Introduction
-
While immunotherapy has transformed treatment for non-small cell lung cancer (NSCLC), only 20-40% of patients experience durable benefits.
-
Emerging evidence highlights the tumor microenvironment (TME) as a critical determinant of immunotherapy response; however, a deeper and more nuanced understanding of the TME is essential to improve patient stratification and outcomes.
-
We developed HistoTME, an artificial intelligence (AI) tool that leverages routine hematoxylin and eosin (H&E)-stained slides to infer TME characteristics. HistoTME integrates a foundational tile embedding model (UNI), an attention-based multiple instance learning (AB-MIL) module, and a multi-layer perceptron (MLP) head trained to predict 30 distinct molecular signatures derived from bulk RNA expression profiles.
-
HistoTME predicted subtypes (Immune-desert vs immune-inflamed) add critical prognostic value for those who may benefit from undergoing immunotherapy. However, it does not capture spatial cellular interactions and lacks the interpretability of the model prediction at the single-cell level.
-
PathExplore , an AI platform developed by PathAI, characterizes the TME at single-cell resolution using standard H&E slides, extracting over 400 human-interpretable features (HIFs). These features include cell counts, densities, area measurements, and spatial relationships for five different cell types within cancer and stromal areas providing a more granular and interpretable view of the TME landscape.
Conference
ASCO 2025
Authors
- Rad et al. (SUNY Upstate)