Clinical pregnancy prediction method after frozen embryo transplantation based on ultrasonic radiomics

Abstract

The invention discloses a method for predicting clinical pregnancy after frozen embryo transplantation based on ultrasonic radiomics, and relates to the technical field of embryo transplantation prediction.The method comprises the steps that S1, clinical data and ultrasonic images of female participants receiving frozen embryo transplantation are collected to form a data set of the female participants receiving frozen embryo transplantation; s2, data exclusion is carried out to obtain a female participant data set which is included in analysis, and female participant data samples which are included in analysis are divided into a training group and an external test group; s3, endometrial radiomics feature extraction and modeling are carried out; s4, performing radiomics feature extraction and model establishment on the endometrial peripheral area; s5, constructing a combined radiomics model; and S6, comparing the endometrial radiomics model with the endometrial peripheral region radiomics model and the combined radiomics model. According to the method, the AUC is used for evaluating the performance of the three models, and the optimal effect of predicting the pregnancy outcome after the frozen embryo transplantation by the combined radiomics model is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of embryo transfer prediction, and more specifically to the technical field of a clinical pregnancy prediction method after frozen embryo transfer based on ultrasound radiomics. Background Art

[0002] Since the advent of assisted reproductive technology, clinicians and embryologists have been trying to develop methods to improve the pregnancy rate. However, the success rate remains unsatisfactory. Endometrial receptivity and embryo quality are two important factors determining the success of transplantation. As a simple and non-invasive examination method, ultrasound has been widely used in reproductive medicine. Researchers have tried to identify an ultrasound index that can predict the outcome of embryo transfer. It has been reported that indicators such as endometrial thickness (EMT), volume, blood supply, and peristaltic waves affect endometrial receptivity (ER). However, other research evidence has confirmed that these indicators have no direct relationship with the implantation rate. All ultrasound indicators have limitations in evaluating ER. In addition, endometrial receptivity markers evaluated using endometrial biopsy and hysteroscopy are described as "poor in ability" and invasive. Therefore, it is necessary to develop a non-invasive alternative method to evaluate ER.

[0003] Radiomics is an emerging method that can be used to extract a large amount of imaging data from medical images that are not recognizable by the naked eye and has been used in various classification tasks. Under the action of estrogen, the junction zone (JZ) between the endometrium and the myometrium can undergo cyclic changes. Studies have shown that it is closely related to the peristalsis of the endometrium and can provide important information about the pregnancy-related microenvironment. We hypothesized that combining EN and JZ could contain more comprehensive microenvironment information and could obtain better diagnostic ability. It has been reported that the average thickness of JZ is 5 to 8 mm, and the average maximum thickness on magnetic resonance imaging (MRI) is 8 mm in the normal state. The pregnancy rate decreases with the increase in thickness. However, MRI is expensive and time-consuming. In addition, the contour of JZ is neither linear nor uniform. The images are usually not very clear, especially on ultrasound. In a previous study, we applied ultrasound radiomics to predict the pregnancy outcome after frozen embryo transfer (FET), and the region of interest (ROI) of the endometrium in the junction zone (JZ) was obtained by enclosing the hypoechoic band around the visible endometrium on the ultrasound image as fully as possible. However, in the clinical practice of the later model, we found that in the ultrasound images of many participants, the region of interest (ROI) of the endometrium in the junction zone (JZ) often showed poor visibility, which limited the clinical applicability of the prediction model we established. Therefore, we tried to observe the optimal perimetrial endometrial zone (PEZ) closely related to pregnancy as a substitute for the junction zone (JZ).

[0004] It is of practical significance to assume that machine learning (ML) algorithms are used to predict endometrial receptivity (ER) after frozen embryo transfer (FET) using ultrasound images of the endometrium (EN) and the optimal endometrial zone (PEZ). Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting clinical pregnancy after frozen embryo transfer based on ultrasound radiomics in order to solve the above technical problems.

