Image segmentation is a challenging part in medical imaging due to the difficulty of the image and the structure of the anatomical models where is use to diagnose some disease. Segmentation has been use in medical imaging such as glaucoma detection. Nowadays it is become a very important part in glaucoma detection research. The main purpose of this study is to segment and extract the glaucoma abnormalities images in order to detect the glaucoma disease. Therefore, this proposal is use to discover the application of evolutionary algorithm techniques which is the Selfish Gene Algorithm in the image segmentation. This is includes the feature extraction of the glaucoma disease. Digital image processing techniques, such as preprocessing, morphological operations and segmentation are widely used for the automatic detection of optic disc, blood vessels and computation of the features. This study use to extract features such as cup to disc (c/d) ratio, ratio of the distance between optic disc center and optic nerve head to diameter of the optic disc, and the ratio of blood vessels area in inferior-superior side to area of blood vessel in the nasal-temporal side. These features are validated by classifying the normal and glaucoma images using the selfish gene algorithm. The proposed technique presented in this paper indicates that the features are clinically significant in the detection of glaucoma. It may help ophthalmologists to make faster decision in future.
Problem statement / thesis statement / problem identification
Get your grade
or your money back
using our Essay Writing Service!
Glaucoma is characterized by increased pressure within the eyeball and therefore it can cause progressive damage to the optic nerve. The most blindness is cause by a glaucoma disease. Glaucoma cannot be cured but if it is discovered early enough, it can be prevented with the right treatment . Glaucoma detection test also is time consuming and need specialist to handle the equipments. Although, present method of the glaucoma diagnosis such as Optical Coherence Tomography (OCT) and Heidelberg Retinal Tomography (HRT) can be use in detecting the glaucoma disease, but the cost of both method is very high and most of the hospital cannot afford them .
Presently, most of the computer assisted analysis of fundus images done manually segmentation for glaucoma estimation. Therefore, the small problem in localization occurs. This will effect the measurements and consequently in the diagnosis .
Aims and objectives of the research
The project aim of this study is to implement image processing techniques for segmenting and extracts the glaucoma abnormalities images. There are several objectives that should be achieved in this study.
The objectives are:
To segment, extract and classify the glaucoma abnormalities images.
To design an algorithm for segmenting and feature extraction the glaucoma based on selfish gene algorithm.
Scope and limitation of the study
The scope for this project covers the segmentation and extraction of the glaucoma abnormalities images. Digital fundus camera is chosen as the imaging modality for capturing images data in this project. Next, it's due to the cost and availability outstanding as compared to other imaging modalities.
In developing this stand alone prototype, the supported software that will be used is Borland C++ Builder 6.0. It can be run on a personal computer with a standard WINDOWS platform. Besides, the digital fundus images format is limited to 24-bit Bitmap (.bmp).
Additionally, the prototype is specially designed for the expert users that refer to ophthalmologists. Therefore, there will be a few medical terms in a prototype that can only be appreciated by these target users.
There are many types of eye disorder such as glaucoma, cataract and conjunctivitis. This study is focus on glaucoma disease. The glaucoma is an eye disorders which is characterized by progressive optic nerve damage at least partly due to increased intraocular pressure (IOP). It is cause by increasingly the optic nerve. In worldwide, glaucoma is the 3rd most common cause of blindness and the 2nd most common cause of blindness in the US. There are at least 3 million Americans and 14 million people worldwide have glaucoma, but only half are aware of it. Glaucoma occurs 6 times more common among people 60 years and above but it also can occur at any age . Based from the explanation by ophthalmologist and glaucoma specialist in Solano Regional Medical Group in Fairfield, glaucoma screening is something that should give high priority same level as prostate cancer screening and breast cancer screening. Thus, their teams are trying to give awareness among other people such as the patients, primary care providers and other eye care professionals . Yet, there is no cure for glaucoma disease. Only by detecting and prevention can avoid us from loss vision .
Always on Time
Marked to Standard
There isÂ two kinds'Â glaucomaÂ namely primary open angle glaucoma and angle closure glaucoma. Primary open angle glaucoma is the most commonly occurs. In this case, the eye's drainage system becomes would be clogged all the time and eventually brought about optic nerve damage. Angle closure glaucoma is much rarer. It may also be silent and chronic because the normal flow of eye fluid becomes suddenly blocked .
