Comparison of the deep learning tool iDAScore® and the current annotation model, KIDScore™ D5, for finding the embryo with best chance of pregnancy after elective single blastocyst transfer – a randomized multi-centre prospective pilot study

• Published in: Reproductive biomedicine online 2024-05, Vol.48, p.104025, Article 104025
• Main Authors: Bohlin, T., Kurkkio, E., Adolfsson, E., Berntsen, J., Johansen, M. Nygård, Apter, S., Caisander, G., Kristensen, K., Schmidt, J.
• Format: Article
• Language: English
• ISSN: 1472-6483
• DOI: 10.1016/j.rbmo.2024.104025

This study aimed to compare two embryo decision support tools from Vitrolife, iDAScore® and KIDScore™ D5, using EmbryoScope for embryo culture and assessment. The deep learning tool iDAScore® (Intelligent Data Analysis for Embryo Evaluation) has learned to rank embryos according to their likelihood of implantation. KIDScore™ D5 uses information on selected morphokinetic events annotated by the user in the software to calculate a score that reflects the implantation potential of embryos. In this prospective multi-centre randomized controlled study, the embryo decision tools are compared regarding clinical pregnancy rate as the primary endpoint. Patients undergoing IVF/ICSI treatments at two Swedish IVF clinics were asked to participate by signing informed consent. After IVF/ICSI, embryos were cultivated in EmbryoScope following each clinic´s practise. On day 5, an embryologist scored all normally fertilised embryos by visual assessment according to the Gardner and Schoolcraft criteria using the EmbryoScope software. If ≥ 2 good quality blastocysts (≥3BB) were available, the cycle was randomized to the control group and assessed with KIDScore™ D5 (n = 62) or the study group and assessed with iDAScore® (n = 64). The highest scored blastocyst was transferred and surplus good quality blastocysts were frozen. If a freeze-all strategy was decided, the first frozen embryo transfer was performed with the highest ranked embryo according to group allocation. Elective single blastocyst transfer resulted in similar rates of positive hCG-results (KIDScore D5 64.5% vs iDAScore® 59.4%, p = 0.59), and the same clinical pregnancy rates, 50.0 % in both groups, p = 1. Clinical pregnancy was defined as ultrasound evidence of an intrauterine pregnancy with a fetal heartbeat at > 6 weeks of gestation. See Table 1 for details. Embryos selected using the deep learning tool iDAScore® resulted in the same pregnancy rate and clinical pregnancy rate as KIDScore™ D5. Since iDAScore® is independent of user annotation and hence independent of the embryologist´s knowledge, expertise and workload, the embryo selection is less subjective. Using iDAScore® is timesaving and hence could benefit clinics with a high workload or with less experienced embryologists.