AI model to predict ovarian function developed by the University of Tokyo

A research group led by Professor Miyuki Harada of the Graduate School of Medicine, the University of Tokyo, Project Research Associate Hiroshi Koike of the University of Tokyo Hospital (at the time of the research, currently affiliated with the Graduate School of Medicine), Katsuhiko Noda and Kaname Yoshida of SIOS Technology, Inc. (Minami-azabu, Minato City, Tokyo Prefecture), and their colleagues has announced the development of an artificial intelligence (AI) model to predict ovarian function. Using simple methods of medical interviews and blood tests, it can predict not only the number of eggs with higher accuracy than conventional techniques, but also the quality of eggs, for which prediction methods have not been established. It is expected to be utilized for preconception care and personalization/optimization of infertility treatment. The results were published in the Journal of Ovarian Research on July 18.

In recent years, age-related infertility has increased due to lifestyle changes such as late marriage, and the number of assisted reproductive technology (ART) treatment patients is on the rise. The decline in fertility associated with aging is thought to be mainly due to the decline in ovarian function.

The pregnancy rate per embryo transfer after in vitro fertilization is relatively high at around 45% from the twenties to the early thirties, but it declines rapidly with age, dropping to 25% at around age 40 (the age of many treatment patients) and, significantly, to 8% at age 45. The number of eggs in the ovaries continues to decrease from 2 million at birth to 300,000 at puberty and 1,000 at menopause.

Therefore, the research group aimed to develop a tool that can predict ovarian function with high accuracy using available clinical information before starting infertility treatment. They collected detailed data on ovarian function and other parameters from the University of Tokyo Hospital and affiliated facilities and constructed multiple prediction models. Furthermore, they selected useful indicators to improve model accuracy and narrowed down the target items.

As a model to predict the number of eggs, they adopted a random forest model, which is a machine learning model. It was confirmed that a model that uses 5 items as input showed the highest prediction accuracy and this accuracy was higher than the prediction accuracy of the anti-Müllerian hormone (AMH) values currently in use.

In addition, a random forest model that uses 14 items as input showed the highest prediction accuracy as a model to predict egg quality. Although research has been conducted on egg quality using various data obtained after starting infertility treatment, no method has reached clinical application and no such method exists. This showed the possibility of predicting egg quality with less burden on the body.

The utilization of this model will enable early health management (preconception care) with future pregnancy in mind.

The ovarian function prediction model developed in this research can predict "the number and quality of eggs" with high accuracy by inputting interview content such as age and menstrual cycle and items determined from a small amount of blood sampling. If highly accurate prediction of ovarian function becomes possible for individual patients, optimized medical care can be provided according to each person's condition.

Harada commented: "The most groundbreaking aspect of this research is that we showed the possibility of predicting ovarian function from medical interviews and blood test results alone. Although further research is needed to improve the accuracy of the model, we are proceeding with research with the expectation that in the future, this will become a tool that allows women to know their ovarian function through reliable tests and can be used for preconception care, or enables the provision of personalized medicine in clinical settings with a thorough understanding of each patient's individual ovarian function."

Journal Information
Publication: Journal of Ovarian Research
Title: Assessment and prediction models for the quantitative and qualitative reserve of the ovary using machine learning
DOI: 10.1186/s13048-025-01732-0