How ART Fertility Clinics Are Redefining AI in IVF

ART Fertility Clinics has developed one of the largest and most detailed IVF datasets from a single group worldwide. The Assisted Reproductive Technology (ART) Dataset includes treatments performed at ART Fertility Clinic in Abu Dhabi between 2015 and 2022.

It comprises 14,776 embryos from 1,810 patients and 2,500 treatment cycles, combining time-lapse imaging from the Vitrolife EmbryoScope system with rich electronic health record (EHR) data.

• The dataset captures morphology, morphokinetics, stimulation protocols, fertilisation methods, and key clinical outcomes such as blastocyst development, genetic results, and pregnancy endpoints.

The structure and pre-processing pipeline of this dataset are described in a dedicated medRxiv report, providing transparency on how the data are collected, cleaned and labelled.

On top of this dataset, ART Fertility Clinics and NYU Abu Dhabi have co-developed a cutting-edge AI system specifically designed for IVF, known as IVF-TFT (In Vitro Fertilization – Temporal Fusion Transformer).

IVF-TFT is a multimodal deep learning model that simultaneously learns from time-series embryo images and patient-specific clinical features.

• It was trained on more than 6 million images from 14,776 embryos, representing one of the largest AI training efforts in human IVF.
• In internal testing, IVF-TFT achieved an AUROC of 0.968 for multi-class embryo development stage classification and 0.977 for blastocyst development prediction.
• In an external test set, it maintained high performance, with AUROCs of 0.950 and 0.947 for these tasks, demonstrating strong generalisability.

IVF-TFT is designed to automate time-consuming, subjective elements of embryo monitoring and staging, supporting embryologists in making more informed, consistent decisions.

An important practical question in the IVF laboratory is predicting when each embryo will reach a usable blastocyst stage to optimise biopsy, cryopreservation and transfer. ART Fertility Clinics has specifically addressed this through AI-driven research.

At the 2024 ESHRE Annual Meeting, a selected oral communication titled “A cutting-edge artificial intelligence system tracking embryo development and pinpointing the optimal day for blastocyst utilization” showcased this work.

• The study evaluated whether an AI model can predict the exact day (day 5, 6 or 7) at which each blastocyst will reach the expansion level required for biopsy, cryopreservation or transfer.
• The AI system achieved high accuracy in tracking developmental timings and forecasting when an embryo would reach a usable blastocyst stage.
• These predictions support more efficient lab workflow planning and more precise scheduling of PGT-A biopsies and transfers.

This illustrates how ART Fertility’s AI research is built around clinically relevant, measurable questions that directly impact patient care and laboratory efficiency.

ART Fertility Clinics follows an evidence-first pathway for integrating AI into IVF.

AI models are developed in collaboration with academic partners at NYU Abu Dhabi and NYU Tandon School of Engineering, ensuring they are designed and evaluated by independent data scientists.

• Model development includes external validation and detailed analysis of performance across different subgroups, such as maternal age, body mass index and fertilisation methods.
• AI tools are explicitly positioned as decision-support systems that complement, rather than replace, the expertise of clinicians and embryologists, with careful recognition of their limitations.

This framework ensures that AI is integrated into care only where it has demonstrable value and a strong scientific basis.

Because ART Fertility Clinics generates and manages its own research dataset, it can apply stringent governance to data use, privacy and ethics.

• All embryo images and EHR records in the ART Dataset are de-identified and processed through defined pre-processing pipelines prior to model development.
• The extraction, cleaning and labelling procedures are documented in detail in the data report to facilitate reproducibility and transparency.
• Publications clearly state that preprints, such as the medRxiv report, should not be used to guide clinical practice until fully peer-reviewed and prospectively validated, reflecting a cautious and responsible stance on clinical translation.

For patients, this means their data contribute to a controlled research programme whose primary objective is to improve safety, transparency and outcomes in IVF.

For individuals and couples seeking fertility treatment, AI at ART Fertility Clinics is not a slogan but a research-driven capability grounded in large, well-described datasets and rigorously validated models.

• ART has built and documented a large IVF dataset linking embryo time-lapse imaging with detailed clinical data.
• ART has co-developed IVF-TFT, a state-of-the-art AI system for embryo development staging and blastocyst prediction, trained on more than 6 million images and externally validated with high performance.
• ART has presented peer-reviewed conference work on using AI to pinpoint the optimal day for blastocyst utilisation, directly supporting better workflow and patient care.

This combination of deep clinical expertise, robust in-house data and academically validated AI systems is central to how ART Fertility Clinics is redefining the role of AI in IVF.