A multiday machine learning framework based on improved-NEQsi for predicting embryo quality and pregnancy outcomes in...

Current embryo selection is limited by fragmented morphological assessments, high resource demands of image-based Artificial Intelligence, and inadequate integration of clinical factors. This study aims to establish a unified multi-day assessment framework by developing an Improved Numerical Embryo Quality scoring index (INEQsi) and to provide a lightweight and robust tool for predicting embryo quality and pregnancy outcomes by integrating these standardized scores with clinical data through machine learning.

This study developed a multidimensional embryo scoring system based on established embryo grading standards and integrated other clinical variables to build a machine learning prediction model. By systematically evaluating and comparing multiple algorithms for predicting embryo quality scores and post-implantation pregnancy outcomes, the best prediction model was ultimately determined.

The random forest model with INEQsi achieved high accuracy in predicting embryo quality for both cleavage-stage (RMSE: 1.26) and blastocyst-stage embryos (RMSE: 1.11-1.13), extending its predictive capability to Day 3. It outperformed NEQsi in pregnancy prediction, with a sensitivity of 0.99-1.00 (95% CI) on Day 5 and 93% accuracy on Day 6.

This study proposed a multitage embryo scoring system and combines it with machine learning to form a lightweight analytical framework. In this single-center retrospective analysis, the approach showed reasonable performance in assessing embryo quality and predicting clinical pregnancy outcomes, suggesting its potential as a supplementary tool in IVF decision-making where advanced imaging systems are not routinely used.