Artificial intelligence-powered oocyte evaluation: Correlating cytoplasmic features with blastocyst development
Article information
Abstract
Objective
This study aimed to establish a quantitative and interpretable method for assessing oocyte quality by analyzing cytoplasmic morphology and intensity features using artificial intelligence.
Methods
A total of 695 oocyte images were collected from hormonally stimulated young and aged mice. The cytoplasmic region was manually annotated to exclude polar bodies, and radiomics analysis was performed to extract morphological and intensity-based features.
Results
Clustering with a Gaussian mixture model identified three distinct oocyte subtypes with unique cytoplasmic characteristics. Cluster 2, with the most spherical and compact oocytes, demonstrated the highest blastocyst formation rate (42.9%), followed by clusters 3 (35.3%) and 1 (20.4%). Cluster 2 oocytes also showed the highest mean intensity and lowest variability, suggesting uniform cytoplasmic structure. Notably, some aged oocytes in cluster 2 exhibited developmental potential comparable to that of young mice, indicating that cytoplasmic quality may be a more informative predictor than age alone.
Conclusion
These findings underscore the value of cytoplasmic features as objective indicators of developmental competence. This artificial intelligence-driven approach may improve embryo selection by providing a standardized, non-invasive method for evaluating oocytes, ultimately contributing to enhanced clinical outcomes in assisted reproductive technologies.
Introduction
Oocyte quality is a critical determinant of success in assisted reproductive technologies (ART), including in vitro fertilization (IVF), embryo development, and implantation. Current approaches to evaluating oocyte quality rely on morphological characteristics observed under a microscope, such as cytoplasmic texture, zona pellucida thickness, polar body appearance, and overall shape and size [1,2]. Although these morphological markers are widely used to assess maturity and developmental potential, they are limited by inter-embryologist variability, absence of standardized grading systems, and poor reproducibility [3,4]. To address these limitations, alternative indicators—such as meiotic spindle integrity, mitochondrial activity, cumulus-oocyte complex (COC) density, maternal RNA and protein content, apoptosis in surrounding cells, and metabolic enzyme activity—have been investigated to better characterize oocyte competence [5-7]. However, the clinical application of these markers is constrained by long processing times and the need for specialized techniques. Establishing reliable and standardized criteria for oocyte quality assessment is therefore essential, given the profound influence of oocyte quality on pregnancy outcomes.
Recent advances in imaging technology and artificial intelligence (AI) have enabled more objective and non-invasive assessments of oocyte quality [8,9]. Morphological features such as cytoplasmic granularity, zona pellucida integrity, and polar body structure remain central to evaluation, and several AI-based strategies have targeted specific oocyte regions. Targosz et al. [10] used deep neural networks to segment human oocytes into cytoplasm, polar bodies, and the perivitelline space, laying the groundwork for quantitative morphology analysis, although clinical outcomes remain to be validated. Barucic et al. [11] classified oocytes into viable and non-viable groups based on segmented components, but the study was limited by a small sample size and subjectively defined viability. Fjeldstad et al. [4] developed machine learning algorithms that predicted blastocyst development directly from oocyte images, without segmentation. Despite these advances, concerns remain about the black-box nature of AI and its lack of interpretability. More recently, AI approaches have been applied in ART for donor oocyte quality assessment, demonstrating improved prediction of blastocyst formation and suggesting potential for treatment personalization [12]. Likewise, AI-driven analyzes of oocytes during IVF have introduced new perspectives on objective morphology evaluation [13]. A recent systematic review further synthesized evidence on AI applications in ART, reinforcing its expanding role in improving reproductive outcomes [14]. Collectively, these findings highlight the potential of AI to establish standardized and interpretable methods for objective oocyte assessment.
