What is it about?
Providing an improved technique which can assist pathologists in correctly classifying meningioma brain tumours with a significant accuracy.
Featured Image
Why is it important?
A technique for histopathological meningioma tumour classification based on texture measures combination, which aims to overcome intra and inter-observer variability, has been proposed in this study. The morphological gradient of the RGB colour channel that best discriminates the cell-nuclei from the cytoplasm background is selected, and then feature extraction is performed by four statistical and model-based texture measures for discrimination using a Bayesian classifier. The pre-processing phase represented by the appropriate colour channel selection and morphological processing proved to be necessary for increasing texture feature separability, and hence can improve classification accuracy.
Perspectives
It was concluded that certain selected texture measures play a complementary role to each other in the process of quantitative texture characterisation. In other words, a certain texture measure can represent a pattern better than another depending on the region of interest frequency of occurrence and noise in the examined structure. This also applies to certain combinations which might outperform other texture measure fusions. However, combining more than two texture measures would not necessarily give a better accuracy even with the removal of highly correlated features. This will increase feature complexity, hence having a negative effect on the classifier’s performance. It was found that the combination of the Gaussian Markov random field & run-length matrix texture measures are the best for characterizing meningioma subtypes of grade I, these two measures outperformed other measures in the study individually and combined.
Dr Omar S Al-Kadi
University of Jordan
Read the Original
This page is a summary of: Texture measures combination for improved meningioma classification of histopathological images, Pattern Recognition, June 2010, Elsevier,
DOI: 10.1016/j.patcog.2010.01.005.
You can read the full text:
Resources
Contributors
The following have contributed to this page
