Beyond the Inner Cell Mass: Evaluating the Impact of Inner Cell Mass Visibility on Implantation Prediction in Static Blastocyst Images - A Retrospective Cohort Study.

In vitro fertilization (IVF) has significantly improved the success rates of couples facing infertility, with advances in embryo selection and culture techniques playing a crucial role in optimizing outcomes.[ 1 2 ] Among these advancements, elective single-embryo transfer (eSET) has gained prominence as a strategy to reduce the risks associated with multiple gestations while maintaining high success rates.[ 3 ] Studies have demonstrated that eSET, particularly at the blastocyst stage, effectively minimizes the incidence of twin pregnancies without compromising the overall live birth rates, leading to improved maternal and neonatal health outcomes.[ 4 5 6 ] Blastocyst assessment, is a key step in selecting the most viable embryo for transfer. Extending embryo culture to the blastocyst stage enhances selection by allowing the evaluation of developmental competence and alignment with the uterine environment. Morphological assessment, particularly of the ICM and TE, helps to identify embryos with the highest implantation potential, ultimately improving success rates and reducing the time to pregnancy.[ 7 8 ] Ideally, embryo assessment should be performed without removing embryos from their stable culture environment to maintain optimal developmental conditions. Time-lapse incubators equipped with built-in cameras enable continuous monitoring without disrupting incubation, preserve stable culture conditions, and potentially improve embryo quality and selection.[ 9 10 ] However, owing to the lack of conclusive evidence showing improvement in live birth rates and the high cost of these systems,[ 11 12 ] many IVF laboratories worldwide continue to rely on traditional benchtop incubators and single time-point evaluations. To facilitate rapid and standardized assessment, embryologists capture high-resolution digital images of blastocysts using a High Definition (HD) Complementary Metal-Oxide-Semiconductor (CMOS) digital camera or similar imaging systems. This allows embryologists to perform detailed assessment and grading at a later time point by evaluating digital images of blastocysts using the Gardner scoring system, which remains the most widely accepted method for blastocyst evaluation.[ 13 ] Embryologists rank embryos by evaluating their blastocoel expansion, ICM quality and TE quality. This assessment helps identify the most viable embryo for transfer or rank embryos for vitrification for later transfer.[ 14 15 16 ] The ICM within a blastocyst is pivotal as it differentiates into the fetus, making it crucial for embryo selection in IVF procedures. Morphologically, the ICM is evaluated based on its size, shape and cell density.[ 17 ] A well-formed, dense, and cohesive ICM is often indicative of a higher implantation potential.[ 18 ] Studies have demonstrated a direct correlation between ICM morphology and implantation outcome. Blastocysts that successfully implant and result in positive outcomes have been shown to possess higher-quality ICMs than those that fail to implant.[ 19 ] Quantitative assessments, such as the ratio of ICM diameter to blastocyst diameter, have also been identified as effective predictors of successful implantation in single blastocyst transfer cycles.[ 20 ] These findings highlight the importance of meticulous ICM evaluation for enhancing embryo selection strategies and reducing the time to pregnancy. Static images captured during blastocyst assessment vary in resolution and focus. As shown in Figure 1 a-f, the visibility of the ICM in static blastocyst images can vary significantly depending on its orientation and the focal plane. Several studies have reported that ICM may often appear out-of-focus or poorly visible in two-dimensional (2D) static images, limiting accurate assessment.[ 21 22 ] This observation is further supported by recent consensus guidelines, which acknowledge that while TE grading is relatively straightforward, ICM assessment can be more challenging due to variability in its shape and position.[ 17 23 ] This is because blastocysts have a spherical structure. ICM is a compact mass of pluripotent cells attached to it’s inner boundary. As the blastocoel expands, the ICM, with its relatively high cell density, may shift towards the lower surface of the culture dish due to gravity. However, its position is not always consistent.