AI-driven clustering techniques support biomedical research by enabling the systematic classification and analysis of complex biological data. In medical imaging, AI has been used to analyze heterogeneous healthcare data and aid in disease diagnosis [15]. Radiomics, a quantitative imaging approach, extracts features related to intensity, shape, and texture from biological or medical images. These features capture subtle patterns that may not be apparent to human observers and can be leveraged to identify phenotypic differences or predict clinical outcomes. In oncology, radiomics has been applied to extract imaging biomarkers to assess tumor heterogeneity, contributing to outcome prediction and personalized treatment strategies [16]. When clustering is performed on radiomics-derived features, the resulting groups provide quantitatively validated diagnostic insights [17,18]. Such outcomes are instrumental in advancing personalized medicine by offering objective and reproducible assessments of disease characteristics, thereby improving both precision and diagnostic accuracy.
A key component of oocyte quality assessment is cytoplasmic evaluation, as cytoplasmic characteristics directly influence early embryonic development. Previous studies have linked features such as granularity, coloration, and organelle clustering to oocyte competence and developmental potential [19-21]. However, quantifying cytoplasmic properties remains challenging due to subjectivity and the absence of standardized evaluation tools. Standardized AI-based approaches could address these limitations by providing reproducible and consistent metrics for assessing cytoplasmic quality.
To the best of our knowledge, this is the first study to associate cytoplasmic characteristics with blastocyst development using AI. We performed a comparative analysis of blastocyst development rates by cluster, based on quantitative features extracted from mouse oocyte cytoplasm. The primary aim was to establish an objective and quantitative method for assessing oocytes by classifying them according to cytoplasmic radiomic features correlated with blastocyst formation using AI-driven analyzes.
Methods
1. Ethics approval
All animal procedures were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. This study was approved by the Animal Research Ethics Committee of the Maria Fertility Research Institute (approval no. 2023-003) and was carried out in compliance with the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines and all applicable national regulations.
2. Oocyte collection and image preparation
B6D2F1 (C57BL/6×DBA/2) female mice were purchased from Koatech. Young mice were obtained at 4 weeks of age and acclimated under controlled conditions for 2–4 weeks, then used at 6–8 weeks of age. Aged mice were also acquired at 4 weeks of age and maintained in the animal facility of the Maria Fertility Research Institute until they reached 66–70 weeks of age. The young group received 5 IU of gonadotropin, while the aged group received 10 IU. Each female was administered an intraperitoneal injection of pregnant mare serum gonadotropin (Prospec), followed 48 hours later by human chorionic gonadotropin (hCG; Sigma-Aldrich). In total, 20 young mice and 75 aged mice were used, yielding an average of 20 oocytes per young mouse and four oocytes per aged mouse. Oocyte quality is strongly influenced by maternal age: younger oocytes typically show higher developmental competence, whereas aged oocytes are prone to cytoplasmic structural abnormalities, mitochondrial dysfunction, and increased oxidative stress. To assess the impact of age on oocyte morphology and cytoplasmic features, both young and aged mice were included in this study.
COCs were collected from the oviducts at 14 hours after hCG injection into prewarmed MRC#OS medium (Maria Medical Foundation), as previously described [22]. Cumulus cells were removed (denudation) by treating the oocytes with hyaluronidase (300 IU/mL) in MRC#OS medium for 3 to 5 minutes, followed by gentle pipetting to fully disperse surrounding cells. A total of 695 oocyte images were obtained at a single research center between January 2023 and September 2024. One day before the experiment, microdroplet dishes were prepared with 20-µL droplets of MRC#OS medium covered with mineral oil. Oocytes were transferred individually into droplets (one oocyte per droplet). Images were then captured using a Nikon microscope (Nikon).
3. IVF and culture
After imaging, IVF was performed within the same droplets. Sperm were collected from the caudal epididymis of 8-week-old B6D2F1 male mice. Five hours after insemination, zygotes were thoroughly washed in medium and transferred into 10-µL droplets of MRC#OS medium, with one zygote per droplet. The embryos were incubated at 37 °C in a 6.0% CO₂ atmosphere. Embryo morphology and development were monitored daily until day 4 after oocyte collection, corresponding to the blastocyst stage.