In many recent years, imaging devices or equipment such as Optical Coherence Tomography (OCT), Heidelberg Retinal Tomography (HRT) and Digital Fundus Camera have been developed in order to diagnose glaucoma . Chen et al (2008) found that Optical Coherence Tomography (OCT) is similar to ultrasound imaging, but it uses light instead of sound to acquire high-resolution images of ocular structures. OCT also have same characteristic to ultrasound where an image is backscattering of different ocular structures or created owing to differences in the reflectivity. OCT can directly capture both the Optic Nerve Head (ONH) and the retinal nerve fiber layer (RNFL). But ultrasound imaging have limitation where sound travels more slowly and light travels so fast that a fundamentally different way to process information from the eye is needed in OCT. Different with another method which is the Heidelberg Retina Tomography (HRT). It is a confocal laser scanning system where is designed for acquisition and analysis of three dimensional images of the posterior segment. It can give the quantitative assessment of the topography of ocular structures and can see the precise follow-up of topographic changes .
Although, both methods can be used in glaucoma detection, it is need more time and specialist to handle the equipments. Furthermore, the equipments are very high cost. To overcome this problem, digital fundus image can be use to replace the present equipments. The easy ways to capture the anatomical features are by using digital fundus camera. It is produce digital fundus images. The images will be captured on film or stored in digital form and displayed on a monitor . Fundus images are easy to manipulate and it give useful information that ophthalmologist needed. It also consists with light source. This fundus image can be used to diagnose eye diseases like diabetic retinopathy, glaucoma .
In the last decade, diagnostic medical imaging has started on a transformation. Until recently, specialists such as radiologists, cardiologists and ophthalmologist primarily worked with images captured on film and videotape, but many healthcare facilities are now contemplating making the move to digital or film less imaging systems. Diagnosis is used to determine and identifying a medical condition or disease by its signs, symptoms, and from the results of various diagnostic procedures. Digital methods are used in medical diagnosis for better result and expectation.
In computer vision, segmentation refers to simplify and change the representation of an image into something that is more meaningful and easier to analyze. For medical imaging purposes, segmentation is used to extract specific organs, vascular structures, tissue types or lesions. There are many methods for image segmentation has been presented . The main propose in this study is use to discover the application of new evolutionary algorithm techniques which is the Selfish Gene Algorithm in the image segmentation. This is includes the feature extraction of the glaucoma disease. Image segmentation is widely used for the automatic detection of optic disc, blood vessels and computation of the features. This study use to extract features such as cup to disc (c/d) ratio, ratio of the distance between optic disc center and optic nerve head to diameter of the optic disc, and the ratio of blood vessels area in inferior-superior side to area of blood vessel in the nasal-temporal side. These features are legitimized by classifying the normal and glaucoma images using the selfish gene algorithm.
This research may involve new algorithm for the image segmentation and feature extraction purposes. Finally, it is believed that this study can provide better framework to solve the image segmentation problem and significant in the detection of glaucoma.
Research methodology / research design
Defining a methodology is very important in every project development. Methodology gives some guideline and flow on how the particular project or system is being implemented. In this research, research methodology that being used consists of four phases. Each phase for this project is show briefly in table 1.
This Essay is
a Student's Work
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.Examples of our work
SYSTEM DESIGN AND DEVELOPMENT
Figure 1 Proposed research design
Table 1 Proposed Research Framework
Search journals/articles & do literature review
Create a citation table based on journals that had choose
Read journals, Articles and thesis
Citation table from 100 journals
Find out all techniques that had been used by other researchers
Determine the hardware /software to be use.
Discuss with supervisor
which software/ hardware
(Aser Travelmate 4050,
RAM 1.6GB, 80GB HDD
-Study programming skill in the programming software.
Collect digital fundus Images
Digital fundus images
are collected from
Faculty of Health
Sciences,UiTM and existing databases.
-50 fundus images will be collected from Faculty of Health Sciences
-40 fundus images from DRIVE database.
SYSTEM DESIGN &
. Determine technique to be used
Identify image pre-processing method
Determine segmentation method to detect glaucoma
Identify the feature extraction
*Genetic Algorithm *Evolution Strategies *Genetic Programming *Memetic Algorithm *Selfish Gene Algorithm
Study the specific feature to determine glaucoma
Prototype of glaucoma detection
-Choose Histogram Equalization.