[ 27 ] Factors, such as surface adhesion, zona pellucida integrity, ICM shape and blastocoel expansion, can influence its orientation. The spatial positioning of the ICM may influence its visibility and impact morphological assessments, particularly in conventional imaging settings where only a single focal plane is used.[ 23 24 ] Given that the visibility of the ICM is not always optimal in 2D static images, it is essential to investigate whether the clarity of ICM visualization during assessment affects the accuracy of prediction of implantation and pregnancy outcomes. As shown in Figure 1 (i–l), even the same blastocyst can display varying ICM visibility when captured at different rotational orientations, highlighting how image acquisition angles can influence assessment. Static blastocyst images showing the ICM distinctly visible at different positions, at 1 o’clock in (a), 5 o’clock in (b), and 7 o’clock in (c). Static blastocyst images where the ICM is out of focus and not clearly visible. The ICM is positioned at 7 o’clock in (d), 10 o’clock in (e), and 3 o’clock in (f). Static 2D blastocyst images immediately post warming (g), two to three hours post warming (h). Impact of blastocyst orientation on ICM visibility. Static images of the same blastocyst captured at different rotational angles showing a clearly visible ICM positioned at (i) 7 o’clock, (j) Out of focus ICM positioned roughly at 5 o’clock, (k) Out of focus ICM positioned between 8 o’clock and 9 o’clock and with focus on TE cells, and (l) Out of focus ICM positioned between 8 o’clock and 9 o’clock with focus on ICM cells

A total of 300 blastocysts were analysed by two senior embryologists to evaluate the relationship between the ICM visualization status and the accuracy of clinical pregnancy predictions. Of these, 189 blastocysts (63%) exhibited good ICM visibility, whereas 111 (37%) were classified as having poor ICM visibility. The accuracy of clinical pregnancy predictions was assessed separately for cases with good and poor ICM visibility conditions, as summarized in Supplementary Table 2 . Accuracy of clinical pregnancy prediction based on inner cell mass visualization status ICM=Inner cell mass In blastocysts with good ICM visibility, embryologists achieved an overall accuracy of 39.68% (75/189) in predicting subsequent fetal cardiac activity. In cases where ICM was Poor, the accuracy was slightly higher at 44.14% (49/111). The overall prediction accuracy across both conditions was 41.33% (of 124/300 cases). The marginally higher accuracy in poor ICM visibility cases (44.14% vs. 39.68%) was not statistically significant, suggesting that ICM visualization status may not be a decisive factor in accurate clinical pregnancy predictions. A Chi-square test was conducted to determine whether the ICM visualization status influenced the accuracy of negative clinical pregnancy predictions, as detailed in Supplementary Tables 3 and 4 . Contingency table reporting the total number of negative predictions made by each embryologist under two conditions: Good inner cell mass visibility and poor inner cell mass visibility The corresponding clinical outcomes for these predictions are also shown. ICM=Inner cell mass Accuracy of negative clinical pregnancy predictions based on inner cell mass visualization status ICM=Inner cell mass Interpretation of negative predictions: The Chi-square test (χ² = 0.0096, P = 0.9220) found no significant association between ICM visualization status and the accuracy of negative pregnancy predictions, as shown in Supplementary Table 5 Supplementary Table 5 Chi-square test results evaluating the association between inner cell mass visualization status (good vs. poor visibility) and the accuracy of negative pregnancy predictions made by two embryologists Analysis group χ 2 P Interpretation Combined (both embryologists) 0.0096 0.9220 Not significant Embryologist 1 0.0097 0.9214 Not significant Embryologist 2 0.0663 0.7968 Not significant Chi-square test results Accuracy rates were similar for both Good ICM (60.00%) and poor ICM visibility (60.61%), suggesting that embryologists were equally capable of making reliable negative predictions regardless of ICM visibility This finding is clinically relevant, as it implies that