Deep learning applications for human embryo assessment using time-lapse imaging: scoping review.

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Embryology

Background:

• The synergy of deep learning (DL) and time-lapse imaging (TLI) presents novel opportunities for refining embryo assessment and selection within clinical in vitro fertilization (IVF).
• Traditional embryo evaluation methods can be subjective; TLI provides continuous monitoring, and DL offers advanced analytical capabilities to interpret this complex data.
• The increasing application of these technologies signifies a shift towards more objective and data-driven approaches in assisted reproductive technologies.

Purpose of the Study:

• To conduct a scoping review exploring the diverse applications of deep learning models in evaluating and selecting embryos using time-lapse imaging systems.
• To synthesize current research trends, methodologies, and findings in the field of DL-assisted embryo analysis.
• To identify key areas where DL has been applied and to highlight potential future research directions.

Main Methods:

• A comprehensive search of six electronic databases (Scopus, MEDLINE, EMBASE, ACM Digital Library, IEEE Xplore, Google Scholar) was performed for peer-reviewed literature published up to May 2024.
• The review adhered to the PRISMA guidelines for scoping reviews, ensuring a systematic and transparent methodology.
• Seventy-seven articles were included after reviewing 773 initially identified studies.

Main Results:

• Deep learning applications in embryo analysis have surged in the last four years, with primary uses in predicting embryo development/quality (61%) and clinical outcomes like pregnancy (35%).
• Convolutional Neural Networks (CNNs) were the dominant DL architecture (81%), utilizing time-lapse video images (100%) as training data, often supplemented with clinical parameters.
• Studies varied significantly in sample size, with a mean of 10,485 embryos, and predominantly used blastocyst-stage images (47%).

Conclusions:

• Deep learning demonstrates significant potential to enhance embryo evaluation and selection in clinical IVF, offering more objective and predictive insights.
• The findings underscore the growing importance and diverse utility of DL in interpreting complex embryological data from time-lapse imaging.
• Future research should focus on standardizing data collection, exploring diverse DL architectures, and validating models for improved clinical application and patient outcomes.