Using deep learning to predict the sex of human embryos

Area of Science:

• Developmental Biology
• Genetics
• Artificial Intelligence

Background:

• Sex differences in human pre-implantation development are not well understood.
• Previous observational studies have yielded inconclusive results.
• Early embryonic development is crucial for understanding developmental abnormalities.

Purpose of the Study:

• To investigate if birth sex influences early human embryo development dynamics.
• To determine if subtle sex-related differences exist in pre-implantation embryos.
• To explore the potential of AI in predicting embryonic sex.

Main Methods:

• Analysis of 515 time-lapse human embryo movies.
• Combined manual annotation and deep learning approaches.
• Training and testing a deep learning model for sex prediction.

Main Results:

• Manual assessment found no reliable developmental timing differences between sexes.
• A deep learning model achieved 61% statistically significant sex prediction accuracy.
• The period after the eight-cell stage (around day 3) was identified as critical for prediction.

Conclusions:

• Subtle sex-related differences in human embryogenesis may emerge by day 3.
• Deep learning can identify early sex-related developmental dynamics.
• Findings may aid in understanding embryo failure, skewed sex ratios in IVF, and developing non-invasive detection tools for abnormalities.