The application of morphokinetic algorithms to predict ploidy status during assisted conception

Bamford, Thomas James (2024). The application of morphokinetic algorithms to predict ploidy status during assisted conception. University of Birmingham. Ph.D.

Preview

Bamford2024PhD_Redacted.pdf
Text - Redacted Version
Available under License All rights reserved.
Download (21MB) | Preview

Abstract

Aim: To undertake a series of studies to answer 5 key questions within assisted conception:
1. Are morphokinetic variables and morphological features associated with the ploidy status of pre-implantation human embryos?
2. Are artificial intelligence or machine learning algorithms superior to logistic regression for predicting ploidy status?
3. Are morphokinetic model risk scores associated with live birth and miscarriage?
4. Should clinical factors be incorporated into embryo selection models?
5. Are morphokinetic models better at prioritising a euploid embryo for transfer over morphological selection by a senior embryologist?

Methods: The above questions were addressed throughout four studies, first, a systematic review and meta-analysis investigated the association of ploidy status and abnormal cleavage, morphokinetic variables, fragmentation, multinucleation and embryo contraction. Second, a model development study collected data on the prognostic variables investigated in the systematic review from nine IVF clinics. Here, a sample of 8148 biopsied blastocysts was used to develop and compare 12 machine learning models to predict ploidy status. This was using four different algorithms, logistic regression, random forest classifier, extreme gradient boosting and deep learning. One model for each algorithm was built with euploidy as target outcome, a second with aneuploidy and a third using a smaller dataset which incorporated embryo Gardner’s classification. Third, the best performing model was retrospectively externally tested on a total of 3587 single embryo transfers. This determined association between three different model derived aneuploidy risk scores (low, moderate and high) and live birth and miscarriage. The final study used a separate cohort of 1958 biopsied blastocysts to compare the ability of morphokinetic models to rank euploid embryos first, given that these models will not be asked to classify embryos but only prioritise.

Results: Meta-analysis demonstrated that ten morphokinetic variables were significantly delayed in aneuploid embryos. It is uncertain whether the morphological components investigated have prognostic potential. On comparing 12 different models, logistic regression performed the best (AUC=0.61). Including predictors such as age resulted in no variability in the ranking within a patient’s cohort of embryos. Incorporating morphological Gardner’s classification resulted in no improvement in the discriminatory ability of the model. A ‘morphokinetics only’ approach was therefore investigated by adjusted logistic regression analysis that demonstrated the model was not associated with miscarriage when comparing the ‘high’ to the ‘moderate risk’ (OR 0.87; 95% CI 0.63-1.63; p=0.39) or ‘high’ to ‘low risk’ embryos (OR 1.07; 95% CI 0.79-1.46, p<0.63). However, an embryo deemed ‘low risk’ was significantly more likely to result in a live birth than those embryos graded ‘high risk’ (OR 1.95; 95% CI 1.65-2.25; p<0.001). The final cohort study reported that arbitrary embryo selection would rank a euploid embryo first 37% of the time, embryologist selection 39%, and the ploidy morphokinetic model 47% of the time.

Conclusions: Morphokinetic variables and the risk scores derived from morphokinetic models are significantly associated with ploidy status. Including predictors such as age, results in a clinically ineffective model; a ‘morphokinetics only’ approach is therefore advised. Logistic regression was the best performing algorithm in this dataset for predicting ploidy status, with aneuploidy as the target variable. The application of this model resulted in an improved chance of a euploid embryo being selected for transfer over that by a senior embryologist.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):