Research On Segmentation And Recognition Of Medical Parasitic Images

In recent years,the image segmentation technique based on graph theory becomes a new hotspot because of its good characteristics.Because there are some good corresponding characteristics between image segmentation and graph theory. The technique can combine the global segmentation with local information and avoid the error caused by discreting image. So it can get a good segmentation result.This dissertation mainly studies two kinds of image segmentation methods based on graph theory:normalized cuts and graph cut,and analyzes their applications on medical image parasites.These methods are aimed at different application scenarios and show better segmentation performance due to incorporate with prior knowledge. On the basis of segmentation by further researches and extracting the features of schistosome eggs, we use LS-SVM to classify schistosome eggs. The main contributions of this dissertation are summarized as follows:(1) First, this dissertation discusses medical image processing technology in common use,and then analyzes the commonly used type of image segmentation based on graph theory and characteristics.In addition,this dissertation describes domestic and overseas research on the parasite cell recognition and segmentation.(2) Because the hospital and CDC(Centers for Disease Control and Prevention)’s current bilharziasis’ s detection method is inefficiency and has low accuracy, a novel algorithm ANcut based on normalized cut and combined with prior knowledge is proposed. The ANcut’s weight matrices used in evaluating the graph cuts are based on the gray levels of an image,rather than the commonly used image pixels.So it avoids eigenvalues and eigenvectors’ complex computing and possesses much higher running efficiency. In addition, ANcut combines with the schistosome eggs prior knowledge, so the optimal segmentation blocks can be calculated automatically. Compared with the classical threshold segmentation algorithm,ANcut can segment schistosome egg margin accurately,and it is beneficial to extract schistosome egg features.(3) According to the characteristics of the malaria parasite image, this dissertation puts forward a new graph cut algorithm GC-SBP based on the similarity between pixels and prior knowledge. ANcut does not require the participation of the user interactive segmentation. Compared with the classical graph cut algorithm,ANcut can segment the parasite from malaria parasites image better than the traditional segmentation algorithm. GC-SBP algorithm has three major improvements as follows: it can get a better segmentation result because of taking good advantage of prior knowledge in the segmentation; Second, it uses YCb Cr color space to substitute the traditional RGB color space. YCb Cr color space has a better clustering performance than RGB,and YCb Cr’s three color components are independent of each other,so it can be efficient to reflect the image information and calculate the similarity between the regional block. Third,it uses regional similarity to substitute the similarity between the pixels,so it reduces the computation amount and raise the computation speed greatly.(4) On the basis of ANcut algorithm, the schistosome egg features are extracted accurately. We develop a binary classifier based on LS-SVM. The experiments show that the recognition rate is high that meets the requirement of CDC’s detection,so it has a high practical value.