limited ICM visibility does not adversely affect embryologists’ ability to make accurate negative assessments. The Chi-square test (χ² = 0.0096, P = 0.9220) found no significant association between ICM visualization status and the accuracy of negative pregnancy predictions, as shown in Supplementary Table 5 Supplementary Table 5 Chi-square test results evaluating the association between inner cell mass visualization status (good vs. poor visibility) and the accuracy of negative pregnancy predictions made by two embryologists Analysis group χ 2 P Interpretation Combined (both embryologists) 0.0096 0.9220 Not significant Embryologist 1 0.0097 0.9214 Not significant Embryologist 2 0.0663 0.7968 Not significant Chi-square test results Chi-square test results evaluating the association between inner cell mass visualization status (good vs. poor visibility) and the accuracy of negative pregnancy predictions made by two embryologists Chi-square test results Accuracy rates were similar for both Good ICM (60.00%) and poor ICM visibility (60.61%), suggesting that embryologists were equally capable of making reliable negative predictions regardless of ICM visibility This finding is clinically relevant, as it implies that limited ICM visibility does not adversely affect embryologists’ ability to make accurate negative assessments. A similar Chi-square test was conducted for positive pregnancy predictions based on ICM visualization status for both embryologists, as shown in Supplementary Tables 6 - 8 . Contingency table reporting the total number of positive pregnancy predictions made by each embryologist under two inner cell mass visualization conditions: Good and poor visibility ICM=Inner cell mass Accuracy rates of positive clinical pregnancy predictions under two inner cell mass visualization conditions: Good visibility ( n =208) and poor visibility ( n =123), along with the corresponding Chi-square value and P -value for each group ICM=Inner cell mass Chi-square values and P -values assessing the relationship between inner cell mass visualization status (good vs. poor visibility) and the accuracy of positive pregnancy prediction, presented for combined and individual embryologist assessments The accuracy of positive predictions was slightly higher in cases where ICM was Poor (36.59%) compared to those with good ICM visibility (26.92%) as calculated in Supplementary Table 7 . However, this difference was not statistically significant (χ² = 3.40; P = 0.0652). Similar trends were observed when analyzed separately for Embryologist 1 (χ² = 1.21, P = 0.2710) and Embryologist 2 (χ² = 2.23, P = 0.1354). These findings challenge the conventional assumption that better ICM visualization enhances predictive accuracy, suggesting that embryologists may be using compensatory evaluation strategies when ICM is obscured.

ICM remains an important marker in blastocyst evaluation, though its visibility in static images is often compromised by embryo orientation, focal depth or imaging limitations. Our findings suggest that prediction accuracy may not be significantly affected by the clarity of ICM visibility, even when the ICM appears vague or difficult to grade. These preliminary results indicate the potential value of emphasizing other morphological and morphometric parameters when ICM visibility is suboptimal. Further validation in larger and diverse datasets is essential to confirm these observations. The use of automated AI tools may aid in supporting this approach by enabling rapid, objective, and consistent embryo assessment. DB was responsible for conceptualisation, methodology design, investigation, data curation, formal analysis, original draft preparation, and manuscript editing. SKV provided overall supervision and performed critical review of the manuscript. HD contributed to clinical data capture, image sorting, data curation, and compiling data into structured tables. GST assisted in collecting clinical metadata and first review of the manuscript. There are no conflicts of interest. The data analysed in this study consisted of anonymized blastocyst images with known implantation outcomes. The dataset was not publicly available because of institutional policies and confidentiality agreements. However, de-identified data may be shared upon reasonable request to the corresponding author, subject to ethical and regulatory approval.