4. Image segmentation and quantification
Radiomics is a widely used approach for extracting quantitative features from a region of interest (ROI) in medical imaging [23,24]. In this study, the ROI was defined as the cytoplasmic region, excluding the polar body. Cytoplasmic regions were annotated and reviewed independently by three experienced embryologists. To ensure consistency, all images were rotated among the embryologists for cross-checking. Segmentation was then performed on all 695 oocyte images based on these ROIs, with polar bodies excluded (Figure 1). Using the ROI obtained through this process, a binary mask was generated and applied to the radiomics pipeline alongside the original image. From these, we calculated eight morphological features and 16 intensity features, which were quantified using the Python Pyradiomics package [25]. The definitions of all analyzed factors are summarized in Supplementary Tables 1 and 2.
Examples of cytoplasmic segmentation in mouse oocytes and representative images of each cluster (captured at ×200 magnification). (A) Annotated image of an oocyte showing the cytoplasmic region defined as the ROI (excluding the polar body). (B) Representative image of cluster 1, characterized by the largest cytoplasmic area with an elongated shape. (C) Representative image of cluster 2, displaying the most compact morphology with the highest sphericity. (D) Representative image of cluster 3, exhibiting irregular boundaries and heterogeneous cytoplasmic contours. These morphological differences formed the basis for clustering, highlighting distinct patterns of cytoplasmic organization across the three groups. Although representative images are shown for each cluster, the visual distinctions are often subtle and not easily recognized by the human eye; quantitative radiomic features were necessary to capture these differences.
5. Gaussian mixture model clustering
The Gaussian mixture model (GMM) is a widely used clustering algorithm due to its ‘soft’ clustering properties and statistical framework, which enable the determination of cluster number [26]. We used the GMM to analyze cluster characteristics, assuming each cluster followed a Gaussian distribution [27]. Clustering was performed on radiomic features extracted from cytoplasmic regions, with each oocyte assigned to the cluster with the highest posterior probability. This approach provided probabilistic classification rather than rigid partitions, allowing identification of oocyte subgroups with distinct morphological and intensity-based characteristics without predefined labels. Such an exploratory method enabled unbiased analysis of heterogeneity within the dataset.
6. Statistical analyzes
The chi-square and Student t-test were used to assess differences between clusters. All analyzes were conducted using Python 3.8.0. The optimal number of GMM clusters was determined by minimizing the Akaike information criterion (AIC) and Bayesian information criterion (BIC) [28]. The overall analysis workflow is summarized in Figure 2.
Overall workflow for the statistical analysis.
Results
1. Clustering results using GMM
To determine the optimal number of clusters, we applied both the AIC and BIC. The minimum values for both criteria indicated that three clusters were optimal. Using morphological and intensity features, 695 oocytes were classified into three clusters by the GMM approach: cluster 1 (54 oocytes), cluster 2 (324 oocytes), and cluster 3 (317 oocytes). To visualize the distribution of the clusters, we used principal component analysis (PCA) to reduce the features into three dimensions [29]. The resulting three-dimensional plot showed that the clusters formed distinct groups, each separated on the basis of cytoplasmic morphology and intensity features. Oocytes within a given cluster exhibited similar feature distributions, whereas those from different clusters were clearly separated (Figure 3).
Three-dimensional principal component analysis (PCA) plot illustrating Gaussian mixture model clustering of oocytes. Each cluster formed a distinct group based on cytoplasmic morphology and intensity features. Oocytes within the same cluster exhibited similar feature distributions, whereas those in different clusters demonstrated significant variation. 3D, three-dimensional.
2. Comparison of fertilization and blastocyst development rates across clusters
We next evaluated the relationship between cluster classification and fertilization or blastocyst development using the chi-squared test. Cluster 2 exhibited the highest fertilization rate (88.8%), followed by clusters 3 and 1, a pattern similar to that observed for blastocyst development. However, fertilization rates did not differ significantly among clusters.
