Retinal vessel extraction and diameter calculation based on tensor analysis

The segmentation of retinal vessel is widely used in the ophthalmopathy to assist the doctor in diagnosing. However, most existing segmentation methods are sensitive to background noise, or not full of capability on capillary recognition, or inclined to distort vessel width. In this paper, a new retinal vessel segmentation algorithm is proposed, which generates the eigenvalues by computing Hessian matrix of image, then creates a vessel probability function using Tensor invariant feature to determine the vessel pixels. This Retinal Vessel Extraction algorithm achieves an improvement in vessel segmentation accuracy, averagely 95.36 percent of pixels are extracted exactly, relatively is robust in terms of noises and could detect capillary with more authentic width as for most vessels. Meanwhile, a vascular diameter calculation method is proposed according to the gray value of vascular cross-sectional model, which can find out and classify the center point of blood vessel by analytic hierarchy process(AHP). Finally, we evaluate our algorithm on DRIVE dataset and the results show that it is effective in segmenting retinal vessels and improves the prevalent insufficiencies.