Noninvasive Testing of Preimplantation Embryos in Assisted Reproductive Technology

Abstract

One approach to improving the success of assisted reproductive technology (ART) is the careful selection of embryos prior to implantation. Although preimplantation genetic testing (PGT) is widely employed for embryo selection, it needs embryo biopsy and is detrimental to embryos. Thus, noninvasive testing of preimplantation embryos offers new possibilities for evaluating embryo quality. Here, we reviewed current progression of noninvasive embryo testing technologies, including the use of microscopy images combined with artificial intelligence (AI) to select embryos based on morphology, minimally invasive and noninvasive PGT of blastocoel fluid and spent embryo culture medium, and omics analysis of molecules in the culture medium to assess the developmental potential of embryos. More importantly, using the AI technology based on various type of data of each embryo will greatly improve the noninvasive embryo assessments. Thus, these cutting-edge technologies offer fresh insights into noninvasive testing of preimplantation embryos and have the potential to enhance the quality and efficiency of ART procedures.

1. Introduction

Infertility is a critical aspect of reproductive health and remains a highly prevalent global condition, affecting approximately 12.6–17.5% reproductive-aged couples worldwide [1]. In many cases, couples with infertility may opt for assisted reproductive technology (ART) to increase their chances of conception. ART encompasses various procedures, such as the retrieval of gametes from both the woman and man, the fertilization of multiple oocytes through in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI), the culture of embryos, and the transfer of embryos back to the woman’s uterus for implantation [2]. The growing use of ART treatments worldwide has sparked scientific and public interest in its effectiveness. The low success rate of ART can be partially attributed to the assessment of preimplantation embryos for selection [3, 4], as the majority of transferred embryos fail to implant in the uterus. Therefore, the development of noninvasive testing methods for embryo assessment would greatly enhance ART treatments.

2. Microscopy images coupled with artificial intelligence for embryo selection based on morphology

Under the current practice, ART treatment requires the fertilization of multiple oocytes. One of the main challenges is to select suitable embryos for transfer. Time-lapse (TL) imaging is a powerful tool for assessing embryos. The TL system monitors embryos in incubators with a built-in microscope and captures images every 5–20 minutes. These images are then assembled into a video, which is used for morphologic evaluation [5]. Typically, embryo selection involves a visual analysis by embryologists, who discard poor-quality and slow-growing embryos. A grading system for assessing blastocyst morphology quality is considered. The main morphologic criteria used are those described by Gardner and Sakkas [6], which are evaluated using three parameters: the developmental stage of embryos, the quality of the inner cell mass (ICM), and the quality of the trophectoderm (TE). However, embryo grading relies on the analysis of embryologists and is known to be subjective and susceptible to human error due to significant inter- and intra-observer variability [7].

In recent years, artificial intelligence (AI) has been extensively utilized to enhance and automate the procedure of ranking embryos. A convolutional neural network (CNN) is a type of deep learning algorithm that excels in processing 2-dimensional image data and analyzing extensive datasets in a grid pattern [8]. This neural network is commonly employed for medical imaging tasks and embryo assessment. By extracting pertinent information from embryo microscopy images and monitoring videos, AI has the potential to decrease the variability among embryologists and offer a precise and dependable interpretation of embryo morphology, as well as potentially predicting embryo viability [9, 10].

However, it is still crucial to assess the performance of these AI systems in comparison to the evaluations made by embryologists when selecting embryos. By employing a deep neural network (DNN)-based AI approach, Khosravi et al. [11] developed a framework (STORK) to predict blastocyst quality using more than 50,000 human embryo images, with an area under the receiver operating characteristic (ROC) curve (AUC) greater than 0.98. Additionally, they invited five embryologists from three different clinics to evaluate over 300 embryos from various laboratories. As anticipated, there was a low level of agreement among the embryologists, while the STORK accurately predicted the majority vote of the embryologists (at least three out of the five) with a precision of 95.7%. Their findings suggest that deep learning approaches can provide precise quality assessments of blastocysts in different clinical conditions. Similarly, another AI model for ranking blastocyst embryos using deep learning, trained on a large and diverse dataset, demonstrated the potential for improved prediction of clinical pregnancy [12]. A recent review article has encapsulated around 20 research papers on the use of AI in grading embryos. It has been concluded that the AI model achieved a median accuracy of 75.5% (range 59–94%) in predicting the grade of embryo morphology [10]. Furthermore, when clinical information inputs were incorporated, the AI models exhibited a higher median accuracy of 81.5% (range 67–98%).

