Examining the efficacy of six published time-lapse imaging embryo selection algorithms to predict implantation to demonstrate the need for the development of specific, in-house morphokinetic selection algorithms

Assessing the impact of delayed blastulation using time lapse morphokinetics and preimplantation genetic testing in an IVF patient population

PURPOSE

There is clinical evidence that early cleavage timing parameters predictive of blastocyst development also correlate with embryo implantation potential. The aim of this study is to determine the developmental competency of embryos with delayed blastulation.

METHODS

Retrospective study performed from 2015 to 2016 at the Division of Reproductive Endocrinology and Infertility at Northwestern University.

RESULTS

A total of 2,292 embryos from 524 patients were included. Day 6 blastocysts had statistically significant longer times for every time point analyzed than day 5 blastocysts (p < 0.001). We found no statistically significant difference in euploidy rates between day 5 (44%) and day 6 (41%) embryos (p = 0.573). t7 and t8 time points were independent predictors of euploidy after controlling for day of biopsy (p < 0.015 and p < 0.014, respectively). Intrauterine pregnancy (IUP) and live birth (LB) were less likely to occur after transferring day 6 embryos (p = 0.0033 and p = 0.0359) without previous genetic testing. However, in embryos that undergo preimplantation genetic testing for aneuploidy (PGT-A), there were no significant differences in IUP or LB rates.

CONCLUSION

Early time-lapse points can be used to predict embryo development. Day of blastulation may be an independent predictor IUP, with day 6 blastocysts having lower pregnancy and live birth rates. Our data suggests that day 5 and day 6 PGT-A tested embryos show similar rates of euploidy, suggesting that differences in PR seen in the non-PGT-A tested group may be caused by factors other than aneuploidy. Genetic testing technologies in combination with time-lapse microscopy may provide further information to improve IVF outcomes.

Deep learning as a predictive tool for fetal heart pregnancy following time-lapse incubation and blastocyst transfer

STUDY QUESTION

Can a deep learning model predict the probability of pregnancy with fetal heart (FH) from time-lapse videos?

SUMMARY ANSWER

We created a deep learning model named IVY, which was an objective and fully automated system that predicts the probability of FH pregnancy directly from raw time-lapse videos without the need for any manual morphokinetic annotation or blastocyst morphology assessment.

WHAT IS KNOWN ALREADY

The contribution of time-lapse imaging in effective embryo selection is promising. Existing algorithms for the analysis of time-lapse imaging are based on morphology and morphokinetic parameters that require subjective human annotation and thus have intrinsic inter-reader and intra-reader variability. Deep learning offers promise for the automation and standardization of embryo selection.

STUDY DESIGN, SIZE, DURATION

A retrospective analysis of time-lapse videos and clinical outcomes of 10 638 embryos from eight different IVF clinics, across four different countries, between January 2014 and December 2018.

PARTICIPANTS/MATERIALS, SETTING, METHODS

The deep learning model was trained using time-lapse videos with known FH pregnancy outcome to perform a binary classification task of predicting the probability of pregnancy with FH given time-lapse video sequence. The predictive power of the model was measured using the average area under the curve (AUC) of the receiver operating characteristic curve over 5-fold stratified cross-validation.

MAIN RESULTS AND THE ROLE OF CHANCE

The deep learning model was able to predict FH pregnancy from time-lapse videos with an AUC of 0.93 [95% CI 0.92-0.94] in 5-fold stratified cross-validation. A hold-out validation test across eight laboratories showed that the AUC was reproducible, ranging from 0.95 to 0.90 across different laboratories with different culture and laboratory processes.

LIMITATIONS, REASONS FOR CAUTION

This study is a retrospective analysis demonstrating that the deep learning model has a high level of predictability of the likelihood that an embryo will implant. The clinical impacts of these findings are still uncertain. Further studies, including prospective randomized controlled trials, are required to evaluate the clinical significance of this deep learning model. The time-lapse videos collected for training and validation are Day 5 embryos; hence, additional adjustment would need to be made for the model to be used in the context of Day 3 transfer.

WIDER IMPLICATIONS OF THE FINDINGS

The high predictive value for embryo implantation obtained by the deep learning model may improve the effectiveness of previous approaches used for time-lapse imaging in embryo selection. This may improve the prioritization of the most viable embryo for a single embryo transfer. The deep learning model may also prove to be useful in providing the optimal order for subsequent transfers of cryopreserved embryos.

