Automatic detection of retinal capillary nonperfusion via a new active contour model

Capillary nonperfusion (CNP), which is one of characteristic features in diabetic retinopathy patients, is an important judgement of the appearance of retinal diseases. This paper presents a novel, fast and hybrid method to detect CNP. This model combines region-based active contour model(ACM), Graph-Cut and Fuzzy Possibilistic C-Means(FPCM). We modify ACM with a new energy function to reduce suspicious pixels, and employ Graph-Cut to replace the level set method to get a faster arithmetic speed, which also make our method insensitive to the initial contour. And FPCM is applied to average grey levels over different clusters instead of the grey levels of the pixel in order to alleviate the influence caused by noise and capillaries. Experimental results demonstrate that the proposed method outperforms other ACM methods in detecting CNP with better accuracy and sensitivity.