Artificial Intelligence (AI) in the Fertility Clinic: AI-Driven Embryo Blastulation and Implantation Prediction Can Improve the Performance of Assisted Reproductive Technology

Introduction: Conventional embryo evaluation methods rely on morphology assessments and manual annotations. Though standard in practice, this method is time-consuming for embryologists and fraught with intra- and inter-observer variability. AI models that predict the probability of blastulation and implantation would allow for accurate assessments of embryo viability.

Aim: To assess the performance of two AI models for predicting embryo viability and implantation.

Materials and Methods: Time lapse monitoring sequences from 8,000 blastocyst with known ASEBIR and implantation outcomes were included in this study. Both models used a non-overlapping dataset of 4,000 embryos; 2,800 were used for the training set and 1,200 were used for the test-set. The training set was mainly used for model training and the test-set was used for unbiased performance optimization. The blastulation prediction model employed an artificial neural network (ANN) on embryo day 3 time-lapse data. The implantation prediction model relied on bio-feature selection technique on day 5 time-lapse data. The predictive performance of both models was evaluated using ROC curve analysis.

Results: The ANN model for blastulation prediction reached 0.85 accuracy and 0.89 AUC when predicting blastocyst formation using day 3 time-lapse images only. The implantation prediction model reached 0.72 accuracy and 0.74 AUC. This is clearly superior to the embryologists’ predictive accuracy of 0.51, demonstrating improved overall accuracy in diagnosing embryo viability and implantation using AI.

Conclusions: The implementation of broadly applicable and scalable AI models into the IVF clinic may optimize reproductive outcomes and should be prioritized for further evaluation.