Automatic Segmentation of Abdominal Aortic Aneurism (AAA) By Using Active Contour Models

Abdominal aortic aneurysm (AAA) is a disease that is caused by dilation of the aortic wall. Dilation of the aortic wall will affect the size of the diameter of lumen and the aorta. In this study we use T1 and T2 images on 4 patients with AAA which generated from MR Imaging to calculate the diameter of the abdominal aortic aneurysm (AAA). To calculate the diameter of lumen and the aorta, the first step is image registration using Laplacian eigenmap method. After that we propose an automatic segmentation method on region of the aorta by using active contour models to get the contour of lumen and the aorta. The last step, we calculate the diameter of lumen and the aorta by using contour of lumen and the aorta. In our experiment, active contour model is very good method for segmentation AAA. In the result, our proposed model give the accuracy rate of lumen is 96.41% and accuracy rate of aorta is 95.22%.


INTRODUCTION
Abdominal aortic aneurysm (AAA) consists of two parts. The inside is known as Lumen, a place where the blood flows and the outside is known as thrombus which attaches in the aortic wall. Thrombus is formed through a process called thrombosis. Thrombosis occurs when platelets attach to the surface of aortic wall. If the blood flows, then platelets will attach to the aortic wall. Consequently can be formed a mass that stands out in the aortic wall, and will cause dilation of the aorta. Dilation of the aortic wall is called an aneurysm.
If the thrombus enlarge, it will cause the dilation of aortic wall and will affect the size of the diameter of the lumen. Abdominal aorta usually have diameter about 2 cm. But it can widen into more than 5.5 cm [1-2]. The dilation of the aortic wall is caused by weakening of the aortic wall because of blood pressure through aorta. If a large aneurysm occurs, it will cause rupture. In a survey in the United States, rupture occurred about 2% of patients with AAA with a diameter less than 4cm. Meanwhile rupture occurred more than 25% of patients with AAA with a diameter of 5cm [1].
Several studies have been done to identify the AAA such as Mussa in 2015, which stated a large AAA surgery indication if the diameter> 5.5 cm or if the diameter of the growth rate value is greater than 1 cm / year [3]. AAA generally considered to be observed when diameter of aorta reaches or exceeds 3 cm. So it is very important to know diameter of the AAA.
The first step in this research is the detection lumen and aorta. Magnetic resonance imaging (MRI) is one of digital examination tools used in the aorta detection. One of the images which is produced by MRI is T1 and T2 weighted images. At T1 and T2 images there are differences in the position of aortic aneurysm as shown in Figure 1, so we need a method that can normalize the position of T1 and T2 weighted images such that we can do detection or automatic segmentation [4].

T1
T2 T1 T2 In this research we use Laplacian eigenmap method. Laplacian eigenmap is a method used to find the same structure of two or more images. Kosasih et al found that Laplacian eigenmap is a very good method to determine optimal position in AAA as shown in Figure 2  In this study we performed a segmentation of the aortic and lumen area. Furthermore, we calculate maximum diameter of the aortic and lumen. In previous studies have been conducted in aortic aneurysm segmentation by Karyati. Karyati get the value of the maximum diameter of the aorta but there have been limited studies concerned on manual segmentation is shown in Figure 3 [5].
The weakness of manual contour is needs time longer and needs a high concentration to get contour lumen and aorta. Therefore, the objectives of this research are to perform automatic segmentation methods in the area of the aorta using active contour models.
(a) (b) (c) Figure 3. Manual segmentation on lumen and the aortic area a. Input Image. b. Lumen area. c. Aorta Area

METHODS
In this section, the process of segmentation to be discussed. The image of Abdominal Aortic Aneurysm was collected using MRI including T1 and T2 weighted derived from 4 patients ( Figure 1). Each image was initially rescaled to compensate the different scaling of the intensity values. After that, we apply Laplacian Eigenmap to find the optimal position of aortic aneurysm. Laplacian Eigenmap is used to extract information from a set of data using the maximum eigenvector [6]. The result of Laplacian eigenmap as shown in Figure 2. After getting the optimal position of aortic aneurysm, we use an active contour model for segmentation of the lumen and aorta.

Image Segmentation Using Active Contour Models
Image segmentation is a fundamental problem in the field of image processing and objects recognition in computer vision [7]. Segmentation is a method to divide an image into several areas [8]. Lots of segmentation techniques that can be used, one of which is a method of active contour models (ACMs). Active contour models are one of a very good method for segmentation [9]. The basic idea of active contour models is the evolution of the curve with some constraints to extract the desired object. Based on the constraints, ACMs method is divided into two types: edge-based models and regionbased models [10][11].
Edge-based models are popular GAC models that use a gradient to construct edge stopping function to stop the evolution of the boundary contour of the object, but if the boundary of this method is weak or non-existent, then it becomes less good [12].
Another method is the method of region-based models. The method of region-based models has many advantages compared to edge-based models. method of region-based models utilize statistical information within and outside the contour to control the evolution and has a better performance expected in a weak edge or without edges [13]. One popular method is Chan Vese models based on the Mumford Shah segmentation techniques [14].

