Texture Segmentation on Synthesized Vascular Image

As one of the main diseases causing blindness, there is a trend in studying vascular segmentation for diabetic retinopathy (DR). Even though the texture feature has been widely applied to medical image applications, it is limited to offering pattern segmentation only based on the texture feature. As the key pattern of DR, the blood vessel covers a limited number of pixels compared to the other pattern on the retinal image, resulting in an imbalanced distribution. It brings a major problem in pixel-wise classification, which may lead to a failure in distinguishing the small pattern area. However, based on a good texture extraction technique, it is possible to solely offer pattern segmentation by texture. In this paper, we present texture segmentation by our proposed texture descriptor, C-BRIEF. We would show our experiment with the synthesized vascular image, which tackles the segmentation of small patterns. Our preliminary result shows 93.4% accuracy, 72.6% sensitivity and 95.1% specificity on the synthesized vascular image, that the texture segmentation can be achieved on the texture feature. This technique can also be extended to other segmentation of vessel-like patterns.