P. Spyridonos, V. Zolota, D. Cavouras, G. Zenebissis, D. Glotsos, and G. Nikiforidis (Greece)
computer-based grading of astrocytomas; Bayesian classification
Purpose: A computer-based image analysis system was developed for assessing brain tumours (astrocytomas) as low risk (Grade I and II) or high risk (Grade III and IV) according to the WHO grading system using morphological and textural features of the cell nucleus. Materials and Methods: Tissue samples from 46 cases of astrocytomas were classified from two independent pathologists. 19 cases labeled as low risk and 27 as high risk. Images from tissue samples were digitized and an adequate number of nuclei per case were segmented for the generation of morphological and textural nuclear features. Automatic brain tumor characterization as low grade or high grade was performed using the Bayesian classifier. An exhaustive search based on classifier performance indicated the best feature combination that produced the minimum classification error. Results: The best feature combination comprised roundness, energy, inertia, cluster prominence and range of roundness cell nucleus features. This combination optimized the classification performance of a Bayesian classifier and resulted in an overall accuracy of 87%. Classification success for low risk discrimination was 84,2% and for high risk 88,9%. Conclusions: The high classification performance proved that nucleus features carried relevant information concerning astrocytomas malignancy.
Important Links:
Go Back
