Limitations of ploidy prediction by time-lapse morphokinetics and artificial-intelligence-based embryo selection algorithms: trisomies fly under the radar

Abstract

Research question: How does the genetic constitution of embryos impact the accuracy and effectiveness of time-lapse ploidy detection?

Design: A retrospective analysis of chromosomal constitution, morphokinetic characteristics and embryo grading was conducted on 1012 embryos, originating from 386 intracytoplasmic sperm injection cycles at a single clinic. Morphokinetic checkpoints of pronuclear fading, cleavage stages and post-cleavage stages – including start of compaction; time to compacted morula; time to start of blastulation; and time to full, expanded and hatching blastocyst – were recorded for all analysed embryos. Morphokinetic profiles of 363 euploid embryos were used as reference to analyse 649 embryos with aneuploidies, according to their level of gain or loss of chromosomal material. Embryo grading was performed using commercially available time-lapse-based algorithmic embryo selection tools.

Results: Embryos with loss of genetic material and embryos with multiple aneuploidies exhibited consistent developmental delays that accumulated and became significant at the time of blastulation. In contrast, embryos with chromosomal gains displayed a morphokinetic profile almost identical to that of euploid embryos. Subset analyses suggested that aneuploidies of large chromosomes tend to have a greater impact on the morphokinetic profile. Time-lapse-based algorithmic or artificial intelligence scoring downgraded blastocysts with loss of chromosomal material or with complex aneuploidies, but discrimination potential was not observed between embryos with gain of chromosomes and euploid embryos.

Conclusions: Embryos with extra chromosomes show similar morphokinetic patterns to euploid embryos, reducing the effectiveness of non-invasive selection tools. However, embryos with monosomies or multiple aneuploidies experience significant delays in reaching blastulation, and generally score lower on time-lapse selection algorithms.