Image Processing Reference
In-Depth Information
9.1
Introduction
Reliable image segmentation is an important yet challenging task in medical im-
age analysis [16]. However, an entirely autonomous method can vary significantly
in accuracy. Even the ground truth segmentation may differ between physicians.
Therefore, interactive segmentation methods have increased in popularity, such as
the intelligent scissors [23] [22] and the classical region growing method [13]. An
improved region growing method has been proposed for lung nodule analysis [7],
which combines fuzzy connectivity, distance and intensity information as the grow-
ing mechanism and peripheral contrast as the halting criterion.
In addition, watershed algorithm [18], fast level set [28] and graph cut method
[1-3] for the segmentation of volumetric data set, can provide fast and accurate
excellent results in medical applications. Figure 9.1 demonstrates a flowchart of in-
teractive segmentation in a surgery system.
Fig. 9.1
Flowchart of interactive Segmentation
Additionally, methods that allow the use of multiple label classifications include
Growcut [27] with intensity based transition CA rules for image segmentation, and
Seeded ND Cellular Automata [15] are popular due to the robustness. The Seeded
Search WWH ::
Custom Search