Research On Segmentation Algorithms For Ultrasound Images Of The Prostate

Ultrasound has been increasingly used in surgical procedures of the prostate in recent years. Segmentation of the prostate boundary from the ultrasound image is clinically useful in such situations as accurate volume measurement, tumor margin estimation, and real-time image guided biopsy. Traditional manual outlining is tedious and time-consuming. Automatic segmentation of the prostate is a challenging task due to the high level of speckle noises and very low signal-to-noise ratio in the ultrasound image. As a result, semi-automatic segmentation is studied broadly.The segmentation of the prostate ultrasound image is divided into two steps: preprocessing and image segmentation. The main purpose of the preprocessing is to reduce noises and enhance contrast. Median filtering and Gaussian smoothing are often used. The algorithm called"sticks"can reduce speckle and enhance the contrast around the prostate boundary. Different models are used in image segmentation. Active contour model can control an active contour moving to the target by minimizing its energy. Gradient vector flow snake can constrain some noises and reduce the sensitivity to the initial curve. Balloon snake may create knots in the curve because of the strong speckle noise. The goal of segmenting the prostate boundary can be achieved by removing the knots and adjusting the internal energy parameter according to the movement of the curve. The computation of discrete dynamic contour is easier than snake. It requires giving only four points to create the initial curve. Since the capture range of the model is narrow which means that the initial curve should be close enough to the target, we accelerate the speed of the curve moving to the target by incorporating an additional inflation force and changing the mass parameter of different vertexes, in order to improve the result of the segmentation.The performance of the algorithm is compared to manual outlining. The experiment results show that GVF snake may miss some weak edges and can not avoid the interfering of strong noises; improved balloon snake may cause leaking at the weak edge. It shows that improved discrete dynamic contour model performs better segmentation, by computing distance-based metrics and area-based metrics.