INTRODUCTION:
Glaucoma is a common cause of irreversible blindness globally. Clinically, it is characterized by the loss of retinal ganglion cells, neural rim tissue, peripapillary retinal nerve fiber layer and manifests as an enlargement of the optic cup and loss. Gray Level Co-occurrence Matrix (GLCM) and Logistic Regression (LR) classifier based glaucoma assessment using ocular thermal images is discussed in [1]. Linear transformation is used in the pre processing stage. Features are extracted using GLCM and finally, LR classifier is used to classify the given ocular IR thermal image.
Energy features based on wavelet for glaucomatous image classification is implemented in [2]. Discrete Wavelet Transform (DWT) is used to extract energy signatures and subject these signatures to diverse feature ranking and feature selection strategies. Daubechies (db3), symlets (sym3), and biorthogonal (bio3.3, bio3.5, and bio3.7) wavelet filters are used to obtain the discriminatory potential features At last, extracted features are classified using SVM, Random Forest (RF), sequential minimal optimization and naïve Bayes classifier. Independent Component Analysis (ICA) of retina images for glaucoma classification is introduced in [3]. ICA and K-Nearest Neighbor (KNN) classifier are used to analysis the image and classify the image respectively for glaucoma classification.
Wavelet and moment feature based glaucomatous images automatic classification is discussed in [4]. Image decomposition is done by using db3, sym3 and bior 3.3, bior 3.5, bior 3.7 and feature computation is obtained by using higher order moments. SVM, Back Propagation Neural Network (BPNN) and KNN are the three classifiers used for classification. Empirical Wavelet Transform (EWT) and correntropy features based automated diagnosis of glaucoma from fundus images is presented in [5]. The EWT is used to decompose the image and correntropy features are obtained from decomposed EWT components. Then normal and glaucoma images are classified using these features by Least Squares SVM classifier.
Regional wavelet features of Optic Nerve Hypoplasia (ONH) and its surroundings are used for glaucoma classification in [6]. The important regions are selected through visual observation only. Wavelet energies are extracted from these regions for the classification. Glaucoma classification based on the analysis of texture is described in [7]. It uses the whole retinal image and binary robust independent elementary features are extracted. SVM classifier is used with seven features for the glaucoma classification.
Discrete Orthogonal Stockwell Transform (DOST) based glaucoma image classification is explained in [8]. DOST distribute its coefficients based on model spacing paradigm where high frequencies have higher sampling rate and low frequencies have lower sampling rate. RF classifier considers all DOST coefficients for glaucoma classification. Wavelet energy features and Artificial Neural Network (ANN) based automated glaucoma detection system is discussed in [9]. Glaucomatous image is classified by analyzing both structural and energy features. Energy distribution over wavelet sub bands are applied to find the significant texture energy features. Finally, the extracted energy features are applied to multilayer perceptron and BPNN for effective classification.
Wavelet transform based glaucomatous image classification is presented in [10]. Single level DWT based image classification is discussed. Db3, sym3 and rbi03.3, rbi03.5, and rbi03.7 wavelet filters are used to obtain the discriminatory potential of wavelet features. KNN classifier is used to classify the images between normal and glaucomatous images. Haralick texture features based automated diagnosis of glaucoma from digital fundus image is implemented in [11]. Thirteen Haralick features are extracted from the constructed GLCM. Finally, KNN classifier is used to perform supervised classification.
NN based glaucoma classification based on texture features is described in [12]. The extracted texture features around the optic cup are localized. Then NN classifier classifies the extracted features for abnormality. Computer aided diagnosis of glaucoma using digital fundus image is explained in [13]. Optic disc is segmented automatically by using K means clustering, where K is automatically selected by hill climbing algorithm. Fuzzy C-mean clustering is used to extract the optic cup.