This study examined whether the visibility of the ICM in static two-dimensional blastocyst images had any impact on the implantation prediction scores provided by embryologists. ICM is widely recognized as a key parameter in evaluating blastocyst viability, and numerous studies have highlighted its predictive value for implantation and live birth outcomes.[ 18 19 ] Therefore, it is assumed that better visibility of the ICM would enhance embryo assessment and improve the accuracy of predicting clinical pregnancy. Interestingly, in our study, blastocysts with poor ICM visibility showed slightly higher overall prediction accuracy (47.30%) than those with clearly visible ICM (41.80%). However, this difference was not statistically significant, indicating that ICM visibility alone may not be a strong determinant of the predictive accuracy. It is important to note that our visibility classification was based on internally defined descriptors, rather than an externally validated scoring system. While this approach was consistent within our lab, the absence of standardized criteria for assessing ICM clarity in static images may limit reproducibility and comparability across different settings. Several studies have acknowledged the practical challenge of assessing the ICM in static 2D images. The updated Istanbul Consensus (2025) notes that evaluating the ICM is often more difficult than assessing the TE.[ 17 ] Similarly, Flores Saiffe Farias et al . (2023) observed that the ICM often lacks clear visual separation from surrounding structures in standard micrographs, especially when it is not in the same focal plane as the TE.[ 23 ] Also, significant proportion of blastocysts are not present in an optimal orientation for assessment under standard culture conditions.[ 23 25 ] In our study, 111 of 300 embryos (37%) were categorized as having poor ICM visibility, likely due to asymmetric expansion, differential blastocoel fluid distribution, gravitational settling of ICM cells, or structural variations in ICM-TE development. The density of ICM cells may cause them to settle toward the bottom of the blastocoel owing to gravity, making visualization challenging.[ 26 27 ] This variability in orientation and cellular distribution significantly impacts the ease of ICM evaluation during static image evaluation, as illustrated in Figure 1 g and h . These findings support the need to examine the predictive relevance of ICM in clinical decision making, particularly in settings where static imaging remains the primary method for embryo selection. In clinical practice, embryologists may attempt to adjust embryo orientation using gentle pipetting techniques to improve ICM visibility.[ 22 ] However, our findings suggest that this approach may not always be necessary. In our study, blastocysts with poor ICM visibility showed slightly higher overall prediction accuracy (47.30%) compared to those with clearly visible ICM (41.80%). A similar trend was seen in positive predictions, where accuracy was higher in the poor visibility group (36.59%) than in the good visibility group (26.92%). Negative prediction accuracy remained comparable between both groups (60.61% vs. 60.00%), indicating that ICM visibility alone may not substantially influence an embryologist’s ability to rule out non-viable embryos. Furthermore, repeated handling of embryos carries the risk of introducing mechanical stress. While skilled embryologists may attempt to improve ICM visibility through careful repositioning by manual pipetting, repeated handling may still introduce mechanical stress. Such manipulation can exert shear forces on the blastocyst, and direct contact with the trophectoderm layer may inadvertently damage cells critical for implantation.[ 28 29 30 ] The extent of this risk depends on the technique and experience of the embryologist. Therefore, unnecessary repositioning may be avoided when the benefit of improved ICM visibility is uncertain. The ICM has traditionally been emphasized as an important marker in blastocyst evaluation. However, emerging evidence suggests that TE morphology and the stage of blastocyst expansion may serve as stronger indicators of implantation potential.[ 8 17 31 32 33 34 ] A well structured and tightly packed TE layer with a uniform monolayer of polygonal cells is preferred over irregular or loosely arranged cells, as it is associated with higher implantation potential and better developmental competence.[ 8 17 23 33 34 35 36 ] The structural features of the TE layer such as cell junction integrity, evenness in thickness, and the presence or absence of extruded apoptotic cells offer additional insight into embryo viability. Strong intercellular connections within the TE support blastocyst expansion and implantation,[ 35 ] while uneven TE thickness and apoptotic extrusions between the TE and zona pellucida may reflect abnormal cellular behaviour.[ 36 37 ] These parameters serve as valuable complements to ICM assessment, especially in cases where ICM visibility is limited. Building on these observations, artificial intelligence (AI) has emerged as a valuable tool to support and enhance embryo assessment, particularly in scenarios where traditional evaluation is limited by focal plane, image clarity, or observer variability. While embryologists assess a limited number of embryos during routine practice, AI models can be trained on thousands of annotated images derived from time lapse videos, captured at multiple focal depths and linked to known implantation outcomes.