MEmEBG Beats Human Embryologists at IVF Grading With 150 Images
A multitask embedding model predicts three Gardner grading components of IVF blastocysts from a single day-five image, outperforming single-task baselines.
PRO- MEmEBG predicts Gardner TE, ICM, and expansion grades from single day-five embryo images using a shared embedding.
- Backbone is ImageNet-pretrained ResNet-18 with a DINOv2 embedding and three task-specific heads.
- Trained on 249 blastocysts from the Saeedi et al. dataset, 75/25 train/test split.
- Multi-task learning beats single-task on TE (0.64 vs 0.60) and expansion (0.76 vs 0.72), significant at p≤0.05.
- ICM grading did not benefit from joint training, suggesting task-specific visual cues dominate.
- Minority Grade B classes suffer from imbalance; F-scores drop to 0.44 for ICM Grade B.
Picking which embryo to transfer during IVF still comes down to an embryologist squinting through a microscope and assigning letter grades. A new paper called MEmEBG (Multitasking Embedding for Embryo Blastocyst Grading) tries to remove that subjectivity by training a single network to predict three Gardner grading components at once, from a single day-five image.
Why embryo grading is a messy problem
Most IVF labs worldwide use the Gardner system, which scores three blastocyst parameters: expansion grade, inner cell mass (ICM) quality, and trophectoderm (TE) quality. The ICM is the cluster that becomes the fetus, the TE becomes the placenta, and expansion tracks how much the blastocoel cavity has stretched the embryo. Accurate analysis is difficult because these compartments often share similar textures but differ sharply in structure.
Human grading is subjective, and inter-embryologist agreement is notoriously low. In one large study, individual embryologists agreed with the common consensus 41% of the time on average, while an AI model reached 48%. That variance directly affects which embryo gets transferred, so any push toward consistent, quantitative scoring has real clinical stakes.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
