A Hybrid Approach of Detection of Glaucoma with Optic Disk Segmentation and Microaneurysm Detection in Retinal Fundus Images

A Hybrid Approach of Detection of Glaucoma with Optic Disk Segmentation and Microaneurysm Detection in Retinal Fundus Images

Abisha N

PG Scholar School of CSE

Mar Ephraem College of Engineering and Technology Elavuvilai, Marthandam, India

A. Lenin Fred

Professor School of CSE

Mar Ephraem College of Engineering and Technology Elavuvilai, Marthandam, India

Shobhana Assistant Professor School of CSE

Mar Ephraem College of Engineering and Technology Elavuvilai, Marthandam, India

Ashwin Singerji Assistant Professor School of CSE

Mar Ephraem College of Engineering and Technology Elavuvilai, Marthandam, India

Abstract Diabetes is the root cause for the visual problems of human, the medical treatment is highly efficient in protecting the vision loss for the diabetic patients. Vision test are encouraged for the patients on regular basis for early detection of vision problems. Automated diabetic retinopathy evaluation programs have been developed for the identification of patients with vision losing diabetic eye disease. The non diabetic persons are also encouraged to obtain the evaluation of eye for the prevention of blindness. The proposed system is an automatic detection and evaluation of the eye disease. The detection of micro aneurysms and hemorrhages in retinal image are the base for the automated classification of red lesion. It includes the following contributions a) Extraction of a set of Dynamic shape features b) Classification process for the differentiation of lesions and vessel segments. This method is examined with the fundus images in e-ophtha dataset obtained from the freely availed database called ADCIS. The preprocessing of input images was done by mean filter. The Morphological flooding is applied for the extraction of shape features. Finally, the classification was done using Random Forest (RF) classifier. The experimental results on dataset validate the efficiency of the proposed method in the detection of redlesions.

Keyword Red lesion, Morphological flooding

1. INTRODUCTION

   Image processing is the technique to process and analyze the images to obtain various operation s such as segmentation, clustering and classification etc., Medical image processing is the technique for the automated detection of various diseases. In medical image processing the the specialized cameras were required to capture the medical images. The medical image is the pictorial representations of the human body parts for medical analysis and detection of diseases. In the current technology, the medical image processing is an important technology for the detection and treatment of various diseases occurred in different parts of the body. This supports the efficient maintenance of health care. Fundus imaging is the process of capturing the image of the back of the eye i.e.fundus.

   Specially designed fundus cameras that includes a microscope attached to a visual enabled camera are used to capture the fundus photography. The structures that can be represented on a fundus photo are the central and peripheral retina,optic disc and macula. The Diabetic retinopathy affects blood organs in the light visible tissue called the retina that lies on the back of the eye. It is the main cause of blindness among people with diabetes and the root cause of vision lose and blindness among people. The proposed system mainly concentrate with mainly the major risk factors that affect the eye [2]. An early detection and treatment will reduce the risk of vision loss. Medical authorities encourages an regular examination to diabetic patients. Doctors verify the patients eye fundus images to the detection of red lesions. The rating is based on the red lesions found in the fundus images, as well as on contextual data, as patients .

   Microaneurysms (MA) are first extractable symptom of the Diabetic Retinopathy. The count level of microaneurysms is used to predict the severity level of the disease. Early identification of microaneurysm can help decrease the occurrence of visual problems. Hemorrhage is the interior or exterior problem from the blood vessels. The critical problem of hemorrhage is the injury to a blood vessel.

   The automatic detection of red lesions in color fundus images is obtained by a number of alreasdy exist technologies. Most of the system [6], [7], [9] focus only on the detection of microaneurysms. The MAs are similar round shape and short size range, microaneurysms can be detected using morphological operations such as morphological closing [11] and visual transformation using a concrete structuring element at various scale and directions[6][10]. The other systems use a base shape information and to perform a convolution with a double ring filter [13] or through pttern matching with multiscale Gaussian kernels [12], [14], [15]. The vascular segments, have directional informations, but the microaneurysms have

   a Gaussian maxima value in all orientations. The microaneurysms are the first symptom for Diabetic Retinopathy, hemorrhages are also very valuable for automatic detection and evaluation of redlesions.

   The proposed automated redlesion detection system deals with automatic identification and effective prediction of red lesions. The preprocessing are obtained by filtering process called mean filter and the set of shapes are extracted by morphological watershed operation. The retinal image dataset is given as input for the detection by morphological flooding and classification by Random Forest Classifier. The classification step comprises the trainin of sample dataset and testing with the input dataset. The advantage of the proposed system is to achieve better detection accuracy for multiple dataset with effective classification. The remaining paper is organized as follows: Section II describes the methodology. Section III presents the Morphological flooding algorithm in detail. The experiments and results of this proposed system are presented in Section IV. The section V deals with conclusion and future enhancements of the system.

