JBRA Assisted Reproduction 2025;29(Suppl.2 SBRA 2025):28
Poster Presentation
29^th Annual Congress of the SBRA. São Paulo/SP - Brazil, 2025
doi: 10.5935/1518-0557.20263552

P-016. Artificial intelligence in thawed blastocyst transfer cycles: superior pregnancy prediction than conventional morphological classification

Bruna Campos Galgaro¹, Luiza da Silva Rodrigues¹, Julia Janis Santos¹, João Sabino Lahorgue da Cunha Filho¹

¹Insemine - Porto Alegre - RS - Brasil

Objective: This study aimed to evaluate whether EmbryoAID artificial intelligence is superior to traditional morphology assessment in predicting pregnancy of warmed single blastocyst embryo transfer.
Methods: This case-control study was conducted at a single centre, including day five warmed single blastocyst transfer cycles from January to November 2024, paired 1:1 for bHCG test results. All blastocysts were vitrified and thawed using the same protocol (Ingamed, Brazil) and placed in a low-oxygen incubator (MINC, COOK) until embryo transfer. Two embryologists performed a morphological assessment of the blastocyst using the Gardner scale three to four hours after thawing. Blastocysts with only A and/or B grades for inner cell mass and trophectoderm were classified as high-quality embryos. Images were analysed using AI software (EmbryoAID v1.0, MIM Fertility) and scores ranged from 0.0 to 10.0 according to embryo potential for implantation. Clinical factors such as age, anti-mullerian hormone (AMH) and body-max index (BMI) of patients were analysed. We performed t-test, chi-square test and logistic regression on JASP software. Data were expressed as mean±standard deviation and considered significant when p<0.05.
Results: The study included one hundred and ten patients. Age, AMH and BMI were not statistically different between patients with positive and negative bHCG results (p>0.05). Blastocysts with a Gardner high-quality classification had enhanced implantation potential (p<0.001), even when analysing only inner cell mass (p<0.001) or trophectoderm quality (p=0.017). In contrast, expansion grade was not associated with bHCG outcome (p=0.280). As expected, Gardner high-quality blastocysts had an EmbryoAID score higher than the poor-quality ones [6.43±1.95 and 3.01±1.84 (p<0.001)]. Moreover, the EmbryoAID score was also increased (p <0.001) in the pregnancy group (6.01±2.29) than in the non-pregnant group (4.24±2.47). Furthermore, the logistic regression analysis showed that only the EmbryoAID score positively predicted pregnancy (p=0.049), controlling for Gardner classification, inner cell mass, trophectoderm quality, age, BMI and AMH.
Conclusion: The EmbryoAID score was more predictive of pregnancy outcomes than traditional morphological classification in warmed single blastocyst transfer cycles. While further validation through a randomised clinical trial is still necessary to confirm these results, our findings support the application of this automated embryo selection in the embryology lab.