Application of Hyperspectral Imaging Technique and Artificial Intelligence on Quality Prediction of Embryonic Cells

Abstract Background Assessing embryo quality is crucial for improving in vitro fertilization (IVF) success rates, yet traditional grading methods rely on subjective evaluations by embryologists, leading to potential bias. Hyperspectral imaging (HSI) offers a non-invasive approach to embryo assessment by capturing detailed spectral information beyond conventional grayscale images. Objective This study aims to investigate the effectiveness of HSI in predicting embryo quality and to compare its classification performance with grayscale images, using maternal human chorionic gonadotropin β (β-HCG) levels as labels. Methods HSI of embryos obtained from National Taiwan University Hospital were analyzed, with β-HCG levels categorized into three groups. Machine learning models were trained and evaluated using confusion matrices, Cohen’s kappa coefficient, and receiver operating characteristic (ROC) curves to assess classification performance. Results The findings indicate that HSI significantly improves classification accuracy compared to grayscale images, particularly in identifying embryos with medium β-HCG concentrations. HSI-based models outperformed random classification, demonstrating enhanced predictive capability in embryo quality assessment. Conclusions This study is the first to use β-HCG levels as classification labels, reducing subjective bias in embryo evaluation. The results highlight the potential of HSI as an objective and reliable tool for embryo quality assessment, paving the way for improved decision-making in assisted reproductive technologies.