Skip to main content
. 2022 Jul 18;49(9):6019–6054. doi: 10.1002/mp.15840

TABLE 2.

Summary of common image and dose based similarity metrics

Metric Formula
Image similarity MAE/ME ↓ 1ni=1n|CBCTiCTi|/1ni=1n(CBCTiCTi)
MSE/RMSE ↓ 1ni=1n(CBCTiCTi)2/MSE
PSNR ↑
20log10MAXCTRMSE
SSIM ↑
2μCBCTμCT+c12σCBCT,CT+c2μCBCT2+μCT2+c1σCBCT2+σCT2+c2

with μx=mean; σx2=variance; σx,y=covariance; c1=(k1L)2; c2=(k2L)2; L=luminance; k1=0.01; k2=0.03

DICE ↑
2|AreaCBCTAreaGroundTruth||AreaCBCT|+|AreaGroundTruth|

with =intersection

Dosimetric similarity DPR ↑

Fraction of voxels where DD ⩽ x%

with DD=DCBCTDCTDCTΔ100 where D=dose

DVH ↑ Cumulative histogram of dose–volume frequency distribution for a given volume
GPR ↑ Fraction of voxels where γ1

Note: Arrows indicate better result.

Abbreviations: DPR, dose difference pass rate; DVH, dose–volume histogram; GPR, gamma pass rate; MAE, mean absolute error; ME, mean error; MSE, mean squared error; PSNR, peak signal‐to‐noise ratio; RMSE, root mean square error; SSIM, structural similarity.