To build the system of computer- aided diagnosis for coronary heart disease and provide some parameters for lesion detection, a new method for coronary artery lumen segmentation based on the coronary artery centerline is proposed, which is the result of the algorithm of improved minimal cost path with second order Hessian vessel filter. The method uses the centerline as the tracking path to get every cross section of the vessel, and then ray casting is used for locating the coronary artery lumen along every ray. The disperse coordinates of the lumen location in every ray can be fitted to an outline of the coronary lumen in every cross section by piecewise spline fitting. As a result of all the outlines of coronary artery lumen in cross section reconstruction, a 3D coronary artery lumen can be built by marching cubes algorithm. The result of the method, which has high computing performance, has nice visualization of vessel lumen and can be the basis of the vessel parameter calculation.