This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
Retinal image analysis is a key element in diagnosis retinopathies in patients. The patterns of disease that affect the fundus of the eye are varied. Therefore, a trained human observer such as an ophthalmologist is required to identify these patterns. By analyze some features in a retinal images, ophthalmologist can diagnosis the possible ocular disease that occur in the retinal such as diabetes retinopathy (DR), macular degeneration, glaucoma and etc. Main features to be observed for diagnosis disease are divided into two categories which are bright spots and dark spots. The bright spots in retinal images consist of Optic disc, exudates, and cotton wool while dark spots are blood vessels, haemorrhages, and microaneurysm.
Most of the original retinal images are low contrast and have occurrence of noise that will makes the diagnosis process become tougher because all of the features are unobvious. It is hard for an ophthalmologist to observe a feature accurately in poor quality retinal images.
According to the National Eye Institute, blindness or low vision affects 3.3million Americans age 40 and over. This figure is expected to reach 5.5million by the year 2020 . As the increasing numbers of retinopathy cases, ophthalmologists need to analyze more retinal images which increase their workload. Ophthalmologist might need a system which can assist them to improve the efficiency of diagnosis process.
Therefore, in this project, we are going to build a Decision Support System based on Retinal Images for disease diagnosis process.
Pre-processing on retinal images to enhance the local contrast and reduce the noises in the image. The pre-processing can improve the obviousness of important features in bright spot such as optic disk, exudates and cotton wool spots.
Detect and extract the features in bright spot of retinal images which are optic disk, exudates and cotton wool spots.
Present each main feature in separate images in user interface.
Analyze on each feature based on rules of diagnosis to decide whether it is normal or abnormal.
Diagnosis whether any ocular disease is proven based on the analyzed features.
A simple and user friendly interface is necessary for this system because the user might not familiar in programming code.
To build a decision support system using retinal images for disease diagnosis such as Diabetes Retinopathy. This system will figure out the possible ocular disease based on the features in retinal images.
To process on retinal images using Matlab Image Tools for extracting the features needed in disease diagnosis process. These features include optic disk, exudates, and cotton wool which are the bright spot in the retinal image.
To help the ophthalmologist to improve their productivity, efficiency and cost effective in the disease diagnosis process. Ophthalmologist can identify each feature by using this system rather than manual diagnosis by analyze using the original retinal images which features are not shown obviously.
To do research on methodologies for processing retinal images such as detection and extraction of optic disk, exudates, and cotton wool. Besides, research on certain ocular disease and the rules of diagnosis these diseases are needed in this project.
1.5 Gantt Chart
Background And Literature Search
Retinal images are widely used as tools to diagnose retinopathies by ophthalmologist. Retinal images are obtained using fundus camera. Figure 2.1 shows a fundus camera and figure 2.2 shows a normal human retinal obtained by fundus camera. A fundus camera or retinal camera is a specialized low power microscope with an inner attached camera designed to take photos for the interior surface of the eye . The interior surface of the eyes includes retina, optic disc, macula, and posterior pole.
Figure 2.1 Example of a fundus camera Figure 2.2 A normal retinal image.
The features that can found in majority of retinopathies retinal images are categorize into two groups which are bright spots and dark spots. Bright spots include optic disc, exudates and cotton wool spots while dark spots include blood vessels, haemorrhages and microaneurysms. By analyze these features, certain retinopathies can be detected. Examples of retinopathies that can be detected from retinal images are diabetic retinopathy, glaucoma, macular degeneration, hypertensive retinopathy and etc.
Figure 2.3 shows a retinal image for diabetic retinopathy. Most of the features are shown and labelled in this retinal image.
Features in Retinal Images for Disease Diagnosis
Exudates, cotton wool spots and optic disc are three types of bright spots in retinal images.
Exudates or hard exudates: visible as bright yellowish deposits with sharp margins on the retinal due to the leakage of blood from abnormal blood vessels. The weakened vessels walls causes out-pouching in their walls called microaneurysms, which may also leak. Exudates frequently arranged in circular pattern or crescents surrounding zones of retinal edema or group of microaneurysms. Besides, exudates also possible to arrange as individual dots, sheets, or confluent patches. Exudates represent accumulations of lipid and protein. If exudates encroach on the macula, vision will be affected .
