Skip to main content

Table 3 Ablation results of different inputs and weighted loss functions in the testing cohort

From: A coarse-to-fine cascade deep learning neural network for segmenting cerebral aneurysms in time-of-flight magnetic resonance angiography

Input type

Loss function

DSC

HD (mm)

VS

Single-channel

Initial loss

0.567 ± 0.205

6.433 ± 8.153

0.738 ± 0.217

Dual-channel

Initial loss

0.591±0.201

5.686±7.020

0.738±0.225

Initial loss

0.557 ± 0.197

9.002 ± 10.083

0.709 ± 0.225

Single-channel

WDL (β = 0.1)

0.590±0.194

5.085±5.787

0.741±0.213

WDL (β = 0.2)

0.528 ± 0.164

5.585 ± 8.480

0.674 ± 0.221

WDL (β = 0.3)

0.483 ± 0.162

6.194 ± 9.307

0.657 ± 0.241

WDL (β = 0.4)

0.363 ± 0.149

6.910 ± 5.076

0.554 ± 0.241

WDL (β = 0.5)

0.289 ± 0.113

9.052 ± 11.633

0.513 ± 0.245

WDL (β = 0.6)

0.232 ± 0.111

8.269 ± 5.708

0.508 ± 0.276

WDL (β = 0.7)

0.166 ± 0.101

9.453 ± 5.296

0.369 ± 0.269

WDL (β = 0.8)

0.053 ± 0.164

12.659 ± 5.978

0.202 ± 0.195

WDL (β = 0.9)

0.528 ± 0.164

17.377 ± 8.370

0.163 ± 0.213

WDL (β = 1.0)

0.003 ± 0.009

31.892 ± 5.294

0.455 ± 0.314