Skip to main content

Table 2 Segmentation evaluations and comparisons

From: ID-Seg: an infant deep learning-based segmentation framework to improve limbic structure estimates

Dataset

Metric

Region

Segmentation method

F

p.adj

  

dHCP

ID-Seg without pre-training

ID-Seg with pre-training

  

Internal dataset: ECHO-Dataset1

DSC

L Amyg

0.79 (0.12)

0.76(0.07)

0.86(0.03)

8.81

0.001

(n = 20)

 

L Hippo

0.76 (0.09)

0.75 (0.08)

0.87 (0.03)

14.10

 < 10−4

  

R Amy

0.77 (0.14)

0.76 (0.08)

0.86 (0.03)

6.31

0.003

  

R Hippo

0.74 (0.15)

0.76 (0.08)

0.87 (0.03)

10.67

 < 10−4

 

ICC

L Amyg

0.87 (0.09)

0.86 (0.05)

0.92 (0.02)

3.41

0.05

  

L Hippo

0.86 (0.08)

0.85 (0.06)

0.93 (0.02)

4.75

0.024

  

R Amyg

0.86 (0.13)

0.86 (0.05)

0.92 (0.02)

3.79

0.05

  

R Hippo

0.84 (0.13)

0.86 (0.05)

0.93 (0.01)

5.89

0.02

 

ASD

L Amyg

0.49 (0.34)

0.41 (0.09)

0.32 (0.11)

5.89

0.007

  

L Hippo

0.60 (0.37)

0.61 (0.59)

0.26 (0.11)

10.38

0.001

  

R Amyg

0.53 (0.47)

0.79 (1.3)

0.31 (0.09)

3.97

0.024

  

R Hippo

0.65 (0.43)

0.66 (0.57)

0.26 (0.11)

6.43

0.006

External dataset: M-CRIB

DSC

L Amyg

0.73 (0.02)

0.81 (0.03)

0.88 (0.02)

184.47

 < 10−4

(n = 10)

 

L Hippo

0.67 (0.04)

0.81 (0.03)

0.88 (0.03)

156.2

 < 10−4

  

R Amyg

0.67 (0.03)

0.83 (0.04)

0.87 (0.02)

68.78

 < 10−4

  

R Hippo

0.60 (0.04)

0.83 (0.03)

0.87 (0.03)

109.61

 < 10−4

 

ICC

L Amyg

0.86 (0.05)

0.90 (0.01)

0.93(0.01)

195.48

 < 10−4

  

L Hippo

0.83 (0.06)

0.91 (0.02)

0.94(0.01)

393.73

 < 10−4

  

R Amyg

0.83 (0.06)

0.90 (0.02)

0.93(0.01)

88.13

 < 10−4

  

R Hippo

0.78 (0.09)

0.91 (0.01)

0.93(0.02)

373.74

 < 10−4

 

ASD

L Amyg

0.94 (0.13)

0.55 (0.06)

0.36 (0.11)

190.95

 < 10−4

  

L Hippo

2.5 (0.29)

0.37 (0.17)

0.25 (0.06)

158.59

 < 10−4

  

R Amyg

1.1 (0.11)

0.45 (0.09)

0.41 (0.09)

70.13

 < 10–4

  

R Hippo

3.3 (0.44)

0.41 (0.18)

0.28 (0.06)

112.71

 < 10–4

  1. DSC Dice similarity coefficients, ICC intra-class correlation, ASD average surface distance, measured in mm. Segmentation metrics for each method are shown in mean (standard deviation). Higher DSC and ICC, and lower ASD indicate better segmentation accuracy. Amyg Amygdala, Hippo hippocampus; L left; R right