STUDY FUNDING/COMPETING INTEREST(S)

D.T. is the co-owner of Harrison AI that has patented this methodology in association with Virtus Health. P.I. is a shareholder in Virtus Health. S.C., P.I. and D.G. are all either employees or contracted with Virtus Health. D.G. has received grant support from Vitrolife, the manufacturer of the Embryoscope time-lapse imaging used in this study. The equipment and time for this study have been jointly provided by Harrison AI and Virtus Health.

Blastomere cleavage plane orientation and the tetrahedral formation are associated with increased probability of a good-quality blastocyst for cryopreservation or transfer: a time-lapse study

OBJECTIVE

To determine whether blastomere spatial arrangement in early human embryos is reflective of embryonic potential.

DESIGN

Retrospective analysis of prospectively collected data.

SETTING

Single academic center.

PATIENT(S)

Patients undergoing a single blastocyst transfer.

INTERVENTION(S)

None.

MAIN OUTCOME MEASURE(S)

Developmental kinetics, blastocyst quality, embryo dysmorphisms, and live birth rate.

RESULT(S)

A total of 716 embryos were examined in detail for cleavage plane orientation, blastomere arrangement, and morphokinetic behavior. Tetrahedral (TET) and nontetrahedral embryos (nTET) differed significantly in developmental kinetics. The frequency of dysmorphisms, multinucleation, and irregular chaotic division was higher in nTET embryos. Only 44% of nTET versus 62.9% of TET embryos were scored as top-quality blastocysts. After adjusting for age, our data indicated that having TET embryos significantly increased the odds of having a blastocyst for cryopreservation/transfer (odds ratio, 3.58; confidence interval, 2.42-5.28) when compared with nTET. A total of 164 fresh single ETs were performed with blastocyst-stage embryos. The implantation rate for TET- and nTET-derived blastocysts were similar (64.7% and 62%, respectively). The live birth rate was 55% in both groups. A meridonal first division was noted in 85% of the fresh SET blastocysts.

CONCLUSION(S)

Cleavage plane orientation during the first three divisions appeared to dictate final blastomere spatial arrangement. The TET formation at the four-cell stage was predictive for embryos most likely to develop into good-quality blastocysts for cryopreservation/transfer. Morphokinetic markers of embryo potential were significantly different between TET and nTET embryos.

Time-lapse videography for embryo selection/de-selection: a bright future or fading star?

The recent clinical introduction of time-lapse videography into in vitro fertilization laboratories has offered a novel opportunity for embryologists to explore improved methods for embryo selection. While the concept of uninterrupted culture of embryos provided by such systems is welcomed, the current evidence does not support its full application in routine clinical practice. The issue of whether or not algorithms for embryo selection can be extrapolated between laboratories, which may represent a major hurdle to its wide application, is currently gaining increasing attention amongst embryologists worldwide. In this commentary issues identified in time-lapse embryo selection/de-selection algorithms, such as quantitative versus qualitative parameters, are discussed alongside the reference start point for the timing system, and types of datasets used for developing and validating time-lapse algorithms. Considering these factors, alternative future research directions which could potentially solve current issues are proposed.

Artificial intelligence and machine learning for human reproduction and embryology presented at ASRM and ESHRE 2018

Sixteen artificial intelligence (AI) and machine learning (ML) approaches were reported at the 2018 annual congresses of the American Society for Reproductive Biology (9) and European Society for Human Reproduction and Embryology (7). Nearly every aspect of patient care was investigated, including sperm morphology, sperm identification, identification of empty or oocyte containing follicles, predicting embryo cell stages, predicting blastocyst formation from oocytes, assessing human blastocyst quality, predicting live birth from blastocysts, improving embryo selection, and for developing optimal IVF stimulation protocols. This represents a substantial increase in reports over 2017, where just one abstract each was reported at ASRM (AI) and ESHRE (ML). Our analysis reveals wide variability in how AI and ML methods are described (from not at all or very generic to fully describing the architectural framework) and large variability on accepted dataset sizes (from just 3 patients with 16 follicles in the smallest dataset to 661,060 images of 11,898 human embryos in one of the largest). AI and ML are clearly burgeoning methodologies in human reproduction and embryology and would benefit from early application of reporting standards.

