TIME-LAPSE

Blastocyst selection by time-lapse monitoring no better than by morphology alone in RCT

Time-lapse imaging fails to improve on morphology alone in new RCT.

Published 24 February 2022

A large multicentre RCT has found comparable ongoing pregnancy rates derived from blastocysts selected by time-lapse monitoring and those selected by morphology alone. Will artificial intelligence improve the time-lapse model?

Although the assessment of embryo quality from time-lapse monitoring is recognised as an ‘add-on’ in IVF, its use in many laboratories – especially in Europe – has become routine over the past decade. ESHRE, however, in its forthcoming guideline on the use of add-ons in ART will still include time-lapse (as a ‘selective’ add-on), while the HFEA’s traffic-light rating of add-ons gives it a cautionary amber light. In explaining its reasoning, the HFEA notes ‘there is conflicting evidence from randomised controlled trials (RCTs) to show that it is effective at improving the chances of having a baby for most fertility patients’.

Now, one substantial piece of evidence has been added to the time-lapse equation in publication of a large RCT from ten Nordic IVF centres involving 776 patients - all of whom had at least two good quality blastocysts available at randomisation.(1) Thereafter, two randomised groups proceeded to single blastocyst transfer, the one selected according to morphology alone and the other according to findings from a commercially available time-lapse model. Results showed that the primary endpoint of ongoing pregnancy rate for the time-lapse group was 47.4% and 48.1% for the control group. Thus, say the authors, ‘the findings of our study, alongside many others, demonstrate the limitations of morphology and morphokinetic markers to fully reflect the reproductive potential of developing embryos’. Thus, with a somewhat neutral view of the value of time lapse in the improvement of outcome in ART, they pin their hopes on artificial intelligence technologies ‘to interpret the full scope of [time-lapse] images and identify key features that have been missed or not quantified in current models’.

Despite the notable patient numbers in the trial, there was a shortfall in recruitment. Only around one half of the planned total were finally randomised, mainly because of clinic closures and other restrictions from the Covid pandemic. Nevertheless, even with reduced numbers the authors report that results did not indicate that any significant difference would have been found even with all intended patients. Thus, they add, their results question the rationale on spending resources on such time-lapse models ‘unless there is a gain in workflow and efficiency’. And that, they suggest, may yet emerge from new developments in automatic systems and deep learning models.

In trying to explain their neutral results the authors examined the (commercially available) algorithm model used and noted that ‘patient and treatment characteristics’ may influence model performance, although subgroup analyses in this trial found no preferential results dependent on the included variables. The report also notes that most patients in both arms of the trial were ‘good prognosis’ with at least two good quality blastocysts available, a level of baseline comparability which may well have limited differentiation.

So the study shows once again that embryology is not an exact science; indeed, the detection of a single embryo guaranteed to implant and form a healthy pregnancy remains the holy grail of the IVF lab. In this study, despite the transfer of a good quality blastocyst (according to the Gardner/Schoolcraft grading scheme) in all cases, around half the transfers did not achieve ongoing pregnancy. And how this apparent paradox of biology might be explained continues to drive the IVF lab. Chromosomal status, mitochondrial load, the rate and pattern of cleavage, or simply the look and feel of the cells have all been tested, but still the guaranteed embryo remains elusive – and implantation clearly depends on more than embryo quality.

However, another large RCT of another ‘add-on’, PGT-A, has similarly found no evidence in its results to support routine use, prompting an accompanying editorial in the New England Journal of Medicine to question the rationale for PGT-A.(2) ‘As a ranking tool,’ wrote the editorial authors, ‘PGT-A is unlikely to replace the morphologic selection of embryos for transfer, since the latter has proved to be highly effective in predicting successful implantation.’

It is notable that in this time-lapse study the investigators seem not to give up hope on time-lapse as a concept, though do ‘question its rationale’ without any evident gain in efficiency. They note that many of the earlier studies of its efficacy were ‘underpowered, non-randomized or have a high risk of bias’. Moreover, even the one RCT adequately powered to show a difference in LBR between embryo selection by time-lapse or morphology alone did not allow for a distinction between the possible benefit of the improved culture conditions and the hierarchical selection model used – ie, the algorithm or the environment?(3)

ESHRE’s own good practice recommendations for the use of time-lapse, published in 2020, noted in its preamble that, while time-lapse systems are being widely introduced in IVF labs, a clear clinical benefit of their use ‘remains to be proven’.(4) Similarly, the latest Cochrane review of time-lapse monitoring in embryo selection – in 2019 – included nine RCTs of ‘very low to low’ quality evidence but failed to find any clear distinction in LBR derived from embryos selected via time-lapse monitoring and conventional morphological assessment.(5)

More recently, studies are beginning to emerge in which AI-based models are trained from time-lapse images associated with pregnancy (‘known implantation data’, KID) to predict viability in selected embryos. One study reporting in 2020 found an accuracy of around 60% in a retrospective analysis of the model, a result, said the authors, beyond that expected by chance, while another, from 2019, found a high predictive value for implantation obtained from another deep learning model – which may, said the authors, ‘improve the effectiveness of previous approaches used for time-lapse imaging in embryo selection’.(6,7)

For the moment, however, time-lapse appears dependent for its value on trials like these, though now considered by many labs as less of an optional add-on and more as valuable technology. As Kersti Lundin, one of the study’s investigators told Focus on Reproduction: ‘Time-lapse as an incubator is an excellent tool, providing possibilities for training as well as validation, and for facilitating logistics in the lab.’ The problems arise, Lundin added, when clinics add time-lapse as an extra cost, and promote it as a means to improve delivery rates.



1. Ahlstrom A, Lundin K, Lind A-K, et al. A double-blind randomized controlled trial investigating a time-lapse algorithm for selecting Day 5 blastocysts for transfer. Hum Reprod 2022; doi:10.1093/humrep/deac020
2. Mastenbroek S, De wert G, Adashi EY. The imperative of responsible innovation in reproductive medicine. N Engl J Med 2021; 385: 2096-2100. doi:10.1056/NEJMsb2101718.
3. Rubio I, Galan A, Larreategui Z, et al. Clinical validation of embryo culture and selection by morphokinetic analysis: a randomized, controlled trial of the EmbryoScope. Fertil Steril 2014; 102: 1287–1294. e1285. doi:10.1016/j.fertnstert.2014.07.738
4. ESHRE working group on time-lapse technology. Hum Reprod Open 2020; 1-26. doi:10.1093/hropen/hoaa008
5. Armstrong S, Bhide P, Jordan V, et al. Time‐lapse systems for embryo incubation and assessment in assisted reproduction. Cochrane Database of Systematic Reviews 2019, Issue 5. CD011320.
DOI:10.1002/14651858.CD011320.pub4.
6. Tran D, Cooke S, Illingworth PJ, Gardner DK. Deep learning as a predictive tool for fetal heart pregnancy following time-lapse incubation and blastocyst transfer. Hum Reprod 2019; 34: 1011-1018.
doi:10.1093/humrep/dez064
7. VerMilyea M, Hall JMM, Diakiw SM, et al. Development of an artificial intelligence-based assessment model for prediction of embryo viability using static images captured by optical light microscopy during IVF. Hum Reprod 2020; 35: 770–784.
doi:10.1093/humrep/deaa013.

Get notified of new articles with our ESHRE newsletter.

Sign up and never miss an update