Abstract Study question Can an automatic AI scoring system predict ploidy, live birth and utilization? Are there differences in AI scoring between donor and own gametes? Summary answer CHLOE-EQ Score is directly associated with oocyte quality, ploidy, utilization, live birth, embryo quality, direct uneven cleavage (DUC), blastulation, utilization and selection for transfer. What is known already The integration of AI algorithms, such as CHLOE-EQ, into different Time-lapse systems requires clinical and biological validation. Geri Time-lapse videos have given the embryologists more insight into embryo development. Analyzing this information manually requires time and introduces risk of error. To tackle this issue, AI solutions like CHLOE-EQ (Fairtility) can be used to automatically assess video datapoints. CHLOE-EQ provides an embryo quality score that has been shown to predict embryo viability and ploidy, providing clarity on the underlying biological factors. Before introducing AI tools in clinical practice, it is crucial to confirm their efficacy and validate with clinical data. Study design, size, duration A retrospective cohort analysis was conducted at a private clinic in Spain from April 2021 to November 2022, involving the review of 3196 Geri time-lapse videos with a subset of known ploidy and live birth outcomes. The correlation of CHLOE-EQ score with ASEBIR clinic grading was evaluated. As well as with DUCs, oocyte quality, sperm source, blastulation, utilization, selection for transfer, ploidy and live birth. Participants/materials, setting, methods Geri time-lapse videos were automatically analyzed by CHLOE-EQ (Fairtility). CHLOE-EQ score was assessed in relation to laboratory (ploidy, clinic ASEBIR embryo scoring, utilization, selection for transfer) and clinical outcomes (live birth), as well as between own vs donor gametes (own eggs >40y vs donor eggs and testicular sperm vs donor sperm) using descriptive statistics and t-test. The accuracy of prediction was measured using binary logistic regression (AUC). Main results and the role of chance CHLOE-EQ score was positively correlated with ASEBIR embryo quality (A:8.7±1.9, n = 349 >B:6.8±2.9, n = 470 > C: 5.1±3.0, n = 124 > D:1.3±2.1, n=751; p < 0.05). Non-DUCs had higher CHLOE-EQ Score than DUCs [5.3±3.8, n=1798 vs 1.9±0.38, n=643, p<0.001]. CHLOE-EQ Score was unaffected by the quality of the sperm sample, with similar CHLOE-EQ scores between donor sperm and testicular derived sperm (4.1±3.9, n=335 vs 3.4±4.2, n=56, respectively, NS). Embryos that blastulated (yes vs no: 5.4±3.7, n=1996 vs 0.6±2.1, n=309, p < 0.001), were utilized (7.4±0.28, n=911, vs 1.0±2.1, n = 1309, p < 0.001), selected for transfer (8.7±2.4, n=153 vs 3.3±3.7, n = 2067, p < 0.001), were euploid (7.5±2.5, n=72 vs 6.3±3, n=152, p = 0.001) and resulted in live births (4.4±4.1, n = 332 vs 3.8±4, n = 499, p = 0.02) had a higher CHLOE-EQ score than embryos that did not. CHLOE-EQ Score is higher in embryos derived from oocytes from donors than own eggs, suggesting that oocyte quality affects CHLOE-EQ score (4.0±4, n = 1189 vs 2.7±3.4, n = 356). CHLOE-EQ Score is predictive of utilization (AUC=0.95, n=2220, baseline=41%, p<0.001), euploidy (AUC=0.63, n=224, baseline=32.1%, p=0.003), blastulation (AUC=0.94, n = 2305, baseline=86.6%, p < 0.001) and selection for transfer (AUC=0.89, n = 2220, baseline=41%, p<0.001). Limitations, reasons for caution This is a retrospective single-center study in which embryos for transfer were selected by human embryologists, and forms part of program to validate the responsible integration of AI into clinical practice in each individual clinic. Wider implications of the findings This is the first study presenting the efficacy of prediction of CHLOE-EQ with GERI data. AI tools have the potential to improve consistency, efficiency, and accuracy of embryo assessment and selection. CHLOE-EQ predicts through quantitative and qualitative morphological and morphokinetics information, resulting in more personalized care for each individual embryo. Trial registration number not applicable
Read full abstract