Medizinische Universität Graz Austria/Österreich - Forschungsportal - Medical University of Graz
Gewählte Publikation:
SHR Neuro Krebs Kardio Lipid Stoffw Microb
Kainz, P; Mayrhofer-Reinhartshuber, M; Burgsteiner, H; Asslaber, M; Ahammer, H; .
Echo State Networks for Granulopoietic Cell Recognition in Histopathological Images of Human Bone Marrow.
BIOMED ENG-BIOMED TECH. 2014; 59: S492-S495.
Web of Science
- Co-Autor*innen der Med Uni Graz
- Ahammer Helmut
- Asslaber Martin
- Altmetrics:
- Abstract:
- The assessment of the cellularity in bone marrow is an essential step in diagnostic processes in pathology. The quantification of cell maturity at microscopic level is a tedious and error-prone task and heavily relies on the experience of the pathologists. The inter-observer variability in cellularity estimation may be reduced by employing supervised machine learning methods on digital histopathological images. The main goal of this paper is to examine echo state networks (ESN) for quantitative cell recognition. We show that a properly designed and trained ESN is able to discriminate early and late granulopoietic cells in histopathological images of the human bone marrow with overall mean accuracy (+/- SD) of 0.846 (+/- 0.013). This work gives strong indication that ESNs are able to work even with raw image patches directly and do not necessarily require image pre-processing, or feature extraction prior to classification.