The current findings indicate that AI models can outperform embryologists in terms of assessing embryo morphology, and using clinical data could enhance the performance of these models. Future studies could explore other potential applications of AI models, such as evaluating the ploidy status of embryos and predicting their implantation potential, which are the most relevant clinical outcomes in ART.

3. Noninvasive preimplantation genetic testing of embryos

3.1 Preimplantation genetic testing of embryos

As mentioned above, morphological assessment is the most commonly used method for selecting embryos, and various AI-based grading systems have been developed to enhance the evaluation of human blastocysts. However, it is worth noting that nearly half of embryos with good morphology are aneuploid [13], indicating that relying solely on morphological grading is inadequate for embryo selection, especially in terms of chromosomal assessments. Embryonic aneuploidy is a significant factor that impacts the success rate of assisted pregnancies and is responsible for over 50% of abortions [14]. Clinically, these embryonic chromosomal abnormalities can be prevented by conducting preimplantation genetic testing for aneuploidy (PGT-A). To carry out this test, embryonic biopsies are typically obtained at the cleavage stage (Day 3) or the blastocyst stage (Day 5–7). Subsequently, array-based comparative genomic hybridization (aCGH) or next-generation sequencing (NGS) platform can be utilized to identify numerical chromosome copy number aberrations. Clinical efficacies of PGT-A include increased rates of implantation and decreased risk of clinical miscarriage, particularly for patients with advanced maternal age (AMA) and recurrent implantation failure (RIF) and couples who have experienced recurrent pregnancy loss (RPL) [15, 16].

In addition to the traditional invasive PGT-A using cells from the TE, the technology of this test has evolved to include minimally invasive PGT-A (miPGT-A) using fluid from the blastocoel cavity and noninvasive PGT-A (niPGT-A) based on the spent culture medium (SCM) of embryos [17].

3.2 miPGT-A of blastocoel fluid

Blastocoel fluid (BF) is a fluid that accumulates in the blastocoel cavity. In 2013, Palini et al. [18] first reported the observation of DNA in the BF of embryos. Fragments of DNA shed from the embryo into this fluid result in a potential material resource for genetic analysis of PGT-A. Gathering BF involves inserting a needle between cell junctions into the blastocoel cavity and aspirating the fluid, instead of performing an embryo biopsy. Although it is still considered invasive, it appears to be well tolerated by the embryo [19]. However, the concordance rate between BF and TE biopsy has varied, ranging from no agreement with TE PGT-A to a significant correlation as high as 97.4% [20, 21]. There are various explanations for this wide range of concordance values, such as the unclear source of detected DNA fragments and different protocols used in multiple embryology laboratories. While there is a hypothesis that DNA in the BF may better represent the embryos compared to what is secreted into the culture medium [18], the concordance results differ from those of the SCM. Additionally, the SCM is collected after the embryo has been removed, making it truly noninvasive.

