1. Joshua Kahn, Scott Wiese ECE533 – Fall 2003 December 12, 2003
Automated Determination of a Reference Point for the Diagnosis of Spinal Instability
Joshua Kahn, Scott Wiese
ECE533 – Fall 2003
December 12, 2003

2. Background Spinal Instability
Caused by degradation of spinal support tissues
Painful
Direct diagnosis not possible

3. Current Diagnosis Techniques
Radiograph Analysis
Uncomfortable
Inaccurate
Compresses 3 dimensional space into 2 dimensions
Clarity of radiographs is not always good
Surgical
Painful
Complicated

4. Proposed Technique
UW Neuroradiologist has proposed a new technique based on computed tomography
Rotate patient 12 degrees each way
Capture CT image set of each rotation
Use digital subtraction to measure rotation
Advantages
Three dimensional image set
More accurate than other techniques

5. Project Goals
Image subtraction software package requires a reference point for each image
Our goal is to perform automated selection of the reference point for each image

6. Proposed Algorithm DICOM to JPEG Conversion Sobel Edge Detection
Region of Interest Selection
Closing (dilation followed by erosion)
Black and White Edge Detection
Spinal Canal Extraction
Geometric Analysis

7. 1) Convert DICOM Image to JPEG
Performed using a UNIX utility (convert)
Command:
convert foo.dcm jpg:foo.jpg

8. 2) Sobel Edge Detection Detect all edges in image
Performed using built-in Matlab function

9. 3) Region of Interest Selection
Minimizes area of maximum computing
Does not analyze regions that are not important for algorithm

10. 4) Closing Minimize number of edges surrounding spinal canal
Dilation followed by erosion
Dilation
Erosion

11. 5) B&W Edge Detection Black and White edge detection Label objects
Extracts edges of closed image
Label objects
Each 8-way connected edge is a new object
Each object labeled with individual marker
Shown in different colors

12. 6) Spinal Canal Extraction
Sort objects based on size
Spinal canal always has the third largest area
Extract each pixel from this object

13. 7) Geometric Analysis Spinal canal is approximated by a triangle
Extract the three vertices of the triangle
Estimation of the reference point is midpoint of upper two vertices
Actual reference point is the intersection of the line connecting bottom vertex and estimation point with the spinal canal image

14. 7) Geometric Analysis

15. 8) Final Image Reference point for this image is detected
A white circle is drawn on point
The coordinates of the point are output

16. GUI Implementation

17. Results
Compared automated detection against expert-classified reference point
Average distance of 6.81 pixels
Average distance well within acceptable range

18. Conclusion
Automated detection will allow more accurate diagnosis of Spinal Instability
Proposed algorithm is comparable to expert-classification of reference point
Algorithm is also accurate and consistent in reference point determination