conducted between August 2024 and April 2025. Twenty healthy volunteers aged 18–38 years with regular menstrual cycles were followed for approximately 90 days with serial transvaginal ultrasonographic antral follicle count (AFC) performed using AI-assisted Folliscan (MIM Fertility® Sp. z o.o. (Warsaw, Poland)) software every three days, yielding a total of 598 assessments. Participants/materials, setting, methods A random intercept was defined for time-dependent total antral follicle counts, and the presence of regular cyclical patterns was tested using autocorrelation (ACF) and spectral analysis methods. Linear, polynomial, natural cubic spline, and Fourier/sinusoidal functions were compared. An autoregressive correlation structure (AR(1)) was incorporated to model serial dependence, and model fit was assessed using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Root Mean Square Error (RMSE), marginal R², and residual standard deviation (σ). Main results and the role of chance Participants had a mean age of 27.9 ± 4.4 years, mean BMI of 22.7 ± 3.1 kg/m² and median AMH level of 2.83 (2.04–3.68) ng/mL. Neither visual analyses of AFC by time plots nor time-series analysis of total AFC showed a consistent periodic wave pattern beyond minor short-term oscillations. Autocorrelation analysis demonstrated weak and short-lived positive correlations between consecutive measurements, indicating limited temporal dependence without long-term cyclicity. Spectral analysis revealed no dominant frequency peaks which may support a wave-like model. Linear mixed-effects models comparing linear, polynomial, natural cubic spline and sinusoidal functions showed that an AR(1)+spline model provided the lowest AIC (4293), BIC (4329), RMSE (8.67), and residual standard deviation (8.85). However, explanatory power remained low (marginal R² = 0.07) and the improvement over alternative models was minimal. These findings suggest that observed AFC fluctuations are largely stochastic and unlikely to reflect a structured or chance-driven cyclical process. Limitations, reasons for caution Findings may not be directly generalizable to women with diminished ovarian reserve, endocrine disorders or infertility. Wider implications of the findings Evidence supporting continuous follicular development provides a physiological basis for random-start and dual stimulation strategies, potentially allowing greater flexibility and personalization in assisted reproductive treatments. Yet, the absence of a detectable follicular pattern also indicates that defining an ideal start day for stimulation may not be possible in clinical practice. Trial registration number No