Skip to main content

Table 4 Tumour segmentation results for the 3D U-Net with two and three input channels

From: A whole-body diffusion MRI normal atlas: development, evaluation and initial use

Architecture

2 input channels

3 input channels

Input data

WF, b = 900

WF, vcDWI

WF, b = 900, t-map

WF, b = 900, t-map × vcDWI

Dice, mean (SD) [%]

38.3 (32.5)

35.6 (32.5)

39.4 (32.9)

40.4 (32.4)

Dice, median (IQR) [%]

40.0 (74.0)

29.5 (70.5)

43.8 (74.8)

44.6 (69.0)

TP

104/2.6

114/2.9

122/3.1

135/3.4

FN

215/5.4

205/5.1

197/4.9

184/4.6

FP

55/1.4

71/1.8

59/1.5

62/1.6

Sensitivity [%]

32.6

35.7

38.2

42.3

Precision [%]

65.4

61.6

67.4

68.5

  1. The TP, FN and FP are given as total number of tumours/tumours per scan.