Relationships between the developmental potential of human in-vitro fertilization embryos and features describing the embryo, oocyte and follicle

Abstract

This paper investigates the relationship of features of in-vitro fertilization (IVF) embryos and the associated oocyte and follicle to the outcome of transfer. It differs from previous studies in including a range of features (n = 53) and in using class probability tree analysis. This is a non-parametric multivariate method which expresses relationships as simple rules of features characterizing the 'take home baby' and 'no take home baby' (negative pregnancy test) classes of embryo batches. Data were analysed retrospectively. Fifty-three (embryo, oocyte and follicular) features for each of the three embryos in the transferred batch were collected for 200 IVF patients. Each batch of three embryos was described by a representative value for each feature. The relationship between features and outcome of transfer was analysed. Only four of the 53 features were identified as predictive. However, an appropriate combination of these four (embryo grade, cell number, follicle size and follicular fluid volume) achieved satisfactory predictivities while offering a simplified and more quantitative basis than regularly used criteria. The key component of the composite embryo grading allotted by embryologists turned out to be cell number. The existing predictive use of follicular size was corroborated and an independent additional predictive contribution of follicular fluid volume was found. The study also suggests that the 49 remaining features have little additional predictive value.

Fitting the data from embryo implantation prediction: Learning from label proportions

“…Note that this standard paradigm is the methodology considered by previous machine learning approaches to the embryo selection problem. Following the general procedure to prepare the dataset for standard learning techniques [10,12,13,15,19,21], all the embryos in non-full bags (embryos of unknown fate) are removed. Thus, a fully labeled dataset is obtained and the aforementioned state-of-the-art learning techniques are used to learn NB [29], TAN [30] and 2DB [31] classifiers.…”

“…It can be known only in those cases where none or all the transferred embryos were implanted. In practice, many studies in the related literature discarded the embryos of unknown fate [10,12,13,15,19,21] for model learning. However, as the conclusions drawn from the results of any data analysis are affected by the sample size, ML researchers have focused their effort on incorporating any available example to the analysis.…”

A Bayesian network model for predicting pregnancy after in vitro fertilization

“…Without considering the EU assumption, it is possible using a traditional classifier for predicting the IVF outcome; this is the approach followed for instance in [7] and [9]. In this section we thus compare BN 1 against traditional classifiers such as TAN, (tree-augmented naive Bayes [28]), AODE [29] (which can be seen as an ensemble of TANs), the J4.8 decision tree and the Random Forest (which is an ensemble of decision trees); see [30] and [31] for a textbook presentation of these classifier.…”

“…Reliably predicting the IVF outcome is thus still substantially an open problem [7,8,9]. A pioneering approach for estimating the probability of single and multiple pregnancy after an IVF treatment is the embryo -uterine model (EU) [10], which assumes that, for pregnancy to happen, it is necessary both a receptive uterus and a viable embryo.…”

Vers un transfert mono-embryonnaire : prise en compte conjointe des paramètres embryonnaires et utérins en assistance médicale à la procréation

“…De nombreuses publications ont démontré que la sélection des embryons avant le transfert peut constituer un outil efficace pour améliorer l'implantation. Plusieurs paramètres embryonnaires ont été associés au potentiel implantatoire dont la morphologie du zygote [3,4,6], la morphologie de l'embryon clivé ou sa vitesse de développement [7][8][9][10][11]. Ces paramètres ont été à l'origine de plusieurs systèmes de classification des embryons dans le but de prédire les chances de grossesse [12][13][14][15][16].…”