IVF — Deep Learning for Follicle Growth and Trigger-Day Decisions
Transformer and agent-based ML for assisted reproduction; collaboration with Imperial College London and Prof. Scott Nelson.
A multi-year line of work in collaboration with Prof. Waljit Dhillo’s group at Imperial College London, the Wolfson Fertility Centre, and Prof. Scott Nelson (University of Glasgow), applying modern machine learning to assisted conception.
Agent-based AI for individual follicular growth (first-author, latest work). I led the development of an agent-based artificial intelligence model that simulates the growth of individual follicles during ovarian stimulation — a technically novel approach that overcomes the limitations of population-level models. The framework lets clinicians forecast individualised final follicle profiles from sparse early-cycle scans, with direct implications for trigger-day timing and oocyte yield. Published in The Journal of Clinical Endocrinology & Metabolism (Hramyka, Kelsey, Abbara, Hanassab, Nelson, Dhillo, et al.).
Decoder-only transformer for follicle distribution prediction. Traditional methods for predicting ovarian follicle growth rely on the clinically infeasible assumption of tracking individual follicles between ultrasound scans. We built a decoder-only, GPT-style transformer that autoregressively forecasts the full follicle size distribution from sequential scan data — the first application of a transformer architecture to distribution-level follicle prediction. Trained on 28,125 IVF/ICSI cycles (2005–2023) from 11 clinics in the UK and Poland, with a composite loss combining Earth Mover’s Distance, Chi-Square, and Quadratic Chi-Square distance. Hyperparameters optimised with Optuna. Manuscript in preparation.
Explainable AI for trigger-day decisions. With colleagues at Imperial and Glasgow we developed interpretable ML to determine the follicle sizes on the day of trigger that maximise mature oocyte yield, published in Nature Communications and Fertility & Sterility.
Earlier work characterised endocrine requirements for oocyte maturation following hCG, GnRH agonist, and kisspeptin during IVF (Frontiers in Endocrinology) and the use of anti-müllerian hormone in PCOS diagnosis.