Research Article

Feature Selection Based on Machine Learning in MRIs for Hippocampal Segmentation

Table 7

Details of the features resulting by the forward selection method using Näive Bayes Classifier.

36 features
Forward selection
Haralick featuresHaar-like features Statistical features
Orientation Coordinate Mask size Type Mask size Entry

Contrast135 Y 3
Gradient5
Correlation 135 X 3
PositionCoordinates
Normalized gray levelValue
Correlation45 X 5
Gradient5
Correlation90 Y 9
Correlation 45 Y 7
Skewness7
Homogeneity90 X 9
Correlation 0 Y 5
Correlation 90 Z 5
Correlation45 X 3
Correlation 135 Z 9
Correlation 90 Y 5
Correlation 135 Z 5
Correlation 0 Z 7
Correlation 90 Z 7
Correlation 90 Z 9
Correlation 0 Y 3
Correlation 135 X 3
Correlation 0 Z 9
Template1
Skewness5
Correlation 90 Z 3
Correlation 45 X 5
Gradient 3
Template 2
Correlation45 X 9
Correlation 45 Y 5
Correlation 90 Y 7
Correlation 45 Z 5
Gradient 9
Homogeneity 0 Z 9
Correlation 0 Y 9

The asterisk indicates the entries also present in the list of SBE features. For Haralick features, the orientation in degrees, reference coordinate, and the size of the cubic mask used are reported. In case of Haar-like features, the entry value indicates the template type used (see Figure 2). For statistical/positional kind the size of the cubic mask used or the self-explained value is lister, depending on the specific feature type. In particular for gradients, the column named Entry indicates the segment of the reference diagonal as shown in Figure 3. All the features are listed in top-down order of their inclusion during the SFS procedure execution.