Clinical validation of an automatic classification algorithm applied on cleavage stage embryos: analysis for blastulation, euploidy, implantation, and live-birth potential

STUDY QUESTION

Is a commercially available embryo assessment algorithm for early embryo evaluation based on the automatic annotation of morphokinetic timings a useful tool for embryo selection in IVF cycles?

SUMMARY ANSWER

The classification provided by the algorithm was shown to be significantly predictive, especially when combined with conventional morphological evaluation, for development to blastocyst, implantation, and live birth, but not for euploidy.

WHAT IS KNOWN ALREADY

The gold standard for embryo selection is still morphological evaluation conducted by embryologists. Since the introduction of time-lapse technology to embryo culture, many algorithms for embryo selection have been developed based on embryo morphokinetics, providing complementary information to morphological evaluation. However, manual annotations of developmental events and application of algorithms can be time-consuming and subjective processes. The introduction of automation to morphokinetic annotations is a promising approach that can potentially reduce subjectivity in the embryo selection process and improve the workflow in IVF laboratories.

STUDY DESIGN, SIZE, DURATION

This observational, retrospective cohort study was performed in a single IVF clinic between 2018 and 2021 and included 3736 embryos from oocyte donation cycles (423 cycles) and 1291 embryos from autologous cycles with preimplantation genetic testing for aneuploidies (PGT-A, 185 cycles). Embryos were classified on Day 3 with a score from 1 (best) to 5 (worst) by the automatic embryo assessment algorithm. The performance of the embryo classification model for blastocyst development, implantation, live birth, and euploidy prediction was assessed.

PARTICIPANTS/MATERIALS, SETTING, METHODS

All embryos were monitored by a time-lapse system with an automatic cell-tracking and embryo assessment software during culture. The embryo assessment algorithm was applied on Day 3, resulting in embryo classification from 1 to 5 (from highest to lowest developmental potential) depending on four parameters: P2 (t3-t2), P3 (t4-t3), oocyte age, and number of cells. There were 959 embryos selected for transfer on Day 5 or 6 based on conventional morphological evaluation. The blastocyst development, implantation, live birth, and euploidy rates (for embryos subjected to PGT-A) were compared between the different scores. The correlation of the algorithm scoring with the occurrence of those outcomes was quantified by generalized estimating equations (GEEs). Finally, the performance of the GEE model using the embryo assessment algorithm as the predictor was compared to that using conventional morphological evaluation, as well as to a model using a combination of both classification systems.

MAIN RESULTS AND THE ROLE OF CHANCE

The blastocyst rate was higher with lower the scores generated by the embryo assessment algorithm. A GEE model confirmed the positive association between lower embryo score and higher odds of blastulation (odds ratio (OR) (1 vs 5 score) = 15.849; P < 0.001). This association was consistent in both oocyte donation and autologous embryos subjected to PGT-A. The automatic embryo classification results were also statistically associated with implantation and live birth. The OR of Score 1 vs 5 was 2.920 (95% CI 1.440-5.925; P = 0.003; E = 2.81) for implantation and 3.317 (95% CI 1.615-6.814; P = 0.001; E = 3.04) for live birth. However, this association was not found in embryos subjected to PGT-A. The highest performance was achieved when combining the automatic embryo scoring and traditional morphological classification (AUC for implantation potential = 0.629; AUC for live-birth potential = 0.636). Again, no association was found between the embryo classification and euploidy status in embryos subjected to PGT-A (OR (1 vs 5) = 0.755 (95% CI 0.255-0.981); P = 0.489; E = 1.57).

LIMITATIONS, REASONS FOR CAUTION

The retrospective nature of this study may be a reason for caution, although the large sample size reinforced the ability of the model for embryo selection.

WIDER IMPLICATIONS OF THE FINDINGS

Time-lapse technology with automated embryo assessment can be used together with conventional morphological evaluation to increase the accuracy of embryo selection process and improve the success rates of assisted reproduction cycles. To our knowledge, this is the largest embryo dataset analysed with this embryo assessment algorithm.

STUDY FUNDING/COMPETING INTEREST(S)

This research was supported by Agencia Valenciana de Innovació and European Social Fund (ACIF/2019/264 and CIBEFP/2021/13). In the last 5 years, M.M. received speaker fees from Vitrolife, Merck, Ferring, Gideon Richter, Angelini, and Theramex, and B.A.-R. received speaker fees from Merck. The remaining authors have no competing interests to declare.

TRIAL REGISTRATION NUMBER

N/A.

P–203 Applying artificial intelligence for ploidy prediction: The concentration of IL–6 in spent culture medium, blastocyst morphological grade and embryo morphokinetics as variables under consideration

Study question

Would it be possible to predict embryo ploidy by taking into account conventional morphological and morphokinetic parameters together with IL–6 concentration in spent culture medium?

