13. Classifying Pathological Prostate Images by Fractal Analysis

This study presents an automated system for grading pathological prostate images based on texture features of multicategories including multiwavelets, Gaborfilters, gray-level co-occurrence matrix (GLCM) and fractal dimensions. Images are classified into appropriate grades by using k-nearest neighbor (k-NN) and support vector machine (SVM) classifiers. Experimental results show that a correct classification rate (CCR) of 93.7 % (or 92.7 %) can be achieved by fractal dimension (FD) feature set by using k-NN (or SVM) classifier without feature selection. If the FD feature set is optimized, the CCR of 94.2 % (or 94.1 %) can be achieved by using k-NN (or SVM) classifier. The CCR is promoted to 94.6 % (or 95.6 %) by k-NN (or SVM) classifier if features of multicategories are applied. On the other hand, the CCR drops if the FD-based features are removed from the combined feature set of multicategories. Such a result suggests that features of FD category are not negligible and should be included for consideration for classifying pathological prostate images.