[ 27 29 ] This expansive and multidimensional dataset allows AI to detect and interpret subtle morphological features that may be missed in single static images. By integrating multi depth information from Z stack images, AI Tools have a potential to offer more comprehensive and objective evaluation of blastocyst morphology. Although the Gardner grading system remains central to embryo evaluation, it does not account for all morphological features such as the presence of degenerated or extruded cells in the TE or ICM.[ 17 37 38 ] These may reflect underlying chromosomal abnormalities, but current scoring lacks an objective method to quantify such cellular degeneration.[ 37 ] AI can fill this gap by automatically identifying, quantifying, and correlating the presence or absence of degenerative cells with clinical outcomes. This could provide a more refined assessment of blastocyst viability and offer data-driven insights that go beyond the conventional and widely accepted morphological grading. Features such as cytoplasmic projections in TE cells offer insights into cellular adhesion and migration, which are critical for implantation success.[ 39 ] The ability to precisely quantify blastocyst size, has shown a direct correlation with clinical pregnancy rates, with every micrometre increase in width improving implantation odds.[ 40 ] AI tools can also assess the structural presentation of the zona pellucida, including fibrillar density and organization.[ 41 42 43 ] Although detailed fibrillar structures are not directly observable under standard light microscopy, overall zona morphology can still provide valuable indicators of embryo quality. A densely packed zona, appearing more opaque, has been linked to better developmental potential, whereas a less dense or more transparent zona may correlate with reduced viability.[ 41 44 ] AI models trained with enhanced ICM and TE images show a marked improvement in identifying crucial morphological cues linked to implantation, achieving higher predictive accuracy for clinical pregnancy.[ 25 ] These advanced morphological and morphometric assessment go beyond traditional grading, enabling the development of more precise prediction models that integrate multiple objective parameters beyond ICM, TE and expansion. Rather than replacing the classical morphological grading, AI tools offer a valuable opportunity to augment and refine embryo evaluation by incorporating additional objective, quantifiable features that contribute to a more holistic prediction of embryo viability. This study had certain limitations. The data was collected from a single clinic, which may limit its applicability to various laboratory conditions and patient populations. Additionally, the relatively small dataset and limited number of embryologists involved in the predictions may have introduced bias. Also, embryo evaluations were independently performed by two experienced embryologists following a standardized visual protocol, interobserver concordance was not statistically measured, as the primary focus was on clinical outcome prediction rather than interobserver variability. Another limitation is the absence of a widely accepted reference for categorizing ICM visibility. The classification used in this study was based on predefined internal descriptors and requires further validation. The next step is to expand the scope of this study across multiple IVF clinics, incorporating diverse imaging systems and involving a broader group of embryologists to evaluate whether prediction accuracy is consistent across settings. There is also a need to develop a large, multi-center database of annotated embryo images capturing a wide range of morphological and morphometric parameters, all linked to known implantation outcomes. Making such a dataset publicly available would be valuable not only for training embryologists but also for the benchmarking and validation of AI-based assessment tools. A standardized, open-access resource would support reproducibility, accelerate model development, and promote responsible AI integration in routine IVF practice.

This retrospective cohort study included 300 (SVBT) cycles conducted between 2021 and 2024 at a single fertility center. The sample comprised all eligible SVBTs during this period that met the inclusion criteria: Day 5 blastocysts graded 3BB or higher at 116 ± 2 hpi and showing complete re-expansion within 2 to 3 hours after warming. These criteria were applied to maintain uniformity in embryo developmental stage and imaging conditions. All embryos were cultured in single step media (Vitromed One Step) under identical laboratory conditions. Embryo grading was performed at 116 ± 2 hpi using the Gardner system. Only embryos graded 3BB or above were selected for vitrification. Grading was performed by two senior embryologists by mutual consensus and used solely to determine vitrification eligibility. Vitrification