BLOOD VESSEL SEGMENTATION IN RETINAL FUNDUS IMAGES USING PRE-PROCESSING AND ITERATIVE REFINEMENT
Ms.A.Rajapackiam PG Scholar,
Department of Computer Science and Engineering,
Chandy college of Engineering,
Mullakadu,Thoothukudi(T.N)India.
rajiamrose91@gmail.com Mr.L.Arokia Jesu Prabhu AP/CSE ,
Department of Computer Science and Engineering,
Chandy college of Engineering,
Mullakadu,Thoothukudi(T.N)India.
jlplazer@gmail.com
ABSTRACT
Automatic diagnosis of several diseases such as Diabetic retinopathy, Hypertension etc. can be performed using the help of retinal fundus images. One of the important first steps in retinal fundus image analysis is the identification of blood vessels. Segmentation of blood vessels in fundus images has been well …show more content…
The camera is attached with an intricate microscope and is flash enabled. Special dyes such as fluorescein and indocyanine green are used with color filters to capture the intricate details. The optic disc, peripheral retina and macula regions of the eyes are captured in fundus photography. Retinal Fundus images play a vital role in the early diagnosis of Diabetes, Hypertension etc. The blood vessels in the retina include the arteries which carry oxygenated blood into the eye and the veins that carry deoxygenated blood out of the eye. The thickness of the veins and arteries are important indicators in Type-I Diabetes and Hypertension. In patients with diabetic retinopathy, timely diagnosis and treatment by retinal laser can prevent the loss of vision. In addition to medical diagnosis, fundus images are also used as general research tool.
Several methods have been proposed in literature for segmentation of vessels. Several supervised methods include k-nearest neighbour classification[4], neural networks[6], support vector machines[5], decision tree[7], boosting of weak classifiers etc. Unsupervised methods include morphological processing[8]-[10], model based methods[11]-[13], matched filters or multi-scale methods[14]-[16]. This paper presents a method which is a combination of morphological pre-processing and iterative refinement using region growing methodology. …show more content…
The first category includes wide and thick vessels that are easily distinguishable from the neighboring pixels based on the pixel intensities. The second category includes the fine small blood vessel branches that are not very distinct against the neighboring pixels. The extraction of major vessels ensures segmentation of the first category of blood vessels while the classification of the pixels in the vessel sub-image aids identification of the second category of fine vessels. The proposed approach of separating the methods for identifying the thick and fine blood vessel regions enhances the robustness of vessel segmentation on normal and abnormal retinal images in two ways. First, the major vessel regions comprising of 50-70% of the total blood vessel pixels are segmented in the first stage, thereby significantly reducing the number of sub-image vessel pixels for classification. This reduction in the number of vessel pixels under classification reduces the computational complexity and vessel segmentation error when compared to methods that classify all major and fine vessel pixels alike [2] [1] [3]. Second, the optimal feature set identified for sub-image vessel pixel classification are discriminative for the fine vessel pixel segmentation. These features aid elimination of sub-image vessel pixels from large red lesions and false bright lesion edges in retinal images with pathology. The most
Ms.A.Rajapackiam PG Scholar,
Department of Computer Science and Engineering,
Chandy college of Engineering,
Mullakadu,Thoothukudi(T.N)India.
rajiamrose91@gmail.com Mr.L.Arokia Jesu Prabhu AP/CSE ,
Department of Computer Science and Engineering,
Chandy college of Engineering,
Mullakadu,Thoothukudi(T.N)India.
jlplazer@gmail.com
ABSTRACT
Automatic diagnosis of several diseases such as Diabetic retinopathy, Hypertension etc. can be performed using the help of retinal fundus images. One of the important first steps in retinal fundus image analysis is the identification of blood vessels. Segmentation of blood vessels in fundus images has been well …show more content…
The camera is attached with an intricate microscope and is flash enabled. Special dyes such as fluorescein and indocyanine green are used with color filters to capture the intricate details. The optic disc, peripheral retina and macula regions of the eyes are captured in fundus photography. Retinal Fundus images play a vital role in the early diagnosis of Diabetes, Hypertension etc. The blood vessels in the retina include the arteries which carry oxygenated blood into the eye and the veins that carry deoxygenated blood out of the eye. The thickness of the veins and arteries are important indicators in Type-I Diabetes and Hypertension. In patients with diabetic retinopathy, timely diagnosis and treatment by retinal laser can prevent the loss of vision. In addition to medical diagnosis, fundus images are also used as general research tool.
Several methods have been proposed in literature for segmentation of vessels. Several supervised methods include k-nearest neighbour classification[4], neural networks[6], support vector machines[5], decision tree[7], boosting of weak classifiers etc. Unsupervised methods include morphological processing[8]-[10], model based methods[11]-[13], matched filters or multi-scale methods[14]-[16]. This paper presents a method which is a combination of morphological pre-processing and iterative refinement using region growing methodology. …show more content…
The first category includes wide and thick vessels that are easily distinguishable from the neighboring pixels based on the pixel intensities. The second category includes the fine small blood vessel branches that are not very distinct against the neighboring pixels. The extraction of major vessels ensures segmentation of the first category of blood vessels while the classification of the pixels in the vessel sub-image aids identification of the second category of fine vessels. The proposed approach of separating the methods for identifying the thick and fine blood vessel regions enhances the robustness of vessel segmentation on normal and abnormal retinal images in two ways. First, the major vessel regions comprising of 50-70% of the total blood vessel pixels are segmented in the first stage, thereby significantly reducing the number of sub-image vessel pixels for classification. This reduction in the number of vessel pixels under classification reduces the computational complexity and vessel segmentation error when compared to methods that classify all major and fine vessel pixels alike [2] [1] [3]. Second, the optimal feature set identified for sub-image vessel pixel classification are discriminative for the fine vessel pixel segmentation. These features aid elimination of sub-image vessel pixels from large red lesions and false bright lesion edges in retinal images with pathology. The most