Non-invasive AI embryo selection: AI Predicts Embryo Chromosomes with 0.85 Accuracy

In May 2026, Spanish scientists used an AI model to analyze 2,388 embryos and achieve an accuracy of 0.85 (out of 1.0) in predicting whether an embryo had a normal chromosome set, without any biopsy or invasive procedure. 

Embryo chromosome status directly affects IVF implantation and pregnancy outcomes. Today, PGT-A is the gold standard for detecting chromosomal abnormalities. It requires a biopsy of a few cells, a mature and generally safe procedure, but some families prefer to minimize embryo manipulation.

With advances in AI-driven precision medicine, assisted reproduction is gaining a new, non-invasive decision-support tool.

Time-Lapse Videos: How AI achieves prediction

A time-lapse incubator contains a built-in microscope camera that automatically takes images every few minutes, from fertilization through blastocyst formation. The result is a dynamic video of the embryo’s entire development.

Compared with traditional morphological assessment (in which embryologists remove the embryo for observation at fixed times), time-lapse imaging keeps the embryo in the stable environment of the incubator, reducing disturbance.

It also provides continuous dynamic information, subtle patterns that are often invisible to the human eye.

AI models excel at learning from thousands of embryo videos, whose dynamic patterns correlate with chromosome health, blastocyst quality, and implantation potential, all without physically touching the embryo.

Study 1: Video Vision Transformer (ViViT) Predicts Ploidy

In May 2026, the García-Navarro team (Spain) published a paper in Studies in Health Technology and Informatics. They used a Video Vision Transformer (ViViT) architecture to analyze time-lapse videos of 2,388 embryos.

The model achieved an ROC-AUC of 0.85 for predicting embryo euploidy (normal chromosome number) on the validation set.

ROC-AUC ranges from 0.5 (random) to 1.0 (perfect). A score of 0.85 indicates a strong ability to distinguish between “competent” (euploid) and “incompetent” (aneuploid) embryos.

The researchers noted that ViViT can capture developmental dynamics that human eyes cannot detect, fully automated and without manual annotation.

Study 2: Only Video and Maternal Age Predicts Ploidy and Pregnancy

A second study, published in Scientific Reports in February 2026, went a step further.

An international team developed two deep learning models that used only embryo time-lapse video and maternal age (no embryologist-annotated parameters) to predict:

1. Embryo ploidy status (chromosome normality).

2. Clinical pregnancy (fetal heartbeat after transfer).

The models were trained and externally validated using data from two independent IVF centers. Results on the external validation set:

1. Ploidy prediction: AUROC = 0.759

2. Pregnancy prediction: AUROC = 0.746

Although the accuracy is slightly lower than the ViViT model, the key strength of this study is that no manual annotation was needed, and the models performed well on data from a different center, proving that the AI learned generalizable patterns, not just lab-specific artifacts.

The study also demonstrated AI’s potential for blastocyst quality assessment, offering a non-invasive way to select high-quality blastocysts.

Study 3: Multimodal AI Predicts Singleton/Twin and Miscarriage/Live Birth

The VaTEP model, published in npj Digital Medicine, expanded the data dimension further.

It integrates time-lapse video and clinical variables (maternal age, AMH, endometrial thickness, etc.) using a multi-task learning framework. It can simultaneously predict:

1. Fetal heartbeat (clinical pregnancy).

2. Singleton vs. twin pregnancy.

3. Miscarriage vs. live birth.

The distinctive feature of this model: beyond chromosome prediction and embryo selection, AI can also assess blastocyst quality, predict multiple pregnancy outcomes, and incorporate maternal clinical characteristics.

This allows doctors to estimate before transfer how many embryos to transfer safely and whether miscarriage is likely, enabling more personalized transfer strategies.

Such multimodal integration of clinical data and dynamic imaging represents a cutting-edge application of AI-driven precision medicine in reproductive health.

The Role of AI Embryo Selection: Decision Support, Not Replacement

It is important to note that AI-based embryo selection is still in the clinical validation phase and cannot yet replace PGT-A.

A 2025 study in Human Reproduction cautioned that some AI models show instability across different centers.

Thus, the appropriate role for current AI models is as a decision-support tool for embryologists and patients, helping to quickly triage embryos most worthy of transfer, and potentially reducing unnecessary biopsies, but not as a standalone diagnostic test.

For patients at high risk of chromosomal abnormalities (e.g., advanced maternal age, recurrent pregnancy loss, previous aneuploidy history), PGT-A remains the most reliable chromosome screening method.

CEF Perspective

CEF continuously monitors cutting-edge advances in assisted reproductive technology, including AI-driven precision medicine applications in embryo selection, blastocyst quality assessment, and multiple pregnancy prediction.

We believe that as more multi-center, large-sample prospective clinical studies are completed, AI-assisted embryo selection may become a routine feature in IVF laboratories.

Currently, we primarily help our clients access standard PGT-A and IVF services. If you have any questions about embryo screening options, please feel free to reach out.

>For inquiries, email info@cefivf.com or visit our Contact page.

>Note: This article reviews recent research on AI-based embryo selection. AI is not yet a replacement for PGT-A, and CEF does not currently offer AI embryo assessment services. The content is for educational purposes only.