Analysis of Classification Models Using Image Statistics and Data Miner for Grade Prediction of Astrocytoma

摘要

Astrocytoma is the most common primary tumor which develops from glial cells of brain. They are generally classified as low grade (Grade I and Grade II) and high grade (Grade III and Grade IV), and these classifications are very important in clinical practice which signifies the rate of growth. Grading of astrocytoma relies on magnetic resonant images, and pathological information is also available in clinical settings. In this proposed method, we introduce a novel approach to grade the tumor using first- and second-order image statistical parameters combined with a tool termed as 'XLMiner.' The actual grade of astrocytoma and the predicted grade by the classifiers are compared and the accuracy of the classifiers is summarized based on the classifier-predicted output. Experimental results demonstrate the effectiveness of the method. The accuracy of Naives Bayes, discriminant analysis, regression tree, and classification tree classifiers for the prediction of grades from lower (I, II) to higher (III, IV) are 100, 81, 76, and 78 % for all the views, respectively.