Single embryo transfer: Why and how to identify the embryo with the best developmental potential

Multiple pregnancies with higher risk of preterm birth and the associated higher morbidity have been a major obstacle from the early days of in vitro fertilization. A good strategy to avoid multiple pregnancies is elective single embryo transfer and cryopreservation of spare embryos. Important factors in adopting this strategy are good counselling of the patients and the selection of embryos with high implantation potential. Technical advances in embryo selection have been described during recent years, time lapse monitoring and genetic assessment of the embryos being the most important achievements. With these studies we have gained new information on early embryos. However, at present, there is insufficient evidence to recommend the routine use of these new techniques. The ultimate goal of infertility treatment is a healthy baby.

A comparison of morphokinetic markers predicting blastocyst formation and implantation potential from two large clinical data sets

PURPOSE

To demonstrate whether the standard morphokinetic markers used for embryo selection have a similar relationship to blastocyst formation and implantation in two large clinical data sets.

METHODS

This is a retrospective cohort analysis striving to answer two distinct questions utilizing data sets from two large IVF clinics. Blastocysts (BL) and implanted blastocysts (I) in both clinics, IVI-Valencia (BL = 11,414, I = 479) and WMC (BL = 15,902; I = 337), were cultured in a time-lapse system (EmbryoScope, Vitrolife, Sweden). The study was designed to assess the relationship between early morphokinetic hallmarks and BL development, with a secondary analysis of implantation rates following single-embryo day 3 and day 5 transfers.

RESULTS

We performed a detailed graphical analysis for t3, t5, duration of the second cell cycle (cc2) (t3-t2), and the ratio (t5-t3)/(t5-t2). The t5 timing was not affected between the clinics. However, Weill Cornell Medicine's (WCM) proportions were significantly affected by having BL vs. not. A significant decrease of blastocysts with longer t5 in WCM data, while t5 was more informative in the IVI data set for the implantation rate.

CONCLUSIONS

Morphokinetic intervals for early cleavages were distributed differently between the clinics. Incorporation of embryo-selection algorithms depends on the individual clinic's selected developmental hallmarks, all of which must be validated before incorporation into clinical practice.

The Effect of Advanced Maternal Age on Embryo Morphokinetics

Purpose: To compare the morphokinetic parameters of pre-implantation development between embryos of women of advanced maternal age (AMA) and young women. Methods: Time-lapse microscopy was used to compare morphokinetic variables between 495 embryos of AMA women ≥ age 42 years and 653 embryos of young patients (<age 38 years) who underwent IVF in our unit. Developmental events annotated and analyzed include observed cell divisions in correlation to the timing of fertilization, synchrony of the second (s2) and third cell cycles (s3) and the duration to the second (cc2) and third cleavages (cc3). Results: No significant differences were observed in cleavage times between the embryos of AMA and the control embryos. Interestingly, the older embryos appear to be more prone to developmental arrest (a higher percentage of embryos of older women arrested at 4-7 cells resulting in less embryos reaching the 8-cell stage (66% vs. 72%, respectively), though this difference did not reach a significance at least during the first 3 days of development (p > 0.05). Conclusions: While early morphokinetic parameters do not reflect dynamics unique to embryos of older women, a tendency toward developmental arrest was observed, which would likely be even more pronounced at later stages of development.

Morphology vs morphokinetics: a retrospective comparison of inter-observer and intra-observer agreement between embryologists on blastocysts with known implantation outcome

Objective

Our primary aim was to compare the morphology and morphokinetics on inter- and intra-observer agreement for blastocyst with known implantation outcome. Our secondary aim was to validate the morphokinetic parameters' ability to predict pregnancy using a previous published selection algorithm, and to compare this to standard morphology assessments.

Methods

Two embryologists made independent blinded annotations on two occasions using time-lapse images and morphology evaluations using the Gardner Schoolcraft criteria of 99 blastocysts with known implantation outcome. Inter- and intra-observer agreement was calculated and compared using the two methods. The embryos were grouped based on their morphological score, and on their morphokinetic class using a previous published selection algorithm. The implantation rates for each group was calculated and compared.

Results

There was moderate agreement for morphology, with agreement on the same embryo score in 55 of 99 cases. The highest agreement rate was found for expansion grade, followed by trophectoderm and inner cell mass. Correlation with pregnancy was inconclusive. For morphokinetics, almost perfect agreement was found for early and late embryo development events, and strong agreement for day-2 and day-3 events. When applying the selection algorithm, the embryo distributions were uneven, and correlation to pregnancy was inconclusive.