Chan Vese Models
Chan vese models (CV models) are special cases of the problem Munford Shah. Given image, ∈ Ω, CV models formulated by minimizing the following energy function: where: ≥ 0, ≥ 0, 1 , 2 > 0 and 1 , 2 is a constant which is the average intensity inside and outside the contour, with formulations of Level set as follows: IF | +1 − | < ∆ • ℎ 2 or I > iterasi End Else repeat back to step 4 7. Get the lumen and aortic contour.

Morphological Operation
Morphological operation is an operation imposed on a binary (black and white) image to change the structure of the object shape contained in the image. Core morphology operation involves two arrays of pixels. The first array is the image to be subjected to morphological operation, while the second array is named as the kernel or structuring element [15]. There are three morphological operations used in this study. The dilation operation is commonly used to obtain a widening effect on pixels of value 1. The dilation operations is defined in equation (3).
The erosion operation has the effect of minimizing the image structure. The erosion operation is defined in equation (4) [16].
The opening operation is an erosion operation followed by a dilation using the same structural elements. This operation is useful for smoothing the contour of the object and removing the entire pixel in an area that is too small [17].
Definition of opening operation as follows:

Metric Evaluation
In this study, an evaluation of the model is done by examining the level of accuracy obtained from manual segmentation with automatic segmentation. The formula of Accuracy can be seen in equation (6) [18].
Average of Accuracy = Where is Diameter Manual, ̂ is Diameter Automatic and is the number of dataset testing.

RESULT AND DISCUSSION
In this section, we will discuss about automatic segmentation using active contour models (Proposed Model). In this research, we will find an area and diameter of the lumen and aorta. Based on the algorithm of active contour models, result of image segmentation of AAA on patien 1 as shown in Figure 4. Figure 4. Automaticaly segmentation on patient 1 using proposed model Figure 4((a) -(d)) is an automatic segmentation process to obtain the lumen region. To get lumen area is determined coordinate point A(x, y). In this study, given the initials box A(x± 5, y± 5) in Figure 4b. Initials box aims to get the value of the energy level set, the value of the energy level set continues to evolve until it becomes the contour of the lumen area or according with the number of iterations. In this experiment, the number of iterations as much as 90. The results of the segmentation of lumen area with the number of iterations 90 as shown in Figure 4c. Furthermore, to calculate lumen area then is done by make the mask area, as shown in Figure 4d.  Figure 4f and mask area of aorta as shown in Figure 4g. On the mask area, there is a region which is not an region aorta. So that we use a morphological image to eliminate unwanted area. The result as shown in Figure 4h.
Furthermore, we also do segmentation on patients 2, 3 and 4, which can be seen in Table 1 and Table 2.  Table 1 is the automatic segmentation process to obtain the lumen region of on patient 2, 3 and 4. The first step is to make the initial box A (x ± 5, ± 5 y) which can be seen in the second row. The results of lumen segmentation can be seen in the third row. After that, we create a mask to calculate the diameter of the lumen and aorta which can be seen in row 4. The next step is to perform automatic segmentation to obtain the aortic region as in Table 2. Table 2 is the aortic segmentation process on patients 2, 3 and 4. The process segmentation in the aorta is similar to the process of segmentation in the lumen. The first step, we make the initial box A (x ± 5, ± 5 y) which can be seen in the first row.
The results of segmentation are aortic contours which can be seen in the second row. After that, we create a mask to calculate the diameter of the aorta which can be seen in row 3. To get a better aortic region, the morphological process is used to eliminate regions outside the aorta that can be seen in row 4.
After getting the contours of lumen and the aorta, we will calculate the value of the maximum diameter of lumen and aorta. Furthermore, the calculation results were compared with the results of another researcher. Comparison of the maximum diameter of the automatic segmentation results with manual segmentation performed another researcher is shown in Table 3. In Table 3, the first column is the diameter of the lumen obtained from the manual segmentation and the second column is the diameter of the lumen is obtained from the automatic segmentation using the proposed model. After that the fourth column is the diameter of the aorta is obtained from the manual segmentation and the fiveth column is the diameter of the aorta is obtained from the automatic segmentation using the proposed model. Table 3 it can be seen that the average of accuracy of the measurement of the lumen diameter is 96.41% and the average of accuracy of the measurement of the aortic diameter is 95.22%.

CONCLUSION
Measurement of the diameter of the lumen and the aorta in patients with AAA is one way to observe the growth of abdominal aortic aneurysm. AAA was visible when the diameter of the aorta had reached 3 cm. In this study we used 4 patients with AAA. We propose an automatic segmentation by using active contour models to get the contour of the lumen and aorta. The contour is used to calculate the diameter of the lumen and aorta. In this research, the average of accuracy of the measurement of the lumen diameter is 96.41% and the average of accuracy of the measurement of the aortic diameter is 95.22%. In the future study, we will do extraction on region of lumen to obtain information about the direction of blood flow in the lumen.