Ultrasound Image Segmentation Based On Zernike Moment And Level Set

Ultrasonic images are an important way of medical diagnosis with advantages of no radiation and low cost. However, ultrasonic image has problems of low resolution, low contrast, speckle noise and other inherent problems because of imaging principle.In the clinical diagnosis and treatment, accurate segmentation of ultrasonic images is the foundation of quantitative analysis, qualitative analysis and precision guided treatment. The application of texture in the ultrasonic image segmentation can improve the result of segmentation. Level set method can make use of the image information to get the profile of target.The method that Zernike moments are used to extract the texture feature, the level set method is used to segment ultrasonic image is proposed.First, the method that both the magnitude and phase are used to extract the feature is proposed. Zernike moments were used to extract the image texture features to get the magnitude and phase.After the using of the corresponding non-linear transformation,the Support Vector Machines was used to segment the image. Compared to the method used Gabor filter and method used Zernike Moment magnitude, the proposed method that used both magnitude and phase has lower segmentation error rate.Then, the level set method is introduced which include the GAC model based on the monogenic signal the C-V model based on intensity.These models have proposed different solutions for the shortcomings of ultrasound images. Based on the above method, an ultrasonic image segmentation method based on Zernike moments and level set is presented. First,9 Zernike Moments with different orders and repetitions were used to extract the image features. Both the magnitude and phase were reserved to get 18 feature images. Meanwhile, the inside and outside of target region of each feature image were sampled, the samples were used to count the weights of feature images. Then, the feature images and Gaussian operator were convoluted to count the edge indicator functions. All the edge indicator functions and the corresponding weights of feature images were multiplied. The result of all the products together was the edge indicator function of the ultrasonic image. Finally, the level set method based on variational formulation was applied to segment the ultrasonic image. The results of experiments based on prostate ultrasonic images show that compared to the level set rsethod based on gradient and the level set method based on Zernike Moment magnitude, the proposed method has higher segmentation accuracy, and the dice similarity coefficients are more than 95%.And this method was used on 3-dimensional image to achieve evolution of each layer, improving the efficiency of Segmentation algorithm.