Research On Pathological Object Segmentation And Feature Extraction

With the development of society, the medical treatment level is continuously growing andalso living and eating conditions are in constant improvement.As the public pay moreattention on the diseases such as gastric cancer、liver cancer etc, the diseases from prostatebecome a threat in elderly male killer quietly. Recently, Pathological examinatio n usingpuncture technic is the most reliable and accurate means to diagnose the prostate cancer but aswe know it is not beneficial for the treatment of patients with the examination just by manurethat requires a lot of manpower and material resources and spends a lot of time. A largenumber of studies show that,the organization of the prostate will change a lot once it has beencancerated. At the same time, the use of computer aided diagnosis technology can improve theaccuracy of pathological diagnosis and reduce the time to diagnosis. So, this paper presentsPathological Object Segmentation algorithm and feature extraction algorithm which are usedin prostate cancer diagnosis.This paper uses the way of200times magnification picture to divide the nuc leus andglandular lumen. and, firstly, We apply image color space conversion on the image, fromRGB space conversion to Lab space and then be based on the Gauss model K-Means tocluster initial extraction out of the nucleus and glandular cavity; secondly, process thepreliminary extraction glandular cavity by morphological method to get the whole glandlumen area and at the same time separate adhesion from initial extracted nucleus based on themorphological gradient watershed algorithm to get a single nuc leus. In this paper, weexperiment with50slice images00times magnification picture, the results of all of thepictures are very good to segment the glandular cavity and the nucleus and the accuracy ratereached90%.According to the1000times magnification picture. firstly, the picture is processed by gridand switch color space for every grid image, from RGB space conversion to CMYK space,and then we use multi-channel information (C channel and R channel) to segment thenucleus from the grids; secondly, enhance the brightness for the extracted nucleus andsegment ROI area from brightness enhanced nuclei by the OSTU method; finally, extract intact nucleoli by the improved level set method. In all of42pictures containing nucleolarprostate cells, we have40that extract a good out of the nucleus and nucleolus and theaccuracy rate reached95.2%.Finally, the paper extract the feature parameters form the divided pathological object,such as Glandular cavity size, area, nucleus area variance, nucleolar number etc. which is tobuild the foundation identify prostate cancer.