What is AI-assisted embryology, and does it increase the chance of live birth in embryo selection? Randomised studies and the ESHRE and HFEA assessments.
AI-assisted embryology means using, in the IVF laboratory, software that analyses embryo images and produces a score or a ranking for the embryologist. In the studies conducted, it has not been shown that this approach is superior to the conventional assessment of the embryo or that it increases live birth; in one large multicentre study, it also could not be proven that the software was no worse than the conventional assessment. The software is a supporting tool; your embryologist and your doctor decide together which embryo will be transferred.
How does artificial intelligence produce a score from an embryo image?
This software works with methods such as deep learning, which learn patterns from a large number of embryo images. The input may be a still image taken at a single moment, or a sequence of images recorded by time-lapse systems. The output is a score relating to the embryo's developmental potential, or a ranking of the embryos.
Does the software replace the embryologist's assessment?
In the conventional method the embryologist assesses the shape and rate of development of the embryo under the microscope. The software offers additional data to this assessment. According to the current expert view on this subject, artificial intelligence should not yet replace conventional assessment. The same view reports the observation that artificial intelligence scores are largely related to the existing appearance (morphology) grading; this may mean that the independent information the software provides could be limited. Possible benefits such as shortening the assessment time and reducing the difference between observers are put forward and are being researched; it has not yet been shown that they are reflected in your outcome.
Did artificial intelligence give better results in controlled studies?
One of the most comprehensive controlled studies in this area is a multicentre, double-blind study carried out in centres in Australia and Europe. Embryos selected by a score based on deep learning were compared with those selected by conventional appearance (morphology) grading. No significant difference was found between the two groups in clinical pregnancy; however, it could not be shown, according to the criterion set beforehand, that the software was no worse than the conventional assessment. The live birth rate was found to be numerically lower in the artificial intelligence group, but this difference did not turn out to be statistically significant.
Do the assessments of expert bodies cover artificial intelligence directly?
In current good practice recommendations prepared for Europe, time-lapse imaging is not recommended for increasing live birth. This assessment does not cover artificial intelligence scoring directly, but the results point in a similar direction: no benefit has been shown. Another current assessment states that no improvement in ongoing pregnancy was found in two controlled studies examining algorithmic time-lapse selection and that this is consistent with other reviews.
The regulator in the United Kingdom classifies time-lapse incubation and embryo analysis in the lowest category for most patients; that is, there is evidence that it does not increase the chance of live birth. There is no separate classification specific to artificial intelligence. This classification does not mean that the practice is harmful.
Why is the evidence limited?
The software in the study worked only with images from a particular time-lapse system; for this reason the results cannot be directly applied to other devices and other software. The study was funded by the company that developed the software. Longer-term outcomes such as cumulative live birth and the duration of pregnancy were not assessed. In addition, software may not show the same performance in conditions outside the data it was trained on and the device.
For whom can it be considered?
In the assessments reported above, it is not said that AI-assisted selection is necessary routinely; no separate benefit has been reported for groups such as advanced age, repeated failure or few embryos. Whether the practice is considered depends on the centre's protocol; you can ask your doctor why it is being recommended.
Are artificial intelligence, time-lapse and PGT-A the same thing?
No, the three seek the answers to different questions. Time-lapse concerns how the embryo is monitored and cultured. Artificial intelligence is the software layer that looks at what is to be drawn from the image obtained. PGT-A, on the other hand, is a genetic test that looks at the number of chromosomes in cells taken from the embryo. The use of one does not require the other; one does not take the place of another.
Does using software give assurance about your chance of pregnancy?
The main point to watch for is indirect effects rather than a direct procedure risk. The idea that "if there is technology, my chance is high" can lead to unrealistic expectations. The outcome of IVF is determined together by factors such as your age and the number and quality of embryos; the presence of software does not change them.
What should you learn about the extra fee and the use of your data?
An extra fee may be requested for a practice with limited evidence. You can learn from the centre and from the consent document how your embryo images and the related health data are stored and used.
What can you ask your doctor about artificial intelligence in embryo selection?
- Is a software score used in embryo selection, and if so at what stage?
- Who makes the final transfer decision, and what happens when the score and the embryologist's assessment conflict?
- What can this practice change in my situation?
- Is there an extra fee?
- How are my images and data stored and used?
Frequently Asked Questions
Does artificial intelligence take the place of genetic testing?
In the current assessment on this subject, studies aimed at artificial intelligence predicting the chromosome number were found promising, but it is stated that the diagnostic accuracy is insufficient. Software flagging blastocysts that appear to be at high risk of a chromosome problem is thought of not as a diagnosis, but as an approach to determining which embryos could be candidates for genetic examination.
Is an embryo that gets a low score not used?
What outcome the score leads to, that is, whether the order of transfer will change, whether the embryo will be frozen, or whether it will not be used, depends on the centre's protocol. Asking your embryologist or doctor about this policy before the procedure is a concrete way of reducing your anxiety.
Does artificial intelligence harm the embryo or the baby?
The software performs analysis on images; no additional procedure is applied to you. No additional risk has been stated in time-lapse assessments; comprehensive safety data specific to artificial intelligence software itself, however, are limited. The absence of a reported risk does not mean certainty for every individual situation.
Is it worth paying an extra fee?
Current assessments on this subject remind patients who are considering paying for add-on practices whose evidence is not settled that trying the standard cycle more than once instead may also be an option. You can ask before the procedure about the scope of the fee and what the practice would change in your cycle.
Does the use of artificial intelligence increase my chance of live birth?
In the studies available, neither superiority over the conventional assessment nor an increase in live birth has been shown. Having software does not mean that the outcome will be better.
Does the absence of a significant difference in the study show that the two methods are equivalent?
No. Although no significant difference was found in clinical pregnancy in the large multicentre study, it could not be shown, according to the criterion set beforehand, that the software was no worse than the conventional assessment.
Does the software replace the embryologist?
No. The current expert view is that artificial intelligence should not yet replace the conventional assessment. Your embryologist and your doctor make the transfer decision together.
Do time-lapse, artificial intelligence and PGT-A take the place of one another?
No. Time-lapse is monitoring and culturing, artificial intelligence is image analysis, and PGT-A is the examination of the number of chromosomes in the cells taken. The use of one does not require the other.
Can the study result of one piece of software be applied to all software?
No. The software examined worked with images from a particular time-lapse system. The same performance cannot be assumed in conditions outside the training data and the device.
Has a separate benefit been shown in advanced age or when there are few embryos?
No. In the assessments reported, no separate benefit has been reported for the groups of advanced age, repeated failure or few embryos. You can ask why it is being recommended to you.
Is the body assessment made for time-lapse specific to artificial intelligence?
No. The time-lapse classification reported is not a separate classification specific to artificial intelligence. The time-lapse results and the results in which the software is examined directly should be assessed separately.
Can I learn how my images are used?
Yes. You can learn from the centre and from the consent document how your embryo images and the related health data are stored and used.
Get Your Free Consultation
Fill out the form below and our team will contact you
