Fertility & Reproduction (Dec 2023)

#273 : Comparative Analysis of Pregnancy Predictive Potential Using the Deep Learning Blastocyst Scoring Model Calculated from the Single Focus Blastocyst Image and Time-Lapse Image Sequences

  • Masashi Shioya,
  • Tatsuya Kobayashi,
  • Shun Nakano,
  • Tomoharu Sugiura,
  • Miki Okabe-Kinoshita,
  • Maki Fujita,
  • Keiichi Takahashi

DOI
https://doi.org/10.1142/S2661318223744107
Journal volume & issue
Vol. 05, no. 04
pp. 698 – 698

Abstract

Read online

Background and Aims: Artificial Intelligence (AI) based embryo evaluation may objective and reduce inter-observer disagreement. Most commercially available AI models use blastocyst morphokinetics from time-lapse monitoring (TLM). Lifewhisperer (LW, FUJIFILM) is an AI-scoring model that uses a single-focus blastocyst image at a single-time point. LW may be more acceptable to many facilities than the TLM model as it does not require a time-lapse incubator. This study analyzed the predictive ability of LW for pregnant embryos and compared it to an AI model based on morphokinetics (iDAscore, Vitrolife) to assess their usefulness. Methods: This retrospective single-center study analyzed 666 vitrified-warmed blastocyst transfer cycles from January 2019 to December 2022. Inclusion criteria were blastocysts that developed within 5 days from insemination (maximum 120 h) using EmbryoScope+ (Vitrolife), intracytoplasmic sperm injection (ICSI) cycle, and a single-blastocyst transfer cycle. LW score (1.0–9.9) was calculated from single-focus blastocyst images taken before vitrification, while iDAScore (1.0–9.9) was calculated from time-lapse image sequences taken at 11 focal every 10 min after ICSI. We calculated the odds ratio for pregnancy and ROC analysis to compare these models. Results: The mean score was higher in pregnant blastocysts than in non-pregnant blastocysts in both models (LW: 7.47 vs. 6.72, p<0.01; iDAScore: 8.50 vs. 8.04, p<0.01, Willcoxon-test). The odds ratio (95% confidence interval) for pregnancy was 1.19 (1.10–1.28) for LW and 1.48 (1.27–1.74) for iDAScore, with the odds for clinical pregnancy increasing with higher scores (p<0.01). ROC curve analysis showed that the AUC values were 0.60 for LW and 0.62 for iDAScore. There was no significant difference in the predictive ability for pregnant blastocysts (p=0.51). Conclusions: LW allowed objective score calculation from a single focus at single time point blastocyst images without TLM. Moreover, in day 5 blastocyst, LW and iDAscore had comparable predictive pregnancy ability.