Methods for assessing the quality of mammalian embryos: How far we are from the gold standard?

Abstract

Morphological embryo classification is of great importance for many laboratory techniques, from basic research to the ones applied to assisted reproductive technology. However, the standard classification method for both human and cattle embryos, is based on quality parameters that reflect the overall morphological quality of the embryo in cattle, or the quality of the individual embryonic structures, more relevant in human embryo classification. This assessment method is biased by the subjectivity of the evaluator and even though several guidelines exist to standardize the classification, it is not a method capable of giving reliable and trustworthy results. Latest approaches for the improvement of quality assessment include the use of data from cellular metabolism, a new morphological grading system, development kinetics and cleavage symmetry, embryo cell biopsy followed by pre-implantation genetic diagnosis, zona pellucida birefringence, ion release by the embryo cells and so forth. Nowadays there exists a great need for evaluation methods that are practical and non-invasive while being accurate and objective. A method along these lines would be of great importance to embryo evaluation by embryologists, clinicians and other professionals who work with assisted reproductive technology. Several techniques shows promising results in this sense, one being the use of digital images of the embryo as basis for features extraction and classification by means of artificial intelligence techniques (as genetic algorithms and artificial neural networks). This process has the potential to become an accurate and objective standard for embryo quality assessment.

Distinct Sources of a Bovine Blastocyst Digital Image do not Produce the Same Classification by a Previously Trained Software using Artificial Neural Network

“…Many embryologists select the oocytes/embryos by a noninvasive examination based on simple observation focused on morphology and kinetics of their developmental stage (third day of culture or blastocyst stage) [17]. Among then, embryo metabolism analysis, cellular respiration measurement, evaluation by time-lapse video [18], the quality on in vitro growth of embryos, the integrity of blastomeres membrane [18, reviewed by 19; 20] and electronmicroscopy analysis [18, reviewed by 21]. Other kinds of evaluation have been used in attempt of find a better solution to classify the embryo quality.…”

“…From photographs of oocyte and embryos at the 4-cells stage, Manna et al [17] used AI techniques that are based on (i) segmentation (selection of the correct region of interest in the image) and preprocessing for reducing artefacts due to noise, blur or illumination conditions; (ii) feature extraction (it is usually dependent of numerical descriptors) that represent in a compact way the starting image (such as LBP) [17,18,30]; and (iii) definition of a classification system, in which the classifier is trained using the data stored in the knowledge base.…”

“…Several authors proposed the use of mathematical and statistic tools for the evaluation of the embryo quality, like the multivariate logistic regression with eight predictive factors for the classification of embryos according to implantation potential [18,31], computer-assisted scoring system (CASS) [18,32] that is supposed to have a higher discriminatory power for embryo selection, over the standard scoring system that has intrinsic examiner variability. It was also used the multivariate logistic regression (LR) system together with multivariate adaptive regression splines (MARS) that had shown improvement on predictive model when using the computer assisted scoring system associated with data mining.…”

“…The Artificial Intelligence (AI) is the technique that has potential to develop objective, reproducible and non-invasive methodology to predict embryo quality with high levels of accuracy [18]. A specific method as artificial neural network (ANN) along with genetic algorithm (GA) could be used to simulate an accuracy predictive model [34].…”

“…This is make of interactive way presenting examples with known classification, one before other (it is called learning or training) [35]. GA is a search and optimization method inspired by genetic mechanisms under natural evolution [18,36], whilst the ANN is a technique based on how human neurons transmit and process information to learn and it is indicated for the resolution of complex and nonlinear problems. Once properly trained, the ANN is capable to perform prediction from new data which it has never had access [37,38].…”