Automated grading of prostate cancer using architectural and textural image features

The current method of grading prostate cancer on histology uses the Gleason system, which describes five increasingly malignant stages of cancer according to qualitative analysis of tissue architecture. The Gleason grading system has been shown to suffer from inter- and intra-observer variability. In this paper we present a new method for automated and quantitative grading of prostate biopsy specimens. A total of 102 graph-based, morphological, and textural features are extracted from each tissue patch in order to quantify the arrangement of nuclei and glandular structures within digitized images of histological prostate tissue specimens. A support vector machine (SVM) is used to classify the digitized histology slides into one of four different tissue classes: benign epithelium, benign stroma, Gleason grade 3 adenocarcinoma, and Gleason grade 4 adenocarcinoma. The SVM classifier was able to distinguish between all four types of tissue patterns, achieving an accuracy of 92.8% when distinguishing between Gleason grade 3 and stroma, 92.4% between epithelium and stroma, and 76.9% between Gleason grades 3 and 4. Both textural and graph-based features were found to be important in discriminating between different tissue classes. This work suggests that the current Gleason grading scheme can be improved by utilizing quantitative image analysis to aid pathologists in producing an accurate and reproducible diagnosis.