-Choose Selfish Gene Algorithm (SGA)
a) Cup-disc-ratio b) ratio of the distance between optic disc center and optic nerve head to diameter of the optic disc c) the ratio of blood vessels area
RESULT AND ANALYSIS
Mean of the Cup Disc Ratio
Standard deviation of the cup disc ratio
1. Statistical table
a) Compare with every data
b) Make a conclusion
2. Survey table
a)Compare with every ophthamologist
b) Find good idea from opthamologist
As showed in Figure 1, proposed research design are consists with information gathering, data collection, hardware and software requirement, system design and development and result analysis. Information gathering involve literature review and citation table. In this phase, data & information are collected from many sources.
Data collection is consists of image acquisition task where collecting primary data are done here. Fundus image will be obtained from Dr.Chen Ai Hong, Department of Optometry, Faculty of Health Science, Universiti Teknologi MARA, Shah Alam. Beside that, data from existing databases which is from Image Sciences Institute  also can be used such as from the ophthalmologist website. The output of this phase is fundus images which are used in the next phase.
The third phase is hardware and software requirement. Suitable hardware and software are needed to develop this project.
System design and development is fourth phase. It is involve image enhancement and identifying the features of normal and glaucoma. The fundus image that obtains from phase data collection are use to be enhance. The enhanced fundus image will be input in normal and glaucoma's features identification. During this phase, the architecture of the project is build up. System design is the most importance part which is the implementation of the system. There are two main tasks involve in this phase. The first task is glaucoma feature extraction. This task will come out with the characteristic and features of normal and glaucoma. The second task is glaucoma segmentation. For this project, the method for segmentation that is being used in fundus image segmentation is selfish gene algorithm.
For the final phase, result and analysis will be obtained. Result and analysis is important for future review and references to other researchers. Therefore, documenting of the results analysis is found throughout the development of this research. It is used to see the accuracy for each of the processing outcomes which make this research more reliable. In order to obtain good parameters and the accuracy of objectives, the segmentation using new evolutionary algorithm which is the selfish gene algorithm will be analyze and discuss.
Phase 1: Information Gathering
In this study, literature review and doing the citation table is very important to develop a prototype. From the literature review, we can determine the technique that we are going to use. Every journals and articles have different methods and problems that applied in the research. From there, we can determine which method could be used in the project.
Phase 2: System Requirement
There are specific hardware and software requirements needed in order to make this project more reliable.
Hardware requirements for this project are:
Core 2 Duo processor 1.83 GHz
2 GB SDRAM
160 GB Hard Disk
Software requirement for this project is:
Builder C++ v6
Phase 3: Data Collection
Data collection of the patients is including the patient's criteria which are background of the patient and fundus image of the patient. It includes the image acquisition of human eye images or fundus images that are captured by Digital Fundus Camera. In this part also need to look at data collection instrumentation which is the data must be analyze and classify it before can be use in the implementing the prototype.
There are 50 fundus images will be collected from Faculty of Health Sciences, Universiti Teknologi MARA. Data collection from existing database also will be use as testing inputs. There will be 40 fundus images from DRIVE database . Patient's criteria are needed in order to determine whether the fundus images are accurate or not with the results.
a) Normal 's patient images
All patients' brain must be in normal condition and free from any diseases.
b) Glaucoma's patient images
All patients who are classified having ocular hypertension (OHT) and suspicious discs.
Phase 4: System Design & Development
In this phase, there are two prior tasks which are pre-processing and processing. In pre-processing phase, it is involve image enhancement and identifying the features of normal and glaucoma. The processing phase is design and reconstruction of the algorithm which is the most important phase in the proposed research framework. It is involves primary glaucoma segmentation and feature extraction.
The aim of image enhancement is to improve the perception of the data in images or to provide better images as input for other image processing techniques. It is help diagnostic detail more obvious with sharpening of the object in the images that are useful for analysis such as edges and boundaries. However, when image enhancement techniques are used as pre-processing tools for other image processing techniques, then quantitative measures can determine which techniques are most appropriate. This research will probe some of the potential image enhancement techniques that may be used to enhance the fundus images obtained.
Identifying features between normal and glaucoma images
In order to detect the glaucoma, features of normal and glaucoma's identification are needed. The first feature that have been identify is the cup to disc ratio, which specifies change in the cup area due to glaucoma. For glaucoma, the cup area will increase and shift move the optic nerve head towards the nasal side. The position of optic disc center is computed as a distance. The second feature is the ratio of this distance to the diameter of the optic disc. The ratio of total area of the blood vessels in inferior and superior side of the optic disc to the total area of the blood vessels in nasal and temporal area (ISNT) is taken as the last feature.
