Automated detection of cell division and movement in time-lapse images of developing bovine embryos can improve selection of viable embryos

Abstract

Listen

Translate article

Many studies have demonstrated a correlation between early cell division to the 2-cell stage and subsequent viability of individual embryos. The objective of this study was to: 1) evaluate a novel algorithm for unattended analysis of time lapse images to quantify cellular movement during embryo development; 2) produce a trainable mathematical model to predict the viability of embryos based on the quantified processes; and 3) compare the ability of the model to identify viable bovine embryos with the morphological assessment by a skilled embryologist. 95 bovine embryos were placed in a time-lapse microscope under constant temperature, humidity and CO2 for seven days. Images were acquired twice per hour from 24 hours to 96 hours after fertilization. The ability of the image-analysis procedure to correctly identify the 38 embryos that subsequently (i.e. after 7 days) developed to expanded blastocysts was evaluated and compared to the quality assessments by an embryologist based on the same 145 images for each embryo. Bovine immature cumulus-oocyte complexes were aspirated from slaughterhouse-derived ovaries, matured for 24h before fertilization for 22 h. Cumulus cells were then removed and presumptive zygotes were transferred and cultured in synthetic oviduct fluid medium. Time-lapse images were acquired inside an incubator box fitted onto an inverted microscope stage. The exact onset and duration of cell-divisions could be quantified based on position, shape and size of the recorded peaks in blastomere activity. The blastomere activity pattern of a given embryo could be described by a set of key parameters corresponding to peak height, position and width for prominent peaks as well as similar parameters describing the blastomere activity level between peaks. A total of 55 parameters for each embryo was used in a linear model to classify the embryo as viable or non-viable. Though the model was only a simple model with limited accuracy the fully automated analysis was better at predicting which embryos would develop to expanded blastocysts (Error rate: 20%), than the trained embryologist (Error rate 26%), Moreover the automated analysis also had fewer false positives (29%, as opposed to the manual analysis which had 38%). Even the very simple model used here to interpret the data had lower error rates than a trained embryologist. Furthermore analysis of the training data clearly show that a much better model can be developed when more data are available.