Image segmentation using prior information and its application on medical ultrasound image processing

In Medical Imaging, Ultrasound Imaging is one of the major modalities for Image-Guided surgery and has been playing an increasingly important role in medical diagnosis. This thesis addresses the processing, especially the segmentation, of medical ultrasound imaging. Many traditional segmentation methods fail due to the speckle noise produced by the physical mechanism of ultrasonic devices. This thesis first introduces a preprocessing technique for ultrasound speckle removal using the perception theory. Then the performances of several traditional segmentation methods are studied and a novel method is proposed for fuzzy image segmentation. It is based on the postulate that points with high class uncertainty will have low region homogeneity. Although the proposed method outperforms several traditional segmentation methods in segmenting objects with connective and homogenous regions, it is not adequate for ultrasound image segmentation neither, because of the speckle noise and the weak boundaries between different tissues in ultrasound images. This investigation motivates our further study on the prior knowledge based segmentation approach. Finally, a novel texture and shape priors based method for ultrasound image segmentation is presented. Texture features are extracted by applying a bank of Gabor filters on test images through a two-sided convolution strategy. Meanwhile, the shape constraints are described by an average zero level set function of the signed distance representations of the training data. Segmentation is implemented by calculating the parameters of the model to minimize a novel texture-based energy function. A series of experimental results on simulated images, natural images and real medical ultrasound images are demonstrated and discussed. These results are compared with other image segmentation methods and manual segmentation to evaluate the effectiveness of this novel knowledge-based segmentation approach.