An Optimal Approach in Selecting Embryo for In-Vitro Fertilization (IVF) Based on Deep Learning.

Background: About 14% of couples experience infertility, and In Vitro Fertilization (IVF) has become one of the most widely used treatment options. However, the overall success rate of IVF remains relatively low, at around 30%. At present, viable embryos are typically selected on the fifth day based primarily on morphological assessment, a method that is both subjective and limited in accuracy. Considering the substantial financial, physical, and emotional costs associated with failed IVF attempts, there is a pressing need for more reliable and effective embryo selection techniques. Objective: This study aimed to increase the accuracy of embryo selection in IVF based on a deep learning-based transfer with the GoogLeNet architecture. Material and Methods: In this experimental study, a retrospective dataset of embryo images was used to develop and evaluate a deep learning-based classification model in the following main phases: data preprocessing, model implementation, and evaluation. Embryo images were standardized through cropping and normalization to ensure consistency across different imaging systems. The GoogLeNet architecture, pre-trained on the ImageNet dataset, was utilized and further modified to adapt to the specific task of embryo viability classification. Results: Evaluation on the test dataset demonstrated that the proposed model achieved strong predictive performance, with accuracy, precision, recall, and F1-score all reaching 97%. This performance surpasses that of existing baseline techniques, highlighting the model's effectiveness. Conclusion: The proposed transfer learning-based approach using GoogLeNet shows significant potential for improving embryo selection in IVF, thereby reducing the emotional and financial strain associated with repeated IVF failures.