Categories  |  Inventors

Technology Overview

OHSU # 3124 — Imaging biomarker activation maps for deep learning-aided classifiers


Deep-learning (DL) methods for disease classification have not been widely adopted in clinical care, because it is unclear how AI-based methods arrive at their results. The current technology generates biomarker activation maps for imaging data, allowing for improved clinical interpretation of DL-results and potentially wider-spread use of this methodology in routine clinical care.

Technology Overview

Artificial intelligence (AI) techniques, such as deep learning (DL), can be used with imaging data to assist with disease diagnosis. However, most DL classifiers operate as a ‘black box’, making it unclear how the results were derived and limiting their clinical interpretation.

The current technology is a method for generating biomarker activation maps based on imaging data to improve the clinical relevancy of AI-based results.  Features of this technology include:

  • Accurate highlighting and segmentation of DL-based biomarkers, allowing for improved biomarker interpretability.
  • Ability to differentiate classifier-utilized biomarkers to potentially improve the clinical relevancy of DL-based disease classification.
  • Facilitation of new biomarker discovery.
  • Achieved state-of-the-art performance in providing interpretability to a diabetic retinopathy classifier based on optical coherence tomography and its angiography images.

This technology ultimately could improve the transparency and clinical relevancy of AI-based results, potentially allowing for wider adoption of DL-based disease diagnosis and monitoring in clinical practice.


P. Zang, et al., "Interpretable Diabetic Retinopathy Diagnosis based on Biomarker Activation Map," arXiv preprint arXiv: 2212.06299, 2022. Link

Licensing Opportunity

This technology is available for licensing.