CHAPTER 5 SEGMENTATION OF CALCULI FROM ULTRASOUND KIDNEY IMAGES BY REGION INDICATOR WITH CONTOUR SEGMENTATION METHOD

135 CHAPTER 5 SEGMENTATION OF CALCULI FROM ULTRASOUND KIDNEY IMAGES BY REGION INDICATOR WITH CONTOUR SEGMENTATION METHOD 5.1 INTRODUCTION One of th...
Author: Bernard Wright
0 downloads 4 Views 1MB Size
135

CHAPTER 5 SEGMENTATION OF CALCULI FROM ULTRASOUND KIDNEY IMAGES BY REGION INDICATOR WITH CONTOUR SEGMENTATION METHOD

5.1

INTRODUCTION One of the most common problems that occur in the human urinary

system is renal calculi, which is often called as kidney stones or urinary stones (Ioannis Manousakas et al 2010). Normally, any person affected with kidney stone will suffer from considerable pain resulting in abnormal kidney function and the mechanism for this disease is not clearly understood so far ( Jie-Yu He et al 2010). Kidney is the most crucial and complicated organ in the urinary system, which involves both in production of urine and purification of blood.

The two important functions of kidney are:

(i) Removing toxic substances from blood, and (ii) Keeping the useful components in proper balance. Due to the presence of powerful speckle noises and attenuated artifacts in abdominal US images, the segmentation of stones in these images is very complex and challenging (Abhinav Gupta 2010). Hence, this task is performed by the use of much prior information such as texture, shape and spatial location of organs. Many automatic and semiautomatic methods have been proposed to detect the renal calculi. Even though the performance of such methods are better when the contrast-to-noise ratio is high, they yield poor results, when the structures are inadequately defined and have low

136

contrast like the neuroanatomic structures, such as thalamus, globus pallidus, putamen, etc. (Maulik 2009). Since US kidney images are noisy and contain poor S/N ratio, an alternative effective technique employing a-priori information may be utilized for compensating such problems (Bommanna Raja et al 2006). The segmentation of renal calculi using renal images is a difficult task. In the previous renal calculi IOREWS segmentation method, the Renal calculi have been segmented from the medical US kidney stone images using region indicators and modified watershed segmentation. But, the accuracy of stone detection in IOREWS method does not give adequate result. Moreover, the high complexity of the technique has given only less utility to the medical environment. Hence, to overcome these drawbacks, a new Region Indicator with Contour Segmentation (RICS) method is proposed. In this proposed RICS segmentation method, five major steps are followed to select the exact calculi region in the renal calculi images. In the first and second stages, the region indices library and renal calculi region parameters are computed. Then, the image contrast is enhanced by the Histogram Equalization and the most interested pixel values of enhanced image are selected by the k-means clustering. The most interested pixel values are utilized to find the accurate calculi in the renal images. In the final stage, a number of regions are selected based on the contour extraction process. Subsequently, pixel matching and sequence of thresholding process are performed to find the calculi. In addition, the usage of ANFIS in supervised learning has made the technique more efficient than the previous techniques. Here, the utilization of contour reduces the relative error in between the Expert radiologist and the segmented calculi, which are obtained from the proposed algorithm. Thus, the obtained error is minimized when compared to the existing algorithm which results in high efficiency.

137

The implementation result shows the effectiveness of the proposed RICS method in segmenting the renal calculi. And also, the proposed method improves the calculi area detection accuracy with reduction in computational time. The performance of the proposed RICS technique is evaluated and compared with other segmentation methods. 5.2

CONTOUR BASED SEGMENTATION Contour based segmentation methods are used in the field of image

processing to locate the contour of an object. Trying to locate an object contour purely by using traditional edge based segmentation methods is not particularly successful. There are discontinuities in those methods. There may be holes in between the edges or over edges which can be present because of noise. The contour based segmentation methods try to improve this by striking attractive properties such as link and smoothness to the contour of the object.

The contour based approaches use some degree of prior

knowledge for dealing with the problem of identifying the object contour. One of the most important models in segmentation of medical imaging is contour model because of its merits in both theory and practical applications. One of the important preprocessing steps in image analysis is segmentation. It aims at finding the required region of interest using a-priori knowledge of the exact location and contour of anatomical structure. The main objective of contour model is to partition an image into fixed number of semantically important regions. Here this work is aimed to present a new segmentation method to contour the renal calculi images based on the renal calculi edges and its corresponding homogeneous regions. The combination of boundary and region based information helps to detect objects with accurate results. The multiscale model is used to retrieve the object structure at various scales of observation simultaneously.

