No need to worry about which embryo to transfer for IVF, it can be scored by artificial intelligence!

In the early days of IVF in the clinic, doctors and patients tend to transfer multiple embryos at once, believing that this can increase the chance of a successful pregnancy, but this practice not only leads to the birth of a large number of twins or multiple births, but also makes the mother bear a greater risk during pregnancy and is more prone to pregnancy syndromes
such as diabetes and high blood pressure.
For fetuses in the womb, it increases the risk
of preterm birth, intrauterine growth retardation, low birth weight, and perinatal death.

Therefore, with the continuous development of assisted reproductive technology, more and more countries have set limits on the number of embryo transfers and actively promoted single embryo transfer, which makes it more important
to select the best embryo to improve the success rate of transfer.
This is not only to avoid the physical and mental harm caused by transplant failure, but also to avoid the waste of time costs, especially for
those elderly patients whose fertility is falling off a cliff.

At present, the third generation of IVF, or PGT technology, can detect its euploidy and whether there is a single-gene genetic disease, that is, chromosomal structural abnormalities
, through embryo biopsy.
However, biopsy is difficult to perform, requires manual performance by an experienced embryologist, and is an invasive method
.
Therefore, when biopsy is not necessary, it is necessary
to non-invasively grade embryos.

Embryo transfer can be performed on the third (D3) or fifth (D5) day of embryo development, so there are two different systems
for morphological scoring of embryos at these two time points.

D3 embryos are in the state of blastomere, according to the general law of fertilized egg division, the embryo should have 8 cells, but the actual situation in the embryo development rate is different, so when the number of embryonic cells is between 6-10, it is considered to have a good development potential and can be transferred
.
In addition, whether the size between cells is uniform and transparent, whether there are granules in the cytoplasm, and the degree of cell fragmentation are also important indicators
for judging the quality of embryos.

Figure 1 D3 embryo rating (Source: [1])

D5 embryos have entered the blastocyst stage, and the most widely used clinical application is the Gardner human blastocyst grading system, that is, the embryo is scored
according to the expansion of the blastocyst cavity, the hatching of the inner cell mass, the morphology of trophoblast cells and inner cell mass cells.
In general, if the blastocyst cavity occupies the embryo completely or is hatching from the zona pellucida, and the inner cell mass and trophoblast cells are numerous and closely arranged, the embryo has a high developmental potential
.

Fig.
2 The blastocyst cavity gradually expands and hatches from the zona pellucida (Source: [1])

However, regardless of the day of the embryo, the above evaluation system has traditionally only observed the embryo at a specific time point, and the dynamic information
during embryonic development is lost.
To this end, the Time-lapse imaging system (TLI) came into being, integrating the embryo incubator and optical system to help embryologists obtain continuous dynamic images of embryos from fertilization to transfer without interfering with embryonic development, providing more basis
for embryo screening.

However, the large amount of image data generated by dynamic monitoring brings a huge workload to embryologists, and the subjective evaluation of embryologists is likely to be affected by their past experience and fatigue, while artificial intelligence that is good at integrating and processing massive data can intervene in decision-making, which has the advantages of high efficiency, stable evaluation standards, and even can discover spatiotemporal dynamic characteristics
that are difficult for humans to perceive.

For example, in March 2021, the research team of Tongji Medical College of Huazhong University of Science and Technology developed an ensemble prediction model called STEM and STEM+ based on video preparation algorithms, spatial flow and event flow models, showing an accuracy of more than 70% and AUC of about 0.
8 (an indicator for evaluating the quality of classifiers in machine learning) in predicting the potential of embryonic development into blastocysts [2].

。 Another group developed an euploid prediction algorithm in December 2021 that can predict the euploidy nature of embryos by combining information provided by the jet lag imaging system and data such as patient age and blastocyst days, and predict the result of euploid AUC of 0.
80 on the test dataset [3
].

In the artificial intelligence scoring system that has entered the commercialization, iDAScore, an embryo decision support tool launched by vitrolife, a well-known Swedish company in the field of assisted reproduction, can automatically score embryos through a large and diversified data set training and verification, and is expected to be promoted to all clinics
.
Life Whisperer Viability, developed by Presagen, an artificial intelligence healthcare company, was able to assess embryo implantation potential and the likelihood of clinical pregnancy, outperforming embryologists for manual assessment by 25% and shortening pregnancy time by 15%.

Recently, another article published in Human Reproduction, "Human blastocyst spontaneous collapse is associated with worse morphological quality and higher degeneration and aneuploidy rates: a comprehensive.
" Analysis Standardized through Artificial Intelligence "uses artificial intelligence to comprehensively analyze
the correlation between spontaneous blastocyst collapse and blastocyst morphological quality, degradation probability and aneuploidy rate.

Figure 3 Research results (Source: [5])

Spontaneous blastocyst collapse refers to the outflow of blastocyst fluid, shrinkage of embryo volume, and detachment of the outer trophoblast from the zona pellucida during embryonic development, although the cause of its occurrence is unknown, but it is likely to be a sign
of the decline of embryonic development potential.

In this study, AI software automatically analyzed 720 cycles of 2348 Time-lapse microscopy (TLM) imaging videos, all of which reached the time of starting blastulation (tSB), in which at least one embryo
spontaneously collapsed during 77.
6% of cycles 。 The number of spontaneous blastocyst collapses is related to
the time of expanding blastocyst (tEB) and the delay in biopsy time.
The worse the blastocyst quality, the more spontaneous collapses, the longer the time, and the euploid rate is also significantly reduced, and the corresponding euploid rates from 0 to 4 collapses are 47%, 38%, 32%, 31% and 20%,
respectively.
In the study, data such as tSB, tEB, start and end time of each collapse, between two collapses, shrinkage ratio, embryo:zona pellucida ratio at the time of collapse, etc.
were automatically recorded
by AI software.

The results will help embryologists to incorporate the spontaneous collapse of blastocysts into the embryo implantation scoring system, and there are also articles pointing out that assisted hatching can improve the implantation potential of spontaneously collapsed blastocysts [6], and may be able to "save" these spontaneous collapse "poor births"
on the basis of AI scores in the future.
Assisted hatching is to thin and punch the zona pellucida to help the embryo "break out of the shell" and make smooth contact
with the endometrium.

It is believed that as the mystery of embryonic development is revealed little by little, the non-invasive embryo rating assisted by artificial intelligence will become more and more reliable, although it still cannot replace invasive PGT, but with its low dependence on medical equipment resources, it can be better promoted to fertility centers around the world, and even reach patients in more remote areas through telemedicine!