Feature Extraction and PCA

Following feature extraction of the 316-dimensional feature vector for each sample, PCA is used to map the sample data from a higher dimension to that of

Table 11.15: Sensitivity and average number of false-positive regions per image over all 50 abnormal mammograms

After segmentation After false-positive After segmentation prefiltering reduction

Table 11.15: Sensitivity and average number of false-positive regions per image over all 50 abnormal mammograms

After segmentation After false-positive After segmentation prefiltering reduction

Breast type

Sensitivity

FP/i

Sensitivity

FP/i

Sensitivity

FP/i

1

0.57

163.31

0.51

9.31

0.40

3.26

2

0.58

132.09

0.56

8.26

0.48

4.40

3

0.72

132.55

0.70

6.90

0.66

4.14

4

0.66

158.79

0.64

10.27

0.60

3.56

Mean

0.63

146.69

0.60

8.68

0.54

3.84

After expert segmentation with WGMM'Mrf combined using AWM; after region prefiltering using Tarea — 122; after false-positive (FP) reduction using classifier operating point, by breast type.

After expert segmentation with WGMM'Mrf combined using AWM; after region prefiltering using Tarea — 122; after false-positive (FP) reduction using classifier operating point, by breast type.

a lower one. Using the unbiased PCA strategy described above only eigenvalues > 1.0 are considered, resulting in a 37-dimensional feature vector.

0 0

Post a comment