Photon Absorption Remote Sensing Virtual Histopathology: A Preliminary Exploration of Diagnostic Equivalence to Gold-Standard H&E Staining in Skin Cancer Excisional Biopsies

Abstract

Background: Photon Absorption Remote Sensing (PARS) enables label-free imaging of subcellular morphology by observing absorption interactions. Coupled with deep-learning, PARS produces label-free virtual hematoxylin and eosin (H&E) images in unprocessed tissues. This study evaluates the diagnostic performance of PARS-derived virtual H&E images in skin excisions, including squamous cell carcinoma (SCC), basal cell carcinoma (BCC), and normal skin. Methods: Sixteen unstained formalin-fixed paraffin-embedded excisions were PARS-imaged, then stained with chemical H&E dye. Seven fellowship-trained dermatopathologists assessed all 32 images (16 PARS and 16 H&E) in a masked randomized fashion. Example PARS and chemical-H&E whole-slide images are available at the S-BIAD2324 Archive. Results: Diagnostic sensitivity and specificity were comparable between PARS (sensitivity: 96.7% [95% CI: 89.5–100.0%]; specificity: 95.2% [95% CI: 74.7–100.0%]) and H&E (sensitivity: 95.6% [95% CI: 88.2–100.0%]; specificity: 85.7% [95% CI: 50.2–100.0%]). Primary diagnostic agreement was 95.5% (95% CI: 89.5–100.0%; Cohen’s κ = 0.929 [95% CI: 0.869–0.989]), with high inter-rater reliability (Fleiss’ κ = 0.886 [95% CI: 0.751–1.00] for PARS; κ = 0.805 [95% CI: 0.652–0.958] for H&E). Inter-modality subtyping agreement was 91.8% (95% CI: 81.4–100.0%; Cohen’s κ = 0.731 [95% CI: 0.375–1.000]) for SCC and 100% (κ = 1.000) for BCC. For margin confinement, agreement was 92.2% (95% CI: 85.0–99.4%; Cohen’s κ = 0.718 [95% CI: 0.452–0.984]). Dermatopathologists could not reliably distinguish image origin, and diagnostic confidence was statistically equivalent between modalities. Conclusions: Diagnostic performance of PARS virtual H&E images was consistent with established clinical benchmarks, providing encouraging preliminary evidence of diagnostic concordance with conventional H&E staining. While larger validation cohorts are required to establish true clinical equivalence, these initial findings highlight PARS label-free virtual histology as a promising approach to enhance dermatopathology diagnostics by preserving tissue for downstream analysis while supporting AI integration.