Skip to main navigation menu Skip to main content Skip to site footer

Vol. 11 No. 1 (2025)
Special Issue: Proceedings of CVIS 2025

Articles

Deep Learning-Based Nuclei Segmentation for Label-Free Histology Using Photon Absorption Remote Sensing Microscopy

https://doi.org/10.15353/jcvis.v11i1.10011

Published 2026-04-01

  • Gloria J. So
    • ,
  • Umar Ali
    • ,
  • James Simmons
    • ,
  • James Tweel
    • ,
  • Benjamin Ecclestone
    • ,
  • Parsin Reza

Abstract

As pathology workflows shift toward digitization, deep learning is increasingly being used for tasks such as nuclei segmentation. Photon Absorption Remote Sensing (PARS) is a label-free imaging technique that provides nuclear and extranuclear contrast. PARS images can be used to generate virtual H\&E (VHE) images that emulate conventional H\&E staining. Although many nuclei segmentation algorithms exist for conventional histology, none have yet been applied to PARS. We evaluated three state-of-the-art deep learning models for nuclei segmentation: StarDist, Cellpose, and DeepCMorph. We demonstrated that pretrained models show consistent interpretation across H\&E and VHE, with a mean percent error of 1.54\% in nuclear count. For the first time, segmentation models were trained directly on PARS images. The best model produced masks that preserved cell morphology, with errors of only -3.96\% in nuclear count, -0.29\% in intercellular distance, and 1.53\% in nuclear area. These results show PARS as an effective technique for automated nuclear analysis in label-free histopathology.
PDF