The Detection Of Diabetic Retinopathy Health And Social Care Essay

M.Shamela Rizwana, PG Student, Manonmanium Sundaranar University ,

Prof.Dr.P.ArokiaJansiRani ,M.E, Ph.D,Manonmanium Sundaranar University


Microaneursyms are the earliest sign of diabetic eye disease and so are very important for classifying whether images show signs of retinopathy. So, their detection is necessary for both screening the pathology and follow up. MA detection as a problem of target embedded in a background noise. This paper analyze the implementation of Threshold based Technique and Wavelet Decomposition technique to reject specific classes of noises while passing majority of true Microaneursyms using set of specialized features. The rejection strategy is formulated based on the occurrence frequency and discriminability of the underlying clutter. Threshold based technique separates both classes and further allows various subgroups of clutter class to be handled through a cascade solution. A new set of morphological and appearance based features are introduced to characterize the clutter and MA structures. This gives flexibility to achieve better result of finding Microaneursyms separately from the clutter classes. These two methods have been analyzed based on Microaneursyms detection. Experimental results show that, the Threshold based technique reduces the noise in an efficient way and affected Microaneursyms portion can be obtained clearly.


DIABETIC retinopathy (DR) is a major public health issue since it can lead to blindness in patients with diabetes.

Microaneurysms (MAs) are usually the first clinical symptom of DR. They are swellings of capillaries caused by a weakeningof the vessel wall [1]. Their sizes range from 10 to 125 μm [2].In the clinical scenario, experts rely either on direct manualexamination or on luorescein fundus angiography where MAsappear with high contrast as bright white spots. Given the highcost and the cumbersome requirement of intravenous injectionof a dye for this type of imaging, interest in the recent past hasbeen on detectingMAs from a color fundus/retinal image (CFI).

In CFIs, MAs appear as tiny, reddish isolated dots. Automaticdetection of MAs from digital CFIs can play an important rolein DR screening at large scale .

Early published work attempted to address the problem of MA detection in fluorescein angiogram images of the retina

In this method, MA candidates were obtained using top-hat transformation that eliminates the vasculature structure from the image leaving possible MA candidates untouched. A shade correction technique and a candidate detection method using matched filtering. However, potential mortality associated with the intravenous use of prohibits the application of this technique for large-scale screening purposes. Instead, color fundus imaging

has emerged as a preferred modality due to its noninvasive nature. Extensive clinical studies show the effectiveness of CFI for large-scale DR screening .


Fig 1 : Microaneurysms

Existing methods for MA detection generally consist of two stages where the first stage is aimed at obtaining potential MA candidates while the second stage is used to assign MA or non-MA category to the candidate using features computed around the candidate location. The main processing components include:

1) preprocessing, selection of a candidate MA; and 2)feature extraction, classification. The focus of the early methods has been on preprocessing and candidates selection steps.

Later methods focus more on designing new sets of features and choosing of classifiers.

Numerous algorithms have been proposed to detect early signs of DR (MAs) from CFI. The first such method was presented by Oien et al. [11]. The preprocessing used here is similar to the approach used by [5]. In latermethods, a rule-based classification was added to the processing pipeline [6], [8], [12], [13]. Usher et al. [14] employed a neural network-based classification

on the candidate regions obtained using recursive region growing and adaptive intensity thresholding.

If we consider true MAs and non-MAs (similar structures) as two classes, in a given image, the probability that a candidate belongs to the true MA (PT ) class is substantially smaller, compared to that of belonging to non-MA class (PC ). Here, we formulate the MA detection problem as a problem of detecting a target embedded in a background clutter, where the target occurs with a much lower probability compared to the clutter (PT _ PC ). From this formulation point of view, the earlier methods can be viewed as attempts toward getting better characterization of target class followed by a classification stage. We are interested in exploring whether knowledge of the clutter class can play a positive role in MA detection. Thus, instead of the earlier formulations where MA is the only object of interest, we wish to gain better understanding of objects in the clutter class, in addition to the target class. Here, we propose to model the clutter, attempting to address the discrimination aspect early, and postpone the target modeling.

