Presentations at WCLC 2026 evaluate routine histology slides to characterize genomic alterations, immune patterns, and immunotherapy response in non-small cell lung cancer.
Artificial intelligence models analyzing routine pathology images can extract significant tumor microenvironment characteristics associated with genomic alterations and treatment responses in non-small cell lung cancer (NSCLC), according to a series of studies announced by medical AI company Lunit.
The research, scheduled for presentation at the 2026 World Conference on Lung Cancer in Seoul, South Korea, demonstrates how computational pathology tools can evaluate standard hematoxylin and eosin (H&E) slides to support biomarker discovery and clinical stratification, according to a release from the company.
“These studies demonstrate the expanding potential of AI-based analysis to generate deeper insights into lung cancer biology, from genomic characteristics and the tumor microenvironment to features associated with treatment response,” says Brandon Suh, CEO of Lunit, in a release. “By broadening the range of insights that can be derived from routinely available pathology images, we aim to advance AI-powered biomarker discovery and patient stratification, ultimately contributing to more personalized treatment strategies for patients with cancer.”
Characterizing EGFR Mutation Subtypes
In the first study, investigators utilized the AI image analysis software Lunit SCOPE IO to examine 494 H&E whole-slide images from epidermal growth factor receptor (EGFR)-mutant NSCLC cases. The software identified distinct tumor microenvironment profiles across specific mutation subtypes, according to study findings.
Tumors harboring exon 19 deletions displayed significantly lower intratumoral tumor-infiltrating lymphocyte (TIL) density. Conversely, L858R-mutant tumors exhibited a relative enrichment of both TIL and macrophage infiltration, whereas exon 20 insertion tumors showed higher endothelial cell density. The authors reported that variability in clinical outcomes among EGFR subtypes may stem not only from kinase kinetics, but also from these subtype-specific microenvironmental factors.
Spatial Profiling in Neoadjuvant Therapy
A second study, conducted in collaboration with Paola Nistico, MD, of the Regina Elena National Cancer Institute in Rome, Italy, evaluated tissue samples from 32 patients with NSCLC who received neoadjuvant chemo-immunotherapy. The research was supported through the Lunit Research Program for Society for Immunotherapy of Cancer members.
Researchers paired AI image analysis with spatial transcriptomics and high-plex spatial proteomics. Biopsies from patients who achieved a pathological complete response (pCR) demonstrated a highly inflamed, spatially organized immune landscape marked by prominent tertiary lymphoid structures. In contrast, tumors that did not achieve pCR exhibited immune exclusion and stroma rich in activated fibroblasts. The findings support the utility of spatially resolved tissue biomarkers to understand immunotherapy efficacy, according to the researchers.
Predicting TP53 Mutations from Standard Slides
The third study evaluated an H&E-based AI model designed to predict TP53 mutation status directly from tissue sections in lung adenocarcinoma. Validated in an independent cohort of 462 patient cases, the algorithm demonstrated an area under the receiver operating characteristic curve of 0.759, with 82% sensitivity, and 63% specificity.
Further spatial analysis indicated that TP53-mutant tumors presented an immune-inflamed phenotype more frequently. Wild-type TP53 tumors were typically characterized by immune exclusion, higher endothelial cell densities, and elevated stromal fibroblast counts. The investigators noted that standard slide-based AI models could serve as a preliminary screening mechanism to enrich patient cohorts for clinical trials targeting specific TP53 variants.
ID 452939094 © Ppengcreative | Dreamstime.com