138

When the contour model is applied to the image field, it possesses several advantages over the conventional methods for extracting image contours. Active contours or snakes, are the curves which are generated by the computers to enhance the clarity of the image in the view of discovering the object boundaries. They are often used in image analysis and computer vision to locate the objects and represent their exact shapes. The contour model offers the benefit of extracting the similar features with respect to the correct nature of these features without committing itself and their required implication. Another advantage of the contour model over the conventional edge extraction methods is its native connectivity. In a linked contour, information along the length of the contour is incorporated in an implicit manner. It is also recognized that the power of contour methods is that they are adaptable and can fit a great number of diverse shapes. Various contour based models have been discussed in the literature. The contour models are classi¿ed into two types x x

Region-based models Boundary-based models

Region based models obtain a contour representation from the segmentation of the image into well de¿ned regions. The image is examined point-wise in order to choose if a pixel belongs to an object or it belongs to outside the object. A pixel belongs to the boundary if it is in the object region and has neighbours in the background. This segmentation is then used to create an image force ¿eld which aligns the active contour with object of interest.

139

Edge-based methods use a constant approximation of the input image intensity function so that the boundary can be characterized by a differential property. By point-wise examination of the image, a pixel is said to belong to the boundary if it is a local maximum of the image gradient. In this boundary detection process, the fact that these boundary points constitute a closed geometric contour is not taken into account. Mignotte and Meunier (2000) have devised an algorithm to detect the boundaries of endocardial contour or the middle wall of arteries, in US images. They have used the contour model framework for detecting the objects inside and outside boundaries. This framework minimizes the global energy function and the procedure finds the optimal position of contour models with decreased thickness. This method is speedy, produces good convergence properties and particularly well suitable for extraction of anatomical structure boundaries in US imagery automatically. 5.3

PROPOSED

RENAL

CALCULI

SEGMENTATION

TECHNIQUE The proposed renal calculi segmentation method consists of six major steps namely, (i) Preprocessing (ii) Determining inner region indicators (iii) Determining the region parameters (iv) Enhancing the contrast of the image using Histogram Equalization (v) Finding most fascinated pixels by K-means clustering and (vi) Contour based Region selection Process. The proposed renal calculi segmentation training and testing procedure is shown in Figure 5.1.

140

Testing Image dataset

Preprocessing Training Image dataset Enhancing the image contrast by Histogram Equalization

Preprocessing

Inner Region Indicators

Indices Library

Findinding Most Fascinated Pixels

Determining region parameters

Region Selection by Contour

ANFIS System Training

ANFIS System Testing

Segmented Calculi Image

Figure 5.1

Proposed RICS

Segmentation Training

and

Testing

Procedures 5.3.1

Preprocessing In this RICS method, the input images are resized to the image of

256 x 256. The resized images are given as input to the system. In this preprocessing phase, principal component analysis with local pixel grouping (LPG-PCA) based image denoising algorithm is used to remove the noise

141

from the US renal calculi images. The process involved in LPG-PCA process is given in section 4.2.1. 5.3.2

Determining Inner Region Indicators Let D represents the renal calculi image training dataset, which

contains renal calculi images D {I1 , I 2 , I n }; n 1 N , where N is the number of the renal calculi images which is in the given dataset D . The inner region indicators procedure is given in section 4.2.3, an index value, which can be represented as B {I i } : i 1 L , is allocated. Then, each block included in B is checked to find the edge pixels present in the kidney. If any block is found to be containing edge pixels of the kidney, then the index value

of the corresponding block is kept as K {I l } : l  L . Hence, K which can also

be called as indices library, contains the indices of blocks of all the known images. 5.3.3

Determining Region Parameters Using the renal calculi images in D, the calculi and non calculi

regions are extracted. The extracted regions from the renal calculi images are R

{r1 , r2 ,  rm }, m

1 M , where M represents the total number of

extracted regions. Next the centroids values for all the renal part of images in I

I

I

I

D are found, that is C( x , y) {c11 ( x, y), c 22 ( x , y),  c nn (x , y)} , where c11 (x , y)

is a centroid value of image I1 . Then, the region parameters for the extracted regions are determined from R by utilizing MATLAB function. The region parameters determined for each region are (i) Area (ii) Centroid (iii) Orientation and (iv) Bounding Box. This region parameter values are given to the ANFIS system for training process. In training process, the normal and calculi area is identified by the threshold values t1 and t 2 . The ANFIS

