Experimental results on the recognition of embryos in human assisted reproduction

Abstract The recognition of embryos suitable for transfer in human assisted reproduction is important, and there is evidence that the morphology of the cells may influence the results achievable. A procedure for this recognition problem has been formulated based on morphological attributes of the images of the embryos, and it is therefore useful to compare the recognition of experts with that of a machine programme. The aim of this paper is to compare the precision in the recognition of viable embryos by a group of experts to that of a machine recognition procedure, both for a basic set of embryos and a blind set. Experts were asked to classify the images of 249 embryos transferred to 73 patients, indicated as the training set and another set of 103 embryos transferred to 35 patients called the blind set. A machine programme was used for the same classification. For all the experts the results were statistically not significantly different from independence, which means that viable embryos are not recognized as such for both data sets. Instead, the machine algorithm recognizes in a statistically significant way, the membership class for the embryos submitted. Cell morphology is important for IVF, but differences do not appear to be discernable by the senses, clinical insight, experience and/or training, while classification by machine methods provides more accurate results, which could be improved by enlarging the training set.