Computer - assisted Detection and Grading of Prostatic Cancer in Biopsy Image

Prostatic biopsies provide abundant information for diagnosis of prostatic cancer. Whereas, inspection in the vast biopsy images under the microscope is a heavy loading to pathologists. Besides, human grading is always subjective to interand intra-observer variability. Automatic inspection for prostatic biopsy image is thus necessary. In this paper, we proposed a novel approach to automatically detect prostatic cancer and grade them according to Gleason grading system. The proposed approach contains two stages. First stage is to divide biopsy images into regions and classify these regions into clusters based on their Skeleton-set (SK-set), and each region in the same cluster consists of the similar two-tone texture. In the second stage, multiple Fractal-dimension-based (FD) features extracted from regions are used to analyze the variations of intensity and texture complexity in the boxes with different size. Each region is classified to appropriate grade by using Bayes classifiers, respectively. The leaving-one-out approach is used to estimate error rate. The present experimental results demonstrated that 94.88% of accuracy for a set of 1182 pathological images.