Abstract
Immunotherapy has become part of standard care in lung
cancer and other solid tumors, however only a subset of patients derives
durable benefit. This variability has increased interest in the tumor immune
microenvironment (TIME), but large-scale single-cell spatial proteomics
is rarely available in routine practice. Multiplexed immunofluorescence
(mIF) can resolve the TIME at single-cell resolution, but it is
costly and hard to incorporate into standard pathology workflows. We
introduce histology-derived virtual immune composition as a scalable
alternative to experimental mIF. Using MACSima mIF, we phenotype
1,042,624 single cells, and co-register H&E to mIF at single-cell resolution.
From these paired data, we train models to predict cell-type composition
from H&E alone and show that the resulting virtual immune profiles
outperform existing H&E-based approaches on mIF-derived ground
truth. On our cohort, foundation-based virtual cells show substantially
higher agreement with mIF-based ground truth, with the largest gains for
macrophage, plasma cell, and tumor compartments. They further support
patch-level detection of TLS-like patterns and tumor-infiltrating
lymphocytes (TILs). Beyond this experimental reference, we extend the
virtual immune composition to TCGA lung, melanoma, and head and
neck cancer whole-slide images, where the same signatures show robust
associations with bulk RNA-seq immune transcripts and consistently
stratify overall survival across these immunotherapy-relevant cancers.
Taken together, these results suggest that virtual immune profiling from
routine histology can bring single-cell-level immune context closer to
cohort-scale H&E datasets, while still revealing residual gaps relative
to experimental spatial proteomics.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_047.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{KimKyu_Virtual_MICCAISAT2026,
author = { Kim, Kyung A. AND Yoo, Jongwoo AND Shim, Hyo Sup},
title = { { Virtual immune cell prediction from H&E Whole Slide Images } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17251},
month = {pending},
page = {pending}
}
