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

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Medical Imaging Analysis

Background:

• Infertility affects 14% of couples, with In Vitro Fertilization (IVF) as a common treatment.
• Current IVF success rates are low (~30%), with embryo selection based on subjective morphological assessment.
• There's a critical need for more accurate embryo selection to reduce patient costs and emotional distress.

Purpose of the Study:

• To enhance the accuracy of embryo selection in IVF procedures.
• To develop and evaluate a deep learning model for predicting embryo viability.

Main Methods:

• Utilized a retrospective dataset of embryo images for model development.
• Employed a deep learning approach using the GoogLeNet architecture, pre-trained on ImageNet.
• Standardized embryo images via cropping and normalization for consistent analysis.

Main Results:

• The deep learning model achieved high predictive performance on the test dataset.
• Accuracy, precision, recall, and F1-score all reached 97%.
• The model's performance significantly outperformed existing baseline embryo selection techniques.

Conclusions:

• The transfer learning-based approach using GoogLeNet shows promise for improving IVF outcomes.
• This method has the potential to reduce the financial and emotional burden of failed IVF cycles.
• Accurate embryo selection is crucial for optimizing IVF treatment efficacy.