LWMA-Net: Light-weighted morphology attention learning for human embryo grading

Area of Science:

• Reproductive medicine
• Artificial intelligence in healthcare
• Embryology

Background:

• Embryo morphology assessment is crucial for selection but subjective and time-consuming.
• Inter- and intra-observer variability among embryologists affects grading accuracy.

Purpose of the Study:

• To develop an automated system, the light-weighted morphology attention learning network (LWMA-Net), for assisting embryo grading.
• To evaluate the effectiveness of LWMA-Net in improving embryologists' grading performance.

Main Methods:

• LWMA-Net integrates a morphology attention module (MAM) and a multiscale fusion module (MFM).
• The network was trained on 3599 embryos and validated on an independent set of 691 embryos.
• Five embryologists regraded embryos with and without LWMA-Net assistance.

Main Results:

• LWMA-Net achieved high AUCs of 96.88% (4-category) and 97.58% (3-category) on the training set.
• Embryologists using LWMA-Net showed significant improvements in grading capabilities, with average AUC increases of 4.98%-5.32%.
• The AI tool demonstrated effectiveness in reducing subjectivity and enhancing grading consistency.

Conclusions:

• LWMA-Net shows strong potential as an assistive tool in human embryo grading.
• The AI-powered system can improve the accuracy and efficiency of embryo selection in assisted reproduction.
• LWMA-Net contributes to more objective and reliable embryo assessment, benefiting fertility treatments.