Such a strategy that aims at very early clutter labeling can be beneficial to the overall detection as this can facilitate

progressive rejection of clutter responses (using many rejectors sequentially), and target recognition may be performed when

fewer clutter responses remain.

In each rejection stage, responses classified as clutter can be removed from further consideration, retaining the remaining responses as putative targets. These are to be passed on to the

subsequent rejector for further examination. The objective of such a cascade of rejectors is to reduce PC while maintaining PT .


Fig. 2 illustrates the proposed method where the strategy is to get a set of candidate MAs using a simple threshold, from a preprocessed image, and then culling the clutter among the candidates

using a set of rejectors in cascade. Since the clutter class has multiple objects with different characteristics, the known and frequently occurring clutter objects are rejected first, and a second stage is designed to discriminate the remaining class of (largely unknown) clutter objects. In the final stage, the candidates are assigned a similarity score based on their similarity to true MAs.









Fig 2: Stages of microanurysms detection

The candidate selection method is kept simple since current focus is on rejection of false positives rather than acquiring

good candidates. The first rejection stage is aimed at eliminating candidates originating from dark structures such as vessels and hemorrhages. Candidates occurring on such structures canbe well characterized using local morphological information (for example, elongated structure colocated with the candidate

indicates the possibility of the candidate occurring on a vessel).Thus, a set of shape-based features and a two-class classifier are used to eliminate such clutter candidates.The sources of remaining non-MA candidates could be due to a variety of reasons including local minima formed by image noise, region between two bright regions, OD, etc. Non-MA candidates having shape profile similar to an MA can be discriminated if their local surround is considered. Since it is difficult to obtain a common characterization for this range of clutter candidates, a target-oriented rejection strategy is employed. A coarse model forMAcandidates is learnt using context sensitive local features. The candidates found to be inconsistent with this model are classified into the clutter and are eliminated. Culling of clutter by two stages results in a significant reduction in the number of reported candidates. In the final stage, we compute the degree of similarity of each remaining candidate to a true MA profile, and assign a score that ranges from [0 − 1].This stage uses a composite set of features capturing true MA profile based on morphological and appearance-based information. A final set of MA points can be obtained by applying a threshold on the similarity score. In the following sections, each of the processing stages is elaborated in detail.

3.Preprocessing: Image pre-processing is the term for operations on image a lowest level of abstraction. These operations do not increase image information content but they decrease it if entropy is an information measure .The aim of pre-processing is an improvement of the image data that suppresses undesired distortions or enhances some image features relevant for further processing and analysis task. Image pre-processing use the redundancy in images. Neighboring pixels corresponding to one real object have the same or similar brightness value. If a distorted pixel can be picked out from the image, it can be resorted as an average the size of the pixel neighborhood that is used for the calculation of a new pixel brightness.

4. Candidate selection: This operation is based on morphologically opening the image with a linear structuring element at different orientations. Pixel classification is a supervised method, so it needs to be trained using example pixels. Each of the example pixels was extracted from a training set for which a labeled reference standard is available. All pixels in the training set are, thus, assigned a label and a feature vector. A classifier can now establish a decision boundary in the feature space which will optimally separate the two classes of foreground and background pixels.

Feature Extraction: The features should carry enough information about the image and should not require any domain-specific knowledge for their extraction. They should be easy to compute in order for the approach to be feasible for a large image collection and rapid retrieval. They should relate well with the human perceptual characteristics since users will finally determine the suitability of the retrieved image. Synthesized images are good for evaluating techniques and finding out how they work, and some of the bounds on performance. An image of circles that were specified mathematically. The image is an ideal case: the circles are perfectly defined and the brightness levels have been specified to be constant. This type of synthetic image is good for evaluating techniques which find the borders of the shape (its edges), the shape itself and even for making a description of the shape.


Morphology-based approach is used to extract linear structures in various orientations. The suprema of morphological openings see, Figure 3 (obtained with linear structuring elements of different orientations is used as the marker. Topological and geometrical continuous-space concepts such as size, shape, convexity, connectivity, and geodesic distance, can be characterized by Mathametical Morphology(MM) on both continuous and discrete spaces. MM is also the foundation of morphological image processing, which consists of a set of operators that transform images according to the above characterizations. MM was originally developed for binary images, and was later extended to grayscale functions and images. The subsequent generalization to complete lattices is widely accepted today as MM's theoreticalfoundation.The closing of A by B is obtained by the dilation of A by B, followed by erosion of the resulting structure by B.



