2023 World Conference on Lung Cancer (Posters)-P1.20. Predictability of Actionable Mutations in NSCLC using Attention-Based Multiple-Instance Learning on H&E Images

P1.20. Predictability of Actionable Mutations in NSCLC using Attention-Based Multiple-Instance Learning on H&E Images - PDF(Slides)

Back to course

Pdf Summary

This document describes a study that aims to predict actionable mutations in non-small cell lung cancer (NSCLC) using attention-based multiple-instance learning on haematoxylin and eosin (H&E) histology images. The research focuses on predicting cancer biomarkers from H&E images as an alternative to genomic testing, which can be expensive and limited by the availability of tumour tissue. The study uses the Cancer Genome Atlas (TCGA) lung adenocarcinoma (LUAD) dataset to train models and an internal cohort from St. James's Hospital to validate these models for predicting KRAS and KRAS-G12C status. A deep learning model is employed, using a weakly-supervised approach with slide-level labels. The model extracts patch-level features from the H&E images using a convolutional neural network pretrained on histology datasets. The study finds that the low frequency of actionable alterations in NSCLC limits the ability to train reliable machine learning models. Among the three mutation statuses studied (KRAS, KRAS-G12C, and EGFR), only EGFR can be predicted with an area under the receiver operating characteristic curve (AUROC) greater than 0.75. The study suggests that future work should focus on recruiting larger cohorts and developing new deep learning methods to improve classification performance. The research was funded by Science Foundation Ireland and acknowledges the SFI Centre for Research Training in Genomics Data Science. The document includes performance metrics and figures illustrating the results of the A-MIL model's performance in predicting actionable EGFR, KRAS, and KRAS-G12C mutations from the H&E images.

Asset Subtitle

Pierre Murchan

Meta Tag

Speaker Pierre Murchan

Topic Pathology & Biomarkers: Artificial Intelligence in Pathology

Keywords

non-small cell lung cancer

actionable mutations

H&E histology images

cancer biomarkers

genomic testing

KRAS

deep learning model

EGFR

machine learning models

classification performance