3.3 niPGT-A of spent culture medium

The culture medium is the liquid in which the preimplantation embryo will grow during its crucial early stages of development. In 2013, Stigliani et al. [22] made the initial discovery that human preimplantation embryos release genomic and mitochondrial DNA into the culture medium during in vitro culture. They also confirmed that the contents of the DNA were linked to embryo fragmentation. Since then, the potential clinical use of these cell-free DNA (cfDNA) has become a matter of great concern in the field of ART, especially in the development of noninvasive technology for determining chromosomal abnormalities [21, 23, 24]. Although the presence of cfDNA in SCM has been confirmed, achieving chromosomal screening would require high sensitivity and reproducibility of whole-genome amplification (WGA) for these low-amount DNA in SCM. Xu et al. [25] reported a noninvasive chromosomal screening (NICS) method by sequencing the genomic DNA in SCM of human blastocysts, using multiple annealing- and looping-based amplification cycles (MALBAC). The results of the NICS assay were validated by comparing each result with the chromosome information directly obtained from the corresponding embryos. The NICS assay has a high correlation with the identification of aneuploidy, making it potentially suitable for a wider application of PGT-A due to its noninvasiveness.

The main challenge of developing a niPGT-A technique was to verify the consistency between samples and the traditional embryo TE biopsy. Several recent studies in randomized clinical trials have primarily investigated the clinical effectiveness of niPGT-A. Some studies have highlighted the challenges associated with this technique, while others have demonstrated its potential for the future.

In a clinical study, BF samples exhibited high rates of amplification failure, resulting in an overall concordance rate as low as 37.5%. While samples from SCM performed better than BF samples, genetic analysis of SCM samples showed a high detection of artifacts due to the risk of maternal contamination. Therefore, SCM should not be utilized for the detection of single-gene mutations [26]. Another study also reported a high rate of DNA amplification failure in niPGT-A samples. In addition, there were 42 cases (40.4%) where the whole-chromosome discordance with TE biopsy was observed, rendering the clinical applicability of niPGT-A virtually impossible [27]. With further improvements, DNA can now be detected in the majority of SCM samples. This allows for the quantitative and qualitative measurement of segmental aneuploidy or mosaicism caused by individual chromosomal aberrations. However, TE biopsy still remains a more accurate and reliable methodology for PGT-A [28]. Although cfDNA in SCM shows potential as a safe and simple strategy for PGT-A, its clinical application for diagnostic purposes still requires further validation.

To promote the clinical application of niPGT-A, various validation studies have been conducted and have reported the comparison results of niPGT-A with TE biopsy. Jiao et al. [29] reported that niPGT-A can provide fast and accurate results, with the overall testing time reduced to 7.5 hours and a concordance of approximately 90% with results from embryo samples. These advanced findings not only indicate the effectiveness of niPGT-A but also suggest its potential for achieving fresh blastocyst transfer in the future. In addition to higher concordance rates for embryo ploidy and chromosome copy numbers, niPGT-A also demonstrates higher positive predictive value (PPV) and specificity for embryonic mosaicism compared to TE biopsy [19]. Consistent results were also observed in a large-scale validation study of approximately 260 embryos, promising niPGT-A as a reliable assay for prioritizing transfer of chromosome-normal embryos [30]. Recently, a clinical study has been conducted on a large multicenter series involving more than 1300 embryos, indicating that embryonic cfDNA originated from both TE cells and ICM cells [31]. Their noninvasive analysis of niPGT-A demonstrates a high concordance with TE biopsy results. Despite promising ongoing research on the clinical concordance results of niPGT-A, there are still concerns regarding the investigation of challenges and improvement of methodology.

3.4 Combination of blastocoel fluid and spent culture medium

Another approach to noninvasive testing is a combination of BF and SCM, which increases the amount of assayable cfDNA. By analyzing blastocysts derived from oocytes not showing evidence of fertilization (no pronuclei visible, 0PN) and oocytes of single pronuclei (1PN), the miPGT-A using BF enriched culture medium shows consistent results of chromosomal aneuploidies compared to TE biopsy PGT-A [32]. In a validation study, Kuznyetsov et al. [33] reported that the accuracy of miPGT-A was not correlated with blastocyst morphological grade. Furthermore, the overall concordance rate of aneuploidy with TE biopsy was approximately 97.8%.

