Poster
Artificial intelligence (AI)-based classification of stromal subtypes reveals associations between stromal composition and prognosis in NSCLC
AACR
Study Background
- Cancer-associated stroma (CAS) has long been appreciated as an important histological feature in many cancer types. Recently, single-cell molecular analyses have revealed the heterogeneity of CAS. Furthermore, the cellular composition of CAS has been linked to prognosis in several cancer types, including non-small cell lung cancer (NSCLC)1.
- While pathologists can manually classify CAS based on the architecture of the extracellular matrix and the cells within it, measurement of these regions is difficult and not reproducible.
- To this end, we have developed an artificial intelligence ( AI)-based model to sub-classify CAS in hematoxylin and eosin (H&E)-stained whole slide images ( WSIs). Tissue and cell human interpretable features (HIFs) extracted from our model were assessed for their association with clinicopathologic features (e.g., stage and overall survival) and their ability to predict known stromal gene expression signatures.
1 Lambrechts, D., et al. Nat Med. 2018; 24(8):1277 1289.
Authors
Fedaa Najdawi1, Sandhya Srinivasan1, Neel Patel1, Nhat Le1, Michael G. Drage1, Christian Kirkup1, Ylaine Gerardin1, Chintan Parmar1, Jacqueline Brosnan Cashman1, Michael Montalto1, Andrew H. Beck1, Archit Khosla1, Ilan Wapinski1, Ben Glass1, Murray Resnick1, Matthew Bronnimann1
1PathAI, Boston, MA.