Interpolation In Medical Image Processing And Segmentation Techniques

In order to obtain and implement the interpolation with high quality and efficiency, and high reliable segmentation algorithms with the minimization human interaction in the medical image sequence, interpolators and segmentation are studied in the paper. The interpolation is applied widely in medical image processing. Since the ideal interpolation function spatially is unlimited, several class of practical interpolation kernels have been introduced: piecewise local polynomials, windowed sinc, Lagrange, Gaussian et al.. and their properties have been analyzed in spatial and frequency domain and from evaluation and result. In terms of their frequency responses, the best choices for medical image interpolation: the 6×6 Blackman-Harris windowed sinc interpolator, and the C2-continuous cubic kernels with 6 and 8 supporting point. In terms of the quantitative error and evaluations, in practice the best choice is the cubic kernel with 6 supporting points.Segmentation is an important step in many medical image processing applications. Reliable contours are usually found by a combination of human and computer image analysis. In the field of medical image sequence analysis, we would like to minimize the human interaction. Two interactive segmentation methods, snake(active contour) and live wire, which are very popular, have been introduced for the medical image segmentation. A snake is a spline curve, which is controlled by an energy equation. To minimize the snake's total energy, the finite difference and Greedy algorithm have been implemented. It applied to an image sequence by taking advantage of coherence between neighboring images in the sequence. And Live wire is a pixel-level contour-finding algorithm. It finds a minimal-cost path between a start and a goal point in a directed graph. The Live-snake has been brought forward. Live wire is trained and applied to segment a single image in a sequence of medical image, then copy the contour to the next image in the sequence to construct a snake, where to minimize the snake energy to find an optimal configuration. The new snake is used to construct a new live wire, which is then used to segment the new image. The process repeats on subsequent images until the entire sequence has been segmented. This combines the strengths of both techniques and improves the live wire training efficiency. Further, at the end, the shape-based interpolation is introduced to find the missing contour and avoid segmenting again.