Akademik Veri Yönetim Sistemi
PGDIS - 9 th International Conference on Preimplantation Genetics, Berlin, Almanya, 10 - 13 Nisan 2022, ss.39, (Özet Bildiri)
Introduction: Previous studies using manual annotation of images from Time Lapse Monitoring (TLM) incubators indicated that euploid embryos show blastulation earlier than aneuploid embryos. A deep learningbased Artificial Intelligence (AI) tool, CHLOE (Fairtility), has been demonstrated to simultaneously evaluate large numbers of embryos using TLM images. AI tools, such as CHLOE (Fairtility), eliminate inter-operator variation by automatically extracting standardized time-lapse annotations and allowing more robust and efficient embryo evaluation. The aim of the study was to compare morphokinetic behaviors of euploid and aneuploid blastocysts automatically annotated using AI.
Material&Methods: Single centre study analyzing a retrospective cohort that took place between 2019-2020, at Istanbul Memorial Hospital, ART and Reproductive Genetics Center. Time-lapse videos of 1231 blastocysts (486 euploid, 745 aneuploid) which underwent PGT-A analysis, with 301 euploid single embryo transfers (SETs). CHLOE’s morphokinetic assessments (tPNa,tPNf,t2,t3,t4,t5,t6,t7,t8,t9,tM,tSB,tB,tEB), blastocyst score (calculated at 30hpi) and implantation score were compared using Student’s t-test (SPSS). CHLOE blastocyst and implantation score efficacy of prediction of ploidy and clinical outcomes was quantified using the metric AUC.
Results: When annotated using AI, the average time (Mean hours post insemination ±Standard deviation (SD)) for t4 (39.4±6.5 vs 38.7±4.7,p=0.02), t5 (50.1±7.7 vs 49.0±6.7,p=0.012), t6 (53.0±7.3 vs 52.0±6.1,p=0.018), tSB (98.2±7.7 vs 96.3±7.1,p<0.001), tB (106.1±7.6 vs 103.8±7.2, p<0.001) and tEB (113.1±6.5 vs 109.6±7.6, p<0.001) were significantly later in aneuploid embryos compared to euploids. Implantation score was significantly higher in euploid embryos compared to aneuploids (0.76±0.25 vs 0.67±0.27, p<0.001), and was predictive of euploidy with an AUC (95% CI) of 0.621 (0.589-0.651,p<0.001). Blastulation score was similar (0.96±0.18 vs 0.94±0.21, p>0.05) and was not predictive of euploidy (AUC (95% CI) = 0.529 (0.497-0.562), p>0.05). Following euploid SETs (n=301), blastulation and implantation scores were not predictive of pregnancy (blastulation score; AUC(95% CI)=0.541 (0.463-0.619), p>0.05 and implantation score; AUC(95% CI)=0.532 (0.450-0.613), p>0.05), clinical pregnancy (blastulation score; AUC(95% CI)=0.564 (0.484-0.644), p>0.05 and implantation score; AUC(95% CI)=0.535 (0.452-0.617), p>0.05) and live births (blastulation score; AUC(95% CI)=0.479 (0.391-0.567), p>0.05 and implantation score; AUC(95% CI)=0.472 (0.382-0.562), p>0.05).
Conclusion: AI-based tools have the potential of increasing consistency, efficiency and efficacy of embryo
evaluation. Although there is a need to validate AI tools for ploidy detection before their incorporation into clinical
practice, the information on quantitative and qualitative morphokinetics and predictors such as blastulation,
implantation scores that AI tools such as CHLOE provide, seem very promising for embryo selection and ploidy
prediction for PGT-A analysis.