Abstract
Dynamic Full-Field Optical Coherence Tomography (D-FF-OCT) is a non-invasive imaging technique that provides high-resolution full-field images of kidney biopsies within minutes. However, its clinical adoption remains limited because pathologists are more familiar with conventional histopathology, whose sample preparation requires several hours. In this work, we investigate unpaired D-FF-OCT-to-histopathology-like image translation using deep generative models, with the aim of providing faster access to histopathology-like visualizations. We compare UVCGANv2, a CycleGAN-based framework, with UNSB, a bridge-based generative image translation model, and analyze the influence of D- FF-OCT preprocessing and input-channel configurations. UVCGANv2 produced more structurally coherent and visually plausible histopathology-like images than UNSB in our experiments. We found that FID/KID rankings and cycle-reconstruction losses were not always aligned with expert assessment of structural fidelity. These findings highlight the need for structure-aware evaluation and constraints in unpaired virtual histopathology.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DGM4MICCAI_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=~Nasser_Dandana1
BibTex
@InProceedings{DanNas_Unpaired_MICCAISAT2026,
author = { Dandana, Nasser AND Ansart, Manon AND Legendre, Mathieu AND Bard, Patrick AND Guillet, Christophe AND Paindavoine, Michel AND Maldiney, Thomas AND Merienne, Frédéric},
title = { { Unpaired D-FF-OCT-to-Histopathology Translation for Rapid Kidney Biopsy Visualization } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}