Shape Analysis and Classification Feature Definition

According to the flowchart of Fig. 13.3, the steps following the detection and segmentation of the calcifications involve shape analysis of the segmented

Figure 13.14: Adaptive thresholding of the donut filter output of Fig. 13.11. As in Fig. 13.13, both true and false calcifications were isolated and outlined. No edge effects were generated in this case. Furthermore, more calcifications were preserved in the segmentation stage at the expense of a slightly higher number of false signals.

Figure 13.14: Adaptive thresholding of the donut filter output of Fig. 13.11. As in Fig. 13.13, both true and false calcifications were isolated and outlined. No edge effects were generated in this case. Furthermore, more calcifications were preserved in the segmentation stage at the expense of a slightly higher number of false signals.

objects and selection of the feature set to be used as input to the classifier. For this stage, we took advantage of prior art in the field of classification and our experience in mammographic features [56]. Our starting point was the implementation of the four shape features developed by Shen et al. [57] for individual calcifications and their modification to apply to calcification clusters. We expanded this initial set with two more shape descriptors of individual calcifications [20]. To represent the clusters, we added the standard deviations of the six shape descriptors and a distribution feature. To represent the patient and link the demographic data to the images, we added a demographic feature [58]. The final results was a set of fourteen features for cluster classification in mammography. Table 13.2 lists the selected feature set and the physical interpretation of each feature [59]. Specific definitions and details may be found in the listed references.

Table 13.2: Feature set selected from the shape analysis of the segmented individual calcifications and clusters and demographic dataa

Feature No. Feature Nature of feature

1 Age of the patient Demographic feature;

describes the patient

Individual calcification characteristics

2 Mean—Area of calcification Describes the morphology (shape)

3 Mean—Compactness

4 Mean—Moments

5 Mean—Fourier Descriptor (FD) Describes the margins

6 Mean—Eccentricity

8 Number of calcifications in cluster Regional descriptor;

(median of range) describes distribution

Cluster characteristics

9 SD—Area Describes the morphology

10 SD—Compactness

11 SD—Moments

12 SD—Fourier Descriptor Describes the margins

13 SD—Eccentricity

14 SD—Spread aFeatures are limited to morphological and distributional characteristics (with the exception of "age") in order to reproduce the visual analysis system and indirectly use the classification as a measure of segmentation.

0 0

Post a comment