Despite the combination of BF and SCM, the combination of biopsy and noninvasive testing results could also improve the outcomes of assisted pregnancy. Sun et al. [34] have recently reported a similar effectiveness of niPGT-A in assessing embryos with both normal and abnormal karyotypes. Meanwhile, they divided normal embryos into two groups based on the TE biopsy results: euploid in niPGT-A and aneuploid in niPGT-A. By following up on their clinical outcomes, they found that the clinical and ongoing pregnancy rates were higher in the euploid group compared to the aneuploid group, although there was no statistical difference between them. Thus, solely identifying aneuploidy may result in the wastage of embryos due to the high false positives. These findings also suggest that combining biopsy and niPGT-A results could improve the clinical outcome. With the development of noninvasive methodology, it becomes not only an alternative or replacement for present invasive technology but also an encouraging combination to improve ART treatment.

3.5 New perspectives of niPGT-A

During conventional IVF procedure, there is a risk of potential contamination from spermatozoa and cumulus cells attaching to the zona pellucida. As a result, ICSI is typically performed in order for PGT-A to be clinically applicable [35]. However, ICSI is only used to treat couples with severe male factor infertility and is an invasive procedure. Additionally, for patients who did not undergo PGT-A in a fresh cycle and experienced implantation failure or miscarriage due to chromosomal abnormalities, they need to choose existing cryopreserved embryos from IVF for the next transfer cycle. For these patients, selecting embryos based on morphological assessment alone cannot exclude embryos with chromosomal abnormalities. Moreover, PGT-A of frozen embryos requires a series of invasive procedures, including thawing, biopsy, and refreezing. Therefore, if the chromosome ploidy of embryos can be detected by niPGT-A, the need for invasive biopsy will be eliminated, thereby reducing the risk of embryo damage.

A prospective clinical study utilized a quantitative method to detect parental DNA contamination in conventional IVF [36]. The study’s findings indicated a low rate of maternal contamination and negligible risk of paternal contamination. PGT-A can be applied to vitrified conventional IVF embryos. Additionally, Xie et al. [37] confirmed that sperm does not impact the niPGT-A results as the current WGA amplification system failed to amplify sperm DNA. They also discovered that niPGT-A from traditional IVF embryos showed high rates of agreement with corresponding TE biopsies, for both fresh and thawed blastocysts. Another research study examined the agreement between niPGT-A results obtained from BF of fresh blastocysts, SCM of thawed embryos, and PGT-A from biopsies [38]. It has shown that SCM has a higher concordance rate and superior diagnostic performance in identifying euploid or aneuploid categories. The performance of SCM is comparable to that of ICSI blastocysts, which gives hope for the potential use of niPGT-A in existing cryopreserved embryos.

Besides the diagnostic potential, combining niPGT-A results with morphological assessment could provide an effective tool to guide frozen-thawed single blastocyst transfer, thus improving the clinical pregnancy outcome [39]. Similarly, for patients experiencing RPL or RIF, selecting euploid embryos through niPGT-A can reduce the miscarriage rate and improve the clinical pregnancy rate [40].

Altogether, these results preliminarily confirm that the niPGT-A approach using SCM from thawing embryos or embryos from conventional IVF has comparable performance to fresh ICSI blastocysts. It could serve as an alternative or complementary approach to embryo biopsy for improving the assessment of preimplantation embryos. Additionally, the potential impact of niPGT-A on infertility treatment may extend to pregnancy outcomes.

3.6 AI models in predicting embryonic ploidy

In addition to noninvasive testing for assessing embryonic chromosomes, static embryo imaging, analyzed by AI algorithms along with clinical parameters, has also shown potential in predicting the ploidy status of preimplantation embryos. Several AI models have been developed, employing different machine learning or deep learning algorithms, such as the random forest classifier (RFC), logistic regression (LR), support vector machine (SVM), k-nearest neighbors (k-NN), and the CNN-based ResNet18 algorithm.

