An Automated Graphical User Interface based System for the Extraction of Retinal Blood Vessels using Kirsch’s Template

The assessment of Blood Vessel networks plays an important role in a variety of medical disorders. The diagnosis of Diabetic Retinopathy (DR) and its repercussions including micro aneurysms, haemorrhages, hard exudates and cotton wool spots is one such field. This study aims to develop an automated system for the extraction of blood vessels from retinal images by employing Kirsch’s Templates in a MATLAB based Graphical User Interface (GUI). Here, a RGB or Grey image of the retina (Fundus Photography) is used to obtain the traces of blood vessels. We have incorporated a range of Threshold values for the blood vessel extraction which would provide the user with greater flexibility and ease. This paper also deals with the more generalized implementation of various MATLAB functions present in the image processing toolbox of MATLAB to create a basic image processing editor with different features like noise addition and removal, image cropping, resizing & rotation, histogram adjust, separately viewing the red, green and blue components of a colour image along with brightness control, that are used in a basic image editor. We have combined both Kirsch’s Template and various MATLAB Algorithms to obtain enhanced images which would allow the ophthalmologist to edit and intensify the images as per his/her requirement for diagnosis. Even a non technical person can manage to identify severe discrepancies because of its user friendly appearance. The GUI contains very commonly used English Language viz. Load, Colour Contrast Panel, Image Clarity etc that can be very easily understood. It is an attempt to incorporate maximum number of image processing techniques under one GUI to obtain higher performance. Also it would provide a cost effective solution towards obtaining high definition and resolution images of blood vessel extracted Retina in economically backward regions where costly machine like OCT (Optical Coherence Tomography), MRI (Magnetic Resonance Imaging) are not available. Hence an early detection of irregularity will be possible especially in rural areas. Keywords—Blood Vessel Extraction; Retinal Images; Kirsch’s Templates; Image Processing; Graphical User Interface


INTRODUCTION
The vascular network is an essential anatomical structure in the human retina. An analysis of the retinal vasculature may lead to the diagnosis of various abnormalities such as haemorrhages, micro aneurysms etc. thus, an automated system for retinal blood vessel extraction is a preliminary step in the development of a computer-assisted diagnostic system for ophthalmic anomalies.
The extraction of the human eye vasculature from retinal images involves the essential tool of edge detection. According to Muthukrishnan & Radha [1], the classical methods of edge detection operates on convolving the image through an operator. Yin et al. [2] proposed a probabilistic tracking method to detect blood vessels in retinal images. They categorized the task of vessel extraction into two main groups: Pixel-based methods and tracking methods.
According to Zhang et al. [3], matched filter is a simple yet effective method for vessel extraction. They also proposed an extension to the matched filter approach to detect retinal blood vessels that significantly reduce the false detections produced by the original matched filter. Esmaeili et al. [4], presented an efficient algorithm for automatic extraction of blood vessels that comprises the following four steps: (i) Curvelet-based contrast enhancement, (ii) Matched filtering, (iii) Curvelet based edge-extraction and (iv) Length filtering. In their study, the enhanced image is first reconstructed from the modified curvelet co-efficient followed by match filtering to intensify the blood vessels along with the implementations www.ijacsa.thesai.org of curvelet transform to segment vessels from its background. Finally, they have used length filtering to remove the misclassified pixels. Their experimental results have been evaluated on DRIVE database (Niemaiger & van Ginneken, [5] The software used to develop the GUI is MATLAB which is a multi-paradigm numerical computing environment and 4 th Gen. Programming language.

III. METHODOLOGY
The complete procedure of extracting the blood vessels from the coloured retinal image consist of three steps, beginning with fetching the input image from the system or the camera, then converting the RGB image into a gray scale image and finally using the kirsch's templates to detect the edges of the blood vessels. Fig. 1 illustrates the flow diagram of the proposed method for blood vessel extraction. The edge magnitude of the Kirsch operator is calculated as the maximum magnitude across all directions. Except the outermost rows and columns, every pixel along with its eight neighbouring pixels in a given image is convolved with the eight aforementioned templates respectively [7], providing eight outputs for each pixel, the maximum of which is defined as the edge magnitude (Gao et al. [8]. A pixel's Gray value with its eight neighbours is as shown.
The direction of edge is defined by the related mask that produces the maximum magnitude. The general output of edge detection through Kirsch's Templates is an image containing Gray level pixels of value 0 or 255. The value 0 indicates a black pixel and the value 255 indicates a white pixel. The edge information of the target pixel is checked by determining the brightness levels of the neighbouring pixels [9]. In case no major difference in the brightness level is found, the possibility of the pixel being a part of an edge is ruled out.
Kirsch's Template can set and reset the threshold values to obtain most suitable edge of images. It works well for images having a clear distinction between the foreground and background [10]. Since the retinal blood vessels can be considered as the required foreground information from the background fundus images, Kirsch's algorithm is effectively applicable. Fig. 2 illustrates an original fundus image with the extracted blood vessels.

