Abcd Feature Extraction of Image Dermatoscopic Based on Morphology Analysis for Melanoma Skin Cancer Diagnosis

This research present asymmetry, border irregularity, color variation, and diameter (ABCD) feature extraction of image dermatoscopic for melanoma skin cancer diagnosis. ABCD feature is the important information based on morphology analysis of image dermatoscopic lesion. ABCD feature is used to calculate Total Dermatoscopic Value (TDV) for melanoma skin cancer diagnosis. Asymmetry feature consist information of asymmetry and lengthening index of the lesion. Border irregularity feature consist information of compactness index, fractal dimension, edge abruptness, and pigmentation transition from the lesion. Color homogeneity feature consist information of color homogeneity and the correlation between photometry and geometry of the lesion. Diameter extraction is diameter of the lesion. There are three diagnosis that is used on this research i.e. melanoma, suspicious, and benign skin lesion. The experiment uses 30 samples of image dermatoscopic lesion that is suspicious melanoma skin cancer. Based on the experiment, the accuracy of the system is 85% that there are four false diagnoses of 30 samples. Penelitian ini menyajikan ekstraksi fitur citra dermatoskopik untuk diagnosis kanker kulit melanoma berdasarkan asymmetry, border irregularity, color variation, dan diameter (ABCD). Fitur ABCD adalah informasi yang penting berdasarkan analisis morfologi lesi citra dermatoskopik. Fitur tersebut digunakan dalam perhitungan Total Dermatoscopic Value (TDV) untuk diagnosis kanker kulit melanoma. Fitur asymmetry terdiri dari informasi asimetri dan indeks perpanjangan luka. Fitur border irregularity terdiri dari informasi indeks compactness, dimensi fraktal, edge abruptness, dan transisi pigmentasi dari lesi. Warna fitur homogenitas terdiri dari informasi homogenitas warna dan korelasi antara fotometri dan geometri lesi. Ekstraksi diameter adalah diameter lesi. Ada tiga diagnosa yang digunakan pada penelitian ini yaitu melanoma, diduga melanoma, dan benign skin lesion. Percobaan ini menggunakan 30 sampel dari lesi citra dermatoskopik kanker kulit melanoma yang mencurigakan. Berdasarkan percobaan, akurasi dari sistem ini adalah 85% dan terdapat empat diagnosa palsu dari 30 sampel.


Introduction
Malignant melanoma is the most dangerous human skin disease. It is the deadliest from of all skin cancers and arises from cancerous growth in pigmented skin lesion. If early recognized, the melanoma can be removed and the patient can be recovered completely [1]. More early diagnosis of malignant melanoma is a crucial issue for the dermatologists. The list of specify visual features associated with malignant lesions symptoms. Unfortunately, it can be difficult to interpret visually features and then to recognize malignant pigmented lesion. Even experienced dermatologists have difficulties for distinguishing melanoma from other pigmented lesion of the skin, such as typical whose are benign [2]. This problem raises an interest dermatologist that allows ease of clinical recognition of melanoma, including automatic interpretation of color images dermatoscopic with computerized image analysis. That way, there are interesting developments of the computer system aids (computer-aided systems or CAD) for the clinical diagnosis of melanoma as a support for dermatological experts in different analysis steps, such as the detection limit of injury, the calculation of diagnostic features, classification of the types of injuries difference, visualization, and others.
Stages of the process of melanoma skin cancer diagnosis are preprocessing, segmentation, ABCD feature extraction from the lesion, and the calculation of Total Dermatoscopic Value (TDV). Preprocessing and segmentation research has been done by Chastine, et al [3]. This study covers ABCD feature extraction method of object segmentation result that suspected melanoma lesion to get information whether the injury is non-melanoma or melanoma.
ABCD feature is the important information based on morphology analysis of image dermatoscopic lesion. ABCD feature is Asymmetry, Border Irregularity, Color Variation and Diameter features. The melanoma lesions usually have morphology characteristics such as asymmetrical characteristic, irregular edge of the lesion, different color composition, and a large diameter. Asymmetry feature consist information of asymmetry and lengthening Index of the lesion.Border Irregularity feature consist information of Compactness Index, Fractal Dimension, Edge Abruptness, and Pigmentation Transition from the lesion. Color homogeneity feature consist of Color Information Homogeneity and the correlation between Photometry and Geometry of the lesion. Diameter extraction is diameter of the lesion.
The rest of the paper is organized as follows; Section 2 describes architecture of the system and describes method of ABCD feature extraction. Section 3 describes about the experimental result and evaluation performance, and section 4 describes conclusion of this research.

