Using artificial intelligence to avoid human error in identifying embryos: a retrospective cohort study.

Area of Science:

• Embryology
• Artificial Intelligence
• Bioinformatics

Background:

• Accurate patient identification is crucial in assisted reproductive technologies.
• Current methods for tracking embryos can be prone to errors.
• Novel approaches are needed to enhance the reliability of specimen identification in IVF labs.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural networks (CNNs) in identifying patient-specific embryos solely from image data.
• To determine if CNNs can accurately ascertain patient identity for cleavage and blastocyst stage embryos.

Main Methods:

• A CNN model was trained and validated on 4889 time-lapse embryo images from a retrospective cohort.
• The algorithm generated unique identification keys for embryos at specific timepoints (day 3 and day 5).
• The accuracy of patient ID matching was assessed by comparing generated keys with a library of known patient IDs across 400 embryo cohorts.

Main Results:

• CNN technology achieved 100% accuracy in matching patient identification for day 3 embryo cohorts (n=400 patients).
• The CNN model also demonstrated 100% accuracy for day 5 embryo cohorts (n=400 patients).
• These results were consistent across three independent replicates.

Conclusions:

• An artificial intelligence-based method using CNNs provides a robust solution for embryo identification.
• The technology leverages unique morphological features of embryos for accurate identification.
• This AI approach can be integrated into existing IVF laboratory systems to improve specimen tracking and reduce errors.