|
1
|
Brejl and Sonka [19]
|
2000
|
Manual contouring and landmarking.
|
Shape-variant Hough and edge-based object
|
–
|
MRI thorax
|
55 images (15 training)
|
1.8 ± 0.6 1.0 ± 0.3 1.8 ± 0.5
|
Mean error of approximate location mean error of accurate border detection
|
|
2
|
Tezmol et al. [20]
|
2002
|
Gaussian smoothed image and unsharp masking
|
Customize Hough transform
|
Cervical vertebrae
|
X-ray NHANES II
|
50 images
|
72.06/80 average 4.16 ∘
|
LMP falling in boundary box orientation error
|
|
3
|
Peng et al. [22]
|
2006
|
Model-based search method and intensity profiling polynomial function
|
Center point extended profiling canny edge
|
Intervertebral disc
|
MRI scans
|
5 Sets of images
|
94% successful
|
–
|
|
4
|
Lin [23]
|
2007
|
3D Bezier curves
|
Multilayer feed-forward, back-propagation artificial neural network King classification
|
Spine deformity
|
X-ray
|
37 images
|
Highest rn = 0.83 at 2 hidden layers highest rn = 0.75 at 1 hidden layer
|
Identification rate = correctly identified pattern / total validating patterns
|
|
5
|
Xu et al. [25]
|
2008
|
9 morphometric landmark-point corner guided
|
Dynamic programming, partial shape matching
|
Vertebral shapes
|
X-rays NHANES-II
|
900 images
|
Lowest precision of PSM is above 85%
|
Precision= TP / (TP + FP)
|
|
6
|
Tobias et al. [27]
|
2009
|
Vertebra coordinate system intensity information
|
Generalized Hough transform progressive adaptation method
|
Vertebrae segmentation
|
CT images
|
64 patients
|
1.12 ± 1.04 mm
|
Mean point-to-surface error
|
|
7
|
Ribeiro et al. [28]
|
2010
|
Manually delineated plateaus setting
|
180 gabor filter bank and ANN
|
Cervical
|
X-rays
|
41 images
|
0.91–0.92 high overlap success rate
|
Overlap between detected and manually delineated plateaus
|
|
8
|
Anitha and Prabhu [29]
|
2011
|
Anisotropic filtering
|
Gradient vector field Snake and Hough transform providing slope
|
Scoliosis
|
X-Ray
|
250 images
|
–
|
Intra-observer error is eliminated through true identification of the required end vertebras.
|
|
9
|
Larhmam et al. [30]
|
2012
|
Manual ROI histogram equalization canny and Sobel
|
Modified Hough, template matching, and contrast limited adaptive histogram equalization
|
Cervical vertebrae
|
X-ray NHANES-II
|
200 images
|
89%
|
Global accuracy
|
|
10
|
Sardjono et al. [31]
|
2013
|
Manual Cobb multiple X-ray stitched to get whole spine
|
Auto Cobb angle by charged-particle model (CPM) piece-wise linear curve fitting cubic spline and polynomial curve
|
Scoliosis
|
Frontal radiographs
|
36 images
|
R2 of 0.9124 and 0.9175
|
Mean absolute error
|
|
11
|
Rasoulian et al. [32]
|
2013
|
GMM and PCA
|
Expectation maximization, Gaussian filter, and canny
|
Vertebrae shape
|
CT scans
|
32 images
|
Distance error 1.38 ± 0.56
|
Mean point-to- surface error
|
