Adaptive Thresholding

Following the filtering approaches described in sections 13.3.3 and 13.3.4, calcifications were segmented by either an adaptive thresholding approach or a Canny edge detector. The former method yielded better classification results so far with either the symmlet wavelet or the donut filter and this will be discussed here in more detail. Figures 13.13 and 13.14 show the results of the thresholding process applied on the filter outputs of Figs. 13.10 and 13.11.

To reduce FP signals in either output, a criterion was set on the minimum size of the segmented objects based on empirical observations of calcifications

*

I

Si 1 f

/ V

> ?

»

\

I"

*

_

_

_

Figure 13.13: Adaptive thresholding of the symmlet wavelet filter's output of Fig. 13.10. The true calcifications are isolated in addition to false signals generated by calcified arteries or tissue intersections that "look like" calcification structures and have similar spectral properties. The edge effects shown in Fig. 13.10 remain as white borders in this stage that can be removed at the expense of losing details in calcification morphology, size, and number particularly for very small calcifications.

and visibility limits reported for calcifications in mammography literature [55]. Specifically, segmented spots smaller than 4 pixels (0.0144 mm2) in area, of any configuration, were eliminated from the final segmentation step prior to shape analysis and classification.

Was this article helpful?

0 0

Post a comment