Predicting pregnancy test results after embryo transfer by image feature extraction and analysis using machine...

Area of Science:

• Reproductive medicine
• Artificial intelligence in healthcare
• Embryology

Background:

• Assessing blastocyst viability for pregnancy prediction is currently empirical and lacks reproducibility.
• Existing methods for predicting pregnancy outcomes from embryo morphology are limited.

Purpose of the Study:

• To develop and evaluate an artificial vision and machine learning-based algorithm for predicting pregnancy using embryo morphology and patient age.
• To assess the performance of various classifiers in predicting pregnancy outcomes based on embryo images and clinical data.

Main Methods:

• Development of an algorithm integrating novel morphometric features from digital micrographs and non-morphometric data (patient age).
• Utilized two high-quality databases (n=221) with known pregnancy outcomes.
• Evaluated five classifiers: Bayesian, Support Vector Machines (SVM), deep neural network, decision tree, and Random Forest (RF), using k-fold cross-validation.

Main Results:

• SVM classifier achieved an F1 score of 0.74 and AUC of 0.77 in database A.
• Random Forest classifier obtained an F1 score of 0.71 and AUC of 0.75 in database B.
• The system demonstrated ability to predict positive pregnancy tests from single digital embryo images.

Conclusions:

• The developed AI system provides a novel, reproducible approach to pregnancy prediction in assisted reproductive technology.
• The algorithm is adaptable to different laboratory settings and easily integrated into clinical practice.
• Predicting pregnancy outcomes using embryo morphology and AI offers significant advantages over current empirical methods.