Introduction
Background
Related work
Contribution
Materials and methods
Image synthesis framework
Training and loss functions
Data
Patient data
Parameter | Resolution (x/y/z) [mm] | Windowing | Volumes | Volume size | Arms | Age | |
---|---|---|---|---|---|---|---|
CT patient | 100–130 kVp | 1/1/2 | [−1024, 1500] HU | 22 | \(512\times 512\times [52, 151]\) | No | 66 ± 9 |
CT phantom | 90–120 kVp | 1/1/2 | [−1024, 1500] HU | 56 | \(512\times 512\times [80, 124]\) | No | 51 ± 14 |
CBCT patient | 93–124.7 kVp | 0.486/0.486/0.486 | [−1024, 2000] HU | 24 | \(512\times 512\times 386\) | No | 67 ± 10 |
CBCT phantom | 90–120 kVp | 0.486/0.486/0.486 | [−1024, 2000] HU | 56 | \(512\times 512\times 386\) | No | 51 ± 14 |
MRI patient | 3 T | 1/1/3 | [10th , 90th] percentile | 24 | \(330\times 450\times [48, 93]\) | Yes | 67 ± 10 |
MRI phantom | 3 T | 1/1/3 | [10th , 90th] percentile | 56 | \(330\times 450\times [59, 88]\) | Yes | 51 ± 14 |
XCAT phantom data
Evaluation metrics
Synthetic versus XCAT
Synthetic versus real patient
Proof of principle registration evaluation
Results
Synthetic images
CBCT | CT | MRI | |
---|---|---|---|
SSIM | \(0.85 \pm 0.05\) | \(0.94 \pm 0.02\) | \(0.59 \pm 0.04\) |
FSIM | \(0.82 \pm 0.03\) | \(0.82 \pm 0.02\) | \(0.51 \pm 0.02\) |
EPR | \(0.47 \pm 0.06\) | \(0.43 \pm 0.04\) | \(0.40 \pm 0.03\) |
EGR | \(3.0 \pm 0.7\) | \(1.9 \pm 0.3\) | \(1.7 \pm 0.2\) |
MAE | \(109 \pm 14\) | \(51 \pm 16\) | \(37 \pm 6\) |
NCC | \(0.997 \pm 0.001\) | \(0.980 \pm 0.010\) | \(0.86 \pm 0.04\) |
NM (Synthetic) | \(52 \pm 13\) | \(39 \pm 5\) | \(25 \pm 3\) |
NM (Patient) | \(60 \pm 16\) | \(39 \pm 19\) | \(22 \pm 5\) |
HistCC | \(0.994 \pm 0.003\) | \(0.999 \pm 0.002\) | \(0.94 \pm 0.03\) |