A Study On Diagnosis Technology Of Pathological Images Of Prostate Cancer

In recent years, under the influence of a number of factors such as population aging, Westernized diet and lifestyle and so on, prostate cancer incidence increasingly doubled and it has becoming a biggest threat to the health of middle-aged men. To address this situation of high incidence of prostate cancer and relatively backward diagnosis situation, this paper presents a method of prostate case diagnosis which includes pathological image segmentation techniques, pathological image feature extraction techniques, image optimization techniques and pathological classification techniques. In this paper, we have got a high accuracy on a sample set of 80 cases.We first identify the research area of the prostate cancer diagnosis technology. Here we only take a case of one 100 times magnification picture and one 200 times magnification picture into consideration and we classify these cases into two categories, then different pathological object extraction algorithms are presented. With the 100 times magnification picture we use centroid identification and a a series of morphological operations to extract stoma, while with the 200 times magnification picture we use multi-stage filter to get lumens and use several algorithms such as initial classification algorithm, polygon approximation of the concave point search algorithm, contours matching algorithm on elliptic curve-fitting and so on to extract a single nucleus or segment cell mass.In the research stage of image feature extraction, this paper integrates experts'knowledge and presents four types of features. These features are local texture features of stoma, geometric features of lumens, statistical features of lumens and nuclei and the relevant features among three types of pathological objects. Then this paper use Fisher ratio analysis and principal component analysis to optimize the feature set of 20 features.Finally, support vector machine and 5 cross-validation are used to classify the prostate cases. In this paper we get a highest accuracy rate of 100% in training sets and a highest test accuracy of 99.53% in the test sets over 80 prostate cases.