Development of new methods for explainable machine learning in medicine

TraMeExCo: Transparent Medical Expert Companion

The aim of the BMBF’s TraMeExCo (Transparent Medical Expert Companion) project is to research and develop suitable new methods for robust and explainable AI (XAI) in complementary applications (digital pathology, pain analysis, cardiology) in the field of medical technology.

   

To that end, the project team is investigating options for diagnostic predictions in different data forms (microscopic image pyramids, pain videos, EKG data) using current AI methods (deep learning, Bayesian learning, few shot learning). At the same time, the team employs certain approaches to ensure that the decisions thus made are transparent and explainable to the clinical staff. These include investigating and applying methods such as heat maps, layerwise relevance propagation (LRP) and inductive logic programming (ILP). The team is also looking into modeling uncertainties in model predictions using Bayesian learning.

At Fraunhofer IIS, the Image Processing and Medical Engineering department investigates and implements the few shot learning and heat map approaches for digital pathology.

 

That Image Processing and Medical Engineering department also handles determination of heart rate variability in noisy EKG and PEG data using long short-term memory networks. (https://www.iis.fraunhofer.de/en/ff/sse/health/medical-sensors-and-analytics.html).

The Image Sensors department (https://www.iis.fraunhofer.de/shore) investigates Bayesian deep learning methods on pain videos. Researchers make use of the Facial Action Coding System (FACS) as adopted by Ekman and Friesen, in which every movement of individual muscles (action unit), such as the contraction of the eyebrows, is described, interpreted and detected. The presence of certain action units, automatically detected, indicates pain.



Project partner University of Bamberg focuses on the conception, implementation and testing of methods to explain diagnostic system decisions using LIME, LRP and ILP methods.(https://www.uni-bamberg.de/en/cogsys/research/projects/bmbf-project-trameexco/).



Project partner Fraunhofer HHI is further refining the layerwise relevance propagation (LRP) approaches. (https://www.hhi.fraunhofer.de/abteilungen/vca/projekte/trameexco.html).

Press (german)

Publications