Figure 5. a) Normal eye fundus image and b) glaucomatous eye fundus image
Figure 7. Detail of normal and Glaucoma Optic Nerve
Based from article "Making the diagnosis of glaucoma" by Bores L.D. (2002), ophthalmologist are looking in this three criteria in diagnosis the glaucoma. First is evaluated IOP (intraocular pressure), most ophthalmologist list IOP as suspicious for presence of glaucoma. But elevated IOP is only risk factor and is not prognostic. Second is optic nerve cupping when the optic nerve cups is larger than normal and progressive optic disc cupping without visual field loss is considered an early cause of glaucoma. Lastly are visual fields when a characteristic change in the visual fields.
The main process of feature extraction and segmenting glaucoma fundus image is performed during this stage. It is consists by two main processes which are glaucoma feature extraction and primary glaucoma segmentation. Potential methods are also discussed here.
Primary glaucoma segmentation
Image segmentation is a low-level image processing task that aims at partitioning an image into homogeneous regions. To determine how homogeneous is that regions, it is depend on the application. This study investigates and formulates the segmentation problem such images as an optimization problem and adopt new evolutionary algorithm which is the selfish gene algorithm. Figure 8 shows the complete flowchart for project development that will be implemented.
About 20 fundus images will be used as the input data
Determine the pixel color and do reference table for fitness function
Use adaptive histogram equalization for image enhancement
Use selfish gene algorithm as a method to segment the glaucomatous image and extract the glaucoma's features.
Result analysis will be measured in statistical analysis using ROC technique
Figure 8 Main Flow Chart
Main features and steps of Selfish Gene Algorithm.
There are few algorithm steps of selfish gene algorithm. Below are main features of selfish gene algorithm that are different with others techniques.
No explicit population
Individual are created & destroyed only temporary
Mutation encoded as random gene selection (not according to its frequencies)
Not to assign absolute value of fitness to generated solutions
Compare to generated solutions between each other
Return back to the algorithm which solution was better
Figure 9 The main features of Selfish Gene Algorithm
Figure 10 General illustration of the gene in the Selfish Gene Algorithm
Research by  showed the concept of selfish gene algorithm (SGA). Chromosome is also called genotype and all living organisms have DNA molecules. Genes is part of DNA molecule and gene location in DNA is called locus. Usually for the same locus there can be different versions of gene and it is called alleles. To determine the fittest chromosome or genotype, allele's frequencies or probabilities in population are calculated. Generally evolution means, that organism which succeeds will increases its allele's frequency in population at the expense of children. And organism which fails will decreases its allele's frequency in population.
p1 p2 p3 p4 p5 p6 p7 p8 p9
Locus (Position of the genes)
Alleles (Difference version of genes)
Allele's frequencies or probabilities
Below are few algorithm steps of selfish gene algorithm.
Step 1: Sub divide whole fundus image into smaller window size.
Step 2: Perform Selfish Gene Algorithm segmentation on each sub window.
Selfish Gene Algorithm Segmentation Steps
Step 2.1: Initialize the population (pop)
Step 2.2: Evaluate the gene frequencies (fitness function) (pop)
Step 2.3: Select the parent from pop based on its gene frequencies
Step 2.4: Reproduction parents- Mutation (If Any) takes gene randomly and compare each others.
Step 2.5: Reward and Penalize
Reward - Reward Better Solution by Increase its alleles
frequency in pop
Penalize - Penalyze Worse Solution by Decrease its alleles frequency in pop
Step 2.6: Update the best solution
Step 2.7: Repeat steps 1-6 until required solution found or other needed criteria is met-for example maximum evolve cycles elapsed or solutions genes frequency is above some threshold or whatever we need.
Step 3: Check termination criteria
Step 4: Combine all the segmented sub windows into one complete image. Figure 11 and figure 12 shows steps of Selfish Gene Algorithm.
Perform SGA on each window
Check the termination criteria
Sub divide whole image into smaller window size (5X5 matrix)
Combine all the segmented sub windows into one complete image.
Figure 11. Main flow diagram for proposed design
Represent of chromosome/genotype
Produce a prototype of glaucoma detection.
Generate initial population (Virtual Population)
Individual Selection base gene frequency
(Select 2 individuals)
Evaluate gene frequency (Fittness Function)
Update the best Solution
Figure 12. Flow diagram for Selfish Gene Algorithm (SGA)
2. Glaucoma feature extraction
This study use to extract features such as cup to disc (c/d) ratio, ratio of the distance between optic disc center and optic nerve head to diameter of the optic disc, and the ratio of blood vessels area in inferior-superior side to area of blood vessel in the nasal-temporal side.
