Written by Ram Prakash, Clinical Embryologist
Quick takeaways
- AI scoring tools don’t just look at a single photo — they track the precise timing of cell division across days of development.
- Specific timing milestones (when an embryo reaches 2 cells, 3 cells, starts forming a blastocyst, and so on) carry real predictive signal on their own.
- Embryos dividing notably faster or slower than typical windows are associated with lower viability, even when they look fine in a single snapshot.
- More data points is promising, but it hasn’t yet translated into a proven pregnancy-rate advantage over skilled manual grading.
We’ve written before about whether AI embryo scoring outperforms trained embryologists (it hasn’t been proven to, in the strongest trial to date). The more interesting question here is different: what is the algorithm actually measuring? Understanding that helps explain why AI embryo scoring is taken seriously in modern labs, even without a proven pregnancy-rate edge yet.
Two Different Inputs: What Embryologists See vs. What Algorithms See
Traditional morphology grading, the kind we described in our blastocyst grading guide, is based on a handful of snapshots — checking the embryo at set points and scoring what it looks like at that moment. AI scoring tools built around time-lapse incubators work with a richer input: continuous images captured every few minutes across five to six days. The algorithm isn’t just scoring appearance, it’s scoring the pattern and timing of how the embryo got there — something a human reviewing occasional snapshots can’t fully capture.
The Timing Milestones AI Actually Tracks
Embryology researchers have identified specific developmental checkpoints with predictive weight: time to reach 2 cells (t2), 3 cells (t3), 5 cells (t5), start of blastocyst formation (tSB), and full blastulation (tB), among others. The interval between some of these checkpoints often matters as much as the checkpoints themselves. Earlier algorithms like KIDScore used a decision-tree approach built directly on these timings. Newer deep learning models go further, learning patterns directly from raw time-lapse video without a human manually marking each checkpoint first.
Why More Data Doesn’t Automatically Mean Better Selection?
It’s tempting to assume tracking dozens of extra data points must beat a trained eye checking a few snapshots. In practice, the largest randomized trial of a deep learning scoring tool to date didn’t find a pregnancy-rate advantage over standard morphology grading by embryologists — even though the richer dataset is a genuinely more detailed kind of input. Richer data and better outcomes aren’t automatically the same thing, which is exactly why this remains an active research question rather than a settled one.
What This Means for “High-Quality” Selection in Practice?
In our lab, we treat AI scoring as a second, consistent set of eyes — useful for flagging timing patterns a quick manual check might miss, and for keeping scoring consistent across a busy day. It doesn’t replace morphology grading or genetic testing; it sits alongside them as one more input into a decision an embryologist still makes.
Questions Worth Asking About Any AI Tool Your Clinic Uses
- Is the score based on timing data, static images, or both?
- Was this specific tool validated on a large, multi-clinic dataset?
- How does the lab handle cases where AI and manual grading disagree?
The Bottom Line
AI embryo scoring brings a genuinely different kind of information to the table — continuous timing data rather than occasional snapshots. That’s a meaningful technical advance, even though it hasn’t yet proven to outperform skilled manual grading on the outcome that matters most: a healthy pregnancy. At Embryologist.co.in, we use it as one more lens, not the final word.
This article is for general educational purposes and isn’t a substitute for personalized advice from your fertility specialist or embryologist.
Sources
- Meseguer, M. et al. Development of a generally applicable morphokinetic algorithm capable of predicting the implantation potential of embryos transferred on Day 3. Human Reproduction, 2016.
- Illingworth, P.J. et al. Deep learning versus manual morphology-based embryo selection in IVF: a randomized, double-blind noninferiority trial. Nature Medicine, 2024.