Comparative Analysis of Multi-Channel Feature Extraction Using a Modified K-means and PCA for PARS-to-H&E Image Translation

Marian Boktor; Paul Fieguth

doi:10.15353/jcvis.v10i1.10009

Vol. 10 No. 1 (2024)
Special Issue: Proceedings of CVIS 2024

Articles

Comparative Analysis of Multi-Channel Feature Extraction Using a Modified K-means and PCA for PARS-to-H&E Image Translation

https://doi.org/10.15353/jcvis.v10i1.10009

Published 2025-01-31

Marian Boktor
Paul Fieguth

How to Cite

Boktor, M., & Fieguth, P. (2025). Comparative Analysis of Multi-Channel Feature Extraction Using a Modified K-means and PCA for PARS-to-H&E Image Translation. Journal of Computational Vision and Imaging Systems, 10(1), 54–60. https://doi.org/10.15353/jcvis.v10i1.10009

Download Citation

Abstract

Histological staining, particularly H\&E staining, is essential in pathology for visualizing tissue structures, but traditional methods are time-consuming. Photon Absorption Remote Sensing (PARS), a high-resolution microscopy technique, offers a promising alternative by capturing H\&E-like contrasts directly, enabling virtual staining without the need for chemical reagents. However, differentiating biological structures remains challenging for current models. We propose that channel-specific feature extraction could enhance colorization accuracy. This study investigates the effectiveness of modified K-means algorithm and Principal Component Analysis (PCA) for feature extraction in virtual staining. Results reveal that features produced by the K-means approach more effectively isolate tissue-specific structures, leading to improved labeling compared to PCA and conventional PARS channels. This advantage is demonstrated both quantitatively, through higher Structural Similarity Index (SSIM) scores, and visually, with enhanced colorization outcomes.

PDF