This study also can reveal the potential methods that might be used as segmentation method. There are many potential methods such as artificial intelligence techniques. Many artificial Intelligent (A.I) techniques are applied in image processing. We still try to get better result in order to perform the best application for professional used. The algorithms are Genetic Algorithm, Memetic Algorithm and Selfish Gene Algorithm. Evolutionary Algorithms have been applied to problems in many fields. Yet, in the short space of time, the researchers over the world have had remarkable success in almost every complex search problem with the fields. This success must be due to the problem independence of EAs and there is a new algorithm of EAs should be explored more such as The Selfish Gene Algorithm. Based from the literature, it's never been use for an image processing techniques.
This study also can reveal the potential methods that might be used as segmentation method. There are many potential methods such as artificial intelligence techniques. Many artificial Intelligent (A.I) techniques are applied in image processing. We still try to get better result in order to perform the best application for professional used. The study includes seven algorithms which are Genetic Algorithm (GA), Genetic Programming (GP), Evolution Strategies (ES), Memetic algorithm and Selfish Gene from the Evolutionary algorithm class, Particle swarm Optimization algorithm in the SI class and Artificial Immune system from the BFAs class. All of these algorithms are categorized as Bio-inspired algorithms (BIAs). The algorithms have been applied to problems in many fields. Yet, in the short space of time, the researchers over the world have had remarkable success in almost every complex search problem with the fields. This success must be due to the problem independence of BIAs and there is a new algorithm of BIAs should be explored more such as The Selfish Gene Algorithm. Based from the literature, it's never been use for an image processing techniques.
Aligning with the various different algorithms in the above sections, this section tabulate specifically nine algorithmic features in the Bio-inspired algorithm Matrix shown in Table 2.
Encoding is the stage where chromosomes are represented. The chromosomes can be represented as binary strings, char and vectors in GA and PSO; list of real number in SFGA and MA; symbolic tree in GP and attributes strings in AIS. For the parent selections stage almost all the BIAs can employ selection methods such as random, stochastic technique of roulette and tournament selection. Only Memetic Algorithm uses heuristic search to pick its individuals for it population. Population size is very important all variant BIAs algorithms as limited population size may produce solutions in low quality . SFGA uses virtual population where the individual are seen as only storing of genetic material. The recombination or crossover stage is present in GA, MA, ES, GP and AIS. Among the types of crossover operation are single point, two point, uniform and arithmetic crossovers. Fitness function stage is important in BIAs. It is a heuristic function that measures the performance of an individual chromosome or genotype in the problem domain. The fitness function establishes the basis for selecting chromosomes that will be mated during the reproduction in EAs and BFAs and best particles in PSO. The generation stage is used to repeat the process until it finds the most optimum values. Only SFGA and PSO do not include this stage.
Table 3 shows a comparison of five performance evaluation features of the bio-inspired algorithms that is the convergence rate, the algorithm complexity, the accuracy in finding solutions, the processing speed and the rate of achieving the optimal solutions.
Feature Extraction stage
Adaptive Histogram Equalization
Selfish Gene Algorithm
cup to disc (c/d) ratio calculation
ratio of the distance between
optic disc center and optic nerve
head to diameter of the optic disc
the ratio of blood vessels area
Figure 12. Block diagram of Glaucoma detection using digital fundus image
Phase 5: Result and analysis
The final phase of this is result analysis. The objective of this phase is to evaluate findings obtained from the previous phases. This process is very important since it is used to quantify the accuracy of the primary glaucoma segmentation and glaucoma detection in the research.
There are two potential ways that will be used in analyzing the results for this research which are statistical method and comparison with ground truth.
The significant of abnormalities segmentation will be investigated using statistical methods.
Quantitative performance assessment of the segmentation results will be done by comparing the results with the corresponding ground truth data produced by ophthalmologist s.
Significance of study
The product of this study is expected to be used by the physician and ophthalmologist for early detection of glaucoma abnormalities. Additionally, it could be used as the basis for developing computer aided detection software for automated detection of glaucoma abnormality in full field of Digital Fundus Images.
In the context of research area, this project can be used as guidelines by other researchers. Besides, it also can be used as a platform in producing new findings as well as generating new ideas in image processing field especially in deploying selfish gene in segmentation and classification areas. Finally, it is believed that this study has the potential to greatly improve outcomes for the current technology.