[0006] The present invention specifically adopts the following technical solutions to achieve the above purpose:

[0007] The present invention provides a method for predicting clinical pregnancy after frozen embryo transfer based on ultrasound radiomics, including the following steps:

[0008] S1. Collect the clinical data and ultrasound images of female participants who received frozen embryo transfer to form a dataset of female participants who received frozen embryo transfer, and screen out the eligible dataset of female participants who received frozen embryo transfer from the dataset of female participants who received frozen embryo transfer through data inclusion criteria and primary data exclusion criteria;

[0009] S2. Obtain the dataset of female participants included in the analysis through secondary data exclusion from the eligible dataset of female participants who received frozen embryo transfer. The data samples of the female participants included in the analysis are divided into a training group and an external test group;

[0010] S3. Endometrial radiomics feature extraction: Determine radiomics features from the region of interest of the endometrium;

[0011] Process the radiomics features of the endometrium and the endometrial junction zone as follows: Retain the features with ICCs > 0.75; Perform univariate rank sum tests on the features and discard those features with p-values > 0.05; Perform Spearman correlation analysis on the features and use Spearman correlation analysis with r ≥ 0.6 to eliminate redundancy; Perform logistic analysis on the radiomics features; Optimize the model parameters through ten-fold cross-validation using the least absolute shrinkage and selection operator algorithm, and select the radiomics features with non-zero coefficients from the training group; Use machine learning algorithms including logistic regression, support vector machine, random forest classifier, decision tree classifier, k-nearest neighbor classifier, and backpropagation neural network to model and analyze the combined radiomics features; Use AUC, accuracy, sensitivity, and specificity to evaluate the performance of the endometrial radiomics model in predicting endometrial receptivity; Select the model with the optimal AUC from the training group and the external validation group as the endometrial radiomics model;

[0012] S4. Radiological Feature Extraction and Model Establishment in the Peri - endometrial Region:

[0013] S41. At the image level, first use the B - spline interpolation algorithm to resample the ultrasound images, align images with different spatial resolutions, and ensure that the voxel spacing is standardized to 1×1 mm 2 ;

[0014] S42. Apply gray - level normalization to standardize the intensity levels of gray - scale images from different sources or imaging devices, ensuring consistent brightness and contrast between images. At the feature level, use the z - score to standardize all radiological features in the training group; strictly apply the parameters (i.e., mean and standard deviation) calculated from the training group to standardize the validation set to ensure the standardization consistency of data from different cohorts; automatically expand the boundary of the region of interest of the endometrium outward by 2 mm, 4 mm, 6 mm, and 8 mm along the contour of the endometrium to create annular regions of corresponding sizes around the endometrium;

[0015] S43. Normalize the images and use PyRad iomics to extract features such as the original image, squared image, and wavelet - filtered image, retain features with ICCs > 0.75, use the univariate rank - sum test to analyze the significance of radiological features with a threshold of p < 0.05, and then perform Spearman correlation analysis to determine redundant features in the 2 - mm, 4 - mm, 6 - mm, and 8 - mm regions around the endometrium. Define 0.6 as the redundancy threshold and randomly retain one of them to avoid redundancy; next, perform elastic net analysis on the 2 - mm, 4 - mm, 6 - mm, and 8 - mm regions around the endometrium respectively, and use the AUC curve to evaluate the model performance; use the LASSO algorithm to optimize the model parameters through ten - fold cross - validation and select radiological features with non - zero coefficients in the training group; use machine - learning algorithms to model and analyze radiological features; use AUC, accuracy, sensitivity, and specificity to evaluate the prediction performance of the best PEZ radiological model for endometrial receptivity; select the model with the largest AUC in the training group and the external validation group and name it the PEZ radiological model;

[0016] S5. Construction of the Combined Radiomics Model:

[0017] Perform Spearman correlation analysis on the features combining the endometrial radiomics model and the PEZ radiomics model. Features with a correlation threshold > 0.6 are considered redundant, and randomly retain one of them to avoid redundancy; then perform logistic regression analysis to obtain 15 radiomics features; use machine - learning algorithms to model the combined radiomics features obtained by joint LASSO; name the model with the largest AUC in the training group and the external validation group the combined radiomics model;

[0018] Comparison of the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model:

[0019] SPSS 25.0 and Python 2.7 software were used for statistical analysis: Continuous variables with normal distribution were expressed as mean ± standard deviation and analyzed using the t-test; variables with non-normal distribution were expressed as median ± interquartile range and analyzed using the Mann-Whitney U test; categorical variables were expressed as numbers or percentages and analyzed using the chi-square test; p < 0.05 was considered statistically significant. Univariate logistic analysis and multivariate logistic analysis were used to construct the models; the DeLong test was used to generate the ROC curves of the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model, and the AUC value was calculated to evaluate the diagnostic performance;

[0020] The ROC curves and DCA curves of the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model were compared to obtain the optimal model among the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model, and it was concluded that the combined radiomics model was the most capable of predicting the pregnancy outcome after frozen embryo transfer.