For example, a retrospective study developed an embryo evaluation model, called STORK-A, to predict the ploidy status of embryos [41]. Initially, the study included embryoscope images, blastocyst morphological assessments, morphokinetic parameters, and maternal age and used four different models (XGBoost, k-NN, SVM, and RFC) to determine the importance of variables. Subsequently, it evaluated the accuracy of their deep learning CNN-based algorithm for predicting ploidy. After validating its results in two independent and external datasets, the best-performing models accurately predicted binary aneuploid or euploid embryos with an accuracy of 69.3% and a PPV of 76.1%. For complex aneuploidy, the accuracy was 77.6%. This model is the first to predict complex aneuploidy, expanding the application of AI algorithms in a novel way. Additionally, Jiang et al. [42] demonstrated that combining CNNs of static blastocyst images with clinical characteristics could enhance the accuracy of classifying ploidy status from CNN alone. Therefore, including clinical parameters is essential for improving the predictive performance of AI models.

Besides static images, TL incubators can provide more detailed spatial and temporal information about embryos. By using a deep learning approach to analyze monitoring videos, Paya et al. [43] have identified a noninvasive method for diagnosing aneuploidy. Their model’s accuracy ranged from 0.62 to 0.73. Serrano-Novillo et al. [44] discovered that a new parameter during the initial cell cleavage was highly associated with ploidy status. A logistic regression model incorporating various morphokinetic parameters extracted from videos yielded an ROC value of 0.69 for the prediction of chromosome ploidy status. Interestingly, when comparing 12 machine learning models using a multicenter morphokinetic meta-dataset, it was found that mixed effects logistic regression performed better than all other approaches for predicting ploidy [45].

Various AI models have been developed using embryonic morphology from static images or TL videos. In addition to the studies mentioned here, advancements in this field have been discussed in a recently published review paper [46]. The current PGT-A, whether through traditional invasive biopsy or noninvasive screening methods, has several limitations, including financial burdens and delays in result reporting. Due to their ability to quickly make decisions for embryo selection, AI models have the potential to improve clinical pregnancy rates while reducing the cost of each ART cycle.

4. Omics analysis of the SCM to evaluate the developmental potential of embryos

4.1 Molecular markers of embryonic potential in SCM

The evaluation of embryo quality is a crucial factor in determining the success rates of IVF. In order to increase the chances of success, scientists have been constantly searching for reliable methods to predict and select embryos with the highest potential. Currently, the main methods for assessing embrynic potential are morphological assessment and genetic testing for ploidy status. However, these criteria are not able to accurately identify embryos with a high chance of implantation [47]. Therefore, it is of utmost importance to find a noninvasive, reliable, and fast method for assessing embryos. Recently, researchers have started investigating techniques that link molecular markers in SCM with the quality of embryos. This area of research has attracted significant interest due to its potential to enhance the success rates of IVF, decrease the chances of multiple pregnancies, and reduce the expenses associated with infertility treatments.

In recent years, many studies have focused on identifying molecular markers that are closely linked to embryo quality and developmental potential. These markers include metabolites, proteins, and nucleic acids found in SCM. By analyzing these molecules in SCM, researchers can gain insights into the status of the embryo and predict its developmental potential (Table 1).

	Markers	Samples	Culture time	Outcome	Reference
METABOLITE	Pyruvat, alanine	48	D3	Implantation outcome	Pudakalakatti et al. [48]
	Glucose	50	D3	Embryo selection	Gardner et al. [49]
Protein	ALCAM, EPHB4, JAM-A/F11R, SELE, CCL24, FAS, PDGF-A, PECAM-1, TIMP4, PON3, CSTB, BMH, vWF	42	D5	Implantation outcome	Freis et al. [50]
	sCD146	126	D2/D3/D5	Embryo selection	Bouvier et al. [51]
	LIF	208	D5	Implantation outcome	Li et al. [52]
	GDF9	399	D3	Implantation outcome	Li et al. [53]
	sHLA-G	539	D3/D5	Implantation outcome	Vani et al. [54]
	HptA1	143	D3	Embryo selection	Montsko et al. [55]
DNA	mtDNA, gDNA	699	D3/D5	Embryo selection	Stigliani et al. [56]
	cfDNA	55	D3	Embryo selection	Pan et al. [57]
RNA	miR-142-3p	36	D3	implantation outcome	Borges et al. [58]
	miR-20a, miR-30c	53	D5/D6	Embryo selection	Capalbo et al. [59]
	miR-191-5p, miR-24-1-5p	50	D5	Implantation outcome	Acuna-Gonzalez et al. [60]
	PROK1	34	D2	Implantation outcome	Alfaidy et al. [61]

Table 1.

