AI reveals key follicle sizes to boost IVF success

By Dr. Priyom Bose, Ph.D.Reviewed by Benedette Cuffari, M.Sc.Jan 14 2025

Study uses explainable machine learning to optimize hormone timing and improve IVF outcomes

Study: Explainable artificial intelligence to identify follicles that optimize clinical outcomes during assisted conception. Image Credit: Corona Borealis Studio / Shutterstock.com

A recent Nature Communications study utilizes explainable artificial intelligence (XAI) to identify follicle sizes important for relevant downstream clinical outcomes during assisted conception.

Assisted conception and the use of XAI

Current estimates indicate that one in six couples throughout the world is affected by infertility, which is considered one of the most serious global disabilities by the World Health Organization (WHO). Assisted reproductive technology (ART), including in vitro fertilization (IVF) treatment, has emerged as a viable option to support patients experiencing infertility.

Due to the extensive amount of data produced during IVF treatment, along with the complexity and specificity of IVF protocols for each patient, it can be difficult for clinicians to consider all relevant data while devising treatments. XAI can overcome these challenges by handling large complex datasets, thereby enhancing the quality of personalized ART treatments and preventing the underutilization of crucial data.

About the study

Ovarian stimulation (OS) involves the administration of human chorionic gonadotropin (hCG) or a gonadotropin-release hormone (GnRH) agonist that facilitates the maturation of oocytes that can be subsequently fertilized by sperm. Importantly, the timing of hCG or GnRH agonist administration must be carefully determined to ensure that the ovarian follicles are at an optimal size for oocyte production.

The current study utilizes XAI techniques to identify follicle size on the day of trigger (DoT) administration to increase the likelihood of mature oocyte retrieval. It also studied the effect of follicle size on premature progesterone elevation.

To this end, data were obtained from a multi-ethnic cohort of 19,082 female participants between 2005 and 2023 who were receiving treatment in clinics throughout Poland and the United Kingdom. All patients previously underwent a transvaginal ultrasound scan showing a minimum of three follicles of size greater than 10 millimeters (mm) on the same day as the DoT administration.

All patients were between 18 and 49 years old, with a median body mass index (BMI) of 24.17. The antral follicle count was 15.

As the primary outcome, the assessment of oocyte maturity grade detailing metaphase-II oocytes was used. Downstream outcomes such as the number of high-quality blastocysts and two pronuclear (2PN) zygotes were also examined.

Study findings

Follicle sizes on the DoT that contributed most to the number of mature oocytes were identified using a gradient-boosting regression tree model. Follicles between 13 and 18 mm in size contributed most to the number of mature oocytes retrieved, whereas those between 12 and 20 mm contributed the most to the overall number of oocytes retrieved

Follicles within the size range of 14-20 and 13-18 mm were the most important for high-quality blastocysts and 2PN zygotes, respectively. Follicle size for 2PN zygotes remained the same in sensitivity analyses; however, for high-quality blastocysts, 15-18 mm follicles were the most contributory.

Among patients, 35 years of age or younger, follicles between 13 and 18 mm in size were the most important. Comparatively, follicles between 15 and 18 mm in size provided the greatest contribution in patients 35 years of age and older. Follicles sized 14-20 mm contributed most among patients who received “long” protocol cycles as part of the IVF treatment, whereas follicles within the range of 12-19 mm were most contributory in “short” protocol cycles.

Assessment of a multilayer perceptron model showed a higher mean absolute error (MAE) and identified 14-18 mm follicles as the most important. This improved and reduced to 2.54 when potential aberrant data were excluded from the analysis.

Although follicles of 13-18 mm in size had the highest relative contribution, other follicles were also contributory, although to a relatively lesser extent. The trained model accounting for each follicle size and the relative contribution of a follicle of that size had more predictive performance.

There was a modest improvement in MAE on including other variables such as BMI, age, and IVF protocol. Overall, the most important factor was the knowledge of follicle size on the DoT.

The proportion of follicles 13-18 mm in size on the DoT was positively associated with the live birth rate (LBR) when adjusted for total follicle count, age, and type of trigger administered. A negative association was observed between LBR and mean follicle size.

Progesterone elevation on the DoT was associated with reduced LBR, whilst the mature oocyte yield remained similar. Furthermore, as the number of follicles greater than 18 mm in size on the DoT increased, serum progesterone on the DoT also increased.

Conclusions

Intermediately sized follicles on the DoT were found to significantly contribute to the retrieval of mature oocytes and embryo development. These findings demonstrate the potential of XAI techniques to offer data-driven optimization of IVF treatment, which can lead to improved downstream clinical outcomes.