Intersubject Surface Mapping with Nonrigid Registration for Neurosurgery Vishal Majithia Presented in partial fulfillment of the requirements for the Degree.

Slides:



Advertisements
Similar presentations
Bayesian Belief Propagation
Advertisements

Surgical Planning Laboratory Brigham and Women’s Hospital Boston, Massachusetts USA a teaching affiliate of Harvard Medical School User Interfaces and.
Active Contours, Level Sets, and Image Segmentation
DR.RAVINDRA SRIVASTAVA, CONSULTANT NEUROSURGEON VIMHANS HOSPITAL.NEW DELHI.
Real-time, low-resource corridor reconstruction using a single consumer grade RGB camera is a powerful tool for allowing a fast, inexpensive solution to.
Dynamic Occlusion Analysis in Optical Flow Fields
Image Registration: Demons Algorithm JOJO
Minimally-Invasive Approach to Pelvic Osteolysis Srinivas Prasad, Ming Li, Nicholas Ramey Final Presentation May 10, 2001.
Iterative Relaxation of Constraints (IRC) Can’t solve originalCan solve relaxed PRMs sample randomly but… start goal C-obst difficult to sample points.
Snakes with Some Math.
On Constrained Optimization Approach To Object Segmentation Chia Han, Xun Wang, Feng Gao, Zhigang Peng, Xiaokun Li, Lei He, William Wee Artificial Intelligence.
Active Contour Models (Snakes)
Computer Vision Optical Flow
Motion Tracking. Image Processing and Computer Vision: 82 Introduction Finding how objects have moved in an image sequence Movement in space Movement.
Non-Rigid Registration. Why Non-Rigid Registration  In many applications a rigid transformation is sufficient. (Brain)  Other applications: Intra-subject:
Detecting and Tracking Moving Objects for Video Surveillance Isaac Cohen and Gerard Medioni University of Southern California.
CS CS 175 – Week 2 Processing Point Clouds Registration.
Computing motion between images
3D Augmented Reality for MRI-Guided Surgery Using Integral Videography Autostereoscopic Image Overlay Hongen Liao, Takashi Inomata, Ichiro Sakuma and Takeyoshi.
Yujun Guo Kent State University August PRESENTATION A Binarization Approach for CT-MR Registration Using Normalized Mutual Information.
MSc project Janneke Ansems Intensity and Feature Based 3D Rigid Registration of Pre- and Intra-Operative MR Brain Scans Committee: Prof. dr.
tomos = slice, graphein = to write
July 2010 Image Registration Techniques, Benchmarking, Strategy Surgical Planning Laboratory Center for Neurological Imaging July 2010 Lidwien Veugen Supervision.
A Survey of Medical Image Registration J.B.Maintz,M.A Viergever Medical Image Analysis,1998.
Robotic Needle End Effector for Integration with CT Scan Team Members: David Sun Xuan Truong Chris Willingham Advisor: Dr. Bradford Wood.
ORTHOPEDIC PRODUCT PORTFOLIO. KNEE NAVIGATION KNEE ARTHROPLASTY KNEE ARTHROPLASTY – THE CHALLENGES A lot of revisions need to be done in the first two.
Statistical Color Models (SCM) Kyungnam Kim. Contents Introduction Trivariate Gaussian model Chromaticity models –Fixed planar chromaticity models –Zhu.
Optical flow (motion vector) computation Course: Computer Graphics and Image Processing Semester:Fall 2002 Presenter:Nilesh Ghubade
Image Guided Surgery in Prostate Brachytherapy Rohit Saboo.
Feedback Control for Steering Needles Through 3D Deformable Tissue Using Helical Paths Kris Hauser, Ron Alterovitz, Nuttapon Chentanez, Allison Okamura,
Matching 3D Shapes Using 2D Conformal Representations Xianfeng Gu 1, Baba Vemuri 2 Computer and Information Science and Engineering, Gainesville, FL ,
06 - Boundary Models Overview Edge Tracking Active Contours Conclusion.
Quantitative Brain Structure Analysis on MR Images
Multimodal Interaction Dr. Mike Spann
Multi-Modal Quantitative Analysis of Pediatric Focal Epilepsy Andy Eow Medical Vision Group CSAIL, MIT.
Medical Image Analysis Image Registration Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.