[0021] In one embodiment, in step S1, the methods for obtaining clinical data and ultrasound images are as follows:

[0022] A1. All female participants included in the analysis received transvaginal ultrasound examination within 24 hours before frozen embryo transfer. The instrument used for transvaginal ultrasound examination was the Mindray Nuewa R9 ultrasound machine equipped with a transvaginal volume probe (2.0 - 9.0 MHz);

[0023] A2. The transvaginal ultrasound machine was used to perform two-dimensional ultrasound scanning to evaluate the endometrial thickness and pattern, and the standard mid-sagittal plane images of the endometrium of each participant were saved in the Digital Imaging and Communications in Medicine format;

[0024] A3. Color Doppler ultrasound was used to classify the endometrial and endometrial sub-blood flow types; according to the Applebaum classification standard, they were divided into three types: Type I is that the spiral artery passes through the endometrial area and approaches the uterine cavity; Type II is that the spiral artery reaches the endometrium but does not exceed half of the single-layer endometrial area; Type III is that the spiral artery only reaches the sub-endometrial area and does not enter the endometrium;

[0025] In one embodiment, in step S1, the data inclusion criteria are as follows: accepting frozen embryo transfer, being under 40 years old, transferring one high-quality embryo, and having a clearly visible endometrium. If any one of these criteria is not met, the case will be excluded. Embryos with a score of ≥ 3AA, 3AB, 3BA, 3BB (on the 5th day) or ≥ 4AA, 4AB, 4BA, 4BB (on the 6th day) are considered high-quality embryos.

[0026] In one embodiment, in step S2, the primary data exclusion criteria are as follows: having congenital uterine malformations, submucosal fibroids, endometrial polyps, uterine adhesions, or intrauterine fluid accumulation, or having an endometrial thickness (EMT) less than 8 mm or greater than 14 mm. If any one of these criteria is met, the case will be excluded.

[0027] In one embodiment, in step S3, the specific method for feature extraction is as follows:

[0028] S31. Ultrasonographer A manually demarcates the region of interest (ROI) covering the endometrium on the standard mid-sagittal plane using 3D-Slicer software; the regions around the endometrium on the ultrasound image are obtained by automatically expanding 2 mm, 4 mm, 6 mm, and 8 mm outward from the contour of the endometrium to obtain circular regions of corresponding sizes.

[0029] S32. Radiomics features are extracted from the ROIs of the endometrium and the regions around the endometrium, including original images, square images, and images after wavelet filtering, using the open-source software tool PyRadiomics.

[0030] Half a month later, Ultrasonographer A and Ultrasonographer B use a blind method to re-demarcate the ROIs of the endometrium on the standard mid-sagittal plane images of 30 randomly selected participants and compare the intra-class and inter-class correlation coefficients.

[0031] In one embodiment, all female participants included in the analysis transferred one high-quality embryo; female participants included in the analysis underwent transvaginal ultrasound examination 4 - 5 weeks after frozen embryo transfer. If the embryo survived in the uterus, it was diagnosed as a clinical pregnancy.

[0032] In one embodiment, the frozen embryo transfer protocol is as follows: Cycle treatment is carried out using hormone replacement therapy, which starts on the third day of the menstrual cycle. Oral estradiol valerate tablets (2 mg, twice a day) are taken for 10 days; transvaginal ultrasound examination is performed to monitor the endometrial thickness, and the threshold for endometrial thickness is set to be greater than 8 mm; subsequently, progesterone is intramuscularly injected (60 mg, once a day) and dydrogesterone is orally administered (10 mg, twice a day); after a 5-day progesterone conversion period, the frozen embryo transfer surgery is performed.

[0033] The beneficial effects of the present invention are as follows:

[0034] This study was the first to explore the area most closely related to perigestational endometrium, and combined the radiomics features of the endometrium and the optimal area around the endometrium to predict the pregnancy rate after frozen embryo transfer through a machine learning model.

[0035] The ROC curves and DCA curves of the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model were compared. The optimal model among the endometrial radiomics model, the PEZ radiomics model, and the combined radiomics model was selected, and it was concluded that the combined radiomics model was the most capable of predicting the pregnancy outcome after frozen embryo transfer.

