Table 1 Test results of three models for automatically segmenting multiple BMs.

From: Automated segmentation of brain metastases in magnetic resonance imaging using deep learning in radiotherapy

Diameter (mm)

U-Net

U-Net Cascade

BUC-Net

\({P}_{12}\)

\({P}_{23}\)

\({P}_{31}\)

DSC

      

Binary-class (n = 96)

0.891 ± 0.021

0.894 ± 0.012

0.912 ± 0.022

0.226

0.038

0.019

Multi-class (n = 1152)

0.757 ± 0.016

0.782 ± 0.011

0.797 ± 0.015

0.001

0.397

0.001

 < 3 (n = 60, 5%)

0.571 ± 0.031

0.589 ± 0.059

0.641 ± 0.055

0.011

0.085

0.009

3–6 (n = 714, 46%)

0.667 ± 0.025

0.725 ± 0.018

0.734 ± 0.018

0.001

0.966

0.001

6–9 (n = 258, 27%)

0.842 ± 0.018

0.844 ± 0.014

0.848 ± 0.013

0.437

0.517

0.807

9–12 (n = 70, 9%)

0.893 ± 0.013

0.872 ± 0.018

0.894 ± 0.012

0.061

0.214

0.127

12–15 (n = 20, 6%)

0.933 ± 0.007

0.913 ± 0.012

0.923 ± 0.007

0.249

0.139

0.062

15–18 (n = 8, 2%)

0.931 ± 0.010

0.926 ± 0.010

0.936 ± 0.014

0.123

0.187

0.631

18–21 (n = 10, 1%)

0.932 ± 0.011

0.924 ± 0.012

0.943 ± 0.011

0.201

0.041

0.081

 > 21 (n = 12, 3%)

0.931 ± 0.012

0.931 ± 0.014

0.954 ± 0.012

0.014

0.002

0.031

HD95 [mm]

      

Binary-class (n = 96)

3.812 ± 1.070

3.441 ± 1.378

0.901 ± 0.110

0.381

0.144

0.022

Multi-class (n = 1152)

1.348 ± 0.091

1.171 ± 0.072

0.922 ± 0.041

0.164

0.057

0.001

 < 3 (n = 60, 5%)

1.886 ± 0.351

2.074 ± 0.554

0.924 ± 0.128

0.106

0.045

0.041

3–6 (n = 714, 46%)

1.336 ± 0.145

1.148 ± 0.119

0.824 ± 0.056

0.523

0.039

0.012

6–9 (n = 258, 27%)

0.980 ± 0.122

1.141 ± 0.161

0.941 ± 0.080

0.424

0.206

0.156

9–12 (n = 70, 9%)

1.027 ± 0.151

1.255 ± 0.204

0.904 ± 0.103

0.298

0.079

0.045

12–15 (n = 20, 6%)

1.844 ± 0.127

1.713 ± 0.745

0.860 ± 0.187

0.233

0.023

0.041

15–18 (n = 8, 2%)

1.710 ± 0.103

1.137 ± 0.432

0.950 ± 0.040

0.256

0.003

0.005

18–21 (n = 10, 1%)

0.925 ± 0.009

0.791 ± 0.062

0.639 ± 0.012

0.301

0.068

0.002

 > 21 (n = 12, 3%)

1.664 ± 0.611

1.813 ± 0.546

1.234 ± 0.549

0.232

0.007

0.029

ASD [mm]

      

Binary-class (n = 96)

0.918 ± 0.207

0.964 ± 0.417

0.332 ± 0.142

0.261

0.001

0.001

Multi-class (n = 1152)

0.326 ± 0.021

0.289 ± 0.015

0.210 ± 0.017

0.211

0.013

0.012

 < 3 (n = 60, 5%)

0.315 ± 0.185

0.438 ± 0.106

0.281 ± 0.064

0.587

0.116

0.073

3–6 (n = 714, 46%)

0.373 ± 0.057

0.273 ± 0.020

0.211 ± 0.024

0.079

0.107

0.296

6–9 (n = 258, 27%)

0.255 ± 0.026

0.238 ± 0.034

0.201 ± 0.043

0.623

0.005

0.003

9–12 (n = 70, 9%)

0.275 ± 0.035

0.304 ± 0.029

0.210 ± 0.028

0.249

0.431

0.343

12–15 (n = 20, 6%)

0.227 ± 0.017

0.240 ± 0.015

0.209 ± 0.007

0.44

0.041

0.279

15–18 (n = 8, 2%)

0.200 ± 0.031

0.239 ± 0.025

0.208 ± 0.011

0.111

0.086

0.128

18–21 (n = 10, 1%)

0.234 ± 0.007

0.273 ± 0.009

0.225 ± 0.010

0.303

0.029

0.056

 > 21 (n = 12, 3%)

0.371 ± 0.078

0.410 ± 0.068

0.230 ± 0.066

0.112

0.001

0.001

  1. \({P}_{12}\), \({P}_{23}\) and \({P}_{13}\): U-Net & U-Net Cascade, U-Net Cascade & BUC-Net, U-Net & BUC-Net.
  2. Significant values are in [bold].
Back to article page