142

system result value is represented as F . The final decision is defined by equation 5.1 as shown below F

5.3.4

­t 1 ® ¯t 2

normal

(5.1)

calculi

Contrast Enhancement using Histogram Equalization The contrast enhancement process is given in section 3.4.1.2,

initially the given US image I nt is converted into a grayscale image G nt , as HE process can be used only on grayscale images. HE makes some enhancements to the contrast of the given gray scale US image. In HE all pixel values in gray scale image are adjusted to maximum intensity values of the image. The image that is obtained after the HE process is denoted as G nt ' . 5.3.5

Finding most Fascinated Pixels by K-Means Clustering Mostly required pixels are computed from the image G nt by

utilizing the k-means clustering method. The K-means clustering is a method of cluster analysis which aims at partition of observations into number of clusters in which each observation belongs to the cluster with the nearest mean. The steps involved in the K-means clustering used in the proposed method are described as in section 4.2.4.1 The k-means algorithm aims at minimizing an objective function

¦ ¦ A G

H

a 1g 1

d ag

 Ca

2

(5.2)

143

In Equation (5.2) d ag represents data points and C a means center of the cluster. The resultant of the k-means clustering process has a number of clusters, which form a cluster-enabled image I A . Here the cluster can be detected, with maximum white colour pixel values, and is applied to the newly created mask, I ' . 5.3.6

Contour based Region Selection Process Region selection process performed using renal calculi images are

taken from the testing image dataset D t

{I1t , I 2t , I nt }; n 1 N t , where N

represents the total number of renal calculi images in the dataset D t . The

dataset D t contains the images that are in the dimension of P u Q ; 1 d p d P ,1 d q d Q .

The procedure for contour based region selection Process is as follows Step 1 Initially the contour plot of the given gray scale image G nt is extracted. The contour function is described in the following Equation 5. 3. G ntc

x

x

x

(G nt , k )

(5.3)

G nt is an input renal calculi gray scale image

k is the number of evenly spaced contour levels in the plot In order to find the contour plot, the axis its orientation and aspect ratio are defined automatically.

x

Where G ntc represents the result of the contour method renal calculi gray scale image

144

Step 2 tc

Then, the final group values from the contour result image G n are selected. This group values contain some regions, then the region parameters are calculated for that regions and the region parameter values are given to the ANFIS system that are referred in section 5.3.3. Step 3 Then few numbers of regions are chosen from the image G

tc n

which are greater than the threshold value t1 and this selected region

values are given to the empty mask S . Step 4 The mask S contains m s number of regions, which is represented as R s

{r1s , r2s , r s s }, m s m

1 M s .

Next, compute the centroid values for the regions R s in the mask S , it is represented as C s ( x, y ) {c1s ( x, y ), c2s ( x, y ), c s s ( x, y )} . m

Step 5 There are m s number of regions in the mask ‘ S ’. These mask regions are not optimal to find the exact calculi from the images. So the optimal regions are found among the available regions in S by exploiting Squared Euclidean Distance (SED) between the regions.

145

Step 6 SED is computed between the x coordinates regions centroid values C s ( x, y ) and training images centroid values C ( x, y ) and y coordinates regions

centroid values C s ( x, y ) and training images centroid values C ( x, y ) values individually. Step 7 The Squared Euclidean Distance (SED) difference process is described in the equation 5.4 and 5.5 for both x and y coordinates values. M ( x)

( c1 1 ( x )  c1s ( x )) 2  ( c 2 2 ( x )  c 2s ( x )) 2   ( c nI n ( x )  c s s ( x )) 2

(5.4)

M ( y)

(c1 1 ( y )  c1s ( y )) 2  (c 2 2 ( y )  c 2s ( y )) 2   (cnI n ( y )  c s s ( y )) 2

(5.5)

I

I

m

I

I

m

The values M (x ) and M ( y ) are compared with the threshold value

t 3 . If any one of the result values M (x ) or M ( y ) are greater than the given

threshold value t 3 , that corresponding region are selected. Step 8 Then, the last group values are selected from the contour method result image G ntc and these values are placed into the newly created mask M . After getting the result from contour process, the pixel matching and Multidirectional traversal operation is performed as in IOREWS method

146

x Pixels Matching Here, first step is to divide the mask image M into m number of blocks and the index values I x performed in section 5.3.2. Then I x

x {I m } are allocated as being already x {I m } is compared against K by using

the following conditions (i)

Retain the pixel values in the block m  M ; if an index value x Im

I L , then.