2. MATERIALS AND METHODS

   The block diagram of the proposed algorithm is given below.

   Input Retinal Image

   Rgb To Gray

   Preprocessing

   Candidate Extraction And Optic Disc

   Morphological Flooding

   Classification

   Figure 1: Block diagram of the proposed method

   1. Input Dataset-Retinal fundus images

      The retinal fundus images were obtained from database called ADCIS. It is an open source database. Around 200 Diabetics patients were followed within two years. In proposed framework red lesion affected and unaffected images are obtained from ADCIS. Nearly 233 sets of normal and 148 abnormal retinal images are used as the input of the red lesion detection system. It contains 148 images with microaneurysms or small hemorrhages and 233 images with no lesion. The proposed system provides an efficient and accurate way to analyze the images of the normal and abnormal modalities. It is needed mostly for diabetic retinopathy analysis applications.

   2. Pre-Processing

      Preprocessing is the process which is performed before the analysis and classification process. It enhances the image quality. The feature extraction phase the image features can be obtained in an easier and accurate manner. The

      microaneurysms or small hemorrhages are difficult to identify in low illumination or contrast areas. The images are varying in terms of color information and quality. The preprocessing phase is used to improve the quality of the image by applying various techniques.

      The proposed system uses the following approachesfor the image quality enhancement.

      1. Illumination Equalization: To rectify the low contrast or illumination variation problem, the illumination equalization method is used. The mean filter is applied to every pixel color channels of the input image to identify its color illumination. Finally, the weighted average intensity of the original channel is summed to obtain the same color value as in the original image.

      2. Denoising: Denoising is the process of rectifying the noise components forms the image. The mean filter is applied to the color channels of the illumination equalized image to rectify the noise which is added from the image capturing and digitized processes without affecting the lesion areas.

      3. Adaptive Contrast Equalization: The noise removed image is given to the adaptive contrast equalization process. The contrast changes are calculated using the local standard deviation of every pixel by considering each pixel with the neighboring for each color channel components. To improve the low contrast areas, sharpen the pixel values by sharpening filters in these specified regions.

      4. Color Normalization: It is the process of obtaining the images with standardized color range. It is necessary in to obtain images with a standardized color range to perform the morphological flooding operations. The Histogram stretching and clipping operations are used to perform the color normalization process. The histogram stretching process is used to perform the color improvement operation and the clipping performs the extraction of required color information.

   3. Optic Disc Removal

      The optical disc is the component which will increase the detection of false positives in red lesion detection. The removal of optical disc is important to increase the accuracy of the system. By considering the preprocessed image, entropy based approach is used to calculate the location or the position of the optical disc. Normally the OD is present in the high intensity region where the blood vessels have maximum directional entropy. The optimization processes are applied for the refined detection of the size of the optic disc.

   4. Candidate Extraction

   The blood vessels and red lesions have the greater contrast value in the green color channel, so it can be considered for the candidate extraction and it is obtained from the preprocessed or quality enhanced image. The red and blue color channels images are used to extract color features for the later processes. In the green channel, microaneurysms and hemorrhages appear in the structures with local minimal intensity. The matched ring filtering is an effective method for candidate extraction. This extracted candidate region is used for the identification of red lesions, which are darker than the retinal background.

   1. Dynamic Shape Features

      It is process of extracting the set of shape features. While consider the candidates many regions are fall in the category of non-lesions. The blood vessel and noise in the back layer of retinal. To differentiate between non-lesions and true lesions, a set of dynamic features are extracted, based on shape information. Watershed algorithm is applied to extract the shape features. it from the lowest water source and ending when the retinal background is reached. It is indeed hypothesized that when the flooding reaches the retinal background intensity, the catchment basins degenerate and no longer contextually represent a red lesion. At last flooding level the different shape attributes are extracted from the catchment basin. The following are the different shape attributes. The shape attributes depend on the structure that is identified by the catchment basin. The dynamic shapes are as follows: Relative area, Elongation, Eccentricity, Circularity, Rectangularity, Solidity.

   2. Morphological Flooding

   An effective morphological flooding algorithm depends

   with different levels. The input fundus images were pre- processed by mean filter are depicted in

   The below image shows the image preprocessing and Detect the Red lesion in human eye images

   on,

   Sorting of pixels in increasing order of their gray values.

   Flooding consisting of a fast breadth-first scanning of all pixels in the order of their gray-levels.

   Step 1: In sorting step, a brightness histogram is computed.

   Step 2: Information about the image pixel sorting is used extensively in the flooding step.

   Step 3: If the flooding has completed up to a level k.