Conclusions

Time-lapse annotation is consistent and accurate, but our external validation of a previously published selection algorithm was unsuccessful.

Are computational applications the "crystal ball" in the IVF laboratory? The evolution from mathematics to artificial intelligence

Mathematics rules the world of science. Innovative technologies based on mathematics have paved the way for implementation of novel strategies in assisted reproduction. Ascertaining efficient embryo selection in order to secure optimal pregnancy rates remains the focus of the in vitro fertilization scientific community and the strongest driver behind innovative approaches. This scoping review aims to describe and analyze complex models based on mathematics for embryo selection, devices, and software most widely employed in the IVF laboratory and algorithms in the service of the cutting-edge technology of artificial intelligence. Despite their promising nature, the practicing embryologist is the one ultimately responsible for the success of the IVF laboratory and thus the one to approve embracing pioneering technologies in routine practice. Applied mathematics and computational biology have already provided significant insight into the selection of the most competent preimplantation embryo. This review describes the leap of evolution from basic mathematics to bioinformatics and investigates the possibility that computational applications may be the means to foretell a promising future for the IVF clinical practice.

Cumulative live birth rates and perinatal outcomes with the use of time-lapse imaging incubators for embryo culture: a retrospective cohort study of 1882 ART cycles

OBJECTIVE

Comparison of live birth rates and the perinatal outcomes after fresh and frozen embryo transfer between time-lapse imaging (TLI) and standard culture (SC) incubators.

DESIGN

Retrospective cohort study.

SETTING

A single tertiary level IVF unit.

POPULATION

Women undergoing IVF between January 2014 and October 2015.

METHODS

Comparison was done between 1064 IVF cycles using TLI (TLI cycles) and 818 IVF cycles using SC (SC cycles).

MAIN OUTCOME MEASURES

Cumulative live birth rate per oocyte retrieval and perinatal outcomes including birthweight, gestational age, preterm birth (PTB) (<37 weeks), early preterm birth (PTB; <32 weeks), low birthweight (LBW; <2500 g), very LBW (<1500 g) and macrosomia (>4500 g).

RESULTS

The fresh embryo transfer live birth rate was noted to be higher for TLI cycles [TLI 36.8 versus SC 33.9%, adjusted odds ratio (aOR) 1.28, 95% CI 1.05-1.57], but the frozen embryo transfer live birth rates were not significantly different. The mean birthweight was higher in the TLI group after both fresh [adjusted mean difference (aMD) 174.78 g, 95% CI 64.80-284.77] and frozen embryo transfers (aMD 175.91 g, 95% CI 16.98-334.84). After a fresh embryo transfer, there was a lower risk of early PTB and very LBW in the TLI group. Among frozen embryo transfers, there was a lower risk of early PTB and LBW in the TLI group.

CONCLUSIONS

TLI incubators are associated with improved perinatal outcomes and higher mean birthweight after fresh and frozen embryo transfer.

TWEETABLE ABSTRACT

Time-lapse imaging incubators in IVF improve perinatal outcomes after both fresh and frozen embryo transfers.

Assessing efficacy of day 3 embryo time-lapse algorithms retrospectively: impacts of dataset type and confounding factors

This study investigated the efficacy of four published day 3 embryo time-lapse algorithms based on different types of datasets (known implantation data [KID] and single embryo transfer [SET]), and the confounding effect of female age and conventional embryo morphology. Four algorithms were retrospectively applied to three types of datasets generated at Fertility North between February 2013 and December 2014: (a) KID dataset (n = 270), (b) a subset of SET (n = 144, end-point = implantation), and (c) SET (n = 144, end-point = live birth), respectively. All four algorithms showed progressively reduced predictive power (expressed as area under the receiver operating characteristics curve and 95% confidence interval [CI]) after application to the three datasets (a-c): Liu (0.762 [0.701-0.824] vs. 0.724 [0.641-0.807] vs. 0.707 [0.620-0.793]), KIDScore (0.614 [0.539-0.688] vs. 0.548 [0.451-0.645] vs. 0.536 [0.434-0.637]), Meseguer (0.585 [0.508-0.663] vs. 0.56 [0.462-0.658] vs. 0.549 [0.445-0.652]), and Basile (0.582 [0.505-0.659] vs. 0.519 [0.421-0.618] vs. 0.509 [0.406-0.612]). Furthermore, using KID dataset, the association (expressed as odds ratio and 95% CI) between time-lapse algorithms and implantation outcomes lost statistical significance after adjusting for conventional embryo morphology and female age in 3 of the 4 algorithms (KIDScore 1.832 [1.118-3.004] vs. 1.063 [0.659-1.715], Meseguer 1.150 [1.021-1.295] vs. 1.122 [0.981-1.284] and Basile 1.122 [1.008-1.249] vs. 1.038 [0.919-1.172]). In conclusion, SET is a preferred dataset to KID when developing or validating time-lapse algorithms, and day 3 conventional embryo morphology and female age should be considered as confounding factors.

