Skip to main content
. 2023 Aug 1;544:None. doi: 10.1016/j.neucom.2023.126295

Table 4.

Quantitative comparison of different synthesis-based and synthesis-free methods for whole glioma segmentation. # and denote separated and end-to-end image synthesis and segmentation respectively. denotes synthesis-free methods for the segmentation task. Bold font highlights the best values obtained by synthesis-based methods. Results with no significant difference from the upper bound are denoted by *, according to a paired t-test (p-value > 0.05).

Methods Dice (%) ASSD (mm) HD95 (mm)
#Pix2Pix [15] 84.67±7.93 2.26±2.09 9.12± 11.97
#PGAN [3] 85.79±7.73 1.90±1.29 7.21±7.53
PSCGAN [36] 86.45±8.04 1.99±1.59 8.62±9.62
Wang et al. [32] 83.77±8.48 2.41±1.70 9.97±11.26
UAGAN°[41] 81.55±2.96 2.53±0.29 not reported
UAGAN [41] 86.27±8.10 1.96±1.37 8.04±8.47
w/o FLAIR 84.54±9.62 2.32±2.12 10.75±15.86
Real FLAIR 87.49±8.51 1.57±1.71 5.69±4.62
Ours 87.55±7.62 1.53±0.85 5.67±4.92
Upper bound 88.65±10.01 1.50±1.89 5.54±4.69