In this paper, an efficient DSS for glaucoma diagnosis based on TT is proposed using OCT images. The rest of the paper is organized as follows: Section 2 gives the overview of various approaches that forms the DSS for glaucoma diagnosis. Section 3 proposes the glaucomatous image classification system using TT and SVM. Section 4 provides the outcomes of the DSS for glaucoma diagnosis and conclusion is given in the last section.
2. MATERIALS SECTION:
The DSS for glaucoma image classification using OCT image is built based on TT, GLDM, and binary SVM classifier. This section gives the overview of the materials used for glaucoma diagnosis.
2.1 Tetrolet Transform:
Initially, some of the notations and definitions required to give explanation the idea of the tetrolet transform is given.
Let 
be the index set of a
digital image 
with
. Determine a 4-neighborhood
of an index 
by [14]
(1)
2.1.1 Ideas of Tetrolet:
The two-dimensional classical Haar wavelet decomposition leads to a special tetromino partition. Introducing the discrete tetrolet transformation, we recall the conventional Haar case in a notation consistent to the following tetrolet idea.
(2)
where the coefficients 
are
entries from the Haar wavelet transform matrix.
2.1.2 Tetrolet Filter Bank Algorithm:
Input: Image
with 
Output: Decomposed image 
· Split the image into 4x4 blocks.
· Find the sparsest tetrolet representation in every block.
· Reschedule the low- and high-pass coefficients of every block into a 2 × 2 block.
· Accumulate the tetrolet coefficients (high-pass part).
· Apply step 1 to 4 to the low-pass image for the next level of decomposition.
2.2 GLDM:
GLDM is a traditional statistical method to
extract texture feature of an image. It is based on the occurrence of two
pixels having absolute difference in their intensity value and both are
separated by a specific displacement [14-16]. To describe the GLDM, let
be the
digital picture function. For any given displacement
, where 
and 
are
integers,
let
.Finally, let 
be the
estimated probability density function associated with the possible values of 
, i.e.,
(3)
The four possible forms of displacement
vector 
with
inter-sample spacing of d are (0,d),(d,0),(-d,d),(-d,-d).
Hence, texture features are extracted from the four different probability
density functions.
2.3 SUPPORT VECTOR MACHINE:
By constructing an N dimensional hyperplane, SVMs optimally divides the data into two categories. The classification problem is viewed by SVMs as a quadratic optimization problem which is a set of related supervised learning methods that analyze data and recognize patterns, used for classification and regression analysis. A classification task usually involves with training and testing data which consists of some data instances. Each instance in the training set contains one “target value” (class labels) and several “attributes” (features). The performance of SVM largely depends on the kernel. SVM classifier is employed for glaucomatous image classification [17-20].
3. DSS FOR GLAUCOMA DIAGNOSIS:
The classification system for glaucoma diagnosis using OCT images mainly consists of two different phases; training phase and classification phase. The following sub-sections discusses the abovementioned two phases briefly. Figure 1 shows the DSS for glaucoma diagnosis using TT and SVM.
Figure 1 DSS for glaucoma diagnosis using TT and SVM
3.1 Training Phase:
The first step in the training phase is preprocessing stage in which ROI is extracted automatically from the whole OCT image. The extracted ROI contains only the retinal region and discriminating features are extracted from the ROI for classification. Figure 2 shows a sample OCT and its corresponding ROI region. Before extracting ROI, the colour OCT image is converted into gray scale image.
Figure 2 (a) OCT colour image (b) OCT gray image (c) OCT-ROI image
To extract features, OCT-ROI image is decomposed by TT at a predefined resolution level of decomposition. As the low pass image resembles to original image, the approach extract features from the high pass image only. Two set of statistical features are extracted. One set of statistical features are extracted from the TT decomposed image (SfTT) and another set of statistical features are extracted from the GLDM (GfTT). The extracted SfTT are mean, standard deviation, skewness, and kurtosis. GLDM is computed for the mean of high frequency coefficients in each block of size 2x2. Figure 3 shows the histogram of GLDM with inter-sample spacing of single pixel for OCT-ROI image in Figure 2 (c).
