Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster session 15

1084P - Predicting KRAS G12C subtype from non-small cell lung cancer H&E slides using deep learning


10 Sep 2022


Poster session 15


Tumour Site

Non-Small Cell Lung Cancer


Isabelle Soubeyran


Annals of Oncology (2022) 33 (suppl_7): S448-S554. 10.1016/annonc/annonc1064


I. Soubeyran1, R. Dubois2, M. Jacquemin1, K. Courtet1, Y. laizet1, C. Lucchesi1, B. Allard3, A. Rousset3, A. Jaeger2, J. moreira2, E. KHALIFA4, B. bonhomme1, A. Italiano5

Author affiliations

  • 1 Biopathology, Institute Bergonié, 33076 - Bordeaux/FR
  • 2 R&d, OWKIN France, 75010 - Paris/FR
  • 3 Artifical Intelligence / Medical Devices / Emr Integration, Amgen France, BOULOGNE BILLANCOURT/FR
  • 4 Biopathology Department, Institute Bergonié, 33000 - Bordeaux/FR
  • 5 Early Phase Trials Unit, Institute Bergonié, 33000 - Bordeaux/FR


Login to access the resources on OncologyPRO.

If you do not have an ESMO account, please create one for free.

Abstract 1084P


Testing biomarkers in NSCLC relies on complex sequencing and imaging techniques, leading to financial and operational challenges for health care systems. As such, some testings are not systematically performed, while specific targeted therapies exist for those biomarkers. With a prevalence of 13% in NSCLC patients with lung adenocarcinoma, KRAS G12C mutation is considered among the most essential ones. This study aims at showing that AI techniques applied to digital pathology could offer a fast and cheap mass patient screening solution to complement DNA testing by predicting the KRAS mutation status and its G12C subtype from digital pathology slides used in clinical workflow.


For that purpose, data from 1,076 non-small cell carcinoma patients from Institut Bergonie and partner centers was used to train a DeepMIL deep learning model. For each patient, H&E diagnostic slides from resections or biopsies, as well as the KRAS mutation status and G12C subtype obtained with NGS have been used to run analyses and train models. The model was evaluated through a repeated cross validation scheme on the local cohort, and later evaluated on the TCGA-LUAD public cohort (537 patients).


Our deep learning models succeed in discriminating KRAS G12C mutated patients from wild type or other subtype by using routine H&E histology data with encouraging performance range of 58% (P=.04) and 65% (average of the cross-validated performances) ROC AUC obtained respectively on public and local cohorts. Furthermore, a systematic analysis performed with an experienced pathologist revealed that the regions of interest associated with the G12C status identified by the model contained a significant over-representation of tumoral cells. While requiring further work to reveal its full potential, this experimentation is the first demonstrating that KRAS G12C mutated patient identification would be possible through morphological patterns analysis from annotation-free whole slide imaging.


In future, these innovative approaches could complete existing testing techniques and pave the way for a more systematic, early and potentially cost-effective mass screening approach of KRAS G12C mutations in clinical routine.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.


Owkin Inc.


R. Dubois, A. Jaeger, J. Moreira: Financial Interests, Personal, Full or part-time Employment: Owkin. Inc. B. Allard, A. Rousset: Financial Interests, Personal, Full or part-time Employment: Amgen. All other authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.