P-148 Genetic Algorithm Outperforms Viability Algorithm in Differentiating Euploid and Aneuploid Embryos: A Case for Genetic Precision in AI-Based Evaluation
C Giménez Rodríguez, Sandra Marqueño, A Del Arco, Lorena Bori, B Caparrós, José Bellver, Marcos Meseguer · Human Reproduction · 2025
Abstract Study question How does a genetic algorithm outperform a conventional viability Artificial Intelligence (AI) algorithm in distinguishing euploid from aneuploid embryos, particularly in extreme value ranges? Summary answer The genetic algorithm identifies significant differences between euploid and aneuploid embryos, emphasizing the limitations of viability-based evaluations and providing a non-invasive alternative to PGT. What is known already While conventional viability algorithms assess morphological and developmental quality, they often fail to detect genetic abnormalities, leading to potential misclassification of aneuploid embryos as viable. Genetic abnormalities are a key determinant of implantation success and pregnancy outcomes, making accurate detection essential. Current PGT methods, although effective, are invasive, expensive, and require embryo biopsy. AI-driven algorithms present an innovative, non-invasive approach to embryo evaluation by analyzing video data to predict genetic status, potentially improving embryo selection without the need for physical intervention. Study design, size, duration A total of 706 embryos from PGT cycles were analyzed using two AI-based algorithms: a viability algorithm and a genetic algorithm. The study was conducted from January 2024 to December 2024, with analyses focusing on quartile-based and cutoff value evaluations to assess algorithm accuracy and performance. Participants/materials, setting, methods Embryos were classified as euploid or aneuploid based on PGT and standard genetic testing. Videos of whole embryo development were analyzed using EMA platform which includes a viability algorithm (AIVF Day-5 Algorithm, scoring range: 0-10) and a genetic algorithm (AIVF Genetic Algorithm, scoring range: 0-100). Performance was evaluated using statistical comparisons (chi-square tests), quartile analysis, and cutoff thresholds (50, 80). The algorithms’ ability to predict euploidy rates was measured against established genetic testing results. Main results and the role of chance The viability algorithm showed no significant differences between euploid and aneuploid embryos (mean AIVF Day-5 Algorithm score for euploids: 4.42 ± 1.98; for aneuploids: 4.29 ± 2.16; p = 0.398), underscoring its limitations in identifying genetic abnormalities. Conversely, the genetic algorithm demonstrated significant differences (mean AIVF Genetic Algorithm score for euploids: 45.24 ± 17.55; for aneuploids: 38.42 ± 16.73; p 76 (90.9%) (p 80, the euploidy rate reached 100%. Using a 50-point cutoff, embryos scoring ≤50 were identified as 53.5% euploid, compared to 71.6% for those scoring >50. These results demonstrate the genetic algorithm’s capability to distinguish embryos based on genetic potential, surpassing morphology-based evaluations in accuracy and precision. Limitations, reasons for caution Algorithm performance depends on video quality, patient-specific genetic variability, and the accuracy of the training data. Further validation across diverse populations and clinical settings is required to ensure generalizability. Wider implications of the findings The genetic algorithm identifies euploid embryos, overcoming viability-based limitations. By leveraging genetic markers, it offers a precise, non-invasive embryo selection tool, reducing the need for invasive techniques. It may also prioritize transferring embryos with lower aneuploidy risk in patients not requiring PGTA, enhancing efficiency and personalization in assisted reproductive treatments. Trial registration number No