|
| Study | Database | Dim | Image size | Approach | AL | Running time | GS | Performance |
|
| Hu et al. [3] | 24 datasets from 8 subjects | 3D | 512 × 512, 3 mm thin | Iterative threshold, dynamic programing, morphological operations | A | 2-3 min on a 300 MHz processor, 512 MB RAM (512 × 512 × 120) | 229 manual traced images | RmsD = 0.54 mm (0.8 pixel) |
|
| Mendonca et al. [43] | 47 image radiographs | 2D | NA | Spatial edge detector | A | NA | 47 manual traced data | Sen. = 0.9225,
PPV = 0.968 |
|
| Yim et al. [8] | 10 subjects | 3D | 512 × 512, 0.75–2 mm thin | Region growing, connected component | A | 42.3 sec on a 2.5 GHz processor, 2.0 GB RAM (512 × 512 × 352) | 10 manual traced data | RmsD = 1.2 pixel |
|
| Sluimer et al. [30] | 26 scans | 3D | 512 × 512, 0.75–2.0 mm | Shape-based | A | 3 hr on a 2.8 GHz processor, 2.0 GB RAM (512 × 512 × 400) | 10 manual traced data, each 4 slice | OM = 0.8165,
AD = 1.48 mm,
HD = 13.45 mm |
|
| Campadelli et al. [42] | 487 image radiographs | 2D | 256 × 256 | Spatial edge detector | A | NA | 487 manual traced data | Sen. = 0.9174,
Spec. = 0.9584,
PPV = 0.9197,
Accu. = 0.9437 |
|
| Korfiatis et al. [44] | 23 scans | 3D | 512 × 512 | Wavelet edge detector | A | 3 min on a 2.8 GHz processor, 2 GB RAM (512 × 512 × 50) | 22 manual traced data | OM = 0.983,
AD = 0.77 mm,
RmsD = 0.52 mm |
|
| Gao et al. [13] | 8 subjects | 2D | 512 × 512 × 240 | Thresholding | A | 15–20 min on a 3.0 GHz processor, 1 GB RAM (512 × 512 × 240) | 8 manual traced datasets | DSC = 0.9946 |
|
| Silveira et al. [18] | 1 subject | 2D | 512 × 512, 1 mm thin | Deformable model | A | NA | NA | Qualitative assessment |
|
| Pu et al. [12] | 20 datasets | 2D | 512 × 512, 1.25 mm thin | Thresholding | A | 1 min on a 2.11 GHz processor, 2 GB RAM (512 × 512 × 540) | 20 manual traced datasets | FP/GT = 0.43%,
FN/GT = 1.63% |
|
| Shi et al. [22] | 247 image radiographs | 2D | 256 × 256 | Shape-based deformable model | A | 75 sec per image on a 3 GHz processor, 1 GB RAM (512 × 512) | 247 manual traced images | OM = 0.92,
AD = 1.78 pixel |
|
| El-Baz et al. [35, 36] | 10 image datasets | 3D | 512 × 512 × 182, 2.5 mm thin | Statistical MGRF model | A | 1.65 sec per image on a 3.2 GHz processor, 16 GB RAM | 1820 manual traced images | Accu. = 0.968 |
|
| Annangi et al. [19] | 1130 image radiographs | 2D | 128 × 128 and 256 × 256 | Shape-based deformable model | A | 7 sec per image on a 2.4 GHz processor | 1130 manully traced images | DSC = 0.88 |
|
| Kockelkorn et al. [32] | 22 scans | 3D | 0.9-1.0 mm | Prior training, statistical classifier | UI | 10 min | 12 manual traced data | OM = 0.96,
AD = 1.68 mm |
|
| Besbes and Paragios [28] | 247 image radiographs | 2D | 256 × 256, 1 mm thin | Shape-based | A | NA | 123 manual traced data | OM = 0.9474,
AD = 1.39 pixel |
|
| Sofka et al. [31] | 260 scans | 3D | 0.5–5.0 mm | Shape-based | A | NA | 68 manual traced data | SCD = 1.95 |
|
| Hua et al. [33] | 15 scans | 3D | 0.3–0.9 mm | Graph-search | A | 6 min on a 2.0 GHz processor, 32 GB RAM | 12 semiautomated traced data | HD = 13.3 pixel,
Sen. = 0.986,
Spec. = 0.995 |
|
| Sun et al. [26] | 30 scans | 3D | 512 × 512 × 424–642, 0.6–0.7 mm thin | Shape-based | A | 6 min per dataset on a NVIDIA Tesla C1060 processor
(240 thread), 4 GB RAM | 30 manually corrected traced data | DSC = 0.975,
AD = 0.84 mm,
SPD = 0.59 mm,
HD = 20.13 mm |
|
| Abdollahi et al. [39, 40] | 11 scans | 3D | 512 × 512 × 390, 2.5 mm thin | Statistical MGRF model | A | NA | 11 manual traced data | DSC = 0.960 |
|