In contrast, blastocyst development rates differed significantly among all pairwise comparisons (cluster 1 vs. cluster 2, cluster 2 vs. cluster 3, and cluster 1 vs. cluster 3) at the 5% significance level (p<0.05). Cluster 2 demonstrated the highest blastocyst development rate (42.9%), followed by cluster 3 (35.3%) and cluster 1 (20.4%) (Table 1).
Chi-square test results for blastocyst development and fertilization rates by cluster
3. Morphological differences in radiomics shape-based features
We compared shape-based radiomic features across clusters, with definitions provided in Supplementary Table 1. These features describe the two-dimensional morphological properties of the oocyte cytoplasmic region (ROI) (Table 2, Figure 1).
T-test results for morphological shape-based features by cluster
Elongation increased significantly across clusters (p<0.001) in the following order: cluster 1 (0.9389) < cluster 2 (0.9539) < cluster 3 (0.9607). Sphericity also differed significantly (p<0.001), with the highest value in cluster 2 (0.9389), followed by cluster 1 (0.8348) and cluster 3 (0.7858).
Size-related parameters—including major and minor axis lengths and maximum diameter—varied significantly among clusters (p<0.001). The major axis length was significantly greater in cluster 1 (699.14) compared with cluster 2 (664.82, p<0.001) and cluster 3 (661.81, p<0.001). A similar pattern was observed for maximum diameter (cluster 1: 740.72 > cluster 2: 674.38 = cluster 3: 676.29, p<0.001) and minor axis length (cluster 1: 656.10 > cluster 2: 633.80 = cluster 3: 635.56, p<0.001). No significant differences were found between clusters 2 and 3 in maximum diameter (p=0.5089) or minor axis length (p=0.1376).
Perimeter values also differed significantly (p<0.001), with cluster 2 showing the smallest perimeter (2,170.37). Clusters 1 (2,561.44) and 3 (2,590.97) were similar (p=0.0514). The perimeter-to-surface ratio followed the order: cluster 2 (0.0066) < cluster 1 (0.0072) < cluster 3 (0.0079) (p<0.001). Pixel surface area was significantly greater in cluster 1 (358,821.33) than in cluster 2 (330,560.01) and cluster 3 (329,868.75) (p<0.001), with no difference between clusters 2 and 3 (p=0.5045).
4. Comparison of intensity features across clusters in radiomics
Intensity-based radiomic features also varied significantly among clusters (Table 3, Supplementary Table 2). Cluster 2 exhibited the highest overall intensity, as reflected by significantly higher mean (149.87), median (152.01), and root mean squared (151.58) values compared with cluster 1 (146.81, 148.81, and 148.84, respectively) and cluster 3 (146.58, 148.57, and 148.22, respectively) (p<0.001). Minimum intensity was significantly greater in clusters 2 (49.83) and 3 (49.49) than in cluster 1 (47.43, p=0.0114). By contrast, maximum intensity was lower in cluster 3 (243.03) compared with cluster 1 (246.24) and cluster 2 (246.18) (p<0.001).
T-test results for intensity features by cluster
Variability measures also revealed clear differences. Cluster 1 showed the greatest variability, with significantly higher variance (601.25), total energy (7.95×10⁹), and interquartile range (29.80) compared with cluster 2 (514.30, 7.60×10⁹, and 26.56, respectively) and cluster 3 (484.76, 7.25×10⁹, and 25.90, respectively) (p<0.001).
Evaluation of intensity distribution demonstrated that cluster 3 had the most uniform and stable pattern, reflected by the highest uniformity (0.3284) and lowest entropy (1.8921) (p<0.001).
