SVM Classifier Based Melanoma Image Classification

 

Atul Kulkarni¹, Dr. Debajyoti Mukhopadhyay²

¹Research Scholar, Information Technology, AMET University, Chennai.

²Department of Computer Science, Maharashtra Institute of Technology, Chennai.

 

ABSTRACT:

Melanoma Classification is the most important aspect that is related to the patients who endures melanoma. The melanoma is usually known by measuring the depth given in millimeters (mm) and is evaluated by the pathological assessment. In order to avoid the interference method usage in the surgery, a method is proposed for computational image analysis. In the system the Gray Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) algorithms are used for the features extraction process and those features are classified by using the Support Vector Machine (SVM) classifier. The proposed melanoma classification gives the output accuracy of about 96.7% of classification accuracy value.

 

 

 

1. INTRODUCTION:

Melanoma is a tumor that occurs in melanocytes and also produces the cells pigment that makes the skin, hair and eyes color. The melanoma that appears at the skin is known as cutaneous melanoma. Some of the melanoma classifications that are present are explained below. In [1], analysis of computer aided system for the integration and selection of the features from the textual and border based features are derived by using the wavelet decomposition and a boundary series of models. The classification is done by using SVM, logistic model tree and random forest classifiers.

 

A melanoma classification system based on vitro Raman spectroscopy is explained in [2]. The features based on Raman spectra are extracted and are classified by using the neural network classifier. A melanoma segmentation system based on the very deep Convolutional Neural Networks (CNN) is explained in [3].

 

 

A fully convolutional residual network algorithm is used for segmentation and is classified by using the CNN classifier. Melanoma classification on dermoscopy images using a neural network ensemble model is discussed in [4]. By using the adaptive thresholding and skewness correction to detect gray areas in melanoma in situ images is discussed in [5].

 

The classification of melanocytic tumors into benign and malignant tumors is explained in Image super resolution reconstruction using iterative adaptive regularization method and genetic algorithm [6]. The extraction of feature like texture, color and border are done by using the self generating neural network and are lesions are classified by using the ensemble model neural network. Breslow index method discussed in Image Super Resolution Using Wavelet Transformation Based Genetic Algorithm [7] is used for the measurement of melanoma invasion depth by the help of pathological analysis of the incisional and excisional biopsy of the patients. Melanoma image classification system by NSCT features and Bayes classification is discussed in [8].

 

2. METHODOLOGY:

A method of melanoma classification system is proposed for the textural feature extraction by using the GLCM and LBP algorithm and is classified by using binary classification based SVM classifier. The proposed system framework is shown in Fig 1.

 

2.1  FEATURE EXTRACTION:

The feature extraction step is the most important step in all image classification systems. In this method also the extracted features from the images are more important for the classification of melanoma. The images used in our system are the dermoscopic images. At first, the input images are per-processed for the reduction of noises as well as the color conversion from RGB to Gray. The second step is the extraction of the features like GLCM and the LBP based features for the corresponding input images. GLCM is a distribution of the co-occurrence matrixes obtained over an image for the distribution of gray level pixel values. The LBP is a type of visual descriptor that is combined by the histogram of oriented gradients descriptor.

 

1.2   Block Diagram:

 

Figure 1. Block Diagram for the Proposed melanoma classification System

 

2.3  CLASSIFICATION:

The binary classification process is done by means of the classifier called SVM classifier. The SVM classifier is used in many applications like ECG signal classification [10] and brain image classification [11] systems. The features that are extracted are given as the input to the classifier which acts as a decision making machine. The SVM makes the decision for the testing images by comparing their features with the training image features. At last the classifier will make the decision successfully whether the test images are normal or abnormal.

 

3. RESULTS AND DISCUSSION:

The images used in the system are from the PH² Dataset that is taken from the dermoscopic instrument. The dataset images are of three stages Benign, Atypical and melanoma images. A specific amount of images are taken from each type and are used for the classification of melanoma. For analyzing the proposed approach, the size of the images are resized to 128x128 pixels and processed. Table 1 shows the performance of the system in terms of classification accuracy.

 

Table 1 Overall classification accuracy of the proposed system

 

From the table 1, it is observed that over 94.5% of accuracy is given at the benign images and 91.8% of accuracy is given at atypical images and for the melanoma images 96.7% of accuracy is obtained.

 

4. CONCLUSION:

In this paper, a system for melanoma classification based on GLCM and LBP features and the SVM Classifier is presented. Here the features are extracted at first and the classification is done at lost for the performance measurements. The size of the images used is 128x128 pixels and the GLCM and LBP based features gives 96.7% accuracy at melanoma images.

 

5. REFERENCES:

1.       Garnavi, R., Aldeen, M, Bailey, J. Computer-aided diagnosis of melanoma using border-and wavelet-based texture analysis. IEEE Transactions on Information Technology in Biomedicine. 2012; 16(6): 1239-1252.

2.       Sigurdsson, S, Philipsen, PA., Hansen, LK, Larsen, J, Gniadecka, M, Wulf, HC. Detection of skin cancer by classification of Raman spectra. IEEE transactions on biomedical engineering. 2004; 51(10): 1784-1793.

3.       Yu, L, Chen, H, Dou, Q, Qin, J, Heng, PA. Automated melanoma recognition in dermoscopy images via very deep residual networks. IEEE transactions on medical imaging. 2017; 36(4): 994-1004.

4.       Xie, F, Fan, H, Li, Y, Jiang, Z, Meng, R, Bovik, A. Melanoma classification on dermoscopy images using a neural network ensemble model. IEEE transactions on medical imaging. 2017; 36(3): 849-858.

5.       Sforza, G, Castellano, G, Arika, SA, LeAnder, RW, Stanley, RJ, Stoecker, WV, Hagerty, JR. Using adaptive thresholding and skewness correction to detect gray areas in melanoma in situ images. IEEE Transactions on Instrumentation and Measurement. 2012; 61(7): 1839-1847.

6.       Panda, SS. Jena, G. Image Super Resolution Using Wavelet Transformation Based Genetic Algorithm. In Computational Intelligence in Data Mining.  2012; 2: 355-361. Springer India.

7.       Panda, SS, Jena, G. Sahu, SK. Image super resolution reconstruction using iterative adaptive regularization method and genetic algorithm. In Computational Intelligence in Data Mining. 2015; 2: 675-681. Springer India.

8.       Sonia, R. Melanoma image classification system by NSCT features and Bayes classification. International Journal of Advances in Signal and Image Sciences. 2016; 2(2): 27-33.

9.       Vijaya, AR. ECG Signal Classification Based on Statistical Features with SVM Classification. International Journal of Advances in Signal and Image Sciences. 2016; 2(1): 5-10.

10.     [Mohankumar, S. Analysis of Different Wavelets for Brain Image Classification using Support Vector Machine. International Journal of Advances in Signal and Image Sciences. 2015; 2(1): 1-4.

 

 

 

Accepted on 10.09.2017        © RJPT All right reserved

Research J. Pharm. and Tech 2017; 10(12): 4391-4392.