Summary answer

Our artificial neural network (ANN) trained with blastocyst morphology, embryo morphokinetics and IL–6 concentration distinguished between euploid/aneuploid embryos in 65% of the testing dataset.

What is known already

The analysis of spent embryo culture media represents the protein and metabolic state of the embryo and could be a non-invasive method of obtaining information about embryo quality. The impact of the presence/absence of several proteins in embryo culture samples over clinical results has been widely studied. The IL–6 is one of the most mentioned protein for its effect on embryo development, implantation and likelihood of achieving a live birth. In this initial attempt, we examined the predictive value for euploidy of a model that took into account the concentration of IL–6 in the spent culture medium.

Study design, size, duration

This prospective study included 319 embryos with PGT-A results. Out of the total, 127 were euploid and 192 aneuploid embryos. Concentration of IL–6 in spent embryo culture media (collected on the day of trophectoderm biopsy-fifth/sixth day of development), morphokinetic parameters (division time to 2 cells-t2; to 3 cells-t3, to 4 cells-t4; to 5 cells-t5 and time of blastocyst formation-tB) and blastocyst morphological grade (according to ASEBIR criteria) were considered to predict the embryo ploidy.

Participants/materials, setting, methods

Embryos were cultured in EmbryoScope. The chromosome analysis was performed using next-generation sequence technology. The concentration of IL–6 was measured in 20µL of spent embryo culture media with ELISA kits. Morphokinetic parameters were automatically annotated and the blastocyst morphology was evaluated by senior embryologists based on blastocele expansion, inner cell mass and trophectoderm quality. All the embryos were divided into 70% for training, 15% for validating and 15% for testing our ANN model with MatLab®.

Main results and the role of chance

The general description for the euploid embryo population was the following: 2% of the embryos were graded as A, 71% were graded as B and 28% were graded as C; the means and standard deviations were 25.32±2.97 hours (h) for t2, 35.33±5.15h for t3, 37.30±5.43h for t4, 48.24±6.62h for t5 and 103.93±12.8h for tB; and the average of IL–6 concentration was 1.51±0.70 pg/ml. The general description for the aneuploid embryo population was the following: 1% of the embryos were graded as A, 48% were graded as B and 51% were graded as C; the means and standard deviations were 26.13±3.51h for t2, 36.70±4.29h for t3, 38.20±4.24h for t4, 49.86±6.89h for t5 and 107.10±8.29h for tB; and the average of IL–6 concentration was 1.47±0.71 pg/ml. Our ANN model showed a higher general success rate as we increased the variables considered in the final prediction of euploid embryos. The accuracy, sensitivity and specificity for the testing dataset were: 0.60, 0.12 and 0.87 with morphokinetic parameters; 0.63, 0.24 and 0.93 with morphokinetics and IL–6 concentration; and 0.65, 0.16 and 0.96 with morphokinetics, IL–6 concentration and blastocyst morphological grade.

Limitations, reasons for caution

The low sensitivity and high specificity achieved in our models indicated that they were more capable of detecting aneuploid than euploid embryos. As this was a preliminary study, the small number of embryos included in the test (n = 48) was also a limitation.

Wider implications of the findings: The results showed that our model tended to classify the embryos as aneuploid. More euploid embryos would be necessary to train our model and achieve better results in the prediction of chromosomally normal embryos. Further studies with large number of embryos and additional variables could improve the non-invasive ploidy prediction.

Trial registration number

Not applicable

P–144 Undisturbed culture in time-lapse systems improves embryo development and quality

Study question

Does culture in integrated time-lapse systems (TLS) improve embryo development and blastocyst quality compared to conventional benchtop incubators (CI), within the same IVF laboratory?

Summary answer

Under similar conditions, culture in TLS resulted in a significant increase in blastocyst rate, top quality blastocyst rate and proportion of biopsied embryos per treatment

What is known already

Integrated TLS have the potential of delivering a stable and undisturbed environment throughout the whole embryo culture, avoiding taking them out for assessment. However, there is still lack of quality evidence of the performance of these incubators compared to CI at supporting embryo culture until blastocyst stage. Studies abording this issue are still scarce, heterogeneous and have a small sample size. Although some authors have reported an improvement in embryo development and quality using TLS, global results are inconsistent. To our knowledge, the present study evaluates the effect of TLS on embryo quality on the largest sample size yet.

Study design, size, duration

Unicentric retrospective cohort study including 14248 ICSI treatments from 2016 to October 2020, with both autologous and donated oocytes. We compared blastocyst rate (BR) and proportion of top-quality blastocysts (TQB=Morphology ASEBIR score A) per treatment between those using TLS (N = 7500) and CI (N = 6748), and the proportion of embryos biopsied (EB) in cycles with pre-implantation genetic testing (PGT-A; N = 2642). We performed a sub-analysis in treatments using single-step culture medium (N-TLS=4398, N-CI=1140) in both types of incubators.

