Abstract
In this paper, we present a glandular shape analysis to detect prostatic cancer glands from prostate cancer slides stained by hematoxylin. Clumped nuclear areas in the gland were segmented and skeletonized. Morphometric and features were analyzed at 5× to locate glands and at 10× to classify them into normal or cancer glands. Compared to normal, malignant glands showed significantly shorter skeletons, less number of branching points, and smaller nuclear area (Mann-Whitney test, p<0.01). Pattern classification algorithms detected glands at the accuracy of 99% and classified them at the accuracy of 85%. Our proposed method was proven to be useful in the screening of whole slide images to locate prostate cancer areas.