O-247 Validation of an AI (Artificial intelligence) model assessing oocyte quality in a single donor program with frozen-thawed oocytes for blastocyst prediction

Abstract

Abstract Study question Can an AI model predict blastocyst development in a specific subset population of frozen-thawed oocytes in a single donor program? Summary answer An AI oocyte quality score is predictive of blastocyst formation of frozen-thawed oocytes using post-ICSI images. What is known already The ability to score oocytes based on quality is of paramount importance. A methodology to categorize oocyte quality continues to elude IVF professionals. CHLOE OQ™ is an AI oocyte-quality (OQ) score designed to analyse a single oocyte Timelapse image and provide an OQ score. Responsible use of AI needs the validation of models across different clinical scenarios. The increasing use of frozen-thawed oocytes in IVF has necessitated to objectively score the developmental potential of this oocytes. The objective of this study was to determine if the morphological characteristics of frozen-thawed oocyte images are predictive of blastocyst formation. Study design, size, duration Retrospective cohort study using time-lapse videos of 1318 post-ICSI frozen-thawed donated oocytes between January 2021 and August 2023, with known blastulation outcomes from a single private clinic. Participants/materials, setting, methods Oocytes were evaluated using OQ Score and placed into categories according to their OQ score. Group A: >0.6, Group B: 0.3-0.59, Group C: 0.001-0.29 and Group D: 0. The correlation between OQ score, embryo quality (EQ) score and blastocyst development was measured with binary logistic regression (AUC) and a 2-sample t-test. Chi-square was performed to determine if there were significant differences in blastocyst development rates across these distinct oocyte quality groups. Main results and the role of chance Oocytes that underwent blastulation exhibited a significantly higher OQ Score (7.0 ± 2.3) compared to those that did not blastulate (5.3 ± 3.0, p < 0.001). The predictive capacity for blastocyst formation was observed for OQ score (AUC= 0.67) and EQ Score (AUC= 0.93), with a baseline blastulation of 66% (n = 1318, p < 0.001). Of interest, as early as Day 3 a Blastocyst prediction score was highly predictive of blastulation (AUC= 0.92, p < 0.001). When assessing a subset of transferred oocytes, OQ Score had a higher predictive ability (AUC=0.71, n = 163, p < 0.001).Oocyte quality scores within OQ score groups showed a direct association with blastocyst formation rate: [Group A 74% (656/878), Group B 56% (130/231), Group C 43% (70/164), Group D 13% (3/23), p < 0.05]. The highest-scoring group demonstrated a five-fold increase in blastulation rate compared to the lowest-scoring group (p < 0.001). Furthermore, the highest-scoring group resulted in the highest proportion of good-quality embryos compared to the lowest-scoring group [78% (456/582) vs 0.19% (1/582), p < 0.001]. Limitations, reasons for caution Retrospective assessment of a single clinic. Results reflect the use of frozen-thawed donated oocytes. Therefore, the proportion of data within the four oocyte quality groups is not equally distributed and is biased towards high quality oocytes. This may impact the predictive ability within the smaller subgroups. Wider implications of the findings Developing an objective methodology to categorize oocyte quality post-retrieval is crucial. It is important to validate AI models in different clinical scenarios, for example in oocyte donor programs using frozen-thawed oocytes. AI is a tool that can be used to predict blastocyst development rates within a frozen donor oocyte program. Trial registration number not applicable