The present invention provides a method for reconstruction of 3D vessel image. The method is able to eliminate the noise in volume rendering properly, and adapt to the vessel branches and the scale variation. In the present invention, the image is intensified in different scales, and then the eigenvalues of the image are analyzed with Hessian Matrix to determine the voxelssymmetry of the image. The provided eigenvalue is the symmetry value, Sk. This value describes the detection sensitivity for tubular structure and the smoothness of the cross-sectional contour in the image meanwhile the polarity of the value assists in detecting hollow tubes by the following weighting. The symmetrical eigenvalues are established into multi-dimension with the spherical harmonics. After integrating in the spherical coordinate, the high-dimension eigenvalue can be obtained in different angles on the spherical surface. The results of weighting, N(Sk) from multiple scales, can then be integrated as the polar distribution on the spherical surface to be the symmetrical structureness. The vessel tracking and the 3D image reconstruction can be achieved by using Hessian Matrix eigenvalues to compute the symmetrical structureness obtained from the polar distribution. The result of the present invention has the intensity of the tubular structure, the linear correlation of the image and self-adapting normal orthogonality.