Example of molecular markers predicting embryonic development potential.

4.1.1 Metabolites

Many researches indicate a strong correlation between the metabolomic characteristics of preimplantation embryos and their potential for development. Changes in the metabolomic profile of SCM can to some extent be used as indicators for selecting embryos and predicting their developmental potential, as well as the clinical outcomes of pregnancy [62]. For example, researchers used NMR spectroscopy to discover that the ratio of acetate to alanine in SCM significantly decreased in embryos with successful implantation, while lactate levels remained similar. By combining these parameters, there is potential to identify a single factor closely linked to implantation potential [48]. In a recent study, Gardner et al. [49] have used microfluorometry to analyze the daily glucose consumption of compacted embryos in SCM. The results showed that transplanted embryos had significantly higher glucose consumption on the 4th and 5th days compared to embryos that did not develop after transplantation. Additionally, it was found that glucose uptake was not dependent on embryo grade.

Studies have found that Raman spectroscopy can be used to detect the metabolic profile of third-day embryo culture medium. This detection can accurately predict the potential of embryos to successfully develop into blastocysts, with an accuracy rate of 73.53% [63]. The use of liquid chromatography-mass spectrometry (LC-MS) technology in detecting metabolites in SCM demonstrates the changes of 13 metabolites. This enables noninvasive prediction of the implantation potential of Day 3 embryos [64]. A study has established a method based on metabolomics and data modeling techniques using Fourier transform infrared (FTIR) spectroscopy [65]. This method has successfully identified diverse metabolic profiles in the supernatant of SCM from Day 3 embryos. It has also been able to predict around 60% of non-implanted embryos. Additionally, the study suggests that the oxidative status of SCM may play a role in the developmental health of embryos, specifically in the formation of strong blastocysts and successful implantation. High-quality embryos exhibit a higher level of oxidative metabolism, which results in an increased “oxidative load” on the surrounding culture medium [66].

Consequently, it will be crucial to identify metabolites that can predict the potential for embryonic development, construct reliable predictive models using metabolomics, and integrate them with traditional morphological assessment methods. This will enable genuinely noninvasive, rapid, and accurate evaluations of embryos, making it a pivotal approach for IVF embryo selection in the future.

4.1.2 Proteins

At present, it is possible to accurately and noninvasively determine the proteomic profile of SCM. Targeted studies of specific protein fragments using mass spectrometry and similar methods have also been proven effective in aiding embryo selection [51, 52, 67], thus providing insight into the biochemical condition of the embryos.

Freis et al. [50] utilized the proximity extension assay (PEA) technique for the first time to analyze individual blastocyst culture media influenced by their potential for implantation. They examined the expression of various proteins that could serve as potential biomarkers for noninvasive assessment of individual blastocysts. The expression levels of ALCAM, EPHB4, JAM-A/F11R, SELE, CCL24, FAS, PDGF-A, PECAM-1, TIMP4, PON3, CSTB, BMH, and vWF were found to be significantly higher in the SCM of successfully implanted blastocysts compared to nonimplanted blastocysts. Li et al. [52] discovered that there is a significant increase in LIF levels in the SCM of embryos in the pregnant group compared to the nonpregnant group. The concentration of GDF9 in SCM is correlated with embryo quality and clinical outcomes [53]. Specifically, in 94% of couples, nontransferable embryos show a higher GDF9 concentration in the culture medium compared to transferable embryos. Interestingly, the levels of sHLA-G are positively correlated with blastocyst grading. In the group with live births, the levels of sHLA-G are significantly higher than those in the group without live births, suggesting the potential to predict successful pregnancy outcomes [54]. Montsko et al. [55] employed liquid chromatography combined with mass spectrometry to analyze HptA1. Their research showed that the degree of Hpt cleavage and the concentration of HptA1 in the culture medium can accurately predict the results of embryo transplantation. Montsko et al. [67] first demonstrated that sCD146 can be detected in SCM. Additionally, a high concentration of sCD146 is linked to a decreased likelihood of embryo implantation.

