Article Figures & Data
Figures
- Fig 1.
Manual segmentation and 3D reconstruction on Slicer. A, manual labels. B, 3D reconstruction with a coronal image. C, Illustrations of the Dice score, Intersection-over-Union, and pixel accuracy.
- Fig 2.
Schematic of the network architecture.
- Fig 3.
Measurement of safe and Kambin triangles. A, Schematics of the Kambin triangle. B, Schematics of the safe triangle. C, Measurement of the Kambin triangle on a manually segmented image. D, Measurement of the safe triangle on automatically segmented images.
- Fig 4.
Automatic and manually labeled masks.
- Fig 5.
3D rendering of automatic masks and manually labeled masks of bones and nerves. A, 3D rendering of manual masks before preprocessing. B, 3D rendering of post-preprocessed masks. C, 3D rendering of automatically generated masks. D, Smoothed 3D rendering of automatically generated masks. E, 3D rendering of manual masks before preprocessing (arrow indicates the compressed dura). F, 3D rendering of post-preprocessed masks (arrow indicates the compressed dura). G, 3D rendering of automatically generated masks (arrow indicates the compressed dura). H, Smoothed 3D rendering of automatically generated masks (arrow indicates the compressed dura).
- Fig 6.
3D model–based viability assessment of a transforaminal epidural steroid injection. A, Inaccessible trajectory to the safe triangle on an axial CT slice. B, Minimal space of the safe triangle on the posterior 3D model. C, Accessible oblique trajectory to the safe triangle on the 3D model. D, Inaccessible trajectory to the Kambin triangle on axial CT slice. E, Accessible trajectory on the 3D model. F, Oblique trajectory-guided nonaxial CT plane.
Tables
Algorithm 1: Combined Algorithm Require: X: CT volume, shape = D × H × W Require: xi=X(Li), (i=1, .…, k): CT voxel patch Require: yi = M(xi): yi is the output of the last layer (softmax activation function) of the model M, yi has 1 more dimension than xi, and this dimension has 3 channels. Each channel refers to the probability of the corresponding voxel belonging to background or bone or nerve, respectively. 1) Initialize: Y ← 0 2) For xi ∈ X,(i = 1, …, k) do 3) Y(Li,:) + = yi 4) End for 5) S ← arg max(Y, axis = −1) (find the channel with the largest value in the last dimension) 6) Return S (the automatic mask) Note:—M indicates the model (network); L, location of the CT voxel patch x at the CT volumn X; Y, summed probability; max, maximum.
Structures Pixel Accuracy (%) IoU (%) Dice Score (%) Bones 94.05 ± 6.68 89.73 ± 4.32 94.54 ± 2.43 (82.0–99.9) (82.0–95.2) (90.1–97.5) Nerves 91.43 ± 3.48 82.71 ± 3.25 90.51 ± 1.94 (85.3–94.4) (76.3–87.4) (86.6–93.2) ↵a Data are means and percentages unless otherwise noted.
Structures Pixel Accuracy (%) IoU (%) Dice Score (%) Bones 99.62 ± 0.35 81.40 ± 11.33 89.34 ± 7.28 (99.3–99.9) (60.5–93.3) (75.42–96.5) Nerves 87.74 ± 4.82 80.64 ± 3.31 89.25 ± 2.00 (79.4–93.1) (75.5–82.9) (88.1–93.4) ↵a Data are means and percentages unless otherwise noted.
Area (mm2) Manual Images Automatic Images P Value Kambin triangle 37.80 ± 20.90 (15.11–87.51) 36.41 ± 19.27 (11.46–78.63) .302 Safe triangle 8.69 ± 2.24 (6.04–13.27) 8.56 ± 3.25 (3.18–17.91) .792 ↵a Data are means and percentages unless otherwise noted.
Intraclass Correlation Coefficient Test-Retest Reliability Interobserver Reliability 3D rendering of manual segmentation Kambin triangle 0.983 0.984 Safe triangle 0.881 0.922 3D rendering of automatic segmentation Kambin triangle 0.988 0.982 Safe triangle 0.977 0.959