[0036] The combined radiomics model showed great potential in helping clinicians predict the results of FET, which was helpful for clinicians to make informed decisions before FET, improve the success rate of transplantation, and thus reduce the emotional and economic burdens brought by FET failure. Brief Description of the Drawings

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0038] Figure 1 It is a flow chart of the present invention; wherein, ART represents assisted reproductive technology; EMT represents endometrial thickness; FET represents frozen embryo transfer; TVS represents transvaginal ultrasound examination.

[0039] Figure 2 It is the cutting of the region of interest on the ultrasound image; wherein, ROI represents the region of interest; EN represents the endometrium; PEZ represents the perigestational endometrium region.

[0040] Figure 3 It is a schematic diagram of radiomics feature selection. The least absolute shrinkage and selection operator model adopted a ten-fold cross-validation technique, determined the optimal tuning parameter λ according to the minimum criterion, and gave the feature screening diagrams based on (a) EN, (b) PEZ, and (c) the combination of EN and PEZ. AUC, area under the curve; EN represents the endometrium; PEZ represents the perigestational endometrium region.

[0041] Figure 4ROC and DCA of three models in the training group and external validation group; among them, (a) represents the ROC of the three models in the training set; (b) represents the ROC of the three models in the external validation set; (c) represents the DCA of the three models in the training set. (d) represents the DCA of the three models in the external validation set; DCA represents decision curve analysis; ROC represents receiver operating characteristic curve; EN represents endometrium; PEZ represents the area around the endometrium. Detailed implementation manner

[0042] To make the technical problems, technical solutions and technical effects of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0043] Embodiment 1

[0044] This embodiment provides a method for predicting clinical pregnancy after frozen embryo transfer based on ultrasound radiomics. This method uses clinical data and ultrasound images within 24 hours before frozen embryo transfer (FET) to establish a machine learning (ML) model to predict the reproductive outcome after frozen embryo transfer (FET). An ultrasound examination is arranged within 24 hours before transplantation. This makes the research more clinically valuable. The specific steps are as follows:

[0045] S1. Research population: Obtain a dataset of female participants included in the analysis. The data samples of the female participants included in the analysis are divided into a training group and an external test group;

[0046] The inclusion criteria are as follows: Women under 40 years old who receive 1 high-quality blastocyst FET and whose endometrium is clearly visible on ultrasound examination. Embryos with a score of ≥3AA, 3AB, 3BA, 3BB (day 5) or ≥4AA, 4AB, 4BA, 4BB (day 6) are high-quality blastocysts.

[0047] The exclusion criteria are as follows: Women with congenital uterine malformations, submucous myomas, endometrial polyps, uterine adhesions or intrauterine fluid accumulation and endometrial thickness (EMT) <8 mm or >14 mm ( Figure 1 ), record the medical history of each participant, including age, cause of infertility and body mass index (BMI). Subsequently, transvaginal ultrasound (TVS) evaluation is performed to determine the exclusion criteria; one day before transplantation, transvaginal ultrasound (TVS) measures the endometrial thickness (EMT), endometrial morphology, and the type of color Doppler on the endometrium, and the median sagittal plane of the endometrium is reserved for imaging feature extraction.

[0048] A total of 358 participants (out of 400, excluding 42) who underwent frozen embryo transfer (FET) at the First Affiliated Hospital of Anhui Medical University (Center 1) from August 2023 to October 2024 were used as the training group. In addition, 64 participants (out of 80, excluding 16) from Hefei Maternal and Child Health Hospital (Second Center) were used as the external validation group.

[0049] There were 209 clinical pregnancies (58.4%) and 149 non-clinical pregnancies (41.6%) in the training group. There were 36 clinical pregnancies (56.3%) and 28 non-clinical pregnancies (43.7%) in the external validation group. The baseline characteristics of the study participants are shown in Table 1.

