Artificial intelligence versus the embryologist; could the embryologist be replaced by the machine for embryo developmental assessment?
The objective of this large, multicentre study was to assess whether manual annotation of time-lapse images performed by experienced embryologists could be reliably replaced by fully automated annotation powered by artificial intelligence.
A large proportion of an embryologist’s time is spent assessing embryo quality and making decisions about their use. We decided to analyse automated annotation to see if this could save the scientists among us some valuable time. We compared manual annotation with automated annotation using a simple agreement statistic; a two-way, mixed intraclass correlation coefficient (ICC) for consistency. This test provides five categories of agreement based on the ICC score; very weak (0-0.20), weak (0.21-0.40), moderate (0.41-0.60), strong (0.61-0.80) and very strong (0.81-1.00).
the automated annotation was either in strong or very strong agreement with the manual annotations
This retrospective cohort analysis of 2442 inseminated oocytes used data from four UK fertility clinics. Embryos had previously been annotated, by an embryologist from pronuclei fading (tPNf) to time of expanded blastocyst (tEB) with all cell stages annotated in between (time to two-cell (t2), three-cell (t3), four-cell (t4), five-cell (t5), six-cell (t6), seven-cell (t7), eight-cell (t8), nine-cell (t9), morula (tM), start of blastulation (tSB) and full blastocyst (tB)). Images for each embryo, collated using the EmbryoScope, were analysed by computer software (Fairtility©, Israel) for each developmental time-point.
We found that, across all developmental time-points, the automated annotation was either in strong or very strong agreement with the manual annotations. The lowest ICC was 0.72 (t9) and the highest was 0.92 (tPNf). These results meant that we could conclude that automated annotation is consistent with manual annotation.
The use of automated annotation could lead to lead to significant time-saving in the laboratory with each embryo typically taking several minutes to annotate. Not only could we save some time, but automated annotation also offers consistency and reproducibility of annotations which, as we well know, are highly susceptible to subjectivity.