Presentation is loading. Please wait.

Presentation is loading. Please wait.

Good afternoon everyone! I am Yu hsun wu

Similar presentations


Presentation on theme: "Good afternoon everyone! I am Yu hsun wu"— Presentation transcript:

1 VESSEL SEGMENTATION BASED ON BONE-TO-BONE ELIMINATION IN BRAIN CT ANGIOGRAPHY
Good afternoon everyone! I am Yu hsun wu Today, I would like to present our paper VESSEL SEGMENTATION BASED ON BONE-TO-BONE ELIMINATION IN BRAIN CT ANGIOGRAPHY Authors : Ku-Yaw Chang、Yu-Hsun Wu、Wei-Lu Lin、 Shao-Jer Chen、Lih-Shyang Chen 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

2 Outline Introduction Method Results Conclusions
This is the outline of my presentation Let’s start with the introduction. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

3 Introduction CT(Computer tomography)
CTA(Computed tomographic angiography) Need injection contrast medium CTA Image(Highlight artery) CT Image In this paper, we propose a new method using both CT and CTA images for vessel segmentation., On the left side, it’s a CT image. The bone structure is displayed with larger gray levels. However, the vessel structure is unclear. On the right side, it’s a CTA image. After injecting a contrast medium into vessels, the vessel structure is highlighted so that, we can see these vessels very clearly. Like this…and these small spots. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

4 Introduction Vessel and bone not connected Easily segmented out
Region Growing Set seed point Segmentation result When bone and vessel structures are not connected, they can easily segmented out, using simple algorithms, like region growing. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

5 Introduction Brain case – vessel and bone connected
Difficult to segment vessels out Vessel and the bone connected However, in many cases, like the brain, bones and vessels are usually connected, so that, it becomes quite difficult to segment vessels out. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

6 Introduction Common solution align CTA and CT images
precise registration algorithm time-consuming One common solution to vessel segmentation is to align CTA images with corresponding CT images. Bones can thus be removed by cross-referencing these two series of images, making vessels easier to be identified. This alignment procedure requires a precise registration algorithm, which is quite time-consuming and difficult to have satisfactory results in clinics. sub CTA image CT image Segmented result 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

7 Introduction Our method not rely on perfect image registration
bone-to-bone searching procedure to remove most of the bone pixels Our method does not rely on perfect image registration between these two series of images for bone subtraction. Instead, we adopt a bone-to-bone searching procedure to remove most of the bone pixels. Our method requires less computation and could be incorporated into an interactive segmentation framework. Standard image 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

8 Outline Introduction Method Results Conclusions
Now, let’s look at our method. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

9 Method Step1:Simple Threshold Step2:Preliminary Elimination
Start End Simple Threshold Preliminary Elimination Morphological Erosion Region Growing Conditional Dilation Step1:Simple Threshold Step2:Preliminary Elimination Preliminary removed bone Step3:Morphological Erosion Step4:Conditional Dilation Vessels dilation Step5:Region Growing This is the flow diagram of our method Let’s look at step 1. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

10 Method CTA :target image (contrast material) CT : reference image
First, we define CTA as the target image, and CT image as the reference image. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

11 Method (1/5) Simple Threshold Value 150 Binary target image
Elimination Erosion Dilation Region Growing Simple Threshold Value 150 Binary target image Binary reference image Step 1 Simple Threshold A simple threshold operation is applied to both target and reference images. Threshold value is set to 150, based on experiences 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

12 Method (2/5) Preliminary Elimination Removed bone PT(POI Target)
Simple Threshold Elimination Erosion Dilation Region Growing Preliminary Elimination Removed bone PT(POI Target) PR(POI Reference) Target image(PT) Reference image(PR) Step2. Preliminary Elimination We labeled the pixel of interest in the reference image as PR, and its corresponding pixel in the target image as PT. They have same coordinates. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

