200 likes | 340 Views
Automatic Centerline Extraction for Virtual Colonoscopy. 作者 : Ming Wan, Zhengrong Liang*, Qi Ke, Lichan Hong, Ingmar Bitter, and Arie Kaufman 出處 : IEEE Transaction on Medical Imaging, Dec. 2002, pp. 1450 - 1460 學生 : 林上智 指導老師 : 張顧耀. Outline. Introduction Requirements
E N D
Automatic Centerline Extractionfor Virtual Colonoscopy 作者: Ming Wan, Zhengrong Liang*, Qi Ke, Lichan Hong, Ingmar Bitter, and Arie Kaufman 出處: IEEE Transaction on Medical Imaging, Dec. 2002, pp. 1450 - 1460 學生:林上智 指導老師:張顧耀
Outline • Introduction • Requirements • Brief Review of Existing Algorithms for VC • Description of New Algorithm • Results • Conclusion
Introduction • Virtual endoscopy is an integration of • medical imaging • computer graphics • Advantages: • noninvasive • cost-effective • highly accurate
Requirements(1/2) • 1.Connectivity: • centerline is a sequence of directly connected voxels. • 6- , 18- , 26- connected • 2. Centricity: • centerline should stay away from the colon wall • 3. Singularity: • centerline should be a single path of one-voxel width
Requirements(2/2) • 4. Detectability : • branch area • 5. Automation: • fully automatic procedure • 6. Efficiency: • seconds on PC platform
Outline • Introduction • Requirements • Brief Review of Existing Algorithms for VC • Description of New Algorithm • Results • Conclusion
Brief Review of Existing Algorithms for VC • Manual Extraction: • manually mark the center of each colon region on each image • Topological Thinning: • peels off a volumetric object layer by layer
Outline • Introduction • Requirements • Brief Review of Existing Algorithms for VC • Description of New Algorithm • Results • Conclusion
DFB / DFS • DFS: • distance from a user-specified source point to each voxel • DFB: ( DFB-cost = 1/DFB ) • distance from each inside voxel to the nearest object boundary B DFB A DFS S
Description of New Algorithm • 1.Construction of a MST tree: • minimum-cost spanning tree • First: • converts the CT volume with DFB-distances to a 3D directed weighted graph. • Second: • builds up a MST tree from the weighted graph • Dijkstra’s shortest path technique. DFB-cost
Description of New Algorithm • 2. Extraction of Colon Centerline and Branches • does not specify the end point of the colon centerline • Find inside voxel with the maximum DFS-value
Modified Dijkstra Algorithm Current B Source DFS(C)
圖解 找DFB COST最小的點 也就是DFB最大的點 B1 B2 current B3 start 有相鄰26個點 B26
Branch detection algorithm(1/2) • Step1: • Scan the centerline by tracking back from • end point(E) to start point(S) • Step2: • For each centerline voxel C, check its 24 neighbors and find those voxel Bi • Pathlink pointing to C
Branch detection algorithm(2/2) • Setp3: • for each voxel Bi Record voxel C to be the closet centerline voxel Find the voxel with largest DFS-distance,Ti. Length of branch DFS(Ti) – DFS(C).
Results • Machine • PC platform • CPU :Intel Pentium 700-MHz processor • Memory: 655 MB • Data: • 44 human colon datasets.
Conclusion • Extend their centerline algorithm to study more complicated human organs with tree structures as airways and blood vessels.