Methodology
Dermatologists generally use slides as image storage and benchmarking visual lesion. Each image has one or more lesion that is located in normal skin with a variety of colors. Lesion is varied in sizes, shapes, colors, and saturations. Figure 1 shows the four types of lesion, i.e. a and b is benign nevus, c and d is malignant melanoma [1]. Preprocessing steps required to improve the quality of the image. Consisting of noise reduction and improve hand-side edge to distinguish the area around the lesion with skin.
All of the process of melanoma skin cancer diagnosis is described on architecture system in figure 2. The preprocessing and segmentation process are previous research [3]. This paper present ABCD feature extraction, compute Total Dermatoscopic Value (TDV) and melanoma skin cancer diagnosis.  In this research, method of preprocessing for smoothing image from noise is median filtering. Median filtering is used for minimizing the influence of small structures like thin hairs and isolated islands of pixels like small air bubbles.The median filteris a non-linear digital filtering technique, often used to remove noise from images or other signals. Median filtering is a common step in image processing. It is particularly useful to reduce speckle noise and salt and pepper noise.
Segmentation aims to select and isolate (separate) objects from an overall image.Segmentation consists of down sampling, filtering, and edge detection. Down sampling stage is a process to decrease the number of pixels and eliminate some of the information from the image. With a fixed image resolution, down sampling the image size is smaller [4]. There are two steps of segmentation process are fuzzyset and region growing. Region growing is an approach to determine which neighborhood pixels from a seed and determine whether a pixel added to the seed or not. The principle of this method is determining the first set of seed point then initialized a region of the seed. Region will continue to grow from seed point into the points close together depending on the criteria. Criteria are usually made based on the specified gray level, intensity, or color.
Region growing is a segmentation technique that gathers the pixels into a homogeneous region according to a similarity criterion. This algorithm requires a seed pixel that lies inside the ROI and WKUHVKROG DV D VWRSSLQJ FRQGLWLRQ ,W VWDUWV ZLWK the seed pixel which represented by the first approximation of the ROI.Four connected neighboring pixels that are above the threshold are labeled as one; these neighbors of these pixels areinspected and the procedure continues.
If the connected pixels is less than the threshold, it is labeled as zero, indicating a boundary pixel, and its neighborhood are not processed. The recursive process continues until all the connected pixels fail the test of inclusion in the region.In order to optimize the results of the region growing, it selects the center of a homogeneous area as the seed pixels. ABCD feature extraction is one of the process to extract the important feature. The results of this process are used to distinguish melanoma or non melanoma. There are four important features i.e. Asymmetry, Border Irregularity, Color Variation, and Diameter.
First, Asymmetry feature. There are two value of asymmetry feature i.e. Asymmetry Index (AI) and Lengthening Index. Asymmetry Index value is computed with the equation 1: Where k is mayor and minor axis, ¨A k is nonoverlapping area of lesion. Lengthening Index is used to describe the elongation of a lesion, for example the degree of anisotropy lesion. Elongation injury is related to HLJHQYDOXH IURP WKH LQHUWLD WHQVRU PDWUL[ This is defined by the ratio of moment of inertia DERXW WKH PDMRU D[LV XVLQJ PRPHQW RI LQHUWLD about minor axis.
Second is border irregularity. There are four value of border irregularity feature i.e. Compactness Index,Fractal Dimension, Edge Abruptnessand Pigmentation Transition. Density index (Compactness Index / CI) is the measurement of the most popular form of barrier which 2D objects estimate unanimous. However, this measure is very sensitive to noise along the boundary term amplified by the square of the perimeter PL is perimeter lesion.
