Categories  |  Inventors

Technology Overview

OHSU # 2630 — Software for deep learning-based inference of biomarker distribution in cancer samples

Multiplexed tissue imaging (MTI)  is a powerful tool for cancer pathology, but it is also expensive, which has limited its availability for underserved communities, particularly in the developing world. The current software uses hematoxylin & eosin (H&E) staining, which is more widely available, to predict oncology biomarker distribution, thereby making sophisticated pathology analysis cheaper and easier to perform.

Technology Overview
Multiplexed Tissue Imaging (MTI) Immunofluorescent (IF) imaging is a powerful method of measuring the spatial distribution of critically-important biomarkers within a single tissue section. Although clinically valuable, MTI is expensive, time-consuming, and requires technical expertise to operate effectively. Whereas MTI technology is powerful yet expensive, hematoxylin and eosin (H&E) staining of tissue sections is much more common and widely used as a routine diagnostic measurement available even in underserved communities.

Oregon Health & Science University researchers have developed a deep learning-based application with the following features:

  • Generation of tumor marker IF predictions for whole slide images, based solely on H&E staining.
  • Demonstrated 94.5% accuracy in tumor marker IF predictions.
  • Low computer processing needs, generating tumor marker predictions in a matter of seconds.
  • Currently extending to 3D MTI reconstruction.

Figure depicting comparison of IF with software predicted biomarker labeling based on H&E.

Burlingame et al. “SHIFT: speedy histological-to-immunofluorescent translation of a tumor signature enabled by deep learning” Scientific Reports 10(2020). Link

Burlingame et al. “SHIFT: speedy histopathological-to-immunofluorescent translation of whole slide images using conditional generative adversarial networks.” Proceedings of SPIE--the International Society for Optical Engineering,  10581 (2018): 1058105. Link

Licensing Opportunity
This technology is available for licensing.



Filed United States
Filed United States
Published United States US 2022/058839 A1