A Fast and Accurate Tracking Algorithm of the Left Ventricle in 3D Echocardiography A Fast and Accurate Tracking Algorithm of the Left Ventricle in 3D.
Feature based deformable registration of neuroimages using interest point and feature selection Leonid Teverovskiy Center for Automated Learning and Discovery.
December 9, 2014Computer Vision Lecture 23: Motion Analysis 1 Now we will talk about… Motion Analysis.
A Segmentation Algorithm Using Dyadic Wavelet Transform and the Discrete Dynamic Contour Bernard Chiu University of Waterloo.
A New Method of Probability Density Estimation for Mutual Information Based Image Registration Ajit Rajwade, Arunava Banerjee, Anand Rangarajan. Dept.
Conclusions The success rate of proposed method is higher than that of traditional MI MI based on GVFI is robust to noise GVFI based on f1 performs better.
Motion Analysis using Optical flow CIS750 Presentation Student: Wan Wang Prof: Longin Jan Latecki Spring 2003 CIS Dept of Temple.
Optical Flow. Distribution of apparent velocities of movement of brightness pattern in an image.
Renaissance® Brain Module Sales Presentation. Renaissance Value in Brain Surgery 2 Improves patient care:  Small, frameless platform may improve patient.
2D Texture Synthesis Instructor: Yizhou Yu. Texture synthesis Goal: increase texture resolution yet keep local texture variation.
CS 641 Term project Level-set based segmentation algorithms Presented by- Karthik Alavala (under the guidance of Dr. Jundong Liu)
©2005 Surgical Planning Laboratory, ARR Slide 1 Prostate Image Processing Steven Haker, PhD.
Flexible Automatic Motion Blending with Registration Curves
Application: Multiresolution Curves Jyun-Ming Chen Spring 2001.
Image-Based 3-D Spinal Navigation Using Intra-Operative Fluoroscopic Registration R. Grzeszczuk, S. Chin, M. Murphy, R. Fahrig, H. Abbasi, D. Kim, J.R.
Implicit Active Shape Models for 3D Segmentation in MR Imaging M. Rousson 1, N. Paragio s 2, R. Deriche 1 1 Odyssée Lab., INRIA Sophia Antipolis, France.
1 Chapter-3 (Electric Potential) Electric Potential: The electrical state for which flow of charge between two charged bodies takes place is called electric.
Occlusion Tracking Using Logical Models Summary. A Variational Partial Differential Equations based model is used for tracking objects under occlusions.
CIVET seminar Presentation day: Presenter : Park, GilSoon.
Introduction to Medical Imaging Regis Introduction to Medical Imaging Registration Alexandre Kassel Course
MOTION Model. Road Map Motion Model Non Parametric Motion Field : Algorithms 1.Optical flow field estimation. 2.Block based motion estimation. 3.Pel –recursive.
A 2D/3D correspondence building method for reconstruction of a 3D bone surface model Longwei Fang
Rigid Needles, Steerable Needles, and Optimal Beam Algorithms Ovidiu Daescu Bio-Medical Computing Laboratory Department of Computer Science University.
Registration Techniques in head surgery
Deformation Modeling for Robust 3D Face Matching Xioguang Lu and Anil K. Jain Dept. of Computer Science & Engineering Michigan State University.
Robust Visual Motion Analysis: Piecewise-Smooth Optical Flow
Multi-modality image registration using mutual information based on gradient vector flow Yujun Guo May 1,2006.
Robotic Needle End Arm Effector for Integration With CT Scan
Computational Neuroanatomy for Dummies
Robotic Needle End Arm Effector for Integration With CT Scan
Combining Geometric- and View-Based Approaches for Articulated Pose Estimation David Demirdjian MIT Computer Science and Artificial Intelligence Laboratory.
Presenter: Francesco T Mangano DO FACS (Chief, Pediatric Neurosurgery)
Presenter: Francesco T Mangano DO FACS (Chief, Pediatric Neurosurgery)
MultiModality Registration using Hilbert-Schmidt Estimators
Presentation transcript:

