Validation of the ViewsML vIHC platform for biomarker classification and intensity binning: A pilot collaboration with Histowiz’s PathologyMap platform
Presenter: Akash Parvatikar, MS;PhD Session: Digital Pathology 3 Time: 4/21/2026 9:00:00 AM → 4/21/2026 12:00:00 PM
Authors
Akash Parvatikar 1 , Patrick Savickas 1 , Chris Lee 1 , Jeffrey Shek 1 , Kenneth To 2 , Christopher Jackson 2 , Lawrence Schobs 2 , Rohan Lyons 2 , Rafay Azhar 2 1 HistoWiz, Long Island City, NY, 2 ViewsML Technologies Inc., Vancouver, BC, Canada
Abstract
The purpose of this study was to validate ViewsML’s Virtual Immunohistochemistry (vIHC) platform, a system that generates virtual biomarker predictions based on H&E whole slide images. This collaboration with Histowiz’s PathologyMap platform enables evaluation of vIHC deployment across the PathologyMap database, demonstrating its potential as a scalable and practical alternative for biomarker discovery workflows. Board-certified anatomic pathologists procured 50 lung carcinoma cases for analysis. Formalin fixed paraffin embedded (FFPE) blocks were cut to 4 µm in quadruplicate onto positively charged slides. Each slide was stained with a GLP-regulated H&E stain and digitally scanned on a Leica AT2 onto the Histowiz PathologyMap platform. Slides were then de-stained and re-stained on a Leica BOND Rx platform with CD68, αSMA, and Pan-CK immunohistochemistry markers, and re-scanned. The resulting whole slide images (WSIs) were processed by ViewsML to train AI models to predict IHC expression from H&E images. For each biomarker, paired H&E and IHC WSIs were first spatially registered to ensure precise alignment. The paired data were randomly divided into training (75%), validation (10%), and test (15%) sets. Supervised machine-learning models were trained to predict IHC expression intensity from the corresponding H&E image, using the corresponding physical IHC as supervision. Evaluation was performed on the hold-out test set to verify performance. Once trained, these models enable biomarker predictions directly from a single standard H&E slide. The virtual markers demonstrated high classification performance, with per-cell ROC-AUC values of 0.94 for CD68, 0.92 for PanCK, and 0.95 for SMA. The model predicted per-cell stain intensity across millions of cells, preserving biological variability across expression ranges and tissue compartments.This collaboration validates ViewsML’s virtual staining as a reliable computational alternative to physical IHC, capable of both qualitative and quantitative biomarker assessment. Integration with Histowiz’s digital pathology infrastructure demonstrates the feasibility of embedding vIHC within scalable workflows for biomarker research, WSI repository annotation, and assay standardization. These findings support further multi-institutional expansion of ViewsML to enable robust, reproducible, and cost-efficient biomarker evaluation.
Disclosure
A. Parvatikar, None.. P. Savickas, None.. C. Lee, None.. J. Shek, None.. K. To, None.. C. Jackson, None.. L. Schobs, None.. R. Lyons, None.. R. Azhar, None.
Cited in
Control: 5591 · Presentation Id: 3133 · Meeting 21436