The histologic grade of a tumour as determined by a pathologist is an important prognostic factor in breast cancer, but associated with substantial inter- and intra-observer variability. Computational diagnostic tools for estimating the morphological properties of cancer tissue would enable objective alternative for diagnosis.
Together with Institute for Molecular Medicine Finland (FIMM), we have introduced a texture based algorithm for automated classification of breast cancer morphology in a series of digitized HE-stained histopathological tissue samples. The method represents the morphology of images using texture descriptors which are further classified using machine learning based classifier.
The three morphological classes used in the study:
- Class 1 (top row): morphology resembling normal breast epithelium, clear tubular formation
- Class 2 (center row): intermediate tubular formation
- Class 3 (bottom row): morphology least resembling normal breast epithelium, no tubular formation.
The images were transformed to gray scale and represented by the basic versions of LBP and LPQ texture descriptors. The classification of the images into the three classes was done using three one-versus-rest SVM classifiers with a radial basis function kernel. The final class was chosen by selecting the largest of the scores produced by the individual SVM classifiers.
Ojansivu V, Linder N, Rahtu E, Pietikäinen M, Lundin M & Lundin J (2012)
Automated classification of breast cancer images according to morphological features using LPQ/LBP texture descriptors and an SVM classifier.
Proc. 11th European Congress on Telepathology and 5th International Congress on Virtual Microscopy, Diagnostic Pathology, accepted.
Last updated: 14.4.2015