Intersubject Surface Mapping with Nonrigid Registration for Neurosurgery Vishal Majithia Presented in partial fulfillment of the requirements for the Degree of Master of Science

Invasive Neurosurgery Historically, there have been risks associated with neurosurgery. The brain and spinal cord are sensitive organs. Therefore, neurosurgical procedures must be highly accurate and minimally invasive.

Stereotactic Neurosurgery Use of external device to direct surgical instruments to a specific target in the brain. Create a stereotactic space coordinate system within intracranial space. Ability to localize target, plan trajectory, identify entry point based on preoperative images. Allows for more procedures to penetrate deep regions of the brain safely.

Applications of Stereotactic Neurosurgery Intracranial –Craniotomy –Needle biopsy –Shunt placement –Deep brain stimulation –Draining of blood clots

Stereotactic Neurosurgery The Cosman Robert Wells (CRW) stereotactic system is currently used by the Department of Neurosurgery. Use: –Stereotactic ring is fixed to patient’s skull preceding CT image acquisition. –Target is localized and aligned with arc-center. –The surgeon can then choose a desired entry point. Advantages: Accurate, flexible, simple to use

Image-guided Neurosurgery Image-guided neurosurgery integrates preoperative imaging with the physical space of the patient in the operating room (OR) obtained from 3D fluoroscopy. StealthStation allows intraoperative planning. Tracking devices give the neurosurgeon real-time 3D imaging capabilities. Increased accuracy of navigation for spinal and intracranial neurosurgery.

Image-guided Neurosurgery

Applications of Image-guided Neurosurgery Spinal –Pedicle screw insertion

Frameless Image-guided Neurosurgery Problem: Current image-guidance and stereotactic technology can become obstructive for neurosurgeons in the OR. Challenge: Devise a guide which can be fixed to rigid anatomy and use stereotactic space, but without the large, bulky equipment currently used. Solution: Use rapid prototyping technology to design and create patient-specific frames and guides based on preoperative planning.

Frameless Image-guided Neurosurgery

Proposal The neurosurgeon manually designs each patient’s guidance frame. We would like to automate the procedure by having the surgeon design frames for a reference case. We apply deformation fields obtained from nonrigid registration to transfer frames between patients. We constrain our images to binary intensity. Solution should be fast and robust for clinical use.

Nonrigid Registration Nonrigid registration aligns images by deforming one image to match the other. The registration results in a transformation T, which maps the moving image M to the fixed image F. T(M) = F

Optical Flow An object moving between a set of images can be characterized by its velocity over the time the images are taken. Assume object maintains constant intensity over time. The optical flow, v, gives the movement of a single point:

Demons Registration Originally developed by Thirion, this algorithm employs local forces, or demons, which push the moving object to fit the fixed object. The force with which each demon pushes is based on the optical flow equation.

Demons Registration This is an iterative algorithm driven solely by local forces. Deformation field is smoothed after each iteration with a Gaussian smoothing function.

CRE Algorithm Developed by Wang and Vemuri, the Cumulative Residual Entropy (CRE) algorithm combines nonrigid registration with an automated segmentation algorithm. The solution is a minimization of the following energy function. is the contour in the moving image, is the contour in the fixed image, and the transform we want is T -1 (v).

CRE Algorithm The first term measures the segmentation functional which is an active contour model. The second term measures the quality of the registration. The registration algorithm used is B- spline deformation.

CRE Algorithm The energy function, C, is based on cumulative residual entropy (CRE), ε, a term similar to the entropy term in mutual information.