5. Cluster-based distribution and blastocyst development of young and aged oocytes
To further examine the effects of cluster-specific blastocyst formation, subgroup analyzes were performed within each cluster (Tables 4 and 5). The chi-square test was first conducted to evaluate the distribution of young (6–8 weeks) and aged (66–70 weeks) oocytes across the three clusters and to determine whether the distributions differed significantly between age groups (Table 4). The proportion of oocytes in cluster 1 differed markedly by age: only 3.8% of young oocytes were classified into this cluster, compared with 13.0% of aged oocytes (p<0.001). In cluster 2, 45.5% of young oocytes and 48.2% of aged oocytes were represented, indicating a balanced age distribution. By contrast, 50.8% of young oocytes were classified into cluster 3, significantly higher than the 38.8% of aged oocytes in this cluster (p=0.002).
Comparison of cluster distributions between oocytes from young and aged mice
Comparison of blastocyst development rates of oocytes from young and aged mice
We next compared blastocyst formation rates stratified by maternal age within each cluster (Table 5). In cluster 2, the blastocyst formation rate was 47.2% for young oocytes and 37.5% for aged oocytes, a difference that was not statistically significant (p=0.079). In clusters 1 and 3, however, blastocyst rates were significantly higher in young oocytes (40.0% and 41.8%, respectively) than in aged oocytes (12.8% and 24.1%, respectively). Importantly, aged oocytes in cluster 2 exhibited morphological and intensity characteristics similar to those of young oocytes, yielding comparable developmental outcomes. Despite these age-related differences, the overall developmental trends across clusters were consistent, following the order cluster 2 > cluster 3 > cluster 1 for both young and aged groups. Specifically, blastocyst rates were 47.2% (cluster 2), 41.8% (cluster 3), and 40.0% (cluster 1) among young oocytes, and 37.5%, 24.1%, and 12.8%, respectively, among aged oocytes.
Discussion
This study is the first to demonstrate the relationship between cytoplasmic morphological and intensity features of oocytes and blastocyst development, providing a novel framework for assessing oocyte quality. These findings may help identify cytoplasmic characteristics that can guide individualized treatment strategies to improve clinical outcomes.
This study demonstrates that, in addition to conventional morphological assessment [30-35], cytoplasmic intensity features are essential for predicting blastocyst formation, potentially enhancing the current methodologies used in ART. Incorporating radiomics analysis into ART offers a quantitative and objective method for evaluating oocyte quality, complementing traditional morphology-based evaluations and enhancing predictive accuracy [36-38].
The high variability in intensity observed in cluster 1 oocytes may reflect cytoskeletal dysfunction, which disrupts organelle distribution and reduces developmental potential [39]. Prior studies indicate that uniform cytoplasmic structure is crucial for proper organelle positioning, whereas irregularities in granularity and cytoplasmic texture suggest impaired maturation and reduced embryonic viability [40-42]. In line with this, increased cytoplasmic granularity, uneven brightness, and organelle clustering have been associated with reduced oocyte quality and developmental competence [39,43,44]. Previous ultrastructural studies identified refractile bodies—electron-dense inclusions containing fibrous material, granular vesicles, and lipid deposits—as contributors to increased cytoplasmic granularity and uneven brightness [39]. These structures are linked to oxidative stress, cellular aging, and cytoskeletal dysfunction, which lead to abnormal clustering of organelles, particularly mitochondria and smooth endoplasmic reticulum. Such clustering may impair metabolic activity and reduce developmental potential. Cytoplasmic abnormalities of this type have also been associated with mitochondrial dysfunction and elevated reactive oxygen species, which disrupt adenosine triphosphate production and meiotic spindle organization [45], ultimately compromising oocyte quality and early embryonic development [46]. Reviews emphasize the critical role of mitochondrial processes in oocyte maturation and embryo competence [47] and highlight how oxidative imbalance accelerates reproductive aging [48]. Together, mitochondrial and oxidative stress pathways compromise the structural and functional integrity of the cytoplasm, destabilize the cytoskeleton, and undermine fertilization potential.
