Table 5 Validation results for two models from the DS and DNS pipelines, both using input images of size 512\(\times\)512\(\times\)64 and Z-Score pixel intensity normalization, with orientation standardization applied in the DS model.

From: The impact of pre-processing techniques on deep learning breast image segmentation

 

DSC

HD (in mm)

Model Name

Total

Breast

FGT

Vessels

Total

Breast

FGT

Vessels

DS 512 Orientation Z-Score

0.711 ± 0.019

0.926

0.832

0.374

111 ± 17

150

91

93

DNS 512 Z-Score

0.701 ± 0.018

0.923

0.834

0.346

106 ± 13

145

84

90

Terminology

Pipeline

Orientation

Pixel spacing

Pixel intensity

Resize

DS 512 Orientation Z-Score

DS

Yes

No

Z-Score

512 × 512 × 64

DNS 512 Z-Score

DNS

No

No

Z-Score

512 × 512 × 64

  1. The first part of the table reports the mean DSC ± standard deviation and HD ± standard deviation over 5-fold cross-validation on the Duke-Breast-Cancer-MRI dataset, showing both the overall averages (Total) and values for each label (Breast, FGT, Vessels). The second part of the table provides a legend describing the model names.
  2.  Bold values denote the best Total results.
Back to article page