(ii) Change the block m  M pixel values into 0; or else And hence M ' is generated. Over M ' and I ' an AND operation is performed followed by a morphological dilation operation and hence the resultant image U is obtained. x Multidirectional Traversal Here two major traversals namely bottom-up traversal and top-down traversals have been applied. In each of the traversal, a left-right traversal is applied. The traversals are applied over U , which is binary. At the time of two major traversals, once the pixel with ‘1’ is obtained, then left-right traversal is enabled so that all the regions in the same axis and the region of the first obtained pixel are removed from the mask. The survived pixel values are marked into the original test image and it is subjected to the consequent process of thresholding. x Thresholding Here, a chain of thresholding processes is performed on the original image.

147

x Firstly, the pixel values that are marked by using the previous process are compared against a defined threshold value t 3 . The pixel values those are greater than t 3 are stored in a newly created mask U s .

x The region parameters are determined (described in section 5.3.3) for the regions in the mask U s and the computed region parameters are given to the ANFIS to obtain the ANFIS score. If the ANFIS score is greater than t 4 , then the selection of regions is performed.

x Then, the number of neighbouring pixel values around the selected regions which are greater than the threshold value t 5 is counted, and the number of count value of each region is compared with a threshold value t 6 . If the count value is greater than the threshold value t 6 , then the regions are selected.

x The selected regions from the previous thresholding process are involved in the morphological dilation operation. After the morphological operation, the number of regions that are presented in the image is counted. The region count value is compared with two threshold values t 7 and t 8 .

x If the count value is greater than t 7 , then the traversing down operation is performed once, and if the count value is greater than t 8 , then traversing down operation is performed for multiple times.

x In the final thresholding process, each region area value is calculated and it is compared with the threshold values t 9 and t10 . The regions that are less than t 9 and greater than t10 are

selected. The selected regions are then placed into the original testing image I nt .

148

By performing all the above described process in various renal calculi kidney images, the calculi region is segmented. 5.4

RESULTS AND DISCUSSION The proposed RICS segmentation technique is implemented in

MATLAB platform (version 7.10) and the performance of the proposed RICS is evaluated.

In the proposed RICS segmentation method, five major steps

are performed over these training and testing renal calculi US images. The sample input of normal and calculi images are shown in Figure 5.2.

(i) Normal Renal Image Figure 5.2

(ii) Renal Image with Calculi

Sample Input Renal Images (i) Normal Renal Image (ii) Renal Image with Calculi

The region parameter values are computed for the 110 training images and these parameters result values are given to the ANFIS system to perform the training process. The region parameter values are well trained in the ANFIS system and this performance is evaluated with testing renal calculi images. 50 testing images are involved in the testing process. Figure 5.3 shows the results of the preprocessing, HE, k-means clustering and contour method.

149

(i) Preprocessing

(ii) Histogram Equalization

(iii) K-Means clustering Figure 5.3

(iv) Contour Process

Result Images are obtained from (i) Preprocessing (ii) Histogram Equalization (iii) K-Means Clustering (iv) Contour Process

The HE image contrast is enhanced when compared to the original input image shown in Figure 5.3 (ii). In Figure 5.3 (iii), the same pixel values are grouped into number of clusters values and this can be used to find the most interested pixel values. The result images in Figure 5.3 (iv) shows that the contour process has divided the testing image pixels into three groups by representing three different colours. The selected group value from the contour process result is shown in Figure 5.4.

Figure 5.4 Selected Group Value Result from the Contour Process

150

After the contour process, the chain of dilation and traversing operations are performed in the processed renal calculi image results that are shown in Figure 5.5. The traversing operations eliminate the most unwanted regions from the renal calculi images so as to easily find the calculi from the image. Subsequently, the thresholding process intermediate results are shown in Figure 5.6.

(i) Pixel Matching

(iii) Traversing up

(ii) Morphological Dilation

(iv) Traversing down

Figure 5.5 Result Images from (i) Pixel Matching (ii) Morphological Dilation (iii) Traversing Up and (iv) Traversing Down

151

(i)

(ii)

(iv)

(iii)

(v)

Figure 5.6 Thresholding Process Results Finally, the selected regions from the thresholding process are given to the original image that is demonstrated in the following Figure 5.7. In Figure 5.7, the calculi regions are exactly marked in red color. The result image has shown that the proposed RICS method has exactly found the calculi region from the renal calculi images. The performance of proposed RICS method is analyzed with different images and it is described in the following section.

