Deep learning-based stratification of gastric cancer patients from hematoxylin and eosin-stained whole slide images by predicting molecular features for immunotherapy response.

Researchers

Journal

Modalities

Models

Abstract

Determining the molecular characteristics of cancer patients is crucial for optimal immunotherapy decisions. The aim of this study was to screen immunotherapy beneficiaries by predicting key molecular features from hematoxylin and eosin-stained images based on deep learning (DL) models. An independent data-set from Asian GC patients was included for external validation. Besides, a segmentation model (HoVer-Net) was used to quantify the cellular composition of tumor stroma. The model performance was evaluated by measuring the area under the curve (AUC). The tumor extraction model achieved an AUC of 0.9386 and 0.9062 in the internal and external test sets. The stratification model can well predict the immunotherapy-sensitive subtypes (AUCs ranged from 0.8685 to 0.9461), the genetic mutations (AUCs ranged from 0.8283 to 0.9225), and the pathway activity (AUCs ranged from 0.7568 to 0.8612). In external validation, the prediction performance of EBV and PD-L1 expression status achieved an AUC of 0.7906 and 0.6384, respectively. The segmentation model identified a relatively high proportion of inflammatory cells and connective cells in some immunotherapy-sensitive subtypes. The DL-based models may potentially serve as a valuable tool to screen out the beneficiaries of immunotherapy in GC patients.Copyright © 2023. Published by Elsevier Inc.