13 Method (2/5) Preliminary Elimination Circular search region
Simple Threshold Elimination Erosion Dilation Region Growing Preliminary Elimination Circular search region Original image Range Search Target image For each PT, a circular search region is given and, every pixel inside this region is visited 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

14 Method (2/5) Preliminary Elimination
Simple Threshold Elimination Erosion Dilation Region Growing Preliminary Elimination Removed by assigning background levels Key to avoid registration computation all visited pixels with higher gray levels are considered to be bone structures, and thus removed by assigning background levels. The bone-to-bone search process between the reference and target images is the key to avoid registration computation Target image after preliminary elimination 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

15 Method (3/5) Morphological erosion Remove all the remaining boundary
Simple Threshold Elimination Erosion Dilation Region Growing Morphological erosion Remove all the remaining boundary Vessel region is unavoidably eroded Step 3. Morphological erosion A simple erosion operation is applied to remove all the remaining bone boundary pixels. However, the vessel region is unavoidably eroded, that’s not what we want. For example the vessel after erosion is Gone. Before erosion After erosion 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

16 Method (4/5) Conditional Dilation Keep the vessel region complete
Simple Threshold Elimination Erosion Dilation Region Growing Conditional Dilation Keep the vessel region complete Conditional dilation stops Ratio is smaller than a pre-defined value Step 4. Conditional Dilation In order to keep the vessel region complete, a conditional dilation is then applied. During the conditional dilation process on the target image, only those pixels are added if their corresponding pixels on the reference image have non-zero levels A dilation ratio is defined to be the ‘comparison’ of the area of an object after and before the dilation. The conditional dilation stops when this ratio is smaller than a pre-defined value. Erosion result Dilation result 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

17 Method (4/5) Conditional Dilation
Simple Threshold Elimination Erosion Dilation Region Growing Conditional Dilation Vessel region is identified in the target image Pixels are labeled as background pixels in original images Whenever a vessel region is identified in the target image, its outer pixels are labeled as background pixels in original CTA images, so that, the vessel region is disconnected from its neighbors, especially connected bone structures, Target image 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

18 Method (5/5) Region growing Simple 2D or 3D region growing algorithm
Simple Threshold Elimination Erosion Dilation Region Growing Region growing Simple 2D or 3D region growing algorithm Our method can identify vessels well Once vessel regions are isolated from’ its’ neighborhood, a simple 2D or 3D region growing algorithm can be applied to segment the vessel structure The segmentation results demonstrate our method can identify vessels well. CTA image CT image Result image 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

19 Outline Introduction Method Results Conclusions
Now, Let’s see other results. 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

20 Results Case 1 CTA image CTA(Bone + Vessel) CT(Bone) This CTA image,
On the left side, it’s our segmentation results 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

21 Results Case 2 CTA image CTA(Bone + vessel) CT(Bone)
2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

22 Results Case 3 CTA image CTA(Bone + vessel) CT(Bone)
2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

23 Results Case 4 CTA image CTA(Bone + vessel) CT(Bone)
2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

24 Results Case 5 CTA image CTA(Bone + vessel) CT(Bone)
2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

25 Outline Introduction Method Results Conclusions
Now look at the Conclusions 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

26 Conclusions Vague search vs. precise alignment Conditional dilation
Fist image Second image in this paper, we propose the concept of "vague search", instead of "precise alignment", to eliminate bone structures. we also invent the conditional dilation technique to fix the possible loss of vessel pixels during the vague search process. several experimental results show that our method works quite well, if vessels and bones are connected visually in CTA images. 3D Schematic diagram 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

27 Conclusions Be incorporated into interactive 2D/3D segmentation framework in the future 3D Schematic diagram(vessels + bone) 3D Schematic diagram(vessels) 1 2 3 our algorithm is simple and could be incorporated into interactive 2D/3D segmentation framework in the future 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)

28 Thank you 2016 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW)


Download ppt "Good afternoon everyone! I am Yu hsun wu"

Similar presentations


Ads by Google