Robust Cell Segmentation for Size Distribution Estimation via Synthetic‐Data Training
Biotechnology Journal · 2026-07 · 21 (7)
DOI: 10.1002/biot.70272 ↗Kurz gefasst
Fachartikel in der Zeitschrift Biotechnology Journal, online erschienen am 09.07.2026. Verfasst von Han Bit Kim und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Chemical and Biological Engineering, Seoul National University, Seoul, Republic of Korea. Hauptthemen laut MEDLINE: Image Processing, Computer-Assisted, Polyhydroxyalkanoates, Cell Size. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Seoul National University.
Abstract (Originalsprache)
Polyhydroxyalkanoates (PHAs) are biodegradable and biocompatible plastics, yet large‐scale production remains limited by costly batch operations and the lack of online analytical tools. Monitoring cell size distribution serves as a strong surrogate for intracellular PHA content, but training robust cell segmentation models for industrial bioprocesses requires extensive manual annotation, which is highly labor‐intensive and impractical for densely populated microscopy images. To address this critical bottleneck, an annotation‐free cell segmentation and size estimation pipeline tailored for real time monitoring is presented. Rather than relying on architectural modifications, this framework introduces a system‐level automated training strategy: individual cells are automatically extracted from diluted‐sample microscopy images using edge enhancement and rule‐based binarization, then augmented and composited onto heterogeneous backgrounds to emulate the texture of undiluted, dense cultures. This fully synthetic data generation eliminates the need for manual labeling while enabling robust training of instance segmentation models. When implemented using a Mask R‐CNN backbone, the proposed pipeline consistently tracks flow cytometry forward‐scatter (FSC) distribution trends under diverse imaging conditions and achieves higher correlation with FSC data compared to Cellpose and CellSAM, representative foundation models for cell segmentation. This annotation‐free methodology provides a practical and reliable automated online monitoring solution for intelligent PHA manufacturing.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- Department of Chemical and Biological Engineering, Seoul National University, Seoul, Republic of Korea.
- CJ BIO Research Institute, CJ CheilJedang, Suwon-Si, Gyeonggi-do, Republic of Korea.
ORCID-Profile: Han Bit Kim, Chaeeun Lee, Naeun Lee, Hyeongseok Han, Chanhun Park, Moo Sun Hong
Förderung
- Seoul National University
Zitieren
APA
Kim, H. B., Lee, C., Lee, N., Han, H., Park, C., & Hong, M. S. (2026). Robust Cell Segmentation for Size Distribution Estimation via Synthetic‐Data Training. Biotechnology Journal, 21(7). https://doi.org/10.1002/biot.70272
Vancouver
Kim HB, Lee C, Lee N, Han H, Park C, Hong MS. Robust Cell Segmentation for Size Distribution Estimation via Synthetic‐Data Training. Biotechnology Journal. 2026;21(7). doi:10.1002/biot.70272
BibTeX
@article{Kim2026,
author = {Kim, Han Bit and Lee, Chaeeun and Lee, Naeun and Han, Hyeongseok and Park, Chanhun and Hong, Moo Sun},
title = {Robust Cell Segmentation for Size Distribution Estimation via Synthetic‐Data Training},
journal = {Biotechnology Journal},
year = {2026},
volume = {21},
number = {7},
doi = {10.1002/biot.70272}
}
EUVIMED