CRE Algorithm Cumulative residual entropy for an image with some intensity variation. The total entropy for this image is

CRE Algorithm Cumulative residual entropy for an iso-intensity image. The total entropy is

CRE Algorithm The third term connects the registration and segmentation components. This is done by calculating a total signed distance between the result of the segmentation (black) and the result of the registration (red).

Registration Test Cases We use test cases for which we can predict results. This allows us to measure the quality of accuracy in the registration as well as the mapping of surface points. Let S A be the number of voxels in the ideal mapping of a surface. Let S B be the number of voxels in the actual mapping of the surface with the obtained transform. Let S M be the number of voxels which overlap between the ideal and actual surface mappings. We can then define the quality of the surface mapping Q s as:

Cube Registration TT TT

Ideal patch (yellow) and moving patch (blue)

Results: Cube Registration Demons registration results completed in 64 minutes, 2500 iterations S A = 990, S B = 990, S M = 920, Q S = 0.929

Results: Cube Registration CRE Registration results completed in 87 minutes, 2209 iterations S A = 990, S B = 609, S M = 547, Q S = 0.339

Spinal Vertebral Body Registration: First Image Set Sheared (gray) and original (blue) spinal vertebral bodies (SVB)

Results: Sheared SVB Registration Demons registration results completed in 655 minutes, 3000 iterations S A = 600, S B = 600, S M = 558, Q S = 0.930

Results: Sheared SVB Registration CRE registration results completed in 492 minutes, 4724 iterations S A = 600, S B = 412, S M = 0, Q S = 0

Spinal Vertebral Body Registration: Second Image Set Stretched (gray) and original (blue) spinal vertebral bodies

Results: Stretched SVB Registration Demons registration results completed in 1106 minutes, 4000 iterations S F = 1800, S B = 1729

Results: Stretched SVB Registration CRE registration results completed in 623 minutes, 5580 iterations S F = 1800, S B = 1611

Discussion Accuracy and robustness –Demons algorithm gave good results for registering test images and mapping surfaces from one object to another. –CRE algorithm was unable to register any of the images very well. Speed –Demons algorithm was very slow to converge –CRE algorithm was faster to converge, but did not give good results –Neither algorithm gave good results in a reasonable amount of time

Discussion Multiresolution approach –Perform a crude registration at low resolutions. –Very fast computation times and converges more rapidly. –Loss of surface points due to aliasing during downsampling makes this method impractical for our application.

Discussion: Demons Registration The demons registration took too long for clinical applications. This may be caused by the binary nature of the images we are registering. Recall the definition of the velocity vectors which drive the registration. Points with zero velocity do not contribute to the transform.

Discussion: Demons Registration The above 2D example shows the active demons for a single iteration. The only driving force is these boundary voxels. The combination of few active effectors, constraints to maintain local deformations, and lack of any global, systematic search for a solution leads to very slow convergence.

Discussion: CRE Algorithm The CRE algorithm was unable to accurately register any of the image sets. A possible source of error may lie in the number of grid points. Increasing the number of grid points does increase the quality of the registration, but at the cost of increased runtime per iteration.

Conclusion Nonrigid registration has been shown to accurately map surface points for test cases. The results from test cases were unable to demonstrate clinical viability based on speed and robustness.

Future Work Recent methods have been proposed to speed up the Demons algorithm with the application of external forces. 1 Constraints on the deformation field obtained by the CRE algorithm may help this method be more stable at the expense of increased iterations. Conversion of iso-intensity volume to multi-intensity volume Testing with clinical images 1. Wang et al, “Validation of an accelerated demons algorithm for deformable image registration in radiation therapy,” Phys. Med. Biol. 50, (2005).

Acknowledgements Committee Members –Dr. Frank Bova –Dr. Samim Anghaie –Dr. David Gilland –Dr. William Friedman. Radiosurgery Lab –Dr. Didier Rajon –Barbara Garita –Atchar Sudhyadhom Computer Science –Dr. Baba Vemuri –Fei Wang