Objectives: The reliable evaluation of immunofixation electrophoresis is part of the laboratory diagnosis of multiple myeloma. Until now, this has been done routinely by the subjective assessment of a qualified laboratory staff member. The possibility of subjective errors and relatively high costs with long staff retention are the challenges of this approach commonly used today. Methods: Deep Convolutional Neural Networks are applied to the assessment of immunofixation images. In addition to standard monoclonal gammopathies (IgA-Kappa, IgA-Lambda, IgG-Kappa, IgG-Lambda, IgM-Kappa, and IgM-Lambda), also bi- or oligoclonal gammopathies, free chain gammopathies, non-pathological cases, and cases with no clear finding are detected. The assignment to one of these 10 classes comes with a confidence value. Results: On a test data set with over 4,000 images, approximately 25% of the cases are sorted out as inconclusive or due to low confidence for subsequent manual evaluation. On the remaining 75%, about 3,000 cases, not even one is classified as falsely positive, and only one as falsely negative. The remaining few deviations of the automated assessment from the classifications assigned manually by experts are borderline cases or can be explained otherwise. As a software running on a standard desktop computer, the Deep Convolutional Neural Network needs less than a second for the assessment of an immunofixation image. Conclusions: Assisting the laboratory expert in the assessment of immunofixation images can be a useful addition to laboratory diagnostics. However, the decision-making authority should always remain with the physician responsible for the findings.

ZeitschriftJournal of Laboratory Medicine
Seiten (von - bis)331-336
PublikationsstatusVeröffentlicht - 01.10.2022

Strategische Forschungsbereiche und Zentren

  • Zentren: Zentrum für Künstliche Intelligenz Lübeck (ZKIL)
  • Querschnittsbereich: Intelligente Systeme


Untersuchen Sie die Forschungsthemen von „Automated assessment of immunofixations with deep neural networks“. Zusammen bilden sie einen einzigartigen Fingerprint.