Introduction
Deformable appearance pyramids for object representation
Local feature pyramid
Anatomy decomposition by DAP
Deformable appearance pyramids fitting
Explicit model
Local feature searching
Shape regularisation
Implicit model
Metrics | AAM | ASM | CLM |
\(\hbox {Gauss}+\hbox {LK}\)* |
\(\hbox {Gauss}+\hbox {SDM}\)* |
\(\hbox {Wavelets}+\hbox {SDM}\)* |
---|---|---|---|---|---|---|
PtoBD |
\(3.10\pm 1.29\)
|
\(2.51\pm 1.32\)
|
\(2.34\pm 1.15\)
|
\(2.21\pm 1.07\)
|
\(1.95\pm 0.92 \)
| 1.87±0.73 |
DSC |
\(90.6\pm 4.9\)
|
\(92.1\pm 5.2\)
|
\(92.4\pm 5.2\)
|
\(92.8\pm 4.0\)
|
\(93.9\pm 3.3\)
| 94.7±2.6 |
Appearance reconstruction, pathology modelling and classification
Experiments
Results of landmark detection
Method | Accuracy (%) of classification | MAE of grading | RMSE of grading | ||||||
---|---|---|---|---|---|---|---|---|---|
L3/4 | L4/5 | L5/S1 | L3/4 | L4/5 | L5/S1 | L3/4 | L4/5 | L5/S1 | |
ASM |
\(79.1\pm 4.8\)
|
\(77.4\pm 4.3\)
|
\(81.7\pm 4.5\)
| 0.25 | 0.31 | 0.20 | 0.55 | 0.67 | 0.48 |
AAM |
\(70.1\pm 7.1\)
|
\(69.7\pm 7.3\)
|
\(71.3\pm 8.8\)
| 0.41 | 0.44 | 0.32 | 0.72 | 0.79 | 0.58 |
CLM |
\(81.0\pm 4.9\)
|
\(82.4\pm 4.5\)
|
\(82.7\pm 4.4\)
| 0.23 | 0.25 | 0.23 | 0.53 | 0.56 | 0.52 |
Gaussian DAP |
\(80.7\pm 4.9\)
|
\(82.1\pm 4.6\)
|
\(84.7\pm 4.2\)
| 0.23 | 0.25 | 0.18 | 0.53 | 0.58 | 0.47 |
Wavelet DAP |
\(\mathbf{84.7} \pm {\mathbf{4.6}}\)
|
\(\mathbf{84.5}\pm {\mathbf{4.3}}\)
|
\({\mathbf{85.9}}\pm \mathbf{4.2}\)
|
0.19
|
0.21
|
0.16
|
0.48
|
0.54
|
0.44
|
Results of anatomical classification
Anatomy | ASM | AAM | CLM | Gaussian DAP | Wavelet DAP |
---|---|---|---|---|---|
L3/4 |
\(83.3\pm 3.8\)
|
\(73.3\pm 5.5\)
|
\(83.1\pm 4.7\)
|
\(84.3\pm 4.1\)
|
\(\mathbf{85.0}\pm \mathbf{3.9}\)
|
L4/5 |
\(82.4\pm 4.6\)
|
\(76.2\pm 5.8\)
|
\(83.3\pm 4.3\)
|
\(86.9\pm 3.9\)
|
\(\mathbf{87.8}\pm \mathbf{3.5}\)
|
L5/S1 |
\(81.8\pm 4.7\)
|
\(74.5\pm 5.7\)
|
\(82.9\pm 4.5\)
|
\(85.2\pm 4.3\)
|
\(\mathbf{85.7}\pm \mathbf{4.3}\)
|