[0050] Table 1 Clinical characteristics of participants in the training group and external validation group

[0051]

[0052] S2. Determine the optimal perimetrial endometrial zone (PEZ):

[0053] The contours of the regions of interest (ROI) of the endometrium were automatically expanded outward by 2 mm, 4 mm, 6 mm, and 8 mm respectively to create corresponding circular perimetrial endometrial regions of different sizes. By comparing different perimetrial endometrial circular regions, it was found that the 4-mm region was the optimal perimetrial endometrial zone (PEZ) related to pregnancy (as shown in Table 2). Using six machine learning algorithm models (including logistic regression (LR), support vector machine (SVM), random forest classifier (RF), decision tree classifier (DT), k-nearest neighbor classifier (KNN), and backpropagation neural network (BPNN)), the backpropagation neural network (BPNN) model was superior to other models. The AUC of the optimal perimetrial endometrial zone (PEZ) in the training group was 0.744, and the AUC in the external validation group was 0.715 (Table 3).

[0054] Table 2 Performance of PEZ of different sizes in the training and external validation sets

[0055]

[0056] Table 3 Performance of six machine learning models in the training set and external validation set

[0057]

[0058] S3. Comparison of three radiomics prediction models:

[0059] Comparison of the endometrial radiomics model, the PEZ radiology model, and the combined radiomics model:

[0060] As shown in Table 4, in the training group, the combined radiomics model (AUC: 0.853, 95% CI: 0.811 - 0.890) was superior to the endometrial radiomics model (AUC: 0.770, 95% CI: 0.716 - 0.815) and the PEZ radiology model (AUC: 0.744, 95% CI: 0.691 - 0.794).

[0061] In addition, in the external validation group, the combined radiomics model (AUC: 0.809, 95% CI: 0.696 - 0.909) was superior to the endometrial radiomics model (AUC: 0.715, 95% CI: 0.581 - 0.833) and the PEZ radiomics model (AUC: 0.732, 95% CI: 0.581 - 0.833). The comparison of the ROC and DCA curves of the three models is shown in Figure 4 .

[0062] Table 4. Performance of three radiomics models in the training and external validation sets

[0063]

[0064] The conclusions are as follows:

[0065] In this study, we analyzed the predictive value of the perimetrial endometrial zone (PEZ) of different sizes on pregnancy outcomes after frozen embryo transfer (FET) using ultrasound images. It was demonstrated that the 4.0 mm region had the strongest correlation with pregnancy outcomes. The ultrasonic image features of the endometrium and the optimal perimetrial endometrial zone (PEZ) in the median sagittal plane were integrated to prove that this method had better predictive potential than the methods based only on the endometrial region or the optimal perimetrial endometrial zone (PEZ) assessment. A non-invasive and personalized ultrasound radiomics model was developed and validated for predicting pregnancy outcomes after frozen embryo transfer (FET).

[0066] Radiomics uses high-throughput data to convert radiological images into quantitative features for classification and analysis. It is widely applied in the medical field. The transitional zone located between the endometrium and the myometrium is called the junctional zone (JZ). It appears as a hypoechoic area surrounding the endometrium on ultrasound scans. The thickness of the junctional zone (JZ) exhibits cyclic variations throughout the menstrual cycle. In the non-pregnant state, the junctional zone (JZ) causes endometrial peristalsis and participates in regulating various reproductive processes, sperm transport, and embryo implantation. Its close relationship with pregnancy makes it an important part of radiomics analysis. However, the junctional zone (JZ) cannot be clearly observed in ultrasound images, and the combined area is uneven and non-linear, which poses challenges to their observation. According to previous studies, the upper limit of the normal thickness of the junctional zone (JZ) is 8 mm. In this study, the maximum thickness around the endometrium under investigation was 8 mm, with an interval of 2 mm. Using ultrasound radiomics, a preliminary attempt was made to explore the optimal perimetrial zone (PEZ) within the ranges of 2.0 mm, 4.0 mm, 6.0 mm, and 8.0 mm around the endometrium related to pregnancy outcomes after FET. This is the first investigation of the optimal perimetrial zone (PEZ). It was found that the predictive performance of the optimal perimetrial zone (PEZ) of PEZ4.0 mm was better than that of other zones, indicating that PEZ4.0 mm may contain more extensive microenvironmental information. Previous studies used deep learning of the endometrium to predict pregnancy outcomes after frozen embryo transfer (FET), with an AUC of 0.825. In this study, we combined the radiomics features of the endometrium and the optimal perimetrial zone (PEZ). In the training group, the AUC of the combined model was 0.853, which was better than the models using only EN or PEZ alone. This integration improved the accuracy of predicting pregnancy outcomes after FET and could evaluate ER more comprehensively. By better identifying participants with higher or lower risks of successful implantation, this model can help optimize treatment strategies, such as adjusting the embryo transfer time or exploring alternative interventions for participants with lower predicted success rates. Ultimately, this approach has the potential to improve the treatment outcomes of participants and reduce the emotional and economic burden of failed transfers.

