Skip to main content

Table 3 Pixel totals resulting from different under-sampling methods

From: Vessel segmentation for X-ray coronary angiography using ensemble methods with deep learning and filter-based features

Pixel count

after the method

% of positive class

% of negative class (the majority class)

Total count (training samples)

Original image

(exclude border)

5.559

94.441

25,002,241

Uniform

Under-Sampling

19.611

80.489

7,087,635

Unsupervised

Under-sampling

34.663

65.338

4,010,003

Tomek links

34.677

65.323

4,000,602

Cluster centroid

50.000

50.000

832,000