An integrated global and local thresholding method for segmenting blood vessels in angiography

In clinical practice, digital subtraction angiography (DSA) is widely used to diagnose cerebrovascular disease based on detailed information about blood vessel structure. Challenges remain on accurately find blood vessel abnormalities in a time-limited manner. In this perspective, computer-aided analysis of DSA can assist clinicians in interpreting the images. Provide a method for extracting cerebral blood vessels from DSA images. In this work, we presented a new method for segmenting digital subtraction angiography (DSA) by incorporating both global and local information about an image to adaptively classify each pixel to the foreground and background. The method utilizes the global mean and standard deviation of an angiography and local mean and standard deviation within a sliding window to build two criteria. The two criteria contains both global and local characteristics about an image and individual pixels. The two criteria work together to reduce noise in segmentation and preserve valid details about the foreground. We tested the method on angiography and compared it with several widely used algorithms. In total, there were 72 DSA images in our dataset. Compared to Otsu, Niblack, iNiblack, Sauvola, Wolf, and CNW, our method achieved the best overall performance. The accuracy, Dice coefficient (Dice), and intersection over union (IoU) are 0.9777, 0.8500, and 0.7440, respectively. The results demonstrated that our method can obtain good outcomes, especially in achieving a balance between extracting the correct foreground and reducing incorrect classifications, and had the best performance among the methods being compared with.