Table 4 Pretraining data scalability across datasets

From: Masked autoencoding, generalizable pretraining, and integrated experts for enhanced glioma segmentation

Pretraining volumes (%)

BraTS21

ISLES22

WMH

WMH OOD

43,505 (100%)

60.87

58.45

73.42

70.32

20,000 (46%)

59.87

57.31

72.86

69.45

10,000 (23%)

58.92

56.14

72.08

68.21

5000 (11%)

57.83

54.91

71.19

66.84

0 (Scratch, 0%)

58.28

55.40

72.35

67.57

  1. Performance in Dice (%).
  2. Bold values indicate the best-performing result within each comparison group (i.e., the highest Dice score or the optimal value for the corresponding metric under the same experimental setting).
Back to article page