Automatic prostate segmentation from transrectal ultrasound images

In this paper, a fully automatic slice-based 3D segmentation method is proposed to segment the sequence of the TRUS prostate images. The initial contour was generated on an initial frame of a stack of 2D ultrasound images acquired for each subject. This initial frame and the initial prostate boundary on this frame were automatically determined based on the radial bas-relief method. Then a region-based level set framework was applied to deform the initial contour using the approximate and detailed coefficients generated using dyadic wavelet transform. Finally, the segmented contour on initial frame is propagated to the adjacent frames as initial contour until all the frames are segmented. 50 transverse prostate images from three patients were segmented using the proposed algorithm for validation. The average mean absolute difference (MAD) between the contour segmented using the proposed algorithm and the manually outlined contour is less than 5 pixels.