Participants/materials, setting, methods

Embryo cohorts were cultured until blastocyst stage in one of 3 TLS: EmbryoScope, EmbryoScope Plus (Vitrolife,) and Geri (Genea Biomedx), or in a CI (ASTEC). Embryo quality was assessed following ASEBIR morphological criteria. Culture protocols and media changed during the included time period. For that reason, we did a sub-study in the treatments performed since the implementation of Gems® (Genea Biomedx) single-step (SS) culture medium in all incubators. Statistical analysis was done using ANOVA tests.

Main results and the role of chance

Treatments were differently distributed and heterogeneous in terms of number of oocytes obtained per patient, so we stratified the analysis according to ovum origin and compared mean rates per cycle instead of total number of embryos per group. BR was statistically higher (P &lt; 0,001) in the TLS group, in both autologous (62,98±29,37% vs 59,49±31,09% in CI) and oocyte donation treatments (69,25±22,07% vs 66,27±23,28% in CI). Proportion of TQB was also significantly higher in the TLS in both types of cycles (P &lt; 0,001): 3,60±12,29% in TLS vs 2,27±9,71% in CI in autologous cycles, 8,68±15,31% in TLS vs 7,32±14,02% CI in ovum donation cycles. Results were corroborated in the SS media sub-study (P &lt; 0,05): BR was 63,87±29,23% in TLS vs 57,53±30,61% in CI with autologous oocytes, and 70,76±21,63% in TLS vs 67,39±22,68% in CI with donated oocytes; TQB rates were 3,66±12,06% in TLS vs 2,05±9,26% in CI in autologous treatments and 8,81±15,21% in TLS vs 6,84±12,91% in CI in ovum donation treatments. Regarding PGT-A treatments, we found no significant difference in the biopsy rate in the total comparison, although the rate significantly increased in the TLS group since the implementation of single-step medium (52,36±24,69% in TLS vs 48,63±22,56% in CI; P = 0,007)

Limitations, reasons for caution

Not only culture conditions varied over time, but also the number of TLS in the laboratory, which increased lately. Hence, even though the most recent treatments included in the all-SS sub-study are more homogeneous in terms of culture conditions, they are unbalanced regarding the distribution among incubators.

Wider implications of the findings: Our results demonstrate the superiority of TLS coupled with single-step culture media against traditional embryo culture systems at supporting embryo development. The optimal environment provided by TLS enhances embryo development until blastocyst stage as well as their quality, increasing the cumulative chances of getting a life-birth for each patien.

Trial registration number

Not applicable

O-121 Exploring non-invasive methods to predict Ploidy Status: Combination of blastocyst morphology image analysis and proteomic profiles by using Artificial Neural Networks

Study question

Is the blastocyst morphology image analysis combined with the protein content of spent embryo culture medium a suitable way to predict embryo ploidy?

Summary answer

Morphological variables from blastocyst image analysis combined with IL-6 or MMP-1 concentration in spent culture medium showed more than 80% of accuracy for euploidy prediction.

What is known already

An artificial intelligence model based on the proteomic profile of euploid embryos and morphological data from blastocyst time-lapse images has been recently published (Bori et al., 2020). The most promising artificial neural network (ANN) algorithm considered 20 morphological variables extracted from image analysis and two proteins detected in embryo culture medium (MMP-1 and IL-6). The overall success rate on blind test data was 72.7% for live birth prediction. The main aim of the present study was to check if the same morphological variables combined with MMP-1 or IL-6 with a cost-effective technique could discriminate between euploid and aneuploid embryos.

Study design, size, duration

This prospective study included 120 embryos from the preimplantation genetic testing for aneuploidies (PGT-A) program. A single blastocyst image was obtained for each embryo and their spent culture medium was collected on the day 5/6 of embryo development (day of trophectoderm biopsy). Morphological variables were extracted for all the blastocyst. On the other hand, we quantified IL-6 levels of 67 embryos and MMP-1 levels of 53 embryos. Resulting parameters were used to predict PGT-A results.

Participants/materials, setting, methods

Blastocyst images were imported into Matlab software and segmented into regions of interest. We obtained 20 mathematical variables related to measurements of areas, number of pixels and texture analysis. Chromosome analysis was performed using next-generation sequence technology. In parallel, 20 µL of spent culture medium from each blastocyst was analyzed with ELISA kits (IL-6 or MMP-1). Protein concentrations and morphological variables were used as input data for an ANN associated with genetic algorithms.

