Skip to main content

Table 3 Segmentation accuracy of the 3D U-Net algorithm in the testing set (n = 41)

From: Development and validation of the 3D U-Net algorithm for segmentation of pelvic lymph nodes on diffusion-weighted images

Metrics

All LNs in different scanners

All LNs

Suspicious LNs

Largest LNs

P value

3.0 T Achieva

3.0 T Discovery

3.0 T Interia

P value

All vs Suspicious

All vs Largest

Suspicious vs Largest

Dice

(95% CI)

0.79 ± 0.08

(0.73, 0.85)

0.75 ± 0.17

(0.67,0.81)

0.75 ± 0.17

(0.67, 0.83)

0.779

0.76 ± 0.15

(0.71, 0.81)

0.85 ± 0.13

(0.79, 0.91)

0.88 ± 0.15

(0.81, 0.94)

0.009

0.002

0.496

TPR

(95% CI)

0.77 ± 0.13

(0.68, 0.86)

0.73 ± 0.20

(0.66, 0.80)

0.77 ± 0.20

(0.64, 0.90)

0.813

0.76 ± 0.18

(0.69, 0.83)

0.82 ± 0.26

(0.78, 0.86)

0.89 ± 0.21

(0.80, 0.98)

0.017

0.012

0.045

PPV

(95% CI)

0.72 ± 0.09

(0.67,0.77)

0.74 ± 0.18

(0.69, 0.79)

0.75 ± 0.21

(0.66, 0.84)

0.712

0.73 ± 0.17

(0.66, 0.80)

0.80 ± 0.20

(0.72, 0.88)

0.83 ± 0.16

(0.76, 0.90)

0.014

0.005

0.394

VS

(95% CI)

0.81 ± 0.09

(0.77, 0.85)

0.83 ± 0.15

(0.80, 0.86)

0.84 ± 0.16

(0.78, 0.90)

0.453

0.82 ± 0.14

(0.78, 0.86)

0.86 ± 0.19

(0.79, 0.93)

0.88 ± 0.20

(0.81, 0.95)

0.046

0.013

0.576

HD(mm)

(95% CI)

3.46 ± 0.22

(3.31, 3.61)

3.27 ± 0.18

(2.38, 4.16)

3.51 ± 0.45

(3.21, 3.81)

0.213

3.41 ± 0.67

(2.97, 3.85)

2.56 ± 0.18

(1.69, 3.43)

2.02 ± 0.09

(1.30, 2.74)

0.039

0.021

0.521

AVD(mm)

(95% CI)

3.03 ± 0.07

(2.44, 3.62)

2.98 ± 0.05

(2.31, 3.65)

3.26 ± 0.15

(2.44, 4.08)

0.389

3.09 ± 0.17

(2.40, 3.78)

2.03 ± 0.27

(1.24, 2.82)

2.01 ± 0.07

(1.60,2.42)

0.027

0.031

0.865

MHD(mm)

(95% CI)

2.34 ± 0.14

(2.15, 2.53)

2.51 ± 0.08

(2.26, 2.76)

2.67 ± 0.16

(2.24, 3.10)

0.132

2.51 ± 0.35

(1.67, 3.35)

1.62 ± 0.15

(1.27, 1.97)

1.54 ± 0.12

(1.25, 3.37)

0.035

0.028

0.670

  1. Suspicious LNs indicates the LNs larger than 0.8 cm in the shortest diameter
  2. LN lymph node, PPV positive predictive value, TPR true positive rate, VS volumetric similarity, HD Hausdorff distance, AVD Average distance, MHD Mahalanobis distance
Back to article page

BMC Medical Imaging

ISSN: 1471-2342

Contact us