Skip to main content

Table 3 Semantic and instance segmentation metrics

From: LI-RADS-based hepatocellular carcinoma risk mapping using contrast-enhanced MRI and self-configuring deep learning

 

LR-3

LR-4

LR-5

LR-M

Semantic segmentation

Mean DSCa

  Tr

0.15 (0.23)

0.23 (0.29)

0.57 (0.32)

0.29 (0.33)

  Ts

0.16 (0.24)

0.21 (0.27)

0.58 (0.34)

0.15 (0.27)

  Ex

-

-

0.45 (0.40)

-

Median DSCb

  Tr

0.00 (0.00, 0.21)

0.02 (0.00, 0.41)

0.72 (0.34, 0.83)

0.07 (0.00, 0.59)

  Ts

0.00 (0.00, 0.31)

0.04 (0.00, 0.41)

0.76 (0.28, 0.85)

0.00 (0.00, 0.15)

  Ex

-

-

0.53 (0.00, 0.85)

-

CCCc

  Tr

0.04 (0.00, 0.13)

0.14 (0.06, 0.22)

0.86 (0.62, 0.97)

0.35 (0.11, 0.71)

  Ts

0.06 (0.00, 0.16)

0.07 (0.02, 0.18)

0.94 (0.83, 0.97)

0.08 (0.00, 0.33)

  Ex

-

-

0.53

-

Instance segmentation

(lesions ≥ 10 mm)

Mean DSCa

  Tr

0.56 (0.23)

0.61 (0.21)

0.63 (0.27)

0.59 (0.25)

  Ts

0.53 (0.24)

0.59 (0.23)

0.66 (0.26)

0.45 (0.31)

  Ex

-

-

0.68 (0.29)

-

Median DSCb

  Tr

0.61 (0.42, 0.73)

0.67 (0.52, 0.77)

0.74 (0.54, 0.82)

0.66 (0.44, 0.78)

  Ts

0.58 (0.35, 0.74)

0.66 (0.46, 0.76)

0.77 (0.58, 0.84)

0.52 (0.10, 0.70)

  Ex

-

-

0.84 (0.65, 0.87)

-

CCCc

  Tr

0.28 (0.08, 0.73)

0.54 (0.39, 0.73)

0.91 (0.78, 0.97)

0.89 (0.73, 0.97)

  Ts

0.05 (0.01, 0.26)

0.08 (0.02, 0.64)

0.93 (0.80, 0.97)

0.57 (0.20, 0.95)

  Ex

-

-

0.91

-

  1. In parentheses: astandard deviation, blower, upper quartile, clower, upper bound of 95% confidence interval. CCC concordance correlation coefficient, DSC Sørensen–Dice coefficient, LR-3, LR-4, LR-5, LR-M LI-RADS categories, mm millimeters, Tr training dataset, Ts internal test dataset, Ex external test dataset