d) Other Important Tools:
A fact worth mentioning is that the GUI designed in order to carry out this study on the extraction of blood vessels may also be used to edit and enhance the retinal images for further analysis using the following options.  Fig. 3 demonstrates the image addition operation.
2) Image Clarity: Generally for certain images the pixel distribution is not equally spaced or rather are clotted to a particular intensity level, hence, making the image too dull or too bright [11]. For this reason the following histogram adjustment techniques are being used: Image Adjust, Histogram equalization, Adaptive Histogram equalization.      Fig. 8

IV. DISCUSSIONS
Diabetic Retinopathy is an important cause of blindness. It is an outcome of prolonged accumulated damage caused to the retinal blood vessels. One percent of global blindness is as a result of Diabetic Retinopathy [13]. We have presented in this paper the retinal blood vessel extraction with the help of Kirsch's Templates and a combination of various MATLAB image processing techniques. The combination of these two techniques provided an optimum result. Table1 depicts the comparison between our Results and other Techniques in the extraction of Blood Vessels.
The interface is presented in a user-friendly manner. Individual Save buttons next to each and every image window helps the user to re-use and re-apply the extraction process on edited images. This process enhances the base image at every stage. Our main aim is to provide a cost effective solution to detect Retinal anomalies and provide a better platform. In some cases, minute examination of Diabetic Retinopathy is carried on and analyzed by Fundus Photography and for detailed insight Optical Coherence Tomography is used [17]. Table 2 represents a comparative study between our Graphical User Interface and the present clinical technique for the detection of Retinal diseases.  As per Dr.Somnath Das [18], Associate Professor of Regional Institute of Ophthalmology (RIO), Medical College (Kolkata):-"This automated system for the extraction of Retinal blood vessels can give us important clue regarding alteration in morphological pattern and pathological changes in and around the retinal blood vessels, status of laminar blood flow within the blood vessels and nature of extravasations of plasma, blood cells and lipids in surrounding retinal tissue on the basis of which we can diagnose and plan for the management of a large group of ocular diseases. Not only that, this system will also be a good prognostic indicator for a particular disease".
Availability of a Computer/Laptop will enable the user to use the Graphical User Interface after a quick installation of MATLAB. One of the limitations of this Graphical User Interface is that it requires a high processor speed for smoothness and trouble-free output. Other than that, slight lag time can be experienced during the execution of the program. Takayasu's disease involving retina can be done with this method. Furthermore we can incorporate an algorithm to differentiate between blood clots and lesions and provide a comparative study. This would help in distinguishing various Retinal abnormalities. But for all this we require the fundus photography which is obtained by a fundus Camera which costs almost 3600 USD. To cut down the cost at that level, the next step could be utilizing the Smart Phones' Camera and combining it with an extra lens to capture Fundus images more easily and in a very low cost. This whole setup can then be incorporated in Smart phones as an Application which will be easily accessible and utilized by all.
Further classification of special features detected from the extracted blood vessels along with the use of a trained Probabilistic Neural Network to recognise and report any of the above mentioned abnormalities can be carried on.
Further classification of special features and parameters detected from the extracted blood vessels along with the use of a trained Probabilistic Neural Network to recognize and report any of the above mentioned abnormalities can be carried out automatically. This Automation of disease detection may also be achieved by the implementation of Neuro-fuzzy template, since it provides a simple way to arrive at a definite conclusion based upon vague, ambiguous, imprecise, noisy, or missing information from the parameters of the pre-processed fundus image for each disease. [19].

VI. CONCLUSIONS
Retinal images are being used by ophthalmologists to aid in diagnosis, to make measurements, and to look for changes in lesions or severity of diseases. The appearance of the retinal vasculature particularly acts as an indicator for diagnosis of Diabetic Retinopathy (both Proliferative and Non-Proliferative) and Glaucoma. Therefore, extraction of these features is the key challenge for proper analysis, visualization and quantitative comparison. The present study focuses mainly on this challenge of blood vessel extraction from colour retinal images obtained from fundoscopy. The proposed algorithm proves to be successful and robust in accurately extracting the retinal blood vessels. In this respect, the dataset of 40 test images from the DRIVE database has been used to evaluate this method. Few real time images obtained from the Regional Institute of Ophthalmology (RIO), Medical College, Kolkata are also used for the evaluation. It included fundoscopy of various patients having NPDR (Non www.ijacsa.thesai.org Proliferative Diabetic Retinopathy), PDR (Proliferative Diabetic Retinopathy) and post PRP (Panretinal Photocoagulation). The fact that the proposed algorithm extracts blood vessels with evident accuracy renders it a quite sought after platform for further improvements. An accurate extraction of blood vessels provides the basis for the measurements of a variety of features including micro aneurysms and hard exudates that can then be applied to the tasks of diagnosis, treatment, evaluation and clinical study [20]. More particularly in rural areas of underdeveloped and developing countries the proposed Graphical User Interface can be utilized to overcome the hardships arising due to the shortage of professional observers by this completely automated computer assisted monitoring and diagnostic system.

ACKNOWLEDGEMENT
This work was supported by ESL (www.eschoollearning.net), Kolkata.