Main results and the role of chance

The euploid rate for the set of embryos included in the IL-6 group was 51.4%. The ANN was trained with 49 embryos and blind tested with 18 embryos. Following results correspond to euploidy prediction on the blind test. The sensitivity, specificity, accuracy and area under the ROC curve (AUC) were: 0.56, 0.78, 0.67 and 0.72 considering only IL-6 values; 0.88, 0.78, 0.83 and 0.61 considering IL-6 values and blastocyst morphological data extracted from the image analysis. The euploid rate for the set of embryos included in the MMP-1 group was 51.9%. The ANN was trained with 39 embryos and blind tested with 14 embryos. Following results correspond to euploidy prediction on the blind test. The sensitivity, specificity, accuracy and AUC were: 0.71, 0.57, 0.64 and 0.67 considering only MMP-1 values; 0.86, 0.86, 0.86 and 0.61 considering MMP-1 values and morphological data extracted from the image analysis.

Limitations, reasons for caution

The detection limit in protein quantification is the main limitation of our study. The small number of embryos and the specific culture medium used should be considered for the model application.

Wider implications of the findings

Our preliminary results showed that blastocyst morphology and embryo secretomics could be useful for euploidy prediction by using artificial intelligence techniques. These findings may contribute to the emerging era of non-invasive preimplantation genetic testing (ni-PGT-A).

Trial registration number

not applicable

P-203 Applying artificial intelligence for ploidy prediction: The concentration of IL-6 in spent culture medium, blastocyst morphological grade and embryo morphokinetics as variables under consideration

Study question

Would it be possible to predict embryo ploidy by taking into account conventional morphological and morphokinetic parameters together with IL-6 concentration in spent culture medium?

Summary answer

Our artificial neural network (ANN) trained with blastocyst morphology, embryo morphokinetics and IL-6 concentration distinguished between euploid/aneuploid embryos in 65% of the testing dataset.

What is known already

The analysis of spent embryo culture media represents the protein and metabolic state of the embryo and could be a non-invasive method of obtaining information about embryo quality. The impact of the presence/absence of several proteins in embryo culture samples over clinical results has been widely studied. The IL-6 is one of the most mentioned protein for its effect on embryo development, implantation and likelihood of achieving a live birth. In this initial attempt, we examined the predictive value for euploidy of a model that took into account the concentration of IL-6 in the spent culture medium.

Study design, size, duration

This prospective study included 319 embryos with PGT-A results. Out of the total, 127 were euploid and 192 aneuploid embryos. Concentration of IL-6 in spent embryo culture media (collected on the day of trophectoderm biopsy-fifth/sixth day of development), morphokinetic parameters (division time to 2 cells-t2; to 3 cells-t3, to 4 cells-t4; to 5 cells-t5 and time of blastocyst formation-tB) and blastocyst morphological grade (according to ASEBIR criteria) were considered to predict the embryo ploidy.

Participants/materials, setting, methods

Embryos were cultured in EmbryoScope. The chromosome analysis was performed using next-generation sequence technology. The concentration of IL-6 was measured in 20µL of spent embryo culture media with ELISA kits. Morphokinetic parameters were automatically annotated and the blastocyst morphology was evaluated by senior embryologists based on blastocele expansion, inner cell mass and trophectoderm quality. All the embryos were divided into 70% for training, 15% for validating and 15% for testing our ANN model with MatLab®.

Main results and the role of chance

The general description for the euploid embryo population was the following: 2% of the embryos were graded as A, 71% were graded as B and 28% were graded as C; the means and standard deviations were 25.32±2.97 hours (h) for t2, 35.33±5.15h for t3, 37.30±5.43h for t4, 48.24±6.62h for t5 and 103.93±12.8h for tB; and the average of IL-6 concentration was 1.51±0.70 pg/ml. The general description for the aneuploid embryo population was the following: 1% of the embryos were graded as A, 48% were graded as B and 51% were graded as C; the means and standard deviations were 26.13±3.51h for t2, 36.70±4.29h for t3, 38.20±4.24h for t4, 49.86±6.89h for t5 and 107.10±8.29h for tB; and the average of IL-6 concentration was 1.47±0.71 pg/ml. Our ANN model showed a higher general success rate as we increased the variables considered in the final prediction of euploid embryos. The accuracy, sensitivity and specificity for the testing dataset were: 0.60, 0.12 and 0.87 with morphokinetic parameters; 0.63, 0.24 and 0.93 with morphokinetics and IL-6 concentration; and 0.65, 0.16 and 0.96 with morphokinetics, IL-6 concentration and blastocyst morphological grade.

Limitations, reasons for caution

The low sensitivity and high specificity achieved in our models indicated that they were more capable of detecting aneuploid than euploid embryos. As this was a preliminary study, the small number of embryos included in the test (n = 48) was also a limitation.

Wider implications of the findings

The results showed that our model tended to classify the embryos as aneuploid. More euploid embryos would be necessary to train our model and achieve better results in the prediction of chromosomally normal embryos. Further studies with large number of embryos and additional variables could improve the non-invasive ploidy prediction.

Trial registration number

not applicable