To find PL value, use the surgery Robert edge detector to detect edges. Robert is a differential technique, the differential in the horizontal direction and the differential in the vertical direction, with the added conversion process after the differential binary. Binary conversion technique proposed is the conversion to level the distribution of a binary black and white. Filter kernel used in Robert's method is: Fractal dimention has self-similarity characteristics, and has properties to the scale / size. Each section has a fractal is a different scale has the same nature with the whole fractal. This characteristic causes suitable for fractal compression techniques. Another characteristic is fractal dimension. Dimension size is generally an integer, such as the line has dimension 1, the field has dimension 2, and 3-dimensional cube has, and so on. However, fractal dimension is a strange as it may worth fractions. This fractal dimension can be used as a characteristic of an image.Fractal dimension can be calculated by the method of calculation of the box (box-counting). This method divides the image into the boxes in varying sizes (r).One example of determining the value of r is 2k, with k = 0, 1, 2, ... etc, and 2k, smaller than the size of the image. Figure 3 shows illustration box-counting method.
In general, use the box grid that divides the image into a pixel size rx r. N (r) is evaluated as the number of pixels that contain pieces of barrier injury. Different pixel size and r is obtained as a slop fd regression line log (r) vs. Log (N (r)).
Equation 4 was expanded to: Lesion with irregular boundaries (Abruptness Edge) has a large difference in radial distance (e.g. distance d2 between the centered and the barrier GL C). Barring irregularities estimate by analyzing the distribution of radial distance difference. We obtained a set of gradient magnitude YDOXH RI . H N " N " . ZKHUe K is the limiting sample size) that describes locally the transition between the injury and setting points of skin on each side. To describe more globally, we use the mean m e and variance v e of the gradient magnitude values e (k) which describes the level of steepness and global variations.
HQI:E, F; = Third is color variation. One early sign of melanoma is the emergence of color variations in color. Because melanoma cells grown in grower pigment, they are often colorful around brown, dark brown, or black, depending on the production of melanin pigment at different depths in the skin. To limit further diagnosis, the color variation in a lesion described by C h color homogeneity and the correlation between the geometry and photometry C pg .
Luminance histogram of injuries are divided into three equal-length intervals. Intervals that relate to the three smallest Luminance values defined dark area in the intermediate level to relate to others from injury and is not involved in the quantification of color. Then, the color homogeneity is described as a transition zone of lighter / darker zone and the zone darker / lighter zone when the scan cuts horizontally and vertically.This attribute evaluates the distribution of color on the lesion. Including an explanation of the evolution of the color levels of the barrier centroid GL lesion. This value is larger for nondangerous injury because it has a target aspect, whereas small values indicate danger.  Fourth is diameter. Melanoma tend to grow larger than common moles, and especially the diameter of 6mm. Because the wound is often irregular forms, to find the diameter, drawn from all the edge pixels to the pixel edges through the mid-point and averaged.
After the value of four components is found, then calculate TDV (Total Dermatoscopic Value). To get the TDV values, the formula is obtained as follows: Then the value obtained has the following conclusion that are 1.00 -4.75 -benign skin lesion, 4.75 -5.45 -suspicious, morethan 5.45 ± melanoma.

Results and Analysis
In the experiment, the image that is used as input data is the dermatoscopic image suspected as melanoma. To evaluate performance system, this research uses 30 dermatoscopic images ( figure 4-17).
There are four phases that are used in this research. The first phase is preprocessing of dermatoscopicimage using median filtering. Median filtering is used for minimizing the influence of small structures like thin hairs and isolated islands of pixels like small air bubbles. The result of the first phase is shown in figure 18  and figure 19.   There are4 images that are falsely diagnosed of 30 images. Therefore the performance of system shows that the accuracy is 85%.

Conclusion
The experimental result shows that the ABCD feature extraction can be used to diagnose melanoma skin cancer. This research used 30 dermatoscopic images with various suspicious lesions. There are three diagnosis that is used on this research i.e. melanoma, suspicious, and benign skin lesion.The experimental result is shown that there are 4 imagesfalsely diagnosed of 30 dermatoscopic images. Therefore the performance of system showsthat the accuracy is 85%. Based on the experimental result, the small value of color variation and diameter causes false diagnosis. To improve this accuracy, it requires machine learning approach to diagnose the melanoma skin cancer.