EUVIMEDEUVIMED entdecken
Crossref

L26/P-079 Validation of an AI-based computer vision pipeline for quantitative paper-based semen diagnostics

D. Marcu, L. Sibielak, S. Immler

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.415 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von D. Marcu und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question Can an AI-based computer vision pipeline reliably standardise smartphone-captured user images for quantitative analysis of paper-based semen diagnostic tests under real-world conditions? Summary answer The validated computer vision pipeline reliably corrects imaging variability and enables robust quantitative analysis of paper-based semen diagnostics using user-captured smartphone images. What is known already Paper-based lateral flow diagnostics offer an accessible approach to decentralised male fertility testing. However, quantitative interpretation of colourimetric semen assays remains challenging. Smartphone-acquired images are highly variable due to differences in lighting, camera hardware, orientation, distance, and user handling. This variability can compromise signal detection, analytical precision, and reproducibility, limiting clinical confidence in at-home or decentralised semen testing. Robust image standardisation and quality control are therefore essential to ensure reliable biomarker quantification and clinically meaningful results. Study design, size, duration This technical validation study analysed 1200 smartphone-captured images of paper-based semen diagnostic tests prospectively generated under controlled, simulated user conditions between January and March 2026. Image acquisition was designed to reflect expected at-home use and incorporated predefined variation in orientation, illumination intensity, colour temperature, motion blur, and background noise. The dataset was structured to include both typical and edge-case imaging conditions relevant to real-world user behaviour. Participants/materials, setting, methods Images of paper-based semen diagnostic cassettes were captured using consumer smartphones in a simulated at-home environment. An AI-based computer vision pipeline performed fiducial marker detection with confidence scoring, geometric perspective correction, and two-stage grayscale and colour normalisation to mitigate illumination and device-related variability. Quantitative extraction of control and test line intensities was conducted using reference-based colour mapping. Quality metrics included marker confidence, background uniformity, signal-to-noise ratio, colour accuracy, and intensity tolerance. Main results and the role of chance All images captured within expected user conditions were successfully processed, with no pipeline failures observed. Automated computer vision–based perspective correction consistently aligned test regions to standardised coordinates, correcting for rotation and tilt. Illumination-related variability was reduced through grayscale and colour normalisation, enabling stable signal extraction. Accepted images demonstrated mean marker confidence scores ≥0.80, background coefficient of variation ≤15%, signal-to-noise ratios ≥3:1, and colour accuracy within Delta E ≤ 5.0. Quantitative control and test line intensity measurements were reproducible across typical user conditions, achieving 100% pass rates. Challenging edge-case conditions, including extreme rotation, increased blur, and substantial lighting variation, achieved 85–90% pass rates, indicating robustness beyond normal usage scenarios. Limitations, reasons for caution This study evaluated technical performance under simulated user conditions rather than real-world clinical deployment. Clinical outcome correlations, algorithm generalisability across populations, and user adherence were not assessed and require further prospective validation. Wider implications of the findings AI-based computer vision enables scalable, standardised interpretation of paper-based semen diagnostics, supporting decentralised male fertility testing at population level. This approach may reduce diagnostic delays, enable longitudinal monitoring, and extend access to reproductive health assessment in non-urban, resource-limited, and digitally connected care pathways. Trial registration number No

Zitieren

APA

Marcu, D., Sibielak, L., & Immler, S. (2026). L26/P-079 Validation of an AI-based computer vision pipeline for quantitative paper-based semen diagnostics. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.415

Vancouver

Marcu D, Sibielak L, Immler S. L26/P-079 Validation of an AI-based computer vision pipeline for quantitative paper-based semen diagnostics. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.415

BibTeX
@article{Marcu2026,
  author = {Marcu, D. and Sibielak, L. and Immler, S.},
  title = {L26/P-079 Validation of an AI-based computer vision pipeline for quantitative paper-based semen diagnostics},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.415}
}
Kostenlos starten →