2022 World Conference on Lung Cancer (Posters)-P2.07-01. Deep-Learning Based Prediction of c-MET Status from Digitized H&E-Stained Non-Small Cell Lung Cancer Tissue Samples

P2.07-01. Deep-Learning Based Prediction of c-MET Status from Digitized H&E-Stained Non-Small Cell Lung Cancer Tissue Samples

Back to course

Pdf Summary

Researchers have developed a machine learning (ML)-powered method to identify c-MET overexpression status directly from hematoxylin and eosin (H&E)-stained non-small cell lung cancer (NSCLC) samples. The study used deep learning models trained on H&E slides to obtain tissue and cell-type spatial information. The models were then used to predict c-MET status based on human-interpretable image features. The study found that c-MET overexpression was significantly associated with a cluster of related features in the tumor microenvironment (TME), particularly an elevated density of lymphocytes in the cancer epithelium. The ML models achieved exceptional performance in predicting c-MET status, with an area under the receiver operating curve (AUROC) of 0.78. The models had a sensitivity of 0.85 and specificity of 0.64. A graph neural network (GNN) model trained on CNN model overlays and slide-level c-MET status labels was particularly strong in predicting c-MET status. The study highlights the potential of using H&E-stained slides and ML models to screen patients for c-MET overexpression, enabling more efficient and targeted therapies. The findings have implications for the identification and treatment of NSCLC patients with c-MET overexpression.

Asset Subtitle

Bahar Rahsepar, United States

Meta Tag

Speaker Bahar Rahsepar, United States

Topic Pathology - Tumour Genomics

Keywords

machine learning

c-MET overexpression

hematoxylin and eosin

non-small cell lung cancer

deep learning models

tissue and cell-type spatial information

human-interpretable image features

tumor microenvironment

lymphocytes density

graph neural network