Segmentation and classification of immunohistochemically stained samples based on NordiQC Pan-CK assessment
Translated title
Segmentering og klassifikation af immunohistokemisk farvede prøver baseret på NordiQC Pan-CK vurdering
Authors
Larsen, Søren ; Zoupis-Schoinas, Christos
Term
4. term
Publication year
2017
Submitted on
2017-06-07
Pages
75
Abstract
Patologer bruger ofte immunhistokemiske (IHC) farvninger, særlige farvestoffer der binder til bestemte molekyler i væv, for at hjælpe med at opdage kræft. Resultaterne af farvningen kan dog variere mellem laboratorier og over tid. For at øge ensartetheden leverer NordiQC international kvalitetskontrol af IHC-farvninger. Der findes kommercielle værktøjer, og tidligere studier viser, at computerstøttet diagnostik (CAD) kan anvendes i patologi. Dette studie undersøger, om CAD kan kvantificere eksperters kvalitetsvurderinger af IHC-farvning. Vi analyserede 39 prøver fordelt på fire kategorier af kvalitetsvurdering. I valideringen gjorde billedegenskaber baseret på intensitet og farve det muligt at opnå gennemsnitlige nøjagtigheder på cirka 60–85 %. Resultaterne tyder på, at dele af kvalitetsvurderingen kan automatiseres, men at der er behov for yderligere forbedringer.
Pathologists often use immunohistochemical (IHC) stains, which are special dyes that bind to specific molecules in tissue, to help detect cancer. However, staining results can vary between laboratories and over time. To improve consistency, NordiQC provides international quality assurance for IHC staining. Commercial tools are available, and prior studies show that computer-aided diagnosis (CAD) can be applied in pathology. This study investigates whether CAD can quantify expert quality assessments of IHC staining. We analyzed 39 samples covering four quality assessment categories. In validation tests, features based on stain intensity and color enabled average accuracies of about 60–85%. These results suggest that parts of IHC quality assessment could be automated, while further improvements are needed.
[This abstract was generated with the help of AI]
Keywords
Documents