Towards Automation in IVF: Pre-Clinical Validation of a Deep Learning-Based Embryo Grading System during PGT-A...

Area of Science:

• Reproductive medicine
• Embryology
• Artificial intelligence in healthcare

Background:

• Preimplantation genetic testing for aneuploidies (PGT-A) is effective but resource-intensive.
• Current embryo selection relies on morphological evaluation, which lacks reproducibility.
• There is a need for objective, non-invasive embryo assessment tools.

Purpose of the Study:

• To externally validate the iDAScore v1.0, a deep-learning model for blastocyst assessment.
• To evaluate the association of iDAScore v1.0 with embryo morphology and competence.
• To compare the objectivity and reproducibility of iDAScore v1.0 with human embryologists' evaluations.

Main Methods:

• Retrospective analysis of 3604 blastocysts from 1232 cycles.
• External validation of the iDAScore v1.0 deep-learning model.
• Assessment of iDAScore v1.0's association with euploidy and live-birth prediction.

Main Results:

• iDAScore v1.0 showed significant association with embryo morphology and competence.
• Area under the curve (AUC) for euploidy and live-birth prediction were 0.60 and 0.66, respectively.
• iDAScore v1.0 demonstrated objectivity and reproducibility, unlike human evaluations.

Conclusions:

• iDAScore v1.0 can objectively support embryologists' blastocyst evaluations.
• The model could identify euploid embryos as top quality in 63% of simulated cases.
• Further randomized controlled trials are necessary to determine the clinical utility of iDAScore v1.0.