Employing digital twins in the pursuit of real ones: leveraging machine learning to improve assisted reproductive technology outcomes

About the Project

Infertility is a disease of the reproductive system defined as a failure to achieve a clinical pregnancy after 12 months or more of regular unprotected intercourse. Infertility is common, affecting 1 in 6 heterosexual couples, and rates are rising, especially with the current trend of delayed childbearing. Infertility can lead to distress, depression, discrimination and ostracism. Although assisted reproductive techniques (ART) have advanced considerably since their advent, these technologies remain mostly unsuccessful (live birth rate (LBR) of approximately 27% per embryo transfer (ET)), inaccessible and unaffordable. Therefore, it is of upmost importance that ART protocols continue to be improved to ensure the best outcome for patients. Digital twins are computational models of real-world patients that aim to predict their future outcomes. These twins are based on machine learning (ML) and mathematical models. This project will combine a histological dataset of ovulation timed endometrial biopsies (n=100) with demographic data, ML, and mathematical models to predict which patients will be able to achieve a pregnancy following ART. We hypothesise that factors most influential in the prediction of ART success can be targeted to increase LBR with each ET performed.

Objectives

Predict the likelihood of ART success and identify treatment targets that can be modelled with a digital twin.

WP1: Histology and dataset curation

Dataset of successful (control group) and unsuccessful fertility treatment and a timed endometrial biopsy using retrospective biobank specimens from women treated at primary institution will be curated.

WP2: Population-level ML - Can we predict who will have a live birth?

Using curated data, multi-modal deep learning algorithms used to integrate histological features with longitudinal patient data to predict which women will have successful fertility treatment. Starting with standard ML algorithms, eg.convolutional neural networks, the candidate will test several algorithms to predict whose fertility treatment may succeed.

WP3: Validation ML – Can we validate our prediction model?

Prediction model will be used on prospective samples collected throughout the PhD from primary institute, data will provide external validation of the algorithm.

WP4: Digital twin of the endometrium – Can we explain the predictions?

A digital twin will be developed linking aspects of the endometrium, such as vascularisation and gland formation, to outcomes. This can be personalised to an individual such that we can eventually use the twin to develop and test possible therapies to enhance fertility.

Novelty

This collaborative interdisciplinary project uses in silico modelling and longitudinal clinical data to build digital twins to predict success of fertility treatment with the ultimate aim to provide personalised fertility therapies.

Timeliness

With rising infertility rates and decreasing NHS funding for treatments this project directly aligns with the MRC to drive more comprehensive understanding of causes and progression of a common women’s health conditions that occur with high incidence in advanced reproductive age.

Experimental Approach

This PhD incorporates histological methods, AI, mathematical modelling, and software development. The work to be undertaken will be conducted between the Department of Women and Children’s Health and the Department of Cardiovascular and Metabolic Medicine as a collaboration between biomedical engineers (Dr El-Bouri) and clinical fertility experts (Dr Tempest). This project will ideally suit individuals with a background in applied mathematics, computer science, or engineering.