Figure 5.7 Proposed RICS Method Result Image The Figures 5.8 and 5.9 shows sample input image and the corresponding result.

152

(a) Input Image

(b) RICS Method Result Image

Figure 5.8 Input and Result of RICS Method

(a) Input Image

(b) RICS Method Result Image

Figure 5.9 Input and Result of RICS Method 5.4.1

Performance Analysis The performance of the RICS method is evaluated by utilizing 50

testing images. This performance analysis exploits statistical measures to compute the accuracy of calculi segmentation done by the RICS method. The statistical performance measures, (the statistical measures are detailed in section 3.6.1) which are obtained for sample of ten US image of the proposed RICS method are shown in Table 5.1.

153

Table 5.1 Statistical Performance Measures for Ten US Renal Calculi Image

SV

SC

FPR

PPV

NPV

FDR

MCC

ACC

1

90.42

99.91

0.29

50.26

99.97

49.74

67.32

99.88

2

99.54

99.89

0.51

43.46 100.00 56.54

65.76

99.89

3

94.24

99.85

0.35

57.38

99.97

42.62

73.38

99.88

4

93.31

99.89

0.20

69.79

99.97

30.21

80.77

99.87

5

89.7

99.7

0.70

42.17

99.94

57.83

61.22

99.35

6

96.39

99.57

0.63

43.85

99.98

56.15

64.89

99.86

7

100

99.84

0.46

49.92 100.00 50.08

70.49

99.87

8

100

99.96

0.21

57.49 100.00 42.51

75.74

99.81

9

87.5

99.93

0.07

81.85

99.96

18.15

84.57

99.89

10

92.42

99.87

0.43

51.99

99.96

48.01

69.20

99.88

ID

In Table 5.1, the results show that RICS method produces better results in terms of all the measures. In terms of sensitivity measure, the proposed scheme achieves 100% performance in 7th and 8th images, but the 9th image sensitivity value is lower (87.50%) when compared to performance level of the other images. The high performance of sensitivity measure shows that the stones areas in the input images are identified more accurately. The low performance form of the 9th image does not degrade the RICS method performance because the low level performance is slightly deviated from the high level performance. Moreover, the 9th image gets the 99.93% performance in the specificity measures compared to specificity values of other images. The high performance of specificity measure shows that the normal areas in the input images are identified more precisely. Among specificity measures of the 10 images, the 6th image gets the lower (99.57%) specificity

154

measure than other images. The average performance of the proposed RICS method achieves 94.35% sensitivity and 99.84% specificity measures. The proposed RICS method achieves higher accuracy of 99.89% results in 2nd and 9th input images and gets lower accuracy of 99.35% in 5th image. When compared to other images, the accuracy values of those are slightly deviated from the higher accuracy value. This small deviation does not affect the proposed method performance. Furthermore, the average performance of the proposed RICS method achieves 99.82% accuracy results. Also high accuracy level has been achieved in 1 second computational time. The segmented calculi area by RICS method is compared with previous IOREWS segmentation method and existing segmentation algorithms. Table 5.2 shows the performance of IOREWS with ESRANFIS, SVM and NN using various statistical measures for average of 50 test images. The ESRANFIS, SVM and NN results are given in Table 3.6 of chapter 3. Table 5.2 Statistical Performance Measures of IOREWS, ESRANFIS, SVM and NN Method/ Measures RICS IOREWS ESRANFIS SVM NN

SV

SC

FPR

PPV

NPV

FDR

MCC

ACC

94.35

99.84

0.16

74.72

99.98

25.28

83.17

99.82

85.67

99.72

0.28

60.23

99.95

39.77

70.30

99.67

52.45

99.77

0.23

41.73

99.86

58.27

45.98

99.63

55.22

99.62

0.38

37.89

99.83

62.11

44.23

99.45

51.27

99.66

0.34

42.45

99.78

57.55

45.75

99.45

155

As shown in Table 5.2, the RICS method performance is better when compared to IOREWS, ESRANFIS, SVM and NN methods. The sensitivity measure of RICS method performance is highly deviated from the other methods, but the specificity measure of RICS is slightly deviated from specificity measure performance of other methods. In addition, the False Positive Rate (FPR) of RICS method is lower when compared to that of IOREWS, ESRANFIS, SVM and NN methods. The FPR is the proportion of absent events that yield positive test outcomes. Moreover, the other measures like PPV, NPV, FDR and MCC have shown high performance result for RICS method. By comparing the measures of all the proposed methods, the RICS method produces average results of 94.35%, 99.84% and 99.82% for sensitivity, specificity and accuracy respectively.