[0067] In addition, among the six machine learning algorithm models applied (including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest Classifier (RF), Decision Tree Classifier (DT), k-Nearest Neighbor Classifier (KNN), and Back Propagation Neural Network (BPNN)), the Back Propagation Neural Network (BPNN) has the highest performance. As a powerful algorithm that models complex non-linear relationships by simulating biological propagation behavior, the Back Propagation Neural Network (BPNN) has obvious advantages in dealing with complex non-linearity, automatic feature learning, generalization ability, robustness, and scalability of large datasets. These advantages usually make it superior to traditional ML algorithms in many tasks. This may explain why the Back Propagation Neural Network (BPNN) model is superior to other models in this study.

[0068] The combined radiomics model can accurately predict the pregnancy outcome after frozen embryo transfer (FET). By successfully applying this comprehensive model in clinical practice, the prognosis of the participants can be improved, thus reducing the emotional and economic burden brought by the failure of frozen embryo transfer (FET).

Claims

1. A method for predicting clinical pregnancy after frozen embryo transfer based on ultrasound radiomics, characterized in that: The steps include: S1. Collect clinical data and ultrasound images of female participants who received frozen embryo transplantation to form a dataset of female participants who received frozen embryo transplantation. Screen out a preliminary dataset of eligible female participants who received frozen embryo transplantation from the dataset of female participants who received frozen embryo transplantation according to the data inclusion criteria and primary data exclusion criteria. S2. The final dataset of female participants included in the analysis was obtained by excluding secondary data from the dataset of eligible female participants who underwent frozen embryo transfer. The data samples of female participants included in the analysis were divided into a training group and an external test group. S3, Endometrial radiomic feature extraction: Determine radiomic features from regions of interest in the endometrium; Endometrial radiomic features were processed as follows: features with ICCs>0.75 were retained; features were subjected to univariate rank sum test, and those with p-values>0.05 were discarded; Spearman correlation analysis was performed on the features, and redundancy was eliminated using Spearman correlation analysis with r≥0.6; logistic regression analysis was performed on the radiomic features; model parameters were optimized using the least absolute shrinkage and selection operator algorithm through ten-fold cross-validation, and radiomic features with non-zero coefficients were selected from the training group, and the combined radiomic features were modeled and analyzed; AUC, accuracy, sensitivity, and specificity were used to evaluate the performance of the endometrial radiomic model in predicting endometrial receptivity; the model with the best AUC was selected from the training group and the external validation group as the endometrial radiomic model; S4. Radiological feature extraction and model establishment of the peri-endometrial region: S41. At the image level, the ultrasound images were first resampled using the b-spline interpolation algorithm, and images with different spatial resolutions were aligned to ensure that the voxel spacing was standardized to 1×1 mm. 2 ; S42. Grayscale normalization was applied to standardize the intensity level of grayscale images from different sources or imaging devices to ensure consistent brightness and contrast between images. At the feature level, all radiological features in the training set were normalized using z scores. The validation set was strictly normalized using the parameters calculated from the training set to ensure the consistency of standardization of data from different cohorts. The contour of the endometrium was automatically expanded outward by 2 mm, 4 mm, 6 mm, and 8 mm to obtain the corresponding sized annular areas, namely PEZ. S43. The images were normalized and PyRadiomics was used to extract features such as original images, square images, and wavelet filtered images. Features with ICCs>0.75 were retained. The significance of radiomic features was analyzed using a univariate rank sum test with a threshold of p<0.

