Adaptive Segmentation of Vessels from Coronary Angiograms Using Multi-scale Filtering

In this paper, we propose a novel automatic and effective method for the vessel segmentation based on Hessian matrix. First, to obtain vessel structure more reliably, we select 25 frames of well-contrast angiograms automatically for further vessel segmentation. Second, we define an adaptive feature transform function using the gray value and the scale to improve the feature response. First, we enhance the original image contrast by reducing the gray level of vessels, and then adjust the eigenvalues by taking the enhanced gray level as a rectification factor to better differentiate between the vessel points and the background ones. We also use the scale as a weighting factor. In the way, vessel and background can result in different feature responses. Next, Hessian with Gaussian derivatives at multi-scales was computed due to different vessel widths, and the maximum response is selected as the vessel candidate. Finally, we use a connected components labeling method to detect the largest connected component as our segmentation result. In our experiments, 14 angiogram sequences are used to evaluate the accuracy. The accuracy of well-contrasted angiograms is 95.66%. For the result of segmentation, 40 angiograms are used to compare with other methods. After inspecting by a cardiologist, the experimental results show that our method works well for the enhancement and segmentation of vessels.