The

results

also

demonstrates that the RICS method produces improved results in SV, SC, PPV, NPV, MCC and ACC measures. The FPR and FDR values are minimized compared with other three methods. The RICS method produces best results in terms of all measures. So it has been proved from the results that the RICS method is better in terms of sensitivity and all other measures compared to ESRANFIS, IOREWS, SVM and NN. A relative error is calculated between the segmented calculi area marked by the expert radiologist and the proposed method. The formula for the calculation of relative error is given in Equation (3.45). The calculi area marked by the expert radiologist, the RICS method and its relative error are given in Table 5.3.

156

Table 5.3 Relative Error Performance of RICS Expert Radiologist (mm2) 88.4 61.5 64.0 91.9 72.4 25.0 121.7 36.0

calculi area of RICS method (mm2) 88.3 61.6 64.0 90.3 72.4 24.9 121.1 36.0

Relative error of RICS method 0.113 0.163 0.000 1.741 0.000 0.400 0.493 0.000

The relative error performance of RICS method is compared with the IOREWS method and conventional algorithms values that are shown in Table 5.4. Table 5.4 Comparison Result of RICS segmentation Relative Error with IOREWS Method and Existing Algorithms Performance Image ID

Relative error of Proposed RICS Method

Relative error of Proposed IOREWS Method

1 2 3 4 5 6 7 8 Mean

0.113 0.163 0.000 1.741 0.000 0.400 0.493 0.000 0.323

0.170 0.650 0.625 1.959 2.210 0.244 1.068 0.556 0.579

Relative error Relative error of existing Log of existing Decompression SRAD Technique Model 2.180 3.142 1.441 1.038 0.908 9.207 0.381 0.370 2.333

1.557 2.243 0.153 0.692 0.893 0.929 0.198 0.192 0.857

157

The Table 5.4 shows that the relative error of RICS segmentation method provides 44%, 86% and 62% more calculi area detection accuracy than the proposed IOREWS and existing SRAD and Log Decompression algorithms. The computational times of ESRANFIS, IOREWS and RICS methods are obtained from the calculi detection process.

The average

Computational time of these systems and existing SRAD are shown in the Table 5.5. From the results, it has been observed that proposed RISC method performs calculi detection process in very less time compared to other two proposed methods and other existing methods. Table 5.5 Computational Time of the ESRANFIS, IOREWS and RICS Methods

Methods ESRANFIS IOREWS RICS SRAD

Computational Time(sec) 13.38 197.2 1.63 4.4

Subsequently, the accuracy of proposed RICS segmentation is evaluated via the Receiver Operating Characteristics (ROC) curve analysis. The proposed RICS segmentation method has yielded a higher accuracy results in segmenting the calculi from the renal calculi images. The segmentation accuracy of the proposed method evaluated using ROC curve is shown in Figure 5.10. Here, the diagonal line is entirely a random guess and the blue colored line represents the ROC curve of the proposed segmentation method. A point above the diagonal line signifies good segmentation results and below the points represents poor segmentation result. The ROC curve is

158

plotted between the False Positive Rate and True Positive Rate. In Figure 5.10, the points that are above the diagonal line are the segmentation result.

Figure 5.10 ROC Curve for the Proposed RICS Segmentation Method 5.5

SUMMARY In this chapter, a RICS segmentation method to segment the calculi

from the renal calculi images has been proposed. The proposed method has been implemented and set of renal calculi images are utilized to evaluate the proposed RICS segmentation method. The proposed method has exactly detected the calculi and produced a high segmentation accuracy result when compared to ESRANFIS and IOREWS segmentation methods. The performances of RICS segmentation method are analyzed with the previously proposed IOREWS segmentation and existing conventional algorithms. The proposed RICS produces less relative error than the IOREWS method does. Moreover, the proposed RICS segmentation method has produced 99.82% of accuracy and 44%, 86% and 62% more calculi area detection accuracy than that of IOREWS segmentation method and the existing algorithms.

Suggest Documents