An Artificial Intelligence (AI) Automated Embryo Morphology Grading Model Shows High Accuracy when Compared to Conventional Morphology Grading by Embryologists

Introduction:
Conventional morphology grading by embryologists remains the gold standard for ranking embryos according to their implantation potential. However, there is no universally standardized criteria for optimized morphology assessment. We propose an AI-based method that uses temporal information from time-lapse images to automatically grade the morphological appearance of blastocysts with high accuracy and efficiency.

Aim:
To validate the accuracy of an AI-based embryo morphology grading system and compare its performance to human embryologists.

Materials and Methods:
An artificial neural network (ANN) model was trained on time lapse sequence data derived from 10,000 blastocysts. All blastocysts were individually graded by 20 senior embryologists according to ASEBIR criteria. The ANN model was used to develop an automated multi-class grading model that was trained to distinguish between A, B, and C graded blastocysts. ROC curve analysis was used to evaluate the performance of the model.

Results:
The specificity, sensitivity, and specificity values for the AI ANN model were as follows: for the detection of C versus A/B blastocysts: sensitivity: 0.94, specificity: 0.95, accuracy: 0.92. For the detection of A versus B/C blastocysts: sensitivity: 0.96, specificity: 0.90, accuracy: 0.93. for the detection of B versus A/C blastocysts: sensitivity: 0.90, specificity: 0.82, accuracy: 0.86. Corresponding AUCs were 0.99, 0.97, and 0.98, respectively.

Conclusions:
The AI model demonstrated high accuracy across all classes and was strongly consistent with embryologists’ grading. An automated AI model for morphology grading serves as an objective, consistent embryo evaluation tool that optimizes clinic efficiency and workflow.