TLC, IT and Artificial intelligence
2020年度 年次大会-講演抄録 | タイムラプスシステムと AI 技術の融合
学会講師 : Simon Cooke
For two decades, almost all clinics in the world have used a variation of the original Gardner alpha-numeric blastocyst grading methodology1） , to try and determine which blastocyst should be ranked higher than the rest of the cohort, and be selected for embryo transfer. Some manufacturers have tried to improve on this with inbuilt annotated algorithms 2), with other independent authors demonstrating their own in-house algorithms have limited use 3). The time taken for lab staff to perform these annotation processes （particularly in large labs）, whilst reducing the intra-observer errors during annotation is a major drawback, and has fueled the desire for full automation.
As systems move away from using 2D static images, and into 3D and 4D decision making, and with the advent of timelapse incubators with operating software that captures huge amounts of data, the next challenge is “how to manage and analyse” these mountains of new data. This has pushed researchers into the area of complex mathematics and computer modelling, and into the world of Artificial Intelligence（AI）to build accurate models that are independent of labs, countries, patient age and culture medium used.
Indeed, the world’s first automated timelapse AI analysis system that can accurately predict the fetal heart potential in a cohort of blastocysts with an AUC of 0.93 has recently been published 4） . Deep learning has also been used in other systems, which mostly revolve around the ability for AI to differentiate between embryos of differing morphology 5）. Laboratory methods, and most importantly the usage and degree of automation of AI will be discussed, and the differences explained, and compared to published accuracies. An exciting new era of embryo assessment based on pure mathematics and proof has already begun.