3D Digital Cleansing Using Segmentation Rays Authors: Sarang Lakare, Ming Wan, Mie Sato and Arie Kaufman Source: In Proceedings of the IEEE Visualization Conference, pp.37–44, 2000 Speaker: Wen-Ping Chuang Adviser: Ku-Yaw Chang 2015/11/281
Outline Introduction Segmentation approach Result Conclusion 2015/11/282
Introduction (1/6) Virtual screening techniques Volume rendering techniques have grown rapidly Interactive frame rates generate accurate results Organs have complex structures Segmentation plays a very important role 2015/11/283
Introduction (2/6) Segmentation Simple threshold Get complicated due to partial volume effect Cause unwanted and non-existing surfaces Combine the threshold and flood-fill techniques Flexible Segmentation rays Volumetric contrast enhancement 2015/11/284
Introduction (3/6) Polyps Potentially cancerous More than 5 mm Consider potentially malignant Need to be removed 2015/11/285
Introduction (4/6) Physical colon cleansing Large amounts of liquids Medications Enemas 2015/11/286
Introduction (5/6) A friendly virtual colonoscopy system Bypass the colon physical cleansing Need for segmenting the residual material Give a clean colon to the rendering algorithm 2015/11/287
Introduction (6/6) A new bowel preparation scheme Enhance the stool and fluid densities Take and reconstruct into a 3D dataset Partial volume effect Have not a clear boundary Worsen situation Finite resolution Low contrast 2015/11/288
Outline Introduction Segmentation approach Result Conclusion 2015/11/289
Segmentation approach Threshold Morphological operations Proposed approach 2015/11/2810
Threshold Human abdomen High density materials Bone Fluid Stool Soft tissue Air 2015/11/2811
Threshold Disadvantages Not remove PVE voxels Sensitive for each range of intensities Gives rise to aliasing effects at the inner colon boundary 2015/11/2812 Fig.1Fig.2Fig.3
Segmentation approach Threshold Morphological operations Proposed approach 2015/11/2813
Morphological operations Succession operation Such as dilation and erosion Flood-fill on all the fluid and stool regions A sequence of dilates and erodes to remove the PVE voxels 2015/11/2814
Morphological operations Dilation The dilation of A by B B is the structuring element 2015/11/2815
Morphological operations Erosion The Erosion of A by B B is the structuring element 2015/11/2816
+ + Morphological operations Highly twisted affect the inner contour of the colon Dilate followed by erode Can fill in holes Erode followed by dilate Can remove noise 2015/11/2817
Morphological operations Disadvantages Task considering the large number of such regions Require a lot of human intervention Slow down the entire process of segmentation Result in some fluid/stool regions being ignored 2015/11/2818
Segmentation approach Threshold Morphological operations Proposed approach 2015/11/2819
Proposed approach Approximate intensity based classification Classify the intensity values in the histogram Depend on the number and type of district regions Region boundaries Define by approximate thresholds Flexible Unique intensity profiles at different intersections Study and store 2015/11/2820
Proposed approach Approximate intensity based classification 2015/11/2821
Proposed approach Region growing Detect and mark the interior AIR region A smooth horizontal surface due to gravity Take a seed point to mark all the air voxels Reach no longer belong the air voxels 2015/11/2822
Proposed approach Selecting starting points for segmentation rays Critical to the overall speed of the algorithm Select fewer the voxels get faster the algorithm Assign the boundary voxels are simplest and fastest 2015/11/2823
Proposed approach Detecting intersections using segmentation rays Critical to the detection of the polyps Remove most of the PVE voxels Give an improved colon contour 2015/11/2824
Proposed approach Segmentation rays From each of the AIR boundary voxel 26-connected-neighbor directions Stop and ignore Not find any intersection after traversing a certain distance 2015/11/2825 +
Proposed approach Volumetric contrast enhancement A programmed transfer function Unwanted materials are removed Similar to contrast enhancement A smooth transfer function Get no-aliasing boundaries Improve the quality of volume rendering 2015/11/2826
Outline Introduction Segmentation approach Result Conclusion 2015/11/2827
Result Virtual colonoscopy system Automatic Histogram classification Seed point detection A fully automatic solution Segmentation Digital colon cleansing 2015/11/2828
Result Crux of this paper algorithm Characterizing the intersections Accurate a result as a manual segmentation Not miss even a single intersection 2015/11/2829
Result A cross-section of the CT data showing colon 2015/11/2830 (L)(R)
Result Volume rendered images showing 2015/11/2831 (L)(R)
Outline Introduction Segmentation approach Result Conclusion 2015/11/2832
Conclusion Advantages Fast and accurate segmentation algorithm Remove the partial volume effect General algorithm Use by any application similar to virtual colonoscopy 2015/11/2833
Conclusion Future work Build an interactive segmentation system Pick intersection characteristics using a mouse Find a particular intersection assigning classification/reconstruction tasks to the rays Add visual feedback Render and display the segmented dataset 2015/11/2834
THE END Thank you for listening 2015/11/2835