Time-lapse algorithms and morphological selection of day-5 embryos for transfer: a preclinical validation study

OBJECTIVE

To determine the agreement between published time-lapse algorithms in selecting the best day-5 embryo for transfer, as well as the agreement between these algorithms and embryologists.

DESIGN

Prospective study.

SETTING

Private in vitro fertilization center.

PATIENT(S)

Four hundred and twenty-eight embryos from 100 cycles cultured in the EmbryoScope.

INTERVENTION(S)

None.

MAIN OUTCOME MEASURE(S)

Interalgorithm agreement as assessed by the Fleiss kappa coefficient.

RESULT(S)

Of seven published algorithms analyzed in this study, only one of the 18 possible pairs showed very good agreement (κ = 0.867); one pair showed good agreement (κ = 0.725), four pairs showed fair agreement (κ = 0.226-0.334), and the remaining 12 pairs showed poor agreement (κ = 0.008-0.149). Even in the best-case scenario, the majority of algorithms showed poor to moderate kappa scores (κ = 0.337-0.722) for the assessment of agreement between the embryo(s) selected as "best" by the algorithms and the embryo that was chosen by the majority (>5) of embryologists, as well as with the embryo that was actually selected in the laboratory on the day of transfer (κ = 0.315-0.802).

CONCLUSION(S)

The results of this study raise concerns as to whether the tested algorithms are applicable in different clinical settings, emphasizing the need for proper external validation before clinical use.

Recent advances in in vitro fertilization

The field of assisted reproductive technology is rapidly progressing with many new advances in the last decade. The present review discusses methods to improve oocyte quality in older women and new stimulation protocols that may improve the number of mature oocytes retrieved during an in vitro fertilization cycle. We will discuss the present use of pre-implantation genetic screening (PGS) and finally focus on some new methods to determine endometrial receptivity. The focus of this review is to point out areas of technology that may be controversial or are new enough to require proper controlled studies for validation.

Does time-lapse imaging have favorable results for embryo incubation and selection compared with conventional methods in clinical in vitro fertilization? A meta-analysis and systematic review of randomized controlled trials

OBJECTIVE

The present study aimed to undertake a review of available evidence assessing whether time-lapse imaging (TLI) has favorable outcomes for embryo incubation and selection compared with conventional methods in clinical in vitro fertilization (IVF).

METHODS

Using PubMed, EMBASE, Cochrane library and ClinicalTrial.gov up to February 2017 to search for randomized controlled trials (RCTs) comparing TLI versus conventional methods. Both studies randomized women and oocytes were included. For studies randomized women, the primary outcomes were live birth and ongoing pregnancy, the secondary outcomes were clinical pregnancy and miscarriage; for studies randomized oocytes, the primary outcome was blastocyst rate, the secondary outcome was good quality embryo on Day 2/3. Subgroup analysis was conducted based on different incubation and embryo selection between groups.

RESULTS

Ten RCTs were included, four randomized oocytes and six randomized women. For oocyte-based review, the pool-analysis observed no significant difference between TLI group and control group for blastocyst rate [relative risk (RR) 1.08, 95% CI 0.94-1.25, I2 = 0%, two studies, including 1154 embryos]. The quality of evidence was moderate for all outcomes in oocyte-based review. For woman-based review, only one study provided live birth rate (RR 1,23, 95% CI 1.06-1.44,I2 N/A, one study, including 842 women), the pooled result showed no significant difference in ongoing pregnancy rate (RR 1.04, 95% CI 0.80-1.36, I2 = 59%, four studies, including 1403 women) between two groups. The quality of the evidence was low or very low for all outcomes in woman-based review.

CONCLUSIONS

Currently there is insufficient evidence to support that TLI is superior to conventional methods for human embryo incubation and selection. In consideration of the limitations and flaws of included studies, more well designed RCTs are still in need to comprehensively evaluate the effectiveness of clinical TLI use.