Incorporating label uncertainty during the training of convolutional neural networks improves performance for the discrimination between certain and inconclusive cases in dopamine transporter SPECT.

Journal: European journal of nuclear medicine and molecular imaging
PMID: 39592475

Abstract

PURPOSE: Deep convolutional neural networks (CNN) hold promise for assisting the interpretation of dopamine transporter (DAT)-SPECT. For improved communication of uncertainty to the user it is crucial to reliably discriminate certain from inconclusive cases that might be misclassified by strict application of a predefined decision threshold on the CNN output. This study tested two methods to incorporate existing label uncertainty during the training to improve the utility of the CNN sigmoid output for this task.

Authors

  • Aleksej Kucerenko
    xAILab Bamberg, Chair of Explainable Machine Learning, Faculty of Information Systems and Applied Computer Sciences, Otto-Friedrich-University, Bamberg, Germany.
  • Thomas Buddenkotte
    Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, United Kingdom; Department of Radiology, University of Cambridge, Cambridge, United Kingdom; Department for Diagnostic and Interventional Radiology and Nuclear Medicine, University Hospital Hamburg-Eppendorf, Hamburg, Germany; Jung diagnostics GmbH, Hamburg, Germany. Electronic address: t.buddenkotte@uke.de.
  • Ivayla Apostolova
    Department for Diagnostic and Interventional Radiology and Nuclear Medicine, University Hospital Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany.
  • Susanne Klutmann
    Department for Diagnostic and Interventional Radiology and Nuclear Medicine, University Hospital Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany.
  • Christian Ledig
    Biomedical Image Analysis Group, Imperial College London, UK.
  • Ralph Buchert
    Department for Diagnostic and Interventional Radiology and Nuclear Medicine, University Hospital Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany. r.buchert@uke.de.

Keywords

External Resources

Popular Topics
Recent Journals