4.1.3 Nucleic acids

Cell-free nucleic acids in SCM consist of cfDNA and cell-free RNA (cfRNA). Studies have revealed that in the culture medium of embryos cultured for 3 days, the ratio of mitochondrial DNA (mtDNA) to genomic DNA (gDNA) is correlated with the outcome of embryo transplantation [56]. Besides, embryos that successfully develop into blastocysts exhibit a significantly higher mtDNA/gDNA ratio in the culture medium compared to those that do not develop into blastocysts. Pan et al. [57] assessed the correlation between cfDNA fragment patterns in follicular fluid and SCM of patients with subsequent embryo grading, demonstrating for the first time a significant association between cfDNA in follicular fluid and SCM with both embryo occurrence and embryo grading.

Research has found that microRNAs (miRNAs) are expressed during embryo cell division and genome activation processes. Additionally, small RNAs can be secreted into the SCM [68]. Several researchers have also identified miRNA molecular markers associated with embryo implantation outcomes and pregnancy outcomes [58, 59, 60, 69]. In the subgroup of negative implantation in the SCM, there was a notable rise in the expression of miR-142-3p. This observation confirms the differential expression of miR-142-3p in human embryos as they progress to the blastocyst stage, depending on their implantation status [58]. A study conducted a comprehensive characterization of the miRNAs of SCM from euploid blastocysts that were either implanted or unimplanted. The study highlighted two specific miRNAs, miR-20a and miR-30c, which exhibited higher concentrations in the implanted blastocysts. Additionally, in silico predictions suggested that these miRNAs may play a role in embryonic implantation [59]. Kamijo et al. [69] identified 53 differentially expressed miRNAs in the culture medium of the pregnancy and nonpregnancy groups. By utilizing logistic regression analysis, a high-quality blastocyst prediction model was generated with 8 miRNAs, resulting in an average accuracy of 0.82. Another research indicates that miR-191-5p in SCM could potentially serve as a positive biomarker for pregnancy, while miR-24-1-5p may indicate a poor prognosis [60]. This former miRNA regulates IGF2BP-1 and IGF2R, which are associated with the implantation window. Conversely, miR-24-1-5p may be associated with a negative prognosis for human embryo development.

At the cellular level, the concentration of the PROK1 gene in follicular fluid and SCM was determined using real-time quantitative amplification technology. This study found a correlation between the PROK1 gene content and the success of embryo implantation [61]. It is speculated that high levels of PROK1 may support a strategy of transferring a single embryo, while low levels of PROK1 may serve as an additional criterion for extending embryo culture or performing double embryo transfer in order to enhance IVF outcomes.

4.2 AI models in predicting developmental potential

In addition to the noninvasive testing of molecular markers in SCM, AI models have also been applied to predict clinical pregnancy by utilizing clinical treatment information, along with embryonic images or TL videos.

