Traumatic Brain Injury (TBI) Detection

Download Report

Transcript Traumatic Brain Injury (TBI) Detection

Final Presentation –Winter 2012
Date: 18.05.2013
Presenters:
Project Advisor:
Project Initiator:
Malihi Naveh, Fidelman Peli
Aides Amit
Dr. Nakhmani Arie
Motivation
 TBI caused by an acute event
 Severe damage to portions of the brain
 TBI may cause severe disabilities - cognitive
deflects, communication, mental health.
 1.7 million new cases of TBI in the U.S. each year
 50,000 deaths caused by TBI each year in the U.S.
 Need for automatic tools for TBI clinical practice
and patient monitoring
MRI Imaging
 Magnetic Resonance Imaging (MRI)
 A medical imaging technique used in radiology to
visualize internal structures of the body
 Good contrast between the different soft tissues of
the body
 MRI uses non-ionizing
radiation (unlike CT)
 Different types of MRI
scans: MP_RAGE,
FLAIR, T1-weighted etc.
 Expansive
Literature survey
 Main approaches for TBI detection:
 Population based atlases
 Symmetry based analysis
 We chose the symmetry based approach
 Doesn’t require statistical analysis of large data bases
 More robust – age & population independent
 Symmetry Axis detection:
 Preprocessing – brain segmentation3
 PCA – Principal Component Analysis1,5
 PSD – Phase Based Symmetry detection1
 Gravitational torque4
Literature survey
 Symmetry Analysis
 Gabor8,10
 Edge matching
 Flow vector8
 Energy medians comparison (boxing)11
 Methods fusion
 Active Contours12
[1] http://en.wikipedia.org/wiki/Magnetic_resonance_imaging#Other_specialized_MRI_techniques
[2] http://www.na-mic.org/Wiki/index.php/DBP3:UCLA#What_is_traumatic_brain_injury.3F
[3] L. Smith, A Tutorial on Principal Components Analysis, www.cs.otago.ac.nz/cosc453/student_tutorials/principal_components.pdf, 2002.
[4] Zhito Xiao and Hun Wu, “Analysis on Image Symmetry Detection Algorithms”, (FSKD 2007. V.4, pp 745-750.
[5] E Song ,et al, ” Symmetry analysis to detect pathological brain in MRI”, MIPPR 2007. Proc. of SPIE Vol. 6789,.67891F, (2007).
[6] Stiven Schwanz Dias, “Improved 2D Gabor filter”, Matlab Central – File exchange,
http://www.mathworks.com/matlabcentral/fileexchange/13776-improved-2d-gabor-filter.
[7] H. Khotanlou, O. Colliot, I. Bloch, “Automatic brain tumor segmentation using symmetry analysis and deformable models”, in: Internat. Conf. on Advances in
Pattern Recognition ICAPR, Kolkata, India, January 2007.
[8] Y. Sun, B. Bhanu, and S. Bhanu, “Automatic Symmetry-Integrated Brain Injury Detection in MRI Sequences”, Proc. IEEE CS Conf. Computer Vision and Pattern
Recognition Workshop, 2009.
[9] Valentina Pedoia, Elisabetta Binaghi, Sergio Balbi, Alessandro De Benedictis, Emanuele Monti and Renzo Minotto, "Glial brain tumor detection by using symmetry
analysis", Proc. SPIE 8314, Medical Imaging 2012: Image Processing, 831445 (February 23, 2012); doi:10.1117/12.910172; http://dx.doi.org/10.1117/12.910172.
[10] J. Movellan, “Tutorial on Gabor Filters”, technical report, MPLab Tutorials, Univ. of California, San Diego, 2005.
[11] N. Ray, R. Greiner and A. Murtha, “Using Symmetry to Detect Abnormalities in Brain MRI”, Computer Society of India Communications, 31(19), pp 7-10, 2008.
[12] Lior Deutch & Marina Kokotov, “Sobolov Active Countours without edges”, a Student project in course “Introduction to Medicl Imaging”, Spring 2010.
Proposed solution
PCA
Symmetry Axis
Symmetry affinity
Edge
Matching
Gabor
Thresholding
Empiric
Threshold
Clustering
Morphological
Contour
Active Contours
3D model
3D modeling
PCA
Symmetry
Axis
gray scale image, 2D calc -frame 1
Binary image, 2D calc. Err=7.65 deg- frame 1
gray scale image, 3D calc. Err=9.36 deg- frame 1
Symmetry
affinity
Thresholding
Clustering
 Without Brain segmentation
 Offset
 Improvement Needed
Contour
 Continued with manual Axis
3D modeling
Gabor
Symmetry
Axis
Symmetry
affinity
Thresholding
 Used As a BandPass
 Gaussian size -> resolution
e  S
2
R2
 0.5
 S 2 R 2  ln  0.5   0.7
Clustering
Contour
0.47 R 1cm
S
 S  0.47[cm 1 ]
R
 Circle shaped filter
 Direction Variant
3D modeling
 DC compensation
Gabor
Left hemisphere
Flipped Right hemisphere
Symmetry
Axis
Symmetry
affinity
1
 pixels 1 
20
1
 pixels 1 
F
40
S
Thresholding
Left
Clustering
Contour
3D modeling
Right
Diffrence
Actual solution
Manually
Symmetry Axis
Symmetry affinity
Edge
Matching
Gabor
Thresholding
Empiric
Threshold
Clustering
Morphological
Contour
Active Contours
3D model
3D modeling
Edge Detection
Symmetry
Axis
 Bilateral Symmetry
 Manual Axis
Symmetry
affinity
 Used: Canny Edges
Thresholding
Original Scan
Edge detection
50
Clustering
100
150
Contour
200
250
50
3D modeling
100
150
200
250
Edge Detection
Symmetry
Axis
 Edges Flipped on each other
 Later used: bwdist
Symmetry
affinity
Thresholding
Clustering
Contour
3D modeling
Left edge
Right edge
Combined edges:
Left, Right, Overlap - Red, Green,Blue
Masking and Clustering
Symmetry
Axis
Symmetry
affinity
 Skull removal
X X
 Elipse Shaped mask  a
 Removal of small objects
  Y  Yc 
 
 1
b
 

2
c
Thresholding
Distance between Edges
Clustering
Contour
3D modeling
Thresholding
Removal of small objects
2
Active Contour
Symmetry
Axis
 Use initial detection edges as initial
Snake.
Symmetry
affinity
Thresholding
Original Scan
50
100
Clustering
150
Contour
3D modeling
200
250
50
100
150
200
250
Active Contour
Symmetry
Axis
Symmetry
affinity
Thresholding
Clustering
Contour
3D modeling
 Get final Contour using an active
contour method.
 Sobolev Snake
Click to
See
Movie
Active Contour
Symmetry
Axis
Symmetry
affinity
 Get final Contour using an active
contour method.
 Sobolev Snake
Original Scan
Thresholding
50
Clustering
100
150
Contour
200
3D modeling
250
50
100
150
200
250
3D modeling
Symmetry
Axis
Symmetry
affinity
Thresholding
Clustering
Contour
3D modeling
 Use Snakes from different frames as
initial 3d data.
 Apply continuity conditions.
 Smoothing the data using 3d gaussian.
 Determine blood pool surface in 3d
space.
3D modeling
Symmetry
Axis
Symmetry
affinity
Thresholding
Clustering
Contour
3D modeling
 Plot Blood Pools Using Patch:
Conclusions
 A working Algorithm to detect Brain Blood Pools
was presented.
 Novelties in this work:
 Symmetry Affinity Based on Edge comparison
 3D continuity conditions