br RatioAspect ObjectWidth br ObjectLength br Histogram is worked by


RatioAspect ¼ ObjectWidth ð9Þ
ObjectLength
Histogram is worked by calculating the spread of intensity value. Histogram has some features used to describe texture pat-tern of image. This study used eight features for representing tex-ture pattern. These were mean, standard deviation, skewness, energy, entropy, smoothness, variance and kurtosis (Jain, 1989). GLCM, which stands for grey level co-occurrence matrix is used to represent texture patterns in first order using statistical opera-tion based on pixel values. GLCM expresses the correlation between two pixels from various angular directions (Haralick et al., 1973). GLCM is initiated by calculating the number of related pixels. An example of GLCM process is illustrated in Fig. 6. This pro-cess produces a framework matrix. The framework matrix is used to produce GLCM matrix by counting it up with the transpose value.
Table 1
Fig. 7. Laws’ vector multiplication of E5 and L5 to produce Doxorubicin maps.
Table 2
Nine Laws’ energy maps.
R5R5 Fig. 9. Multilayer perceptron (MLP) architecture.
Fig. 8. Illustration of SVM, (a) alternative hyper-planes, (b) best hyper-plane with optimum margin m.
Please cite this article as: H. A. Nugroho, Zulfanahri, E. L. Frannita et al., Computer aided diagnosis for thyroid cancer system based on internal and external characteristics, Journal of King Saud University – Computer and Information Scienceshttps://doi.org/10.1016/j.jksuci.2019.01.007
6 H.A. Nugroho et al. / Journal of King Saud University – Computer and Information Sciences xxx (xxxx) xxx
GLRLM stands for grey level run length matrices (GLRLM) and represents the relationship of more than two pixels on the image. The concept of the method is to map the texture patterns that have the same intensity of pixel length or called as run length. This study used seven features of GLRLM with different angle direc-tions, i.e. 0L, 45L, 90L and 135L. There are short run emphasis (SRE), long run emphasis (LRE), grey level no-uniformity (GLN), run length non-uniformity (RLN), run percentage (RP), low grey level run emphasis (LGRE) and high grey level run emphasis (HGRE) (Galloway, 1975; Chu and Sehgal, 1990).
Table 3
Diagnosis rule external characteristics.
Classification Rule
Diagnosis Rule
Margin Shape Orientation
Smooth Round to Oval Parallel Benign
Round to Oval Non-Parallel Malignant
Irregular Round to Oval Parallel Malignant
Round to Oval Non-Parallel Malignant
Irregular Parallel Malignant
Irregular Non-Parallel Malignant
Table 4
Diagnosis rule internal characteristics.
Classification Rule
Diagnosis Rule
Content Echogenicity
Cystic Anechoic
Benign
Solid Markedly Hypo Malignant
Hypoechoic
Malignant
Isoechoic
Benign
Complex Markedly Hypo Malignant
Hypoechoic
Malignant
Isoechoic
Benign
Lacunarity expresses the fractal dimension of image. Lacunarity works by implementing the gliding box method. Lacunarity has three features as presented in (10)–(12) (Kadir et al., 2013).
MN
MN
M N
M mn
MN
P
P P
M
N
P
A
MN
Here, Ls, La and Lp represent features of lacunarity, whilst p is size of fractal used to calculate the texture feature. Value of p is 2, 4, 6, 8 and 10. In image processing, components R, G, B and grey level are used for analysing lacunarity features. Ls, La and Lp are cal-culated in each component. However, F1 generation study only used grey level components because this study used grey level image as the input (Kadir et al., 2013).