Researchers used a semantic segmentation neural network model to measure embryonic morphometric parameters, including blastocyst size, ICM size, and ICM shape. The study supported the correlation between larger blastocyst size and increased potential for implantation [70]. Using colored blastocyst images, a novel AI system called FiTTE has the potential to offer more accurate predictions for clinical pregnancy compared to the conventional Gardner scoring system [71]. A research team has developed iDAScore v2.0, a fully automated deep learning model, using an analysis of TL sequences from a diverse and extensive dataset collected across 22 IVF centers in order to rank embryos. The accuracy rate for predicting implantation outcome varies between 0.621 and 0.707, depending on the day of transfer [72]. In a subsequent retrospective observational study carried out at two additional IVF centers, iDAScore v2.0 was able to distinguish between embryos that led to live birth or no live birth. The AUC for embryos transferred at Day 2 was 0.627, while for embryos at Day 3, the value was 0.607. These findings indicate that iDAScore v2.0 holds some predictive value for live birth [73]. Based on six nonredundant morphodynamical features of TL images, Amitai et al. [74] developed a decision-support tool for identifying embryos at high risk of first-trimester miscarriage. Prioritizing embryos based on their predicted risk of miscarriage, in combination with their implantation potential, is expected to improve live birth rates and shorten the time to pregnancy. Recently, Liu et al. [75] have trained an AI model using CNN to process blastocyst images and a multilayer perceptron to process clinical features. Their work suggested that including clinical features of patient couples along with blastocyst images increased the accuracy of predicting live birth. They achieved an AUC of 0.77. Meanwhile, Duval et al. [76] trained algorithms on multicentric clinical data using a hybrid model. The average AUC of predicting fetal heart rate during pregnancy significantly increased when using their hybrid model compared to the algorithms solely based on the TL videos. This AI model can assist embryologists in various clinical scenarios.

In summary, the AI models can predict the potential of embryonic development in various aspects. On average, the AI models achieved a median accuracy of 77.8% (range 68–90%) in predicting clinical pregnancy. When combining both TL images or videos and clinical information inputs, the AI models demonstrated a higher median accuracy of 81.5% (range 67–98%) [10]. The evolving AI models hold significant promise for the IVF field and embryo selection.

5. Discussion

Noninvasive testing brings new possibilities in evaluating preimplantation embryos, such as examining their morphology, conducting genetic testing for aneuploidy, and predicting their developmental potential. While the substances released into the BF and SCM are not confined to cfDNA, other omics analyses of the SCM can also enable the simultaneous assessment of chromosome ploidy and provide additional information. A recent study has successfully utilized the DNA methylation approach of cfDNA in the SCM to accurately detect chromosome aneuploidy of embryos, trace the cellular origins of the SCM, and confirm the suspicion of cumulus contamination in the SCM. This breakthrough aims to establish a DNA methylation-based niPGT-A method [77]. The development of a methodology to achieve simultaneous quantification and characterization of cfRNA [78] also provides opportunities to comprehensively profile cfRNA in SCM and discover additional molecular markers associated with developmental potential.

In addition, instead of treating the noninvasive parameters for embryo selection mentioned above as separate variables related to the normality and quality of embryos, it would be wise to consider them collectively. Utilizing AI models along with a comprehensive set of information for each embryo can result in a more precise prediction. For instance, integrating machine learning with metabolomic and embryologic data could enhance the accuracy of predicting the potential for embryo implantation [79]. Developing a grading system with AI models using omics data has the potential to enhance the quality and efficiency of IVF procedures, resulting in real-time clinical benefits for patients.

6. Conclusion

Although the invasive embryo biopsy technology for preimplantation genetic testing (PGT) has been widely used for embryo selection in human IVF practice, it is detrimental to embryos. Thus, searching a method of the noninvasive testing of preimplantation embryos has gradually become a hot project in ART. Here, we reviewed the current progression of noninvasive embryo testing technologies, including morphokinetics from TL images, miPGT-A of BF, niPGT-A of SCM, combination analysis of BF and SCM, and molecules in SCM and clinical information associated with developmental potential of embryos. The combination of these methods will be helpful to achieve comprehensive assessment of preimplantation embryos so that the normal embryos could be selected for transfer. It is important that we think that integrating AI technology into these noninvasive embryo assessments based on vast amounts of different data types for each embryo will help us to more accurately select normal genetic embryos and with great developmental potential. This combined technology will make ART more effective and cheaper for patients undergoing infertility treatment.

Acknowledgments

This project is supported by the Guangdong Basic and Applied Basic Research Foundation (No. 2023A1515110478).

Conflict of interest

The authors declare no conflict of interest.