Key points are not available for this paper at this time.
Abstract Study question Comparison of the positive pregnancy predictibility between 2 deep-learning models of iDAScore (iDA-1 and iDA-2) in the same set of embryos. Summary answer The performances (AUC) to predict implantation are similar in both models with slightly higher sensitivity for iDA-2. Scores 6.4 display the best chances for implantation. What is known already The use of artificial intelligence (AI) is dramatically increasing in automatic annotation of the embryos as an aid to embryologists for evaluation and selection of the best implanting embryo. AI-based methods appear to be objective, standardized, and efficient tools for evaluating human embryos. AI-based automatic annotations need high accuracy and reliability to help securely standardizing and automatizing the process. iDAScore (Vitrolife) is a commercial deep-learning model for scoring embryos based on their implantation potential. Two versions of this model are currently available, iDAScore-1 and iDAScore-2. In this study we compare these two versions with the same cohort of embryos. Study design, size, duration Retrospective evaluation by the two models of 814 fresh SET cycles, from 758 patients, during 2020-2023 in a private hospital setting. Blastocyst selection for transfer was based on morphological grading and morphokinetic annotations (KID-Score, Gardner criteria) by experienced embryologists. Both version of iDAScore for each selected embryo was recorded retrospectively for not to influence the decision-making. Participants/materials, setting, methods The ICSIs incubated in Embryoscope + (Vitrolife) followed by fresh SET during the study period were enrolled. All the selected blastocysts for transfer were retrospectively annotated by well-trained experienced embryologist to iDA-1 and iDA-2. Pregnancy outcome was recorded and statistical analyses were performed with SPSS software using appropriate test, ROC analysis and AUC comparison to find predictibility for implantation and cut-off scores for poorest and best chances of pregnancy.Main results and the role of chance The mean female age was 36.43±0.14. The 814 SET achived 349 pregnancies and resulted in 199 live birth. The mean iDAScore in positive pregnancy, whatever the age, was different between the IDAScore versions. In iDA-1 the mean pregnancy score was 8.54 ± 0.98 and in iDA-2 the score was 6.2 ± 1.9. The mean iDAScore for ongoing pregnancy, regardless of age, was 9 ± 0.64 and 7.2 ± 1.51 and for live birth 8.8 ± 0.67 and 6.8 ± 1.74 respectively for each version. Comparing both versions: the AUC for predicting implantation/pregnancies was 0.652 0.645-0.679 for iDA-1 and 0.678 0.659-0.685 for iDA-2, establishing the best cut-off for implantation above 8.8 for iDA-1 and 6.4 for iDA-2. 7 score groups were defined to study pregnancy probability (1.1-2.9, 3-4.9, 5, 6, 7, 8, 9) and our results showed that blastocysts with iDA-1 6 resulted in 7.1% pregnancy per fresh transfer and with iDA-2 3 resulted in 15% pregnancy per transfer. Chances of pregnancy and live birth were equivalent in scores 6-8.7 for iDA-1 and 3-6.3 for iDA-2: respectively 31.6% vs 31.6% and 19.7% vs 20.4%. Limitations, reasons for caution The main limitation of the study was the small size of the sample. However, our findings are concordant with the largest validation cohort of iDAScore-2. Wider implications of the findings iDA-2 appears to have a better pregnancy predictibility compared to the first version. However, the AUCs are quite comparable between the two versions. Setting the cut-off scores under which or above which the pregnancy chances are minimal or maximal is crucial for better management of patients communication and clinical decision-making. Trial registration number not applicable
Garcia et al. (Mon,) studied this question.