INTRODUCTION
IN retinal images, blood vessels are landmarks for localizing the optic nerve, the fovea and lesions, which are useful for medical diagnosis. However, in these images, many vessels are narrow and close to each other, forming a network-like structure. Also, due to the reflection on the tiny uneven surface of the soft tissue in the image, the low contrast between the vessel and background, and the pathological variations, detecting blood vessels automatically from a retinal image is a challenging problem. A number of techniques have been proposed to solve this problem. They can be classified into unsupervised and supervised methods. In an unsupervised method, a pixel is assigned to a candidate vessel according to several predefined criteria. Chaudhuri et al. propose a matched filter response (MFR) method , which applies rotated Gaussian filters to the image.
If the pixel has a large filtered value, it is a part of a vessel. Jiang and Mojon propose an adaptive thresholding technique for vessel segmentation. The detection is conducted in different levels of image intensities. For example, the pixels with intensity values from 80 to 100 are grouped into one level while the pixels with the intensity values from 110 to 140 are grouped into another level. In each level, candidate vessels are obtained by thresholding. In a supervised method, the criteria are determined by the ground truth data based on given features. However, a prerequisite for a supervised method is the availability of the ground truth data that are already classified, which may not be available in real life applications. An average of 2 h is needed to label a single retinal image. Staal et al. employ more than 10 features, including width of the vessel, intensity, and edge strength.
Soares et al. make use of the Gabor wavelet transform. As supervised methods are designed based on preclassified data, their performance is usually better than that of unsupervised ones and can produce very good results for healthy retinal images.Although existing methods are robust for many retinal images,there is still room for further improvement, especially for pathological retina images. A pathological retina may suffer from a certain disease and there may contain some spots (light or dark). Existing methods may recognize those spots as part of the vessels. Due to the unknown characteristics of a pathological region, widely used features such as intensity are not effective for solving the problem. The supervised method of Soares et al. has the same limitation. In their paper, the authors stated “Though very good ROC results are presented, visual inspection shows some typical difficulties of the method that must be solved by future work.
The major errors are in false detection of noise and other artifacts. False detection occurs in some images for the border of the optic disc, haemorrhages, and other types of pathologies that present strong contrast”.Researchers have made many proposals to analyze pathological retina images. Chanwimaluang et al. suggest that more constraints should be added in order to remove the spots .However, there is no discussion on how we can select the constraints.One of the widely used constraints for noise removal is the split-and-merge system. If the size of an object is small
enough, it will be treated as noise.
An implicit assumption for this pruning operation is that the size of the vessel should be larger than that of noise. However, many blood vessels after splitting are very short and can be removed easily. Staal et al. suggest to solve the problem by removing a pathological region in a preprocessing step or by selecting more training data sets that include pathological features in a supervised approach .The removal of a pathological region in a preprocessing step can be difficult. Actually, as we do not know where a pathological region will be, it is not easy to remove it in advance. One way is to use an adaptive thresholding technique proposed by Jiang and Mojon . They separate the image into several levels by a thresholding method based on pixel intensities.
However, if there are large intensity variations within the spots, this method may not work well. To produce more training data, one possible solution is to trace the vessels from the optic nerve, a user-defined point or a given labeled vessel on the image. However, if the spots are close to the blood vessels, which are not connected to the optic nerve or the given information, it may not be possible to remove them.
IN retinal images, blood vessels are landmarks for localizing the optic nerve, the fovea and lesions, which are useful for medical diagnosis. However, in these images, many vessels are narrow and close to each other, forming a network-like structure. Also, due to the reflection on the tiny uneven surface of the soft tissue in the image, the low contrast between the vessel and background, and the pathological variations, detecting blood vessels automatically from a retinal image is a challenging problem. A number of techniques have been proposed to solve this problem. They can be classified into unsupervised and supervised methods. In an unsupervised method, a pixel is assigned to a candidate vessel according to several predefined criteria. Chaudhuri et al. propose a matched filter response (MFR) method , which applies rotated Gaussian filters to the image.
If the pixel has a large filtered value, it is a part of a vessel. Jiang and Mojon propose an adaptive thresholding technique for vessel segmentation. The detection is conducted in different levels of image intensities. For example, the pixels with intensity values from 80 to 100 are grouped into one level while the pixels with the intensity values from 110 to 140 are grouped into another level. In each level, candidate vessels are obtained by thresholding. In a supervised method, the criteria are determined by the ground truth data based on given features. However, a prerequisite for a supervised method is the availability of the ground truth data that are already classified, which may not be available in real life applications. An average of 2 h is needed to label a single retinal image. Staal et al. employ more than 10 features, including width of the vessel, intensity, and edge strength.
Soares et al. make use of the Gabor wavelet transform. As supervised methods are designed based on preclassified data, their performance is usually better than that of unsupervised ones and can produce very good results for healthy retinal images.Although existing methods are robust for many retinal images,there is still room for further improvement, especially for pathological retina images. A pathological retina may suffer from a certain disease and there may contain some spots (light or dark). Existing methods may recognize those spots as part of the vessels. Due to the unknown characteristics of a pathological region, widely used features such as intensity are not effective for solving the problem. The supervised method of Soares et al. has the same limitation. In their paper, the authors stated “Though very good ROC results are presented, visual inspection shows some typical difficulties of the method that must be solved by future work.
The major errors are in false detection of noise and other artifacts. False detection occurs in some images for the border of the optic disc, haemorrhages, and other types of pathologies that present strong contrast”.Researchers have made many proposals to analyze pathological retina images. Chanwimaluang et al. suggest that more constraints should be added in order to remove the spots .However, there is no discussion on how we can select the constraints.One of the widely used constraints for noise removal is the split-and-merge system. If the size of an object is small
enough, it will be treated as noise.
An implicit assumption for this pruning operation is that the size of the vessel should be larger than that of noise. However, many blood vessels after splitting are very short and can be removed easily. Staal et al. suggest to solve the problem by removing a pathological region in a preprocessing step or by selecting more training data sets that include pathological features in a supervised approach .The removal of a pathological region in a preprocessing step can be difficult. Actually, as we do not know where a pathological region will be, it is not easy to remove it in advance. One way is to use an adaptive thresholding technique proposed by Jiang and Mojon . They separate the image into several levels by a thresholding method based on pixel intensities.
However, if there are large intensity variations within the spots, this method may not work well. To produce more training data, one possible solution is to trace the vessels from the optic nerve, a user-defined point or a given labeled vessel on the image. However, if the spots are close to the blood vessels, which are not connected to the optic nerve or the given information, it may not be possible to remove them.
No comments:
Post a Comment