Evaluation of an Artificial Intelligence Model for the Prediction of Human Blastocyst Ploidy Without Invasive Procedures

Overview

The goal of this clinical trial is to evaluate an artificial intelligence model for the prediction of human blastocyst ploidy without invasive procedures in couples that receive preimplantation genetic testing. The main questions it aims to answer are: * Is an artificial intelligence model able to predict the ploidy status of a human blastocyst based on its 3D morphology? * Do quantitative 3D morphological parameters of trophectoderm cells and inner cell mass have strong correlations with human blastocyst ploidy status? Videos that include multi-view images of each blastocyst from participants will be collected on Day 5/6 of culture, and preimplantation genetic testing results of these blastocysts will be collected 4-8 weeks after trophectoderm biopsy.

Background: A pilot study has been conducted at The Affiliated Nanjing Drum Tower Hospital of Nanjing University Medical School from August 2023 to September 2024. Videos and preimplantation genetic testing (PGT) results from 144 blastocysts were retrospectively collected. The artificial intelligence (AI) model first reconstructed the 3D surface of the blastocysts from the videos, and then measured their 3D morphological parameters. Based on these parameters, the model predicted the ploidy status of the blastocysts, and the prediction outputs were compared with the PGT results. The prediction sensitivity, specificity, accuracy and AUC were 90.5%, 91.3%, 90.9% and 0.946, respectively. Study design: This is a multi-center, prospective, non-randomized, non-blinded, and single-group study. After being informed about the study and potential risks, all participants will write the informed consents. Videos and PGT results of Day 5/6 blastocysts from each participant will be collected. Blastocysts will be classified as euploid, mosaic, and aneuploid corresponding to \<30%, 30-80%, and \>80% aneuploidy, respectively. Embryo culture, biopsy, and transfer will follow the standard operating procedure (SOP) in the laboratory. The study is non-interventional, and results will not be used to make treatment decisions. Sample size: We plan to enroll \~1408 Day 5/6 blastocysts in this trial based on one-sample sensitivity and specificity analysis. Meta-analysis shows that the sensitivity and specificity of the existing AI models are 73.4% (3702/5047) and 69.6% (4892/7028), and those of the non-invasive chromosomal screening methods are 80.3% (678/844) and 73.3% (908/1238) for non-invasively predicting blastocyst ploidy status. This study is presumed to achieve a sensitivity of no less than 85% and a specificity of no less than 80% with a significance level of α = 0.05. A total of 1126 blastocysts are required to achieve a statistical power of 0.9. Assuming a \~20% dropout rate, a total of 1408 blastocysts are anticipated to be enrolled. This sample size calculation is based on the analysis of statistical power and will be regularly revisited/adjusted during the trial period to ensure a high statistical power is achieved. Data management: The electronic data capture (EDC) system will be used for data collection. A clinical research coordinator will be assigned at each hospital, and they are responsible for recording the videos and clinical data via the EDC system. A senior clinical research associate will inspect the data in the EDC system regularly among 5 hospitals. The Data Safety and Monitoring Committee (DSMC) is responsible for overseeing the entire research process and the EDC system. For incomplete or missing data in the EDC system, the DSMC will contact the investigators for clarification. Statistical analysis: Statistical analysis will be conducted using IBM SPSS Statistics 26. Categorical variables will be described by number and percentage, and numerical variables will be described by mean, standard deviation (SD) and range. The Chi-squared test will be performed to analyze trends in categorical variables, and the t-test will be performed to compare numerical variables among different groups. Pearson correlation will be used to analyze the linear relationship among numerical variables. All statistical tests are two-tailed. P-values of \<0.05 will be considered statistically significant, and odd ratios (ORs) with 95% confidence interval (CI) will be calculated. Logistic regression will be used for multivariate analysis to calculate the adjusted odd ratios (aORs). The performance of the AI for blastocyst ploidy prediction will be evaluated by sensitivity, specificity, accuracy and AUC, with 95% confidence interval. Missing data will be removed if the proportion of samples with missing values is very small relative to the total sample size. Otherwise, the average, maximum, minimum, medium, or regression model will be used to impute the missing values. Outliers will be treated as the missing data and addressed accordingly.