An Artificial Intelligence-Based Algorithm for Predicting Pregnancy Success Using Static Images Captured by Optical...

Area of Science:

• Reproductive medicine and artificial intelligence
• Embryology and computer vision
• In vitro fertilization (IVF) and machine learning

Background:

• Manual grading of human embryos for IVF is standard but subjective.
• AI applied to time-lapse embryo images shows promise for quality assessment.
• Current AI models often require dynamic imaging and cannot predict pregnancy outcomes.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for predicting embryo quality using static IVF images.
• To assess the model's ability to predict pregnancy versus non-pregnancy.
• To explore the prediction of live birth outcomes from static embryo images.

Main Methods:

• Utilized transfer learning with a pretrained Inception V1 network.
• Employed deep learning techniques, including data augmentation and transfer learning.
• Built models using the Tensorflow software package with a standard train/validation/test split.

Main Results:

• The algorithm achieved an area under the curve of 0.657 for predicting pregnancy.
• The model demonstrated an inability to meaningfully predict live birth outcomes.
• The study utilized a limited dataset of 361 static images from four IVF clinics.

Conclusions:

• This is the first study to successfully predict IVF outcomes using only static embryo images.
• The developed model shows potential despite a limited dataset and lower accuracy than conventional methods.
• Further research with larger datasets may enable prospective validation and improved generalizability.