University of Bamberg
open search form
You are here

Information Systems and Applied Computer Sciences

Chair of Explainable Machine Learning

  1.  University of bamberg
  2. Faculties
  3. Information Systems and Applied Computer Sciences
  4. Subject groups
  5. Applied Computer Science
  6. Chair of Explainable Machine Learning
  7. News
  8. Advancing Parkinson's Research: Master’s Student Work Published in Top Nuclear Medicine Imaging Journal EJNMMI
xAILab Bamberg

Advancing Parkinson's Research: Master’s Student Work Published in Top Nuclear Medicine Imaging Journal EJNMMI

Research conducted by Aleksej Kucerenko during his Master’s thesis at our chair has been published with open access in the highly prestigious "European Journal of Nuclear Medicine and Molecular Imaging (EJNMMI)". This work, in collaboration with Universitätsklinikum Hamburg-Eppendorf (UKE), offers valuable insights into the field of nuclear medicine imaging and Parkinson's disease research. Specifically, the research discusses how we can incorporate and benefit from uncertainty in labelling to better assess (and communicate) CNN-measured reductions in DAT-SPECT imaging signals as they are typical in the context of Parkinson's disease.

Addressing Uncertainty in DAT-SPECT Imaging

Parkinson’s disease diagnosis often relies on Dopamine Transporter SPECT (DAT-SPECT) imaging, but differences in interpretation between readers can present challenges. Aleksej’s study explores methods to address these uncertainties, enhancing the reliability of convolutional neural networks (CNNs) for identifying Parkinson’s-related reductions in imaging signals.

Notable findings include:

  • Incorporating Reader Discrepancies in Training: Making CNNs aware of discrepancies between readers during training can help distinguish certain cases from inconclusive ones.
  • Improved Diagnostic Confidence: The method reduced the proportion of inconclusive test cases. For instance, inconclusive cases dropped from 2.8% to 1.2% at a balanced accuracy of 98%.
  • Robust Validation: The findings were validated on two independent datasets (n=640 and 645), ensuring reliability..

Open Access and Open Science

The research promotes transparency through open access publication and a publicly available codebase, thereby perfectly reflecting our lab’s commitment to advancing research and fostering collaboration through open science. For more details, you can access the paper here and the associated code repository here.

Emphasizing Collaboration and Pressing Forward

Looking back, this project highlights the value of interdisciplinary collaboration, combining Aleksej’s dedication with expert co-supervision by Dr. Ralph Buchert from UKE to drive advancements in medical imaging. While further work is needed, this work provides useful insights into how machine learning can enhance diagnostic tools in healthcare. Of course, not all projects can work out like this (and don't need to), but this is a prime example that with the right talent, dedication, and environment, you can have significant scientific impact at any career stage. Thus, please join us in celebrating Aleksej's success and expressing our graditude to Dr. Ralph Buchert and the UKE team for their exceptionally productive collaboration and guidance.

xAILab Bamberg