Segmentation of Pulmonary Fissures on Diagnostic CT Preliminary Experience

1
Segmentation of Pulmonary Fissures on Diagnostic
CT-- Preliminary Experience
Jingbin Wang, MEng, Margrit Betke, PhD Computer
Science Department, Boston University, U.S.A Jane
P. Ko, MD
Department of Radiology, New York University,
U.S.A
2
What is a fissure?

• A boundary between two lung lobes

axial slice
(1.25mm thickness)
axial slice
(5mm thickness, diagnostic CT)
3
Background Objective

• Previous work
• M. Kubo, N. Niki, et al., IEEE Trans. on Nucl
  Sci., 1999.
• Fixed threshold, edge detection, VanderBurg
  operator, 1mm slice thickness
• M. Kubo, N. Niki, et al., IEEE ICIP2000.
• Surface curvature analysis , 1mm slice thickness
• Zhang Li, et al, Proc. SPIE, Medical Imaging
  1999.
• Fuzzy logic, 3mm slice thickness
• Our objective
• Find a method for diagnostic CT (5mm-7mm)
• Potentially extend to a general solution

4
What are current challenges?

• Very small gradient changes

• Density varies

• Discontinuity exists

5
System Overview
6
Fissure Region Detection Thresholding Phase

• Histogram analysis
• Automatic threshold calculation
• P -tile method determines a threshold T (P 25)
• Thresholding result

7
Fissure Region Detection Marching-bar Method

• Initialization
• Parametric bar
• variable length (4-20mm)
• grows vertically
• marches horizontally
• accommodate some amount of
• white pixels.
• Termination of march when either end of bar hits
  the lung border
• Obtained Fissure ROIs

8
Fissure Detection Local Region Analysis

• Histogram statistics inside the fissure ROI
• Initial threshold calculation P -tile method
  determines a threshold T (P 50)

Fissure ROI in left lung
9
Fissure Detection Extraction of a Candidate
Fissure Path
10
Fissure Detection Determination of result
fissure path

• Generation of several candidate paths
• Different starting bars are used
• Line curvature evaluation
• Result fissure path longest path which most
  closely resembles a line

11
Preliminary Results - I
12
Preliminary Results - II
13
Evaluation
Average RMS error 1.2mm
14
Conclusion

• Contribution
• Detect fissures in diagnostic CT images
• Two-step strategy
• region first, then fissure
• Local region analysis
• A potential general solution
• for CT scan data of various resolutions
• Future work
• Improvement of marching-bar method for fissure
  region detection
• Extensive testing

15
The end
Jingbin Wang, MEng, Margrit Betke, PhD Computer
Science Department, Boston University, U.S.A Jane
P. Ko, MD
Department of Radiology, New York University,
U.S.A