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

ePoster Display

1147P - AI-based detection of molecular biomarkers directly from H&E scanned slide

Date

16 Sep 2021

Session

ePoster Display

Topics

Staging and Imaging;  Clinical Research;  Cancer Biology;  Pathology/Molecular Biology

Tumour Site

Presenters

Nir Peled

Citation

Annals of Oncology (2021) 32 (suppl_5): S921-S930. 10.1016/annonc/annonc707

Authors

N. Peled1, J. Zalach2, I. Hayun3, N. Paz-Yaacov3

Author affiliations

  • 1 Oncology, Shaare Zedek Medical Center, 9103102 - Jerusalem/IL
  • 2 R&d, Imagene-AI, Tel Aviv/IL
  • 3 Scientific Research, Imagene-AI, 6713412 - Tel Aviv/IL

Resources

Login to get immediate access to this content.

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

Abstract 1147P

Background

Cancer diagnosis and treatment rely on accurate identification of morphological, molecular, and genetic biomarkers. As technology and science progress, the number of biomarkers and testing required increases. Professional guidelines structure the biomarkers testing methodology. Yet, testing capabilities, options, and interpretation vary significantly between countries and medical centers in practice. Here, we present an alternative to the traditional methodologies for biomarker detection, using neural networks to detect the different biomarkers directly from hematoxylin and eosin (H&E) stained pathology slide images.

Methods

H&E slide images were scanned in X40 resolution to generate a molecular model using advanced computer vision and neural network algorithms. Internal and publicly available databases were used to train and validate different tissue panels and specific biomarkers, including mutations, fusions, expression, and different cancer signatures.

Results

Thyroid panel validation for NTRK, RET-fusions &, BRAF mutation resulted in 100%, 100% and 91.7% sensitivity and 93.9%, 95.7% & 90.7% specificity, respectively (n=102). Similar results were obtained for the lung cancer panel of actionable alterations tested and MSI signature.

Conclusions

Our validation process reveals the potential of combining artificial intelligence methodologies with pathology and oncology domains. It allows accurate, fast, and accessible testing for all patients with optimal tissue utilization and detection of rare alterations. The concept of inferring such alterations directly from the phenotype presents a significant advantage by ignoring variants of uncertain significance (VUS) and looking only at alterations that prove their relevance by changing the cells’ morphological signature.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Imagene-AI.

Funding

Imagene-AI.

Disclosure

N. Peled: Financial Interests, Personal, Advisory Board: Imagene-AI. J. Zalach: Financial Interests, Personal, Ownership Interest: Imagene-AI; Financial Interests, Personal, Full or part-time Employment: Imagene-AI; Financial Interests, Personal, Stocks/Shares: Imagene-AI. I. Hayun: Financial Interests, Personal, Full or part-time Employment: Imagene-AI. N. Paz-Yaacov: Financial Interests, Personal, Full or part-time Employment: Imagene-AI; Financial Interests, Personal, Stocks/Shares: Imagene-AI.

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.