   Step 4: Then every pixel having gray-level less than or equal to k has already been assigned a unique catchment basin label.

   Step 5: Next, pixels having gray-level k+1 must be processed; all such pixels can be found in the list that was prepared in the sorting step.

   Step 6: Pixels that represent potential catchment basin members are put in a first-in first-out queue and awaiting for the extraction of dynamic shape features.

   G. Classification

   The classification process is used perform the differentiation between blood vessel segments and red lesions. The Random Forest (RF) classification approach is used. This efficient method of classification is used in various image processing system since it has numerous advantages. it can perform semi supervised classification with multiple dimensional data. It is efficient against noises and over- fitting. The Random Forest classification is the combination of decision tree classification which is trained using the sample dataset or training set. Each feature is split using the best fit of the tree nodes of a randomly selected subset of features chosen, according to the decrease in the Gini index. The MATLAB tool is used for the implementation.

3. RESULTS AND DISCUSSIONS

   The proposed system is developed with the tool Matlab 2013 and executed in the system with specifications of Intel Core i3. processor with 4GB RAM, 64bit windows 7 operating system The 10 fundus images from data sets were used for analysis, out of which two are normal and the remaining 6 data sets are the subjects with redlesion affected

   Figure 2 :Red lession Detection of Human Eye

   The figure 3 shows the histogram bin for the gray scale image

   Figure 3: Histogram of gray scale image

   The Figure 4 shows the Histogram for the Enhanced grey scale image

   Figure 4: Histogram for the enhanced Image

   The figure 5 shows the histogram equalized image and adaptive histogram equalized image and the Operation of morphological image and detection of red lesion image using random classification process.

   1. (b)

      (c) (d)

      Figure 5: (a) Histogram Equalized image. (b) Adaptive Histogram Equalized image. (c) Morphological Flooding Operation. (d) Output Image for the Detection of red lesion image using random classifier.

4. CONCLUSION

The proposed system is an efficient approach for red lesion detection depends on a set of shape features, the red lesion detection techniques was presented and evaluated on fundus image databases. The results demonstrate the strong performance of the proposed method in detecting both MAs and HEs in fundus images of different resolution and quality and from different acquisition systems. The method out performs different red lesion detection approaches. The shape features proves that they are highly capable for the effective detection of red lesions. Further work focusing on the the implemetation of adaboost algorithm which ranks the feature for lesion classification by using the reduced number of features. A two level classification strategy is used for the classification of non lesions for red lesions.

REFERENCES

1. M.Abramoff, M.Niemeijer, M. Suttorp-Schulten, M. A. Viergever, S. R. Russel, and B. van Ginneken, Evaluation of a system for automatic detection of diabetic retinopathy from color fundus photographs in a large population of patients with diabetes, Diabetes Care, vol. 31, pp. 93198, 2008.

2. A. D. Fleming, K. A. Goatman, S. Philip, G. J. Prescott, P. F. Sharp, and J. A. Olson, Automated grading for diabetic retinopathy: A large-scale audit using arbitration by clinical experts, Br. J. Ophthalmol., vol. 94, no. 12, pp. 16061610, 2010.

3. H. J. Jelinek, M. J. Cree, D. Worsley, A. Luckie, and P. Nixon, An automated microaneurysm detector as a tool for identification of diabetic retinopathy in rural optometric practice, Clin. Exp. Optom., vol. 89, no. 5, pp. 299305, 2006.

4. M. Abramoff, J. Reinhardt, S. Russell, J. Folk, V. Mahajan, M. Niemeijer,and G. Quellec, Automated early detection of diabetic retinopathy, Ophthalmology, vol. 117, no. 6, pp. 11471154, 2010.

5. M. Niemeijer, M. Loog, M. D. Abramoff, M. A. Viergever, M. Prokop, and B. van Ginneken, On combining computer-aided detection systems, IEEE Trans. Med. Imag., vol. 30, no. 2, pp. 215223, Feb. 2011.

6. B. Antal, I. Lazar, A. Hajdu, Z. Torok, A. Csutak, and T. Peto, A multilevel ensemble-based system for detecting microaneurysms in fundus images, in Proc. 4th IEEE Int.Workshop Soft Comput. Appl., 2010, pp. 137 142.

7. B. Antal and A. Hajdu, Improving microaneurysm detection using an optimally selected subset of candidate extractors and preprocessing methods, Pattern Recog., vol. 45, no. 1, pp. 264 270, 2012.

8. A. A. A. Youssif, A. Z. Ghalwash, and A. S. Ghoneim, Comparative study of contrast enhancement and illumination equalization methods for retinal vasculature segmentation, in Proc. Cairo Int. Biomed. Eng. Conf., 2006, pp. 2124.