Page 295 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

286 Chapter 10 Deep neural network in medical image processing

Figure 10.5 Types of segmentation.

interpretation) or obtained externally (e.g., a field boundary
map). Approaches for determining precision based on reference
data are divided into two main categories: geometric and
nongeometric.

4.2 Edge-based segmentation methods
Edge-based methods postulate a rapid change in the margin
between two regions in values of the neighboring pixels, such
as brightness, color, and texture (Adams and Bischof, 1994
[45]). Nonetheless, edge-based strategies are sensitive to noise
or images’ distortion and are prone to oversegmentation in
textured areas, thus making homogenous and contrasting objects
more successful (Janssen and Molenaar, 1995 [46]). Regional
methods are based on the assumption that adjacent pixels within
one region have similar values of intensity, color, and texture. This
leads to a group of algorithms classified as region-growing algo-
rithms (Fig. 10.6).