05. Spearman correlation analysis was performed to determine the redundant features of the 2mm, 4mm, 6mm, and 8mm regions around the endometrium. 0.6 was defined as the redundant threshold, and one of them was randomly retained to avoid redundancy. Next, logistic regression analysis was performed on the 2mm, 4mm, 6mm, and 8mm regions around the endometrium, and the AUC curve was used to evaluate the model performance. The LASSO algorithm was used to optimize the model parameters through ten-fold cross validation, and radiomic features with non-zero coefficients were selected in the training group. Machine learning algorithms were used to model and analyze radiomic features. AUC, accuracy, sensitivity, and specificity were used to evaluate the predictive performance of the optimal PEZ radiomics model for endometrial receptivity. The model with the largest AUC was selected in the training group and the external validation group and named the model as the PEZ radiomics model. S5. Construction of combined radiomics model: The features of the endometrial radiomics model and the PEZ radiomics model were combined for Spearman correlation analysis. Features with a correlation threshold of >0.6 were considered redundant, and one of the features was randomly retained to avoid redundancy. Then, logistic regression analysis was performed to obtain 15 radiomic features. The combined radiomics features obtained by combined LASSO were modeled using a machine learning algorithm; the model with the largest AUC in the training group and the external validation group was named the combined radiomics model; S6. Comparison of endometrial radiomics model, PEZ radiomics model and combined radiomics model: SPSS25.0 and Python 2.7 software were used for statistical analysis. The DeLong test was used to generate the ROC curves of the endometrial radiomics model, the PEZ radiomics model and the combined radiomics model, and the AUC value was calculated to evaluate the diagnostic performance. The ROC curves and DCA curves of the endometrial radiomics model, the PEZ radiomics model and the combined radiomics model were compared to obtain the optimal model among the endometrial radiomics model, the PEZ radiomics model and the combined radiomics model, and it was concluded that the combined radiomics model can best predict the pregnancy outcome after frozen-embryo transfer.

2. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 1, characterized in that: In step S1, the method of acquiring clinical data and ultrasound images is as follows: A1. All female participants included in the analysis underwent vaginal ultrasound examination within 24 hours before frozen embryo transfer. The instrument used for vaginal ultrasound examination was Mindray NuewaR9 ultrasound machine equipped with a transvaginal volume probe; A2. Two-dimensional ultrasound scanning was performed by vaginal ultrasound to evaluate the thickness and type of the endometrium, and standard midsagittal section images of the endometrium were saved in the digital imaging and medical communication format for each participant; A3. Measure the thickness of the endometrium on two-dimensional ultrasound images; use color Doppler ultrasound to obtain the sub-blood flow type on the endometrium.

3. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 1, characterized in that: In step S1, the data inclusion criteria were: receiving frozen embryo transfer, being under 40 years old, having a high-quality embryo transferred, and having a clearly visible endometrium. Patients who did not meet any of the criteria were excluded.

4. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 1, characterized in that: In step S2, the secondary data exclusion criteria are: congenital uterine malformation, submucosal fibroids, endometrial polyps, uterine adhesions or intrauterine fluid, EMT less than 8mm or greater than 14mm. Patients who meet any of the criteria will be excluded.

5. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 1, characterized in that: In step S3, the specific method of feature extraction is as follows: S31, ultrasound physician A manually delineated the region of interest covering the endometrium on the standard midsagittal plane using 3D-Slicer software; the peri-endometrial region on the ultrasound image was obtained by automatically expanding the contour of the endometrium outward by 2 mm, 4 mm, 6 mm, and 8 mm to obtain the corresponding sized annular regions; S32, Radiomic features were extracted from the regions of interest in the endometrium and the corresponding peri-endometrial regions using the open source software tool PyRadiomics; S33. One month later, ultrasound physicians A and B used a blind method to re-delineate the endometrial region of interest on standard midsagittal images of 30 randomly selected participants, and compared the intra-class and inter-class correlation coefficients.

6. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 1, characterized in that: All female participants included in the analysis had a high-quality embryo transferred; the female participants included in the analysis had a clinical pregnancy diagnosed by vaginal ultrasound 4-5 weeks after the frozen-embryo transfer if the embryo survived in the uterus.

7. The method for analyzing embryo transplantation outcomes using an ultrasound-based machine learning model according to claim 6, characterized in that: The frozen embryo transfer protocol is as follows: cyclical treatment with hormone replacement therapy, starting on the third day of the menstrual cycle, with 2 mg of estradiol valerate tablets taken orally twice a day for 10 days; vaginal ultrasound examination to monitor endometrial thickness, with the threshold for endometrial thickness set at greater than 8 mm; subsequently, 60 mg of progesterone is injected intramuscularly once a day; 10 mg of dydrogesterone is taken orally twice a day; after a 5-day progesterone conversion period, a frozen embryo transfer is performed.

Images