[AI-driven EGFR Detection in H&E histopathology slides of NSCLC]
In anatomopathology, identifying lung cancer on H&E slides triggers consideration for molecular testing to reveal mutations. Despite molecular testing challenges, our project leverages AI to detect EGFR mutations in lung tissue H&E slides.
Objective
Developing an AI model to detect Epidermal Growth Factor Receptor (EGFR) mutation from Hematoxylin and Eosin (H&E) slides.
Methodology
Unsupervised tumor segmentation, feature extraction with pretrained ResNet-50 and attention-based multiple instance learning for EGFR mutation classification.
Impact and future directions
For therapeutic decision making in NSCLC, molecular profiling is crucial, as a number of gene alterations are eligible for targeted therapy. However, the molecular testing is expensive, time consuming and requires enough tumor tissue for testing. AI-based detection of gene alterations e.g. pathogenic/likely pathogenic EGFR mutations in H&E stained WSIs of NSCLCs could make a fast estimation of the odds of an actionable target being present and could ideally even predict the type of mutation (L858R, exon19 deletion or other).
General info and contact
Keywords (#):
Non-Small Cell Lung Cancer, Deep Learning, Multiple Instance Learning, EGFR mutation, Histopathology
RADar project research lead: M.Sc. Louise Berteloot
Principal investigator: MD Franceska Dedeurwaerdere
Timeline: 2023-2024
Status: Project finalized with conclusion that the current dataset was not sufficient to confirm the feasablity of the study objective.
Funding: AstraZeneca
RADar - Learning & Innovation Center 