Computer Aided Diagnosis System for

1
Computer Aided Diagnosis System for Lumbar Spinal
Stenosis Using X-ray Images
Soontharee Koompairojn Kien A. Hua School of
EECS University of Central Florida
Chutima Bhadrakom Department of Radiology Thai
Nakarin Hospital Thailand
2
Outline

• Background
• Methodology
• Classifiers Construction
• Automatic diagnosis
• Prototype
• Experimental Studies
• Conclusions

3
Our Back

• Spine is made up of a series of vertebrae (bone)
  and disks (elastic tissue)

Spine
4
Facet Joints

• A joint is where two or more bones are joined
• Joints allow motion
• The joins in the spine are called Facet Joints
• Each vertebra has two set of facet joints. One
  pair faces upward and one downward
• Facet joints are hinge-like and link vertebrae
  together

5
Spine Anatomy

• First three sections of the spine
• Cervical Spine Neck C1 through C7
• Thoracic Spine Upper and mid back T1 through
  T12
• Lumbar Spine Lower back - L1 through L5

6
Spinal Cord

• Each vertebra has a hole through it
• These holes line up to form the spinal canal
• A large bundle of nerves called the spinal cord
  runs through the spinal canal

Jelly-like nucleus
Holes line up
Tough outer shell
Hole
7
Spinal Nerves

• Spinal cord has 31 segments and a pair of spinal
  nerves exits from each segment
• These nerves carry messages between the brain and
  the various parts of the body

8
Link between Brain Body

• Each segment of the spinal cord controls
  different parts of the body

9
Spinal Cord is Shorter

• Spinal cord is much shorter than the length of
  the spinal column
• Spinal cord extends down to only the last of the
  thoracic vertebrae
• Nerves that branch from the spinal cord from the
  lumbar level must run in the vertebral canal for
  a distance before they exit the vertebral column

10
Shape Size of Spinal Segments

• Nerve cell bodies are located in the gray
  matter
• Axons of the spinal cord are located in the
  white matter. They carry messages.
• Spinal segments closer to the brain have larger
  amount of white matter
• Because many axons go up to the brain from all
  levels of the spinal cord

More white matter
11
Spinal Stenosis

• Spinal stenosis is a progressive narrowing of the
  opening in the spinal canal, which places
  pressure on the spinal cord (nerve roots)
• Pressure on nerve roots causes

• chronic pain, and
• loss of control over some functions because
  communication with the brain is interrupted

12
Spinal Stenosis

• Cervical spinal stenosis Stenosis (narrowing)
  is located in the neck
• Lumbar Spinal Stenosis Stenosis is located on
  the lower part of the spinal cord
• 75 of cases of spinal stenosis occur in the low
  back (lumbar spine), and legs are affected
• Produce pain in the legs with walking, and the
  pain is relieved with sitting

13
We focus on Lumbar Spine Stenosis
14
Diagnosis

• Patients with lumbar spinal stenosis may feel
  pain, weekness, or numbness in the legs, calves
  or buttocks
• Other conditions can cause similar symptoms
• Spinal tumors
• Disorders of the blood flow (circulatory
  disorders)
• Spinal stenosis diagnosis is not easy

15
We Try to Detect These Conditions

• Disc Space Narrowing
• Abnormal Bony Growth (Posterior osteophytes)
• Abnormality of FacetJoint (Posterior Apophyseal
  Arthropathy)
• Vertibral Slippage (Spondylolisthesis)

16
Disc Space Narrowing

• As the spine gets older, the discs lose height as
  the materials in them dries out and shrinks
• Causing the middle part of vertebrae to push down
  resulting in bulging discs and herinated discs
• Bulging discs and herinated discs encroach into
  the canal to narrow it and hence producing
  stenosis

17
Posterior Apophyseal Arthropathy (abnormality of
facet joint)

• Disc space narrowing can also cause instability
  between vertebrae
• The body attempts to reduce the instability by
  trying to fuse around the bad disc
• The facet joints enlarge and the edges try to
  fuse together and hence producing stenosis

18
Osteophytes(abnormal bony outgrowth)

• Osteophyte - Small abnormal bony outgrowth (bone
  spurs)
• Anterior Osteophyte - Outgrowth at the front
  side of a vertebrae
• Posterior Osteophyte - Outgrowth in the back
  side of a vertebrae

19
Spondylolisthesis

• A Vertebra is slipping off another

20
Summary

• Disc Space Narrowing bulging and herinated
  discs
• Posterior osteophytes bone spurs
• Posterior Apophyseal Arthropathy abnormal
  growth on facet joints
• Spondylolisthesis vertebral slippage