In contrast, cluster 2 contained the most spherical and compact oocytes, features that may promote intracellular stability and efficient metabolic activity. The distinct separation of cluster 2 in PCA visualization suggests that this group represents optimal cytoplasmic and morphological characteristics, consistent with superior developmental potential.
The structural irregularities distinguishing clusters 2 and 3 may indicate mechanical or cytoskeletal instability that affects fertilization and cleavage. These findings suggest that cytoplasmic uniformity alone is insufficient for optimal embryonic development if structural abnormalities are present. Cluster 2 demonstrated superior developmental potential compared with cluster 3, underscoring the combined importance of morphological symmetry, high brightness, and low-intensity variability as indicators of optimal oocyte quality in ART.
Our subgroup analysis further revealed that blastocyst formation rates in cluster 2 did not differ significantly between young and aged oocytes, suggesting that some aged oocytes retain cytoplasmic and morphological properties similar to younger oocytes. This supports the concept of heterogeneity in oocyte aging, where not all aged oocytes experience equivalent functional decline. Conversely, cluster 1 oocytes from older mice exhibited the lowest blastocyst rates, reinforcing the link between cytoplasmic instability, structural abnormalities, and developmental failure. In cluster 3, aged oocytes also showed lower blastocyst rates than young oocytes, indicating that although the general characteristics of this cluster were favorable, structural instability had a disproportionately negative impact on aged oocytes, similar to cluster 1.
The significant overrepresentation of aged oocytes in cluster 1 suggests that aging is strongly associated with morphological irregularities and cytoplasmic heterogeneity. This trend was also evident in the ratios of aged to young oocytes across clusters. Because cluster 1 was predominantly composed of aged oocytes, these findings further reinforce the impact of maternal age on oocyte morphology and cytoplasmic characteristics. Nevertheless, the presence of some young oocytes in cluster 1 indicates that age alone does not determine developmental competence; instead, a combination of structural and cytoplasmic factors plays a critical role. Moreover, aged oocytes consistently showed reduced developmental potential, even when their morphology appeared similar to that of younger oocytes. This is consistent with prior studies reporting diminished developmental potential in aged oocytes [49]. Together, these observations suggest that oocyte quality should be assessed not solely by age, but through an integrated evaluation of morphological and cytoplasmic parameters.
Although our findings underscore the importance of cytoplasmic features in predicting oocyte competence, prior studies have primarily relied on age and conventional morphological markers, often neglecting pre-fertilization cytoplasmic properties. Time-lapse imaging and machine learning have been applied to distinguish embryos derived from young versus aged oocytes; however, these efforts have focused on post-fertilization morphokinetics rather than pre-fertilization cytoplasmic characteristics [50]. Our study addresses this gap by directly analyzing cytoplasmic morphology and intensity features, thereby providing a quantitative and predictive framework for oocyte quality assessment. Previous research on oocyte maturation kinetics found no significant differences between young and aged oocytes [51]. In contrast, the present findings demonstrate that clustering based on cytoplasmic morphology and intensity correlates strongly with blastocyst formation rates, offering greater predictive value than age alone. This highlights the central role of cytoplasmic and morphological features in determining oocyte developmental potential. While maternal age remains an established factor in the deterioration of oocyte quality, some aged oocytes retain high competence if they possess favorable cytoplasmic characteristics. Identifying such oocytes in advance is crucial, as interventions for enhancing blastocyst formation may differ according to developmental capacity.
This study introduces a novel framework for identifying oocytes with high developmental potential, which may enhance oocyte selection, improve pregnancy rates, and reduce the emotional and financial burden of infertility treatment. The incorporation of data-driven approaches into subjective evaluations can improve the consistency, reliability, and overall success of ART procedures [3,9]. Furthermore, analyzing the morphological and intensity features of high-quality oocytes provides valuable insight into intrinsic and extrinsic factors influencing oocyte quality, including maternal age, ovarian stimulation protocols, and culture conditions [2].
