Improving Frozen Kidney Biopsy Interpretation Using DUET-Generated Virtual PAS Stains
Authors:
Preprint
Abstract
Frozen section biopsies stained with Hematoxylin & Eosin (H&E) are standard for assessing donor kidneys but are often interpreted by general pathologists with limited renal expertise, and hindered by freezing artifacts and poor tissue morphology. There is currently no rapid, cost-effective, and easily deployable method for generating virtual special stains and quantifying clinically significant histopathological features on frozen section H&E slides. This study evaluates the utility of DUET-generated virtual Periodic acid-Schiff (vPAS) stains produced from H&E-stained frozen kidney biopsies. DUET produces virtual PAS by combining pixel-registered brightfield and fluorescence images from the frozen section H&E-stained slide, overlaying extracted collagen masks to generate the virtual stain. Renal and general pathologists evaluated interstitial fibrosis/tubular atrophy (IF/TA), inflammation, and arteriosclerosis on 29 kidney biopsies, comparing their assessments using H&E images alone versus using both H&E and virtual PAS images. Among general pathologists’ evaluations, ICC scores from the H&E to virtual PAS stain increased for IF/TA percentage, inflammation percentage, and arteriosclerosis number. For renal pathologists’ evaluations, ICC scores from the H&E to virtual PAS stain increased for the percentage of sclerotic glomeruli, IF/TA percentage, inflammation percentage, and number of arteriosclerotic lesions. No improvement in ICC score was observed for arteriolar hyalinosis in either pathologist’s group. We have demonstrated that the use of frozen virtual PAS stains enables higher consistency among pathologist evaluations when compared to the use of frozen section H&E alone for various metrics used to assess donor kidney viability. This approach has the potential to reduce diagnostic variability, improve transplant decision-making, and optimize donor kidney utilization.Keywords
This publication has 24 references indexed in Scilit:
- Virtual histological staining of unlabeled autopsy tissueNature Communications, 2024
- Virtual Staining of Nonfixed Tissue HistologyLaboratory Investigation, 2024
- Automated assessment of glomerulosclerosis and tubular atrophy using deep learningComputerized Medical Imaging and Graphics, 2021
- Computational normalization of H&E-stained histological images: Progress, challenges and future potentialArtificial Intelligence in Medicine, 2019
- An innovative neural network framework to classify blood vessels and tubules based on Haralick features evaluated in histological images of kidney biopsyNeurocomputing, 2017
- Nondestructive Diagnosis of Kidney Cancer on 18-gauge Core Needle Renal Biopsy Using Dual-color Fluorescence Structured Illumination MicroscopyUrology, 2016
- Donor kidney biopsies: pathology matters, and so does the pathologistKidney International, 2014
- The reproducibility and predictive value on outcome of renal biopsies from expanded criteria donorsKidney International, 2014
- Practice guidelines for the renal biopsyLaboratory Investigation, 2004
- ROLE OF DONOR KIDNEY BIOPSIES IN RENAL TRANSPLANTATIONTransplantation, 2001