We detect these conditions using X ray
21
Motivation

• Prior studies need manually determined boundary
  for each individual vertebra
• No computer-aided diagnosis (CAD) system for
  spinal stenosis
• Develop a fully automatic CAD for spinal stenosis
• Focus on X-rays as this is often the first test
  for spinal stenosis diagnosis

22
Imaging Technology

• X-RAYS These show (1) disc narrowing, (2) bone
  spurs (osteophytes), and (3) vertebrae slipping
  off another (spondylo-listhesis)
• CAT SCAN This is a computerized X ray that
  shows how much the diameter of the canal is
  reduced and how far out the discs are
• M.R.I. (Magnetic Resonance Imaging) It produces
  picture like the CAT scan but they are generated
  using a magnetic field (instead of radiation)
  not needed if the CAT scan shows the problems.

23
Features
24
Extracting feature
When a vertebra is normal, some of the boundary
points near the canal are at the same
location (e.g., points 4 11 vs. point 1)
A Anterior vertebral height
B Mid vertebral height
C Posterior vertebral height
D Anterior height of intervertebral disc space
E Mid height of intervertebral disc space
F Postrior height of intervertebral disc space
G Upper anteroposterior (A-P) width of usual
spinal canal
H Lower anteroposterior (A-P) width of usual
spinal canal
I Upper anteroposterior (A-P) width of unusual
spinal canal
J Lower anteroposterior (A-P) width of unusual
spinal canal
25
Feature Extraction

• Automatically determine the boundary points
• Using the Active Appearance Model (AAM) technique
• Measure the distances among the boundary points
  to extract the features

Boundary point
26
Active Appearance Model(morphable model)

• An AAM contains a statistical model of the
  appearance of the object of interest (e.g., face)
  which can generalize to almost any valid example
• The AAM can search for the structures from a
  displaced initial position

Initial position After 1 iteration
After 2 iteration Convergence
Face model Built from 400 images
27
Apply AAM to our Environment

• A radiologist manually labels boundary points of
  training images
• Apply the AAM technique to build a lumbar model
  (with boundary points)
• Apply the lumbar model to determine the boundary
  points of the image under investigation
• Measure the distances among the boundary points
  to obtain the feature values

28
Spine X-ray image
29
Result from AAM
posterior osteophyte (bone spur)
apophyseal arthopathy (growth on facet joint)
spondylolisthesis (vertebral slippage)
30
Predicting spinal conditions

• Bayesian framework is used to build a classifier
  for each spinal condition
• Choosing the most probable spinal condition given
  extracted features
• xi Extracted features
• Ci Spinal condition i
• P Posterior probability for each spinal
  condition
• P Highest posterior probability

If P gt threshold ? spinal stenosis
31
Naïve Bayes Classifier (1)

• Prior Probability Prior probabilities are based
  on previous experience

32
Naïve Bayes Classifier (2)

• Likelihood Likelyhood of X given Red/Green

33
Naïve Bayes Classifier (3)

• Posterior Probability combining the prior and
  the likelihood to form a posterior probability
  using Bayes rule

Percentage of Green in the neighborhood
Percentage of Green population
34
Naïve Bayes Classifier (4)
We classify X as RED
35
Multiple Independent Variables

• Posterior probability for the event Cj among a
  set of possible outcomes C C1, C2, , Cd)

Likelihood
Posterior probability of class membership, i.e.,
the probability that X belongs to Cj
Conditional probability of independent Variables
are statistically independent
Likelihood
36
Multiple Independent Variables

• Probability that X belongs to Cj
• Using Bayes rule above, we label a new case X
  with a class level Cm that achieves the highest
  posterior probability

? X belongs to Cm
37
Automatic Stenosis Diagnosis

• Probability that X belongs to Cj
• Using Bayes rule above, we diagnose a new case X
  as follows

If p(CmX) gt threshold ? spinal stenosis
38
System Architecture
Classifiers construction
Automatic diagnosis
39
GUI for Classifier Construction
The user interface for managing training images
and building lumbar spine classifiers
40
GUI for Stenosis Diagnosis
The user interface for submitting X-ray images
for analysis of spinal conditions
41
Data Set for Experiments
86 lumbar spine X-ray images from NHANES II
database 70 cases for training 16 cases for
testing
There are 17,000 spine X-ray images in the NHANES
II database collected by the second National
Health and Nutrition Examination Survey
42
Average Percentage of correct prediction of
training images
43
Average Percentage of Correct Prediction of test
images
44
Average Percentage of correct prediction using
perfect labels
Better labeling improves performance
45
Conclusions

• A fully automatic CAD system