Cotton wool spots: Cotton wool spots or soft exudates appear as white, pale yellow fluffy opaque area with ill-defined edges in retinal. They result from the damage of nerve fibers whereby the blood supply to that area has been impaired. The nerve fibers in that particular area are injured due to the absence of normal blood flow through the blood vessel there. Therefore, swelling will occur at that spot and it appear as cotton wool spots. Diseases such as diabetes and hypertension will affect the retinal and cause the occurrence of cotton wool spots .
Optic disc: Optic disc is the brightest part in the normal retinal image. It is pale, round or vertically oval disc. Normally, the disc is orange to yellowish-pink in colour with well defined margins. An optic disc is the entrance region of optic nerves and blood vessels to the retinal. It always acts as a landmark for other features in retinal image. In retinal image analysis, location of optic disc is important to measure distance and identify some anatomical parts in retinal images. The lack of light-sensitive cells, rods and cones at the optic disc results a physiological blind spot in the visual field of each eye. Glaucoma, a disease cause by degenerative optic nerve is basically related with a sustained increase of the eye pressure .
The dark spots in retinal can be dividing into three features which are blood vessels, haemorrhages, and microaneurysms.
Blood vessels: Blood vessels are the blood supply for retinal. Blood vessels appearance is an important indicator for many diagnoses such as diabetic retinopathy and hypertension. It can reflect different states of numbers of diseases, which also the pre-characteristic for the registration and mosaic of retinal images. Observable features of blood vessels such as diameter, colour, tortuosity (relative curvature), and opacity (reflectivity) can provide information on pathological changes caused by some diseases. The abnormalities of retinal blood vessels include blockages and bleeding (haemorrhages) from them.
Haemorrhages: Haemorrhages is the abnormal bleeding of the damaged blood vessels in retinal. The appearance of haemorrhages may have many kinds of shapes sometimes resembling bundles of straw but they also can be round or flame shaped . The bleeding of vessels which are haemorrhages can cause temporary or permanent loss of visual accuracy. There is various cause of haemorrhage which major causes are diseases such as diabetic retinopathy, hypertension and prematurity retinopathy. Besides, it can also caused by shaking, particularly in young infants.
Microaneurysms: Microaneurysms are included in dark spots in retinal image that appear as small dark reddish dots on retinal surface. Its definition is less than the diameter of the major optic veins as they cross the optic disc. Microaneurysms are small out pouching in capillary vessels. Normally, capillary vessels are not visible in retinal image. Due to the increasing number of microaneurysms, these small dots appear between the visible retinal vasculature. Microaneurysms are caused by weakening of vessels wall or diseases include diabetic retinopathy.
Nowadays, there is an increasing interest for creating system and algorithms that can support for screen a big amount of patients for sight threatening diseases such as diabetic retinopathy, glaucoma, and hypertension retinopathy. These systems and algorithms provide automated detection of these retinopathies. Retinal images are widely used as tools to screen and diagnose retinopathies by ophthalmologist. Currently, digital image processing is very famous and practical for retinopathies diagnosis. By using image processing, features such as blood vessels and exudates can be detected, extracted, and analyzed for the purpose of diagnosis. In the literature, there are some examples of image processing techniques which have been applied in identification and detection of features in retinal image for disease diagnosis. Research is done on several literatures about features detection and extraction in retinal images.
Anantha et al. have implemented an algorithms for automatic detection of exudates by using dynamic thresholding and edge detection (IDTED). In first step, green component of the pre-processed image was divided into blocks of 64x64 pixels with 50% overlap with each other. A dynamic thresholding was selected based on the histogram of each blocks. High threshold value is set if the histogram was unimodal, else threshold value can be find using Otsuâ€™s thresholding algorithms. As each pixels belongs to four blocks, a particular pixels was classified as an exudates pixels if its intensity value was higher than the interpolate of threshold value of the four blocks to which it belongs. The result of this classification process was a binary image. The third step was edges detection which aim to detect all objects with sharp edges included exudates by using Canny edge detector. A binary image containing sharp edges was obtained from thresholding using a defined global threshold value. The last step was false positive region elimination. The binary images obtained from step 2 and 3 were combined by a feature based AND. Pixels that ON in both image were marked as white in a new binary image and became the exudates region. The IDTED algorithm was tested against 25 digital retinal images and compared with the performance of a human grader, has shown a mean sensitivity of 99% and a mean predictivity of 93%.