Compared with other AI-based approaches such as deep neural networks or traditional machine learning models, the radiomics–GMM framework offers distinct advantages. Radiomics generates handcrafted, interpretable features (e.g., shape, intensity, texture) that can be directly linked to biological or clinical attributes. The GMM provides unsupervised clustering without requiring large annotated datasets, making it especially suitable for exploratory studies with limited sample sizes. By contrast, deep learning models automatically learn hierarchical representations from raw images and may achieve superior predictive performance when applied to large, diverse datasets. Traditional supervised machine learning methods, including support vector machines or random forests, can also provide robust classification but depend on predefined labels. Thus, while the radiomics–GMM approach offers interpretability and feasibility in smaller datasets, its predictive power and generalizability may be more limited compared with data-driven deep learning methods. Future research should explore hybrid strategies that integrate radiomics-derived features with advanced AI models, thereby combining interpretability with enhanced predictive performance.
This study has several limitations. First, the experiments were performed using mouse oocytes; thus, the direct applicability of our findings to human oocytes remains to be validated. Whether the cytoplasmic morphology and intensity-based classification system established in mice demonstrates similar correlations with blastocyst development in humans is still unclear. Second, this study was conducted at a single institution under specific experimental conditions. Therefore, multicenter studies and meta-analyzes are required to confirm the reproducibility and generalizability of these results. Future research should incorporate larger, clinically relevant datasets across multiple laboratories and diverse patient populations to further validate this classification approach. Third, there was an imbalance in cluster sizes, particularly the relatively small number of oocytes in cluster 1 compared with clusters 2 and 3. This uneven distribution may affect the robustness of statistical comparisons, and thus, findings related to cluster 1 should be interpreted with caution. In this study, Student’s t-tests were used to compare characteristics among clusters that had already been defined by the GMM. The aim was not to use individual features as criteria for classification, but rather to determine whether predefined clusters exhibited distinct feature distributions. Because only pairwise comparisons were conducted, the number of statistical tests was limited, and analyzes avoided excessive multiple testing. Nonetheless, even a modest number of comparisons increases the risk of type I error, and this potential limitation should be considered when interpreting the results. Fourth, radiomic features are sensitive to image acquisition conditions, including equipment, imaging protocols, and resolution. Because our dataset was obtained from a single center, there remains a potential risk of bias from site-specific imaging parameters. Future studies using multicenter datasets with standardized imaging protocols will be essential to establish reproducibility and ensure generalizability.
In conclusion, this is the first study to quantitatively analyze oocyte cytoplasmic morphology and intensity features and to demonstrate their association with blastocyst formation. Unlike prior studies that primarily emphasized general morphological traits, our approach integrated both structural and intensity-based cytoplasmic parameters. These findings suggest that cytoplasmic features can serve as valuable predictors of developmental potential. Future studies should validate these results in human oocytes and further investigate the biological mechanisms influencing cytoplasmic morphology and intensity, with the ultimate goal of improving IVF success rates.
Notes
Conflict of interest
HJL is a shareholder of Kai Health, but there is no potential conflict of interest relevant to this paper. The others have no conflicts of interest to disclose.
Author contributions
Conceptualization: JL. Methodology: HMK. Formal analysis: HMK. Data curation: JH, HK, KJ. Funding acquisition: HJL. Project administration: HJL, JY. Investigation: JH, HK, KJ, EJ, HP, JJ, JY. Supervision: JL, JY. Writing-original draft: HMK, HJL, JY. Writing-review & editing: HMK, JH, HK, KJ, EJ, HP, JJ, JL, HJL, JY. Approval of final manuscript: HMK, JH, HK, KJ, EJ, HP, JJ, JL, HJL, JY.
Supplementary material
Supplementary material can be found via https://doi.org/10.5653/cerm.2025.08396.
Shape-based features and description
Intensity features and description