was carried out between 117 ± 2 hpi using the Cryotech vitrification kit, with each blastocyst frozen on an individual cryo-device and stored in liquid nitrogen according to standard protocols. Frozen embryo transfers were performed in either natural or programmed cycles, depending on the patient’s clinical profile. In natural cycles, ovulation was monitored using transvaginal ultrasound, and progesterone supplementation (Crinone 8%) was initiated following confirmed ovulation. In programmed cycles, estradiol (2 mg, three times daily) was used to promote endometrial proliferation. Once the endometrium was deemed receptive, progesterone was initiated using either vaginal Crinone 8% or oral Duphaston 10 mg, administered thrice daily. Embryo transfer was scheduled on the sixth day of progesterone administration to align with the expected window of implantation. All embryos were transferred 2 to 3 hours post warming using a Cook 7019 catheter. Embryos were warmed using Kitazato thawing media per manufacturer instructions. Post warming, they were cultured in the same single step media and monitored for 2 to 3 hours. Only those embryos that achieved full re expansion within this time frame were included in the study. Embryos that did not re expand fully by 3 hours post warming were excluded [ Figure 1 g and h]. Fully re expanded embryos were imaged approximately 2 to 3 hours after warming, prior to transfer. Images were captured using an Olympus IX73 inverted microscope with a 20x DIC objective and a 2MP CMOS camera system, processed through TUSCAN 2.0 software. Images were stored at native 2MP resolution on a secure Synology NAS server. Cycles involving severe male factor infertility, uterine anomalies, advanced endometriosis, or adenomyosis were excluded. Patients with technically difficult embryo transfers, such as those with poor cervical visualization or vaginal canal obstruction, were also excluded. In addition, any embryo that did not achieve full re expansion within 3 hours after warming was excluded from the final analysis. Two senior embryologists, blinded to all clinical data (including patient age, BMI, and stimulation protocol), independently assessed each post warming image. Each provided two binary evaluations: ICM Visibility – classified as ‘Good’ or ‘Poor’ based on internal descriptors. An ICM was considered ‘Good’ if well demarcated, compact and in the same focal plane as the TE. It was labelled ‘Poor’ if diffuse, undefined or poorly oriented (e.g., facing downward or appearing as shadow). These criteria were informed by internal consensus and supported by recent literature acknowledging the challenge of ICM assessment in static images. Representative examples are shown in Figure 1 a-f Implantation Prediction – classified as ‘Positive’ or ‘Negative’, indicating the embryologist’s belief regarding likelihood of clinical pregnancy. ICM Visibility – classified as ‘Good’ or ‘Poor’ based on internal descriptors. An ICM was considered ‘Good’ if well demarcated, compact and in the same focal plane as the TE. It was labelled ‘Poor’ if diffuse, undefined or poorly oriented (e.g., facing downward or appearing as shadow). These criteria were informed by internal consensus and supported by recent literature acknowledging the challenge of ICM assessment in static images. Representative examples are shown in Figure 1 a-f Implantation Prediction – classified as ‘Positive’ or ‘Negative’, indicating the embryologist’s belief regarding likelihood of clinical pregnancy. Discrepancies in visibility classification were rare, as the criteria used for classification were highly objective and left limited room for interpretation. Clinical pregnancy, defined as the presence of fetal cardiac activity at 7 to 8 weeks gestation, was used as the ground truth for outcome prediction. Inputs were recorded in an Excel sheet and linked to a unique blastocyst ID. The actual clinical pregnancy outcome was entered as Y (yes) or N (no). The complete dataset of 300 blastocyst images is summarised in Supplementary Table 1 . Dataset of 300 blastocysts with unique identifiers, binary implantation predictions by two embryologists (positive/negative), inner cell mass visibility (good/poor), and cardiac activity outcome (yes/no) ICM=Inner cell mass This retrospective study analysed fully anonymized digital images of blastocysts with known implantation outcomes. All images were de-identified prior to analysis using a randomized alphanumeric coding system, with no associated patient identifiers, metadata, demographic information, or clinical records, thereby eliminating any risk of direct or indirect patient identification. A formal waiver of ethical review and informed consent was obtained from the Institutional Ethics Committee, acknowledging the use of fully anonymized data and confirming that no human subjects were involved. A copy of the waiver certificate is available as supplementary material. The study was conducted as per the Helsinki Declaration. Standard statistical tools were used to analyse the correlation between ICM visibility, blastocyst grading, and predictive accuracy of embryologists for implantation outcomes. IBM SPSS Statistics Version 29 was used for data analysis.