Liver Image Analysis in CT
4.1.3 Post-processing
After segmenting the liver area from the CT image, the segmentation result usually contains not only liver tissue but also other organs, which have similar intensity to the liver. In addition, it has holes and connected pixels with the neighbour organs. Therefore, the post-processing based on morphological operation is applied on the segmented area.
The morphological process is used to delete certain unwanted region and filling the holes, which lead to detect the liver object and enhance the segmentation result. Usually a set of structure elements are used in morphological process, such as dilation, erosion, opening, and closing. The shape and size of structure element is important in selecting or extracting object to avoid of removing too many desirable objects, or keeping too many unwanted ones (Luo et al., 2009). Here a circular structure element with radius 3 is adopted which selected as a compromise between elimination of the unwanted area and the connected organs with each other and preservation of details the object. Therefore, the value of structure element was selected based on the prior- knowledge of the anatomical structure of the abdomen and the resolution of the CT image. Hence, the liver segmentation post-processing is applied as follows:
• Applied Erosion process to delete the fragments of other organs.
• Applied Connected Component Labeling algorithm (CCL) to identify connected pixel areas where the largest connect pixels is selected as liver region based on pre-knowledge (The liver is the largest internal organ in the human body). • Applied Dilation process to fill the holes and reserve the pixels that removed
from the liver by erosion process.
• Filled the remain holes inside the segmented region.
• Obtained the segmented liver by complemented and multiplied the resulting bi- nary mask with the original CT image, as shown in Equation4.7.
S(i, j) = (
CT (i, j), if M (i, j) = 1
0, otherwise (4.7)
Where S(i,j) is a final segmented liver, CT(i,j) is the original CT image and M(i,j) is the liver mask.
Figure4.7depicts the liver segmentation process. The original CT image shown in Figure4.7.a. The pre-processing median filter is used to smooth and remove the image noise to improve the segmentation process. The pre-knowledge and the histogram- based adaptive threshold is used to produce the initial binary mask for the liver, as
shown in Figure4.7.b. The erosion process is used to separate the liver area from other organs, as shown in Figure4.7.c. The CCL algorithm is applied to select the largest connected pixels area as considering the liver is the largest organ in the abdomen, as figured in Figure4.7.d. The dilation and filling the holes is used to refine the segmented liver mask, as shown in Figure4.7.e and Figure4.7.f respectively. Figure4.7.g shows the final results for liver segmentation.
Figure 4.7: Steps of liver segmentation process. (a) The original input CT image. (b) The initial binary liver mask. (c) Operates erosion morphology. (d) Select the largest connected pixels as the liver is the largest organ that appear in CT image. (e) Operates dilation morphology. (f) The final liver mask after filling the holes. (g) The final result of the liver segmentation.
Figure 4.8 illustrates the graphical evaluation results of liver segmentation based on Dice similarity coefficient (DSC) metrics. The vertical axis (x-axis) represents the number of cases that achieved the DSC values in the horizontal axis (y-axis). The highest DSC value means the better segmentation result.
Figure 4.8: Number of cases and obtained DSC values of liver segmentation for eval- uating the method.
Chapter 4. Liver Image Analysis in CT
Figure 4.9:Samples of liver segmentation results (our method versus ground truth). (a) The original input CT image. (b) The ground truth of liver segmentation representing by the red line. (c) Liver segmentation result (green line) with ground truth (red line).
Figure 4.9 depicts the difference in segmentation by our automatic method com- pared with the ground truth obtained from the radiologist (included in the dataset). However, in some CT cases, the segmentation results are not quite satisfactory. The
increase in a false positive segmentation is due to some of the organs may surround the liver and have almost the same intensity value. In addition, the boundary between liver and contact organs may disappear and are difficult to discriminate it and this lead to incorrect segmentation by considering the other organs as a part of the liver.
By inspecting the segmentation results, we can find that some exams presented a high false positive due to the contact organs have similar intensity to the liver with no clear boundaries between of them. Hence, to overcome this problem, the segmented liver boundary was traced based on location and area across CT slices. In order to capture the significant increase in the liver area and irregular change in the boundaries over CT slices, due to overlap between the contact tissues and liver area. By compar- ing the liver boundary slice-by-slice, the unexpected added area will be removed, as depicted in Figure4.10.
Figure 4.10: Refine segmented liver boundary. (a) The CT image in the first slice. (b) The segmentation result(green liner) compared to the ground truth (red line). (c) The CT image in the next slice. (d) The segmentation result(green liner) compared to the ground truth (red line). (e) Refine the segmentation for the first slice.
The segmentation results of the first slice, as depicted in Figure 4.10.b, shows a high false positive due to the surrounding non-liver tissues have short common bound- aries with liver as shown in Figure 4.10.a (inside red circle). In the next slice, the
Chapter 4. Liver Image Analysis in CT
boundaries between liver and surrounding tissues disappeared, as shown in Figure
4.10.c (inside blue circle). By comparing the first slice to the next slice, non-liver areas will be discarded, as illustrated in Figure4.10.e.