From Arno Schaepe’s experience

Linking tissue images to patient outcomes in digital pathology

Definiens set out to turn tissue slides into data that predicts how patients respond. Arno led platform development and architecture, then drove product and innovation in tissue phenomics and digital pathology.

Industry
Health and life sciences
Organization
Definiens
Arno’s role
Chief Architect, later VP Product & Innovation

The business need

Definiens built image-analysis software for pathology and biomarker research. In 2011 the company described its aim as making pathology quantitative: measuring cells in tissue images and correlating them with therapy response and patient outcomes to find predictive or prognostic biomarkers.

It worked with pharma and biotech companies in oncology and immuno-oncology. MedImmune, the biologics research arm of AstraZeneca, acquired Definiens in 2014.

What Arno did

  1. Arno led platform development and architecture at Definiens as Director of Platform Development, Chief Architect and VP & Chief Architect.
  2. Arno is a named co-inventor on a Definiens patent for context-driven image mining, which combines data mining with image analysis to build image-based biomarkers that correlate with clinical endpoints (filed 2011, granted 2013, with a continuation granted in 2016).
  3. Arno is also a named co-inventor on a patent for a user interface that shows tissue images next to their image-analysis values and the patient's health state, such as survival (granted 2014).
  4. Later, as VP Product & Innovation, Arno drove product and innovation in tissue phenomics and digital pathology.

Technology and approach

  • Object-oriented image analysis of tissue slides and tissue microarrays
  • Image-based biomarkers built from weighted sets of image features and tested against clinical endpoints such as survival and disease recurrence
  • Survival analysis, including Kaplan-Meier plots, linked back to the underlying tissue images

Outcome

  • The image-mining and interface patents were granted in the United States in 2013, 2014 and 2016.

Decisions this experience speaks to

  • Should we use classical scientific computing and machine learning, or large language models, for our problem?
  • How do we take our research prototype to a working product in the market?

Related services and use cases

Sources

  1. Arno Schaepe’s profile on CTO Grand Rounds
  2. Definiens releases quantitative digital pathology portfolio (2011) - biospace.com (opens in a new tab)
  3. Definiens forearms organization for strategic expansion (2017) - biospace.com (opens in a new tab)
  4. Cognition Network Technology - wikipedia.org (opens in a new tab)
  5. US8594410B2 Context driven image mining - patents.google.com (opens in a new tab)
  6. US9245337B2 Context driven image mining - patents.google.com (opens in a new tab)
  7. US8711149B2 Graphical user interface for image analysis results - patents.google.com (opens in a new tab)