EUVIMEDEUVIMED entdecken
Crossref

ROI Based Modified Reinhard Stain Normalization for Improved Colorectal Cancer Histopathology Image Classification

Shirley Sheeba S., Vijayalakshmi A.

Journal of Computer Science · 01.11.2026 · 22 (11) · 3469-3478

DOI: 10.3844/jcssp.2026.3469.3478 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computer Science, erschienen am 01.11.2026. Verfasst von Shirley Sheeba S. und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Staining is essential to enhance cellular features in histopathological imaging. however, inconsistent or excessive staining degrades digital image quality and adversely impacts deep learning model performance. Variability arising from differences in staining techniques and imaging devices remains a fundamental challenge in computational histopathology. Existing color normalization methods, including the standard Reinhard technique, operate globally across entire images without distinguishing biologically relevant tissue regions from background, limiting their effectiveness in correcting stain variability. This work presents a novel Region of Interest (ROI) based Reinhard color normalization method that incorporates tissue masking and post normalization smoothing. Otsu thresholding combined with morphological operations isolates biologically relevant tissue regions, confining normalization to those areas. Gaussian smoothing is subsequently applied to reduce boundary artifacts introduced during masking, yielding stain corrected images with improved spatial consistency. The proposed method is evaluated on the EBHI-3 histopathology image dataset using a Resnet-18 deep learning classification model. The proposed ROI Reinhard method achieves a classification accuracy of 97%, outperforming the original Reinhard method's accuracy of 95%. Quantitative image quality metrics SSIM and PSNR further confirm that ROI based normalization produces superior stain color uniformity compared to the global approach. These results demonstrate that region focused normalization enhances the reliability of deep learning based histological image classification.

Zitieren

APA

S., S. S., & A., V. (2026). ROI Based Modified Reinhard Stain Normalization for Improved Colorectal Cancer Histopathology Image Classification. Journal of Computer Science, 22(11), 3469-3478. https://doi.org/10.3844/jcssp.2026.3469.3478

Vancouver

S. SS, A. V. ROI Based Modified Reinhard Stain Normalization for Improved Colorectal Cancer Histopathology Image Classification. Journal of Computer Science. 2026;22(11):3469-3478. doi:10.3844/jcssp.2026.3469.3478

BibTeX
@article{S2026,
  author = {S., Shirley Sheeba and A., Vijayalakshmi},
  title = {ROI Based Modified Reinhard Stain Normalization for Improved Colorectal Cancer Histopathology Image Classification},
  journal = {Journal of Computer Science},
  year = {2026},
  volume = {22},
  number = {11},
  pages = {3469--3478},
  doi = {10.3844/jcssp.2026.3469.3478}
}
Kostenlos starten →