Table 4 Performance of WMH segmentation methods for different WMH lesion loads.

From: Performance of five automated white matter hyperintensity segmentation methods in a multicenter dataset

Method

Fazekas scale

WMH volume reference

WMH volume method

ΔWMH

|ΔWMH|

DSC

ICC

Cascade

1

4 ± 4

12 ± 6

8 ± 6

8 ± 6

0.24 ± 0.16

0.02 (−0.12; 0.27)

2

16 ± 10

18 ± 11

2 ± 12

10 ± 6

0.50 ± 0.15

0.31 (−0.16; 0.67)

3

73 ± 61

26 ± 18

−47 ± 62

49 ± 60

0.54 ± 0.22

0.13 (−0.23; 0.67)

kNN-TTP

1

4 ± 4

5 ± 4

0.4 ± 1

0.9 ± 0.6

0.64 ± 0.10

0.91 (0.67; 0.97)

2

16 ± 10

15 ± 9

−1 ± 3

3 ± 2

0.78 ± 0.06

0.96 (0.90; 0.99)

3

73 ± 61

56 ± 41

−17 ± 22

18 ± 21

0.82 ± 0.06

0.92 (0.62; 0.99)

Lesion TOADS

1

4 ± 4

18 ± 20

13 ± 21

13 ± 21

0.35 ± 0.21

0.11 (−0.13; 0.43)

2

16 ± 10

19 ± 11

3 ± 13

6 ± 12

0.61 ± 0.20

0.50 (0.08; 0.78)

3

73 ± 61

53 ± 37

−20 ± 24

22 ± 22

0.77 ± 0.06

0.90 (0.49; 0.98)

LST-LGA

1

4 ± 4

4 ± 5

−0.3 ± 2

2 ± 2

0.47 ± 0.12

0.76 (0.46; 0.91)

2

16 ± 10

15 ± 10

−0.4 ± 7

5 ± 5

0.61 ± 0.14

0.84 (0.63; 0.94)

3

73 ± 61

53 ± 17

−20 ± 48

31 ± 40

0.70 ± 0.08

0.68 (−0.11; 0.94)

LST-LPA

1

4 ± 4

5 ± 5

0.3 ± 3

2 ± 2

0.49 ± 0.13

0.76 (0.45; 0.91)

2

16 ± 10

14 ± 10

−2 ± 6

4 ± 4

0.64 ± 0.14

0.85 (0.60; 0.94)

3

73 ± 61

62 ± 39

−11 ± 23

16 ± 18

0.78 ± 0.07

0.90 (0.53; 0.98)

  1. Note: WMH, ΔWMH, |ΔWMH| and DSC are shown as means ± SD. ICC is shown as means (95% confidence interval).
  2. ΔWMH: mean difference in WMH volume (mL) between the reference segmentations and segmentations of the methods.
  3. |ΔWMH|: mean absolute difference in WMH volume (mL) between the reference segmentations and segmentations of the methods.
  4. DSC: dice similarity coefficient; ICC: intra-class correlation coefficient.
  5. Seventeen subjects had a Fazekas scale of 1, eighteen subjects had a Fazekas scale of 2 and seven subjects had a Fazekas scale of 3.
Back to article page