JBRA Assisted Reproduction 2025;29(Suppl.2 SBRA 2025):190
Poster Presentation
29^th Annual Congress of the SBRA. São Paulo/SP - Brazil, 2025
doi: 10.5935/1518-0557.20263895

P-178. Prediction of gestational success using Artificial Intelligence applied to the internal cell mass of blastocysts

Bruno Araújo Mendes¹, Mariana Nicolielo², Dóris Spinosa Chéles², Bruna Lourenço², José Roberto Alegretti², Aline Rodrigues Lorenzon², Marcelo Fábio Gouveia Nogueira¹, José Celso Rocha¹

¹ Universidade Estadual Paulista "Júlio de Mesquita Filho" – UNESP - Assis -SP – Brasil
² Huntington Medicina Reprodutiva – Eugin Group -São Paulo, SP, Brasil

Objective: To develop and evaluate an Artificial Intelligence model based on convolutional neural networks to predict gestational success from the analysis of segmented images of the inner cell mass of blastocysts.
Methods: A database with 1,209 images of internal cell mass (ICM) obtained using the EmbryoScope^® system was used. Automatic segmentation of the ICM was performed using a U-Net architecture developed in Python to isolate the region of interest in a standardized manner. Next, the segmented images were submitted to a convolutional neural network (CNN) classification implemented and trained on the Matlab platform using the Deep Learning toolbox, employing the data augmentation technique. The database for CNN implementation was divided into training (70%), validation (15%), and testing (15%). The evaluation included accuracy, F1-score, area under the ROC curve (AUC), and confusion matrix. To increase interpretability, Grad-CAM was applied, allowing visualization of the ICM regions most relevant to the classification decision.
Results: In training, 846 images were used, achieving a performance of 98% accuracy. In the test set, consisting of 181 images, the model showed an accuracy of 65.19%, AUC of 0.71, and F1-score of 69.51%. Grad-CAM activation maps showed that the densest and most central regions of the ICM contributed significantly to the prediction result, which is consistent with the biological relevance of the ICM as the structure that will give rise to the embryo itself.
Conclusion: The Artificial Intelligence-based approach, focused specifically on the inner cell mass, shows potential by concentrating its analysis on a structure that is fundamental to embryonic development. The ICM, being responsible for the formation of the embryo itself, has characteristics directly associated with the potential for implantation and gestational success. The use of automated techniques to evaluate this region can provide embryologists with an objective and standardized tool, contributing to optimizing embryo selection, reducing treatment time, and increasing success rates. Future studies include expanding the database, improving the network architecture, and integrating other variables and other areas of study, such as the trophectoderm, to make the model more robust and applicable to clinical practice.