Research On The Piecewice Constant Level Set Method And The Mumford-Shah Model For Medical Image Segmentation

Retinal optic nerve diseases (such as glaucoma, diabetic retinopathy, age-related macular disease, etc.) are caused by a variety of retinal and neural tissue lesions, ultimately, it leads to blindness. These diseases in the world have high incidence rate, and most of them have no clinical signs at early stage. Therefore, treatment in early stage is very important. The segmentation, reconstruction, and measurement of optic nerve head or optic disc are most challenging problems as for the computer-aided diagnosis of retinal optic nerve diseases in last three decades. However, due to non-uniform illumination, vascular occlusion and other reasons, most of existing method's segmentation results may not be satisfying.We mainly focuses on how to use PCLSM based Mumford-Shah modelto segment the images of the optic nerve head. There are some reasons to choose the model: 1, it deals with the multi-phase image well; 2, it segments the image by only using a single level set; 3, and there is no need of manual intervention, so it has high degree of automation.Firstly, we can get the cup and the disk's morphological edge from segmenting the image of optic nerve head by using the PCLSM based Mumford-Shah model; Secondly, we removed the depressed and broken brink, and retained the edge-points of the cup and disk by combining of the priori knowledge of glaucoma optic nerve head images on cup and disk; And then, we used the technique of smoothing spline curve fitting to reconstruct the edge of the cup and disk obscured partially by vasculars. Finally, we reconstruct the optic nerve head's cup and disk image by using the fitted edge of the cup and disk, and extract the pathological measurement such as cup-disk ratio based on the reconstructed image.The results of the reconstruction, segmentation and measurement of the optic nerve head images in different stages demonstrated that this method is able to effectively segment, reconstruct and measure the optic cup and disk in optic nerve head images of different stages glaucoma patients, which are of poor quality, very low contrast, obscure due to blood vessels, and distinct inter-differences of individuals. We found that the PCLSM based Mumford-Shah model is very effective by comparing with the clinical conclusions and the segmentation results of Multi-layer Mumford-Shah model.