2024 World Conference on Lung Cancer (WCLC) - Posters-P1.06A.07 Prognostic Gene Expression Profiling in Lung Adenocarcinoma Using Deep Learning Applied to Whole-Slide Images

P1.06A.07 Prognostic Gene Expression Profiling in Lung Adenocarcinoma Using Deep Learning Applied to Whole-Slide Images

Pdf Summary

The study explores the potential of using deep learning models to predict prognostic biomarkers for lung adenocarcinoma (LUAD) directly from hematoxylin and eosin-stained whole-slide images (WSIs), thus providing a cost-effective and efficient means for assessing the prognosis of LUAD. Traditional methods require cumbersome procedures, but recent advancements suggest that computational pathology could simplify this by leveraging WSIs.<br /><br />To achieve this, the researchers utilized data from the Cancer Genome Atlas (TCGA) and Clinical Proteomic Tumor Analysis Consortium (CPTAC). They developed and refined their model using 224x224 pixel patches from WSIs, embedded into 1x768 features through a process involving the CTransPath algorithm. Key analytical methods included Cox regression, adjusted for variables like age, stage, and sex, to correlate gene expression with overall survival (OS).<br /><br />Notably, the model successfully predicted the expression of 114 prognostic genes with a Pearson's R  >0.4 in an external validation cohort from CPTAC-LUAD. Furthermore, a third of these genes were identified as significant for predicting OS, establishing a clear stratification of high and low-risk patients in LUAD, even outperforming previous methodologies that evaluated OS straight from WSIs.<br /><br />The results suggest that predicted gene expressions were aligned with particular histological features like necrotic tissue, which emerged as key indicators of poor prognosis. Such advances underscore the potential to leverage WSIs for not only gene prediction but also deeper insights into tumor aggressiveness and genomic stability through enriched processes observed in the study.<br /><br />Thus, the study opens up promising opportunities in clinical application, specifically in using computational models to make informed prognostic evaluations in LUAD directly from WSIs, potentially transforming cancer treatment protocols by integrating such biomarkers into routine diagnostic processes.

Asset Subtitle

Pierre Murchan

Meta Tag

Speaker Pierre Murchan

Topic Pathology & Biomarkers

Keywords

deep learning

prognostic biomarkers

lung adenocarcinoma

hematoxylin and eosin-stained images

Cancer Genome Atlas

CTransPath algorithm

gene expression

overall survival

tumor aggressiveness

computational pathology