Deep learning neural network analysis of human blastocyst expansion from time-lapse image files

Abstract

Research questionCan artificial intelligence (AI) discriminate a blastocyst's cellular area from unedited time-lapse image files using semantic segmentation and a deep learning optimized U-Net architecture for use in selecting single blastocysts for transfer? DesignThis platform was retrospectively applied to time-lapse files from 101 sequentially transferred single blastocysts that were prospectively selected for transfer by their highest expansion ranking within cohorts using a 10 h expansion assay rather than standard grading. ResultsThe AI platform provides expansion curves and raw data files to classify and compare blastocyst phenotypes within both cohorts and populations. Of 35 sequential unbiopsied single blastocyst transfers, 23 (65.7%) resulted in a live birth. Of 66 sequential single euploid blastocyst transfers, also selected for their most robust expansion, 49 (74.2%) resulted in live birth. The AI platform revealed that the averaged expansion rate was significantly (P = 0.007) greater in euploid blastocysts that resulted in live births compared with those resulting in failure to give a live birth. The platform further provides a framework to analyse fragmentation phenotypes that can test new hypotheses for developmental regulation during the preimplantation period. ConclusionsAI can be used to quantitatively describe blastocyst expansion from unedited time-lapse image files and can be used to quantitatively rank-order blastocysts for transfer. Early clinical results from such single blastocyst selection suggests that live birth rates without biopsy may be comparable to those found using single euploid blastocysts in younger, good responder patients.