Biomedical Engineering Reference
In-Depth Information
ExpertKnowledge:
Groundtruth ψ ( x )
Image
I( x )
Plant : Process User Input, Update ψ
Level Set Evolution
Visualization
ψ ( x ,t )= C
User Repaint
ψ ( x ,t k ) →ψ ( x ,t k )
Evolve ψ : Segmentation
ψ t ( x ,t ) = G ( ψ,I )+ H ( ψ,U )
ψ ( x ,t )
+
u
U
ψ
Input Integrator
U ( x ,t ) = 0 u ( x )
u
Controller :
User Input
Fig. 4 Visualization feedback to the user allows him to supervise the automatic segmentation and
exploits his expert knowledge as prior information for the algorithm. This “human in the loop”
structure compensates for a poor initialization or suboptimal choice of segmentation energy (the
appropriateness of both is typically difficult to judge a priori) for a particular scene
3 Results and Analysis
3.1 Results of HSC Interactive Segmentor
Data in this study consists of approximately 300 high-resolution (512 512 200)
MRI scans of the knee that are being segmented to obtain 3D models of the
metaphysis and epiphysis. Models will undergo shape analysis and data will be
grouped according to the patient's age. Using statistical analysis, a model will
determine bone age and growth potential. An interactive segmentor must be
evaluated based on three criteria: the time it requires to segment a case, the
accuracy of the segmentation, and the ease of use. A typical segmentation based
on the method described in Sects. 2.2 and 2.3 result is shown in Fig. 5 .
Search WWH ::




Custom Search