1 / 28

Fast Algorithm for Probabilistic Bone Edge Detection - Background Reading Presentation- FAPBED

Fast Algorithm for Probabilistic Bone Edge Detection - Background Reading Presentation- FAPBED. Team Members Danilo Scepanovic Josh Kirshtein. Advisors Ameet Jain Dr Russell Taylor. Problem. No current method to quickly segment CT images

filia
Download Presentation

Fast Algorithm for Probabilistic Bone Edge Detection - Background Reading Presentation- FAPBED

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Fast Algorithm for Probabilistic Bone Edge Detection- Background Reading Presentation-FAPBED Team Members Danilo Scepanovic Josh Kirshtein Advisors Ameet Jain Dr Russell Taylor

  2. Problem • No current method to quickly segment CT images • CT images contain little data that is relevant to any one process • How can we segment the bone quickly and efficiently for use in intra-operative registration?

  3. Goal • Given an arbitrary CT image, quickly and automatically segment bone edges • Assign each voxel a probability of being a bone surface point • Model the probabilistic surface • Final goal of registering US scans to the CT volume

  4. Alternate Approaches • Approaches: • Conventional thresholding • Semi-Auto Segmentation

  5. Alternate ApproachesThresholding W. Lorensen, H. Cline, “Marching Cubes: A High Resolution 3D Surface Construction Algorithm”, Computer Graphics. Vol 21, Num 4:163,169, 1987

  6. Alternate ApproachesThresholding • Marching Cubes – An application of thresholding • Use a cube that traverses CT volume, locking onto sets of 8 neighboring vertices • Evaluate pixel values at each vertex • Determine if an edge should be present

  7. Evaluate the vertices Vertex value = 1 if value ≥ threshold for surface 0 otherwise Determine edge pattern from voxel values Alternate ApproachesThresholding

  8. Advantages Intuitive Mostly accurate results Repeatable Disadvantages Computationally Intensive Speed-ups require lots of memory or sacrifice accuracy Image could be choppy and lose fine scale detail dependant on cube size Threshold value is set by user Alternate ApproachesThresholding

  9. Alternate ApproachesThresholding • B. Brendel, S. Winter, A. Rich, M. Stockheim, H. Ermert, “Registration of 3D CT and Ultrasound Datasets of the Spine using Bone Structures”, Computer Aided Surgery. Vol 7:146-155, 2002

  10. Alternate ApproachesThresholding • Simple thresholding to locate surface • Set a threshold value • From every voxel • Look at each neighbor’s density • Dependant on current voxel density and neighbor’s, create or do not create an edge • Connect all edge points into a surface mesh

  11. Alternate ApproachesThresholding • From the surface model • Estimate the portion of the surface visible with US and remove all other surface voxels “behind” • Estimate the US beam incidence angle to form probabilistic (estimated) surface • Use for US registration

  12. Advantages Simple algorithm Quick implementation Reproducible surface representation Good enough for some purposes Disadvantages Simple algorithm Sensitive to threshold Computationally intensive for large images Alternate ApproachesThresholding

  13. Alternate ApproachesThresholding • Summary of Thresholding • Very simple, rigid algorithm • Does not allow for probabilistic edge modeling

  14. Relevance • Relevance lies in its choice of threshold • How is the threshold chosen effectively? • ROC curves

  15. Alternate ApproachesAutomatic • Khonen Self-Organizing Maps (SOM) • A form of unsupervised clustering • Inspired by competitive, biological neural networks • Structure: 2 layered, densely connected neural network with topographical information • Associates input vectors with the closest cluster and adjusts weights • Utilized in [4,5]

  16. Alternate ApproachesKhonen Automatic • C. Aldasoro. “Image Segmentation with Kohnen Neural Network Self-Organising Maps”, Instituto Tecnologico Autonomo de Mexico. • C. Aldasoro, A. Aldeco. “Image Segmentation and Compression using Neural Networks”, Abstract, Instituto Tecologico Autonomo de Mexico.

  17. Alternate ApproachesAutomatic • SOM

  18. Alternate ApproachesAutomatic

  19. Alternate ApproachesAutomatic • [4] Method: • Input MRI slice of skull as 3D matrix: 2D for original voxel position, 1D for voxel intensity • Segment (here we go again…) the image • Use this matrix as input to the network and train initially randomized neurons until suitable value of is reached

  20. Alternate ApproachesAutomatic

  21. Alternate ApproachesAutomatic Threshold

  22. Alternate ApproachesAutomatic n=80

  23. Alternate ApproachesAutomatic

  24. Advantages Unsupervised Little information needed Yields good results Adaptable to different portions of any image based on threshold-segmented image Disadvantages Sensitive to threshold Computationally intensive Number of neurons greatly affects accuracy / execution time Alternate ApproachesKhonen Automatic

  25. Relevance • By the nature of neural networks, positions are probabilistic • Automatic, unsupervised method • Adaptable to many identifiable features in both CT and MRI (soft and hard tissues)

  26. New Methods to Come • Receiver operating characteristic – based thresholding • Rule based, probabilistic labeling method based on expected feature and surface characteristics

  27. 1) W. Lorensen, H. Cline, “Marching Cubes: A High Resolution 3D Surface Construction Algorithm”, Computer Graphics. Vol 21, Num 4:163,169, 1987 2) D. Muratore, J. Russ, B. Dawant, R. Galloway, “Three-Dimensional Image Registration of Phantom Vertebrae for Image-Guided Surgery: A Preliminary Study”, Computer Aided Surgery. Vol 7:342-352, 2002 3) B. Brendel, S. Winter, A. Rich, M. Stockheim, H. Ermert, “Registration of 3D CT and Ultrasound Datasets of the Spine using Bone Structures”, Computer Aided Surgery. Vol 7:146-155, 2002 4) C. Aldasoro. “Image Segmentation with Kohnen Neural Network Self-Organising Maps”, Instituto Tecnologico Autonomo de Mexico. 5) C. Aldasoro, A. Aldeco. “Image Segmentation and Compression using Neural Networks”, Abstract, Instituto Tecologico Autonomo de Mexico.

  28. Questions? Many thanks to: Ameet Jain Ofri Sadowsky Dr Taylor

More Related