L26/P-064 Hyperspectral Sperm Imaging Guides Machine-Learning–based Prioritisation of Spermatozoa Associated with Top-Quality Blastocyst Development
Human Reproduction · 01.07.2026 · 41 (Supplement_1)
DOI: 10.1093/humrep/deag083.400 ↗Kurz gefasst
Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von R.M. Pacheco-Rendón und 12 weiteren Autor:innen.
Abstract (Originalsprache)
Study question Can hyperspectral sperm imaging (HS) combined with machine-learning (ML) assign predictive success scores to individual spermatozoa reflecting their likelihood of resulting in top-quality blastocysts? Summary answer HS-derived biochemical signatures enable AI-based scoring of single spermatozoa of achieving top-quality blastocysts, potentially increasing reproductive outcomes when utilized for sperm selection. What is known already Sperm selection in assisted reproduction relies predominantly on subjective assessment of morphology and motility, providing limited predictive value for downstream embryological outcomes. Hyperspectral imaging enables non-invasive acquisition of single-cell biochemical fingerprints. Applied to sperm selection, this approach offers a framework for objective and reproducible prioritisation within clinically viable sperm populations, compatible with ICSI workflows. Our group already reported promising results using HS-based precision algorithms to predict fertilization and embryo development, proving that through its guidance may increase standard laboratory success rates. However, remains essential to evaluate other key reproductive outcomes closer to of reproductive success, such as top-quality blastocyst development. Study design, size, duration Prospective study embedded in routine clinical practice including 93 patients undergoing ICSI with either own or donor, fresh or vitrified oocytes, and autologous fresh ejaculate sperm. All ICSI procedures followed the standard practice; with the only addition being hyperspectral imaging after sperm immobilisation and prior microinjection. Hyperspectral images were acquired from 716 immobilised spermatozoa, of which 699 were used for model development. Data were collected January 2022-August 2024. Participants/materials, setting, methods Hyperspectral images were acquired using a 4200M Microscope system coupled to an IX73-inverted microinjector microscope and TruScope v.1.17 software. Traceability of gametes and embryos was maintained. Top-quality blastocyst was defined as inner cell mass and trophectoderm grades≥B. Rates were calculated per MII-oocyte microinjected, including non-fertilised oocytes and embryos that failed to reach the blastocyst stage. Remaining embryos were classified as poor quality outcomes. Hyperspectral sperm features and clinical variables were used to train XGBoost models. Main results and the role of chance Hyperspectral images were summarised at single-spermatozoon level by aggregating spectra across segmented pixels, applying quality filtering, and reducing dimensionality using principal component analysis (PCA). The resulting components, combined with routine clinical variables, were used to train an XGBoost classifier. Data were split at patient-level into training (n = 459 images), validation (n = 135), and independent test (n = 105) sets to prevent data leakage. Candidate models were ranked using a precision-weighted F-score (Fβ, β = 0.25), and the top-performing model was selected. On the independent test set, the model achieved 33.33% precision (Fβ=29.57%), versus a baseline clinical success rate of 18.10%, calculated per microinjected MII-oocyte. At patient level, predicted probabilities showed substantial intra-patient dispersion, with within-patient score ranges (maximum–minimum) spanning 0.14 to 0.59 across patients, indicating clear differentiation among spermatozoa and enabling top-1/top-k ranking of candidate sperm. This prioritisation effect concentrates successful outcomes within a subset of embryologist-preselected spermatozoa. Clinically, model-based ranking has the potential to support and improve final microinjection decisions by incorporating biological information beyond naked-eye observation, without disrupting standard ICSI workflows. Overall, the enrichment of top-quality blastocyst outcomes exceeds baseline expectations and is unlikely explained by chance alone. Limitations, reasons for caution This non-interventional study evaluated model performance prospectively within embryologist-preselected sperm populations. Results are therefore conditional on prior clinical selection and derived from a single-centre cohort with limited sample size. The clinical impact of implementing model-guided prioritisation, including required workflow adaptations, remains to be prospectively assessed. Wider implications of the findings These findings indicate that hyperspectral imaging combined with ML has the potential to serve as a decision-support strategy for sperm prioritisation within daily routine ICSI workflows. Conditional enrichment of high-probability candidates motivates prospective studies to evaluate its impact on efficiency and clinical outcomes under adapted selection protocols. Trial registration number No
Zitieren
APA
Pacheco-Rendón, R., Lahoz, R. T., Pellicer, D. M., Julia, M. G., Santos, J. D. L., Beltrán, D., Gámiz, P., Escrich, L., Viloria, T., Mossetti, L., Santos, M. D. L., Egea, R. R., & Garrido, N. (2026). L26/P-064 Hyperspectral Sperm Imaging Guides Machine-Learning–based Prioritisation of Spermatozoa Associated with Top-Quality Blastocyst Development. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.400
Vancouver
Pacheco-Rendón R, Lahoz RT, Pellicer DM, Julia MG, Santos JDL, Beltrán D, et al. L26/P-064 Hyperspectral Sperm Imaging Guides Machine-Learning–based Prioritisation of Spermatozoa Associated with Top-Quality Blastocyst Development. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.400
BibTeX
@article{PachecoRendn2026,
author = {Pacheco-Rendón, R.M. and Lahoz, R. Trinidad and Pellicer, D. Muñoz and Julia, M. Gil and Santos, J.M. De los and Beltrán, D. and Gámiz, P. and Escrich, L. and Viloria, T. and Mossetti, L. and Santos, M.J. De los and Egea, R. Rivera and Garrido, N.},
title = {L26/P-064 Hyperspectral Sperm Imaging Guides Machine-Learning–based Prioritisation of Spermatozoa Associated with Top-Quality Blastocyst Development},
journal = {Human Reproduction},
year = {2026},
volume = {41},
number = {Supplement_1},
doi = {10.1093/humrep/deag083.400}
}
EUVIMED