Fig 3: Morphological close operation:

a) original image b)close

linear structuring elements of different orientations [4]) is used as the marker, and with Ibo that as the mask, we perform morphological reconstruction to get Irecon. The final preprocessed image Ipp is obtained by subtracting Irecon from Ibo that , thereby suppressing linear structures. The potential candidate locations in Ipp have a high intensity. Since MAs have high values in Ipp, an empirically chosen threshold is applied to get candidate regions. The local minima in the green color plane of each candidate region are used as candidates (designated as C0 ) in further stages.

Dilation: The value of the output pixel is the maximum value of all the pixels in the input pixel's neighborhood. In a binary image, if any of the pixels is set to the value 1, the output pixel is set to 1(figure 4).

Erosion: The value of the output pixel is the minimum value of all the pixels in the input pixel's neighborhood. In a binary image, if any of the pixels is set to 0, the output pixel is set to 0.

Fig 4: Morphological dilation of image


The simplest method of image segmentation is called the thresholding method. This method is based on a clip-level (or a threshold value) to turn a gray-scale image into a binary image. The key of this method is to select the threshold value (or values when multiple-levels are selected). Several popular methods are used in industry including the maximum entropy method, Otsu's method (maximum variance), and et al. k-means clustering can also be used. The combination of a data-rich experimental system with sophisticated mathematical modeling holds the promise of an improved basic understanding of malignant progression and therapeutic response in humans.


The median filter is a nonlinear digital filtering technique, often used to remove noise. Such noise reduction is a typical pre-processing step to improve the results of later processing (for example, edge detection on an image). Median filtering is very widely used in digital image processing because, under certain conditions, it preserves edges while removing noise.

\begin{displaymath}y[m,n]=median\{ x[i,j], (i,j) \in w \} \end{displaymath}


The main idea of the median filter is to run through the signal entry by entry, replacing each entry with the median of neighboring entries. The pattern of neighbors is called the "window", which slides, entry by entry, over the entire signal. For 1D signal, the most obvious window is just the first few preceding and following entries, whereas for 2D (or higher-dimensional) signals such as images, more complex window patterns are possible (such as "box" or "cross" patterns). Note that if the window has an odd number of entries, then the median is simple to define: it is just the middle value after all the entries in the window are sorted numerically.


Originally known as Optimal Subband Tree Structuring (SB-TS) also called Wavelet Packet Decomposition (WPD) (sometimes known as just Wavelet Packets or Subband Tree) is a wavelet transform where the discrete-time (sampled) signal is passed through more filters than the discrete wavelet transform (DWT).In the DWT, each level is calculated by passing only the previous wavelet approximation coefficients (cAj) through discrete-time low and high pass quadrature filter. However in the WPD, both the detail (cDj (in the 1-D case), cHj, cVj, cDj (in the 2-D case)) and approximation coefficients are

Decomposed to create the full binary tree (fig5)

Fig 5. Wavelet decomposition process


Each wavelet basis has two filters associated with it. In general those filters are expressed in the form y(n) = Ph(n)x(n).The Haar basis is particularly simple. The two filters are H0 and H1 and are defined as follows:

H0: y (n) =1/2x (n) +1/2x (n - 1)

H1: y (n) =1/2 x (n) -1/2 x (n - 1)

For H0, all h coefficients are zero except for h0 (0) = 1/2 and h0 (1) = 1/2. Similarly, in

H1, h1 (0) = 1/2 and h1 (1) = -1/2.

H0 computes a moving average of its input, resulting in a sequence which is smoother than the initial sequence. Hence it is a low pass filter. H1 computes a moving difference and serves as a high pass filter.


The rejector cascade outputs a set C2 of candidates which are likely to be true MAs. This final module assigns a numerical similarity score to each sample in C2 , indicating the chance of it being a true lesion. We choose to perform the score assignment by considering the signed distance of a sample from the optimal hyperplane of a two-class SVM in feature space. A complete representation for a trueMA is obtained by considering features from the previous rejection stages in addition to the features encoding context and structure symmetry information which are explained next.


For the purpose of evaluation, three datasets were considered:

two are the publicly available datasets namely, theDIARETDB1 [27] and ROCd [18] datasets; a custom-built dataset called

CRIAS. Images in each dataset are divided into training and testing sets. Images in each dataset gives a range of image

sizes (768 × 586 to 1500 × 1100), resolution, etc. The detailed specifications of the selected datasets are given next.

DIARETDB1 consists of 89 images, of which 5 images do not contain any DR-indicative lesions. The images were collected from a screening program and taken under a fixed imaging protocol.

The images were selected by the medical experts, but their distribution does not correspond to any typical population. The

ground truth supplied with this dataset is a soft map consisting of regions indicating expert consensus level information averaged from multiple experts. A bright region thus indicates high consensus about the presence of MA.

CRIAS consists of 288 images collected by us from a local hospital. These images are mainly collected for clinical documentation and patient profiling.


Threshold based technique performs better than the Wavelet Decomposition method.


(a) (b) (c) (d)

Fig 6: (a) original image, (b) Green component image, (c)Image after morphological operation, (d) detected microaneursyms

In the Micro aneurysms process, the original color fundus image (fig a) is extracted from the Diabetic Retinopathy Database. In the Micro aneurysms Process the Green Channel is detected from the Original RGB color image. The Green Channel is used to gather the information from the Fundus image. By Histogram equalization, green channel is having highest Contrast, so the green channel image is extracted for the information. The figure 5.3 shows the more information about the affected area. This image gives nerves area clearly to the user.The whitening image takes the green Channel image as an input image, and gives the inverted green component as the output image. It is used to give clear and more information of the affected area in the fundus image. It is a contrast image of the green channel. The most basic morphological operations are dilation and erosion. Dilation adds pixels to the boundaries of objects in an image, while erosion removes pixels on object boundaries. The number of pixels added or removed from the objects in an image depends on the size and shape of the Structuring element used to process the image. The figure (b) shows clear information of the blood vessels. The Candidates extracted from the image by using morphological operation. Morphological opening operation removes small objects from foreground (dark pixels) of an, placing them in background. The simplest method of image segmentation is called the thresholding method. This method is based on a clip-level (or a threshold value) to turn a gray-scale image into a binary image. Threshold based according to global property, usually intensity where the global knowledge is represented by the intensity histogram. It is also possible to extract objects that have specific intensity range multiple thresholds. The figure (d) gives the affected Micro aneurysms portions clearly. After thresholding operation, got micro aneurysms portions with some noises. Median filtering is used to remove that salt and pepper noise from that image, and we got clear Micro aneurysms portion.

Fig 7: Comparision of Threshold and Wavelet Decomposition Technique .

Fig 8: Comparision of Threshold and Wavelet Technique Based On Time

The figure 7 shows that, for threshold based technique, the percentage of microaneursyms is 93% for a total of hundred images, but in Wavelet decomposition method, it is 80%. The

Figure 8 shows, that the time taken by the threshold based technique to detect micro -aneurysms is more than that.

So compared than wavelet decomposition method, threshold based technique is the best one for detecting micro aneurysms in much amount of images.


Microaneursyms are the earliest sign of diabetic eye disease and so are very important for classifying whether images show signs of retinopathy. So, their detection is necessary for both screening the pathology and follow up. MA detection as a problem of target embedded in a background noise. Threshold based Technique and Wavelet Decomposition technique are used to reject specific classes of noises while passing majority of true Microaneursyms using set of specialized features. These two methods have been analyzed based on Microaneursyms detection. By comparing Threshold based technique and wavelet Decomposition method, threshold based technique performs better for detecting microaneursyms in much amount of images.


I Would like to thank Prof.Dr.P.Arokia JansiRani,M.E,Ph.D, Department of Computer Science, Manonmanium Sundaranar university , for her valuable guidance and encouragement given at the right time.