EM Segmentation For Orthopaedic Applications

Objective:

• To utilize the Slicer3 Expectation Maximization Algorithm for segmentation of the phalanx bones of the hand.

Progress:

• We have utilized the Slicer2.7 EM Segmentation Module for segmentation of the phalanx bones
  1. Registration was performed outside of Slicer using ITK registration algorithms that are available in the IaFeMesh software. This includes Thin plate spline, B-Spline and rigid registration algorithms.
  2. Probability map information was created from a single subject used as the atlas image and filtered using a Gaussian filter.
• Initial evaluation has been performed
  • Reliability assessed on fourteen specimens that also had the index finger segmented, and two specimens that had the index, middle, ring and little fingers manually segmented
  • Validation assessed using the laser scanning of the index finger (proximal, middle, and distal) phalanx bones in five specimens
• Work is underway to develop a Slicer2.7 tutorial for this segmentation (File:Draft EM Segment Tutorial 9 12.pdf)

Results:

Relative Overlap with Manual Rater

Finger	Phalanx Segment	Average Relative Overlap
Index (14 subjects)	Proximal	0.87
	Medial	0.80
	Distal	0.70
Middle (2 subjects)	Proximal	0.79
	Medial	0.77
	Distal	0.71
Ring (2 subjects)	Proximal	0.76
	Medial	0.82
	Distal	0.80
Pinky (2 subjects)	Proximal	0.70
	Medial	0.73
	Distal	0.72

• Example of Validation Results is shown below in the figures

To Do:

• Update the the tutorial to support the Slicer3 Workflow
  • Tutorial will be developed by Austin Ramme - In progress
• Slicer3 Segmentation applied to the femur
  • Using BRAINSFit for mapping of probability information to the subject images

Manuscript:

• Ramme AJ, Devries N, Kallemyn NA, Magnotta VA, Grosland NM. Semi-automated Phalanx Bone Segmentation Using the Expectation Maximization Algorithm. J Digit Imaging. 2008.

Key Investigators:

• Iowa: Austin Ramme, Nicole Grosland, and Vincent Magnotta

Links:

• Musculoskeletal Imaging, Modeling and Experimentation (MIMX)

Figures:

Surface models from EM Segmentation generated via Slicer3

Results of the EM Segmentation for the Proximal Pahalanx

Validation of EM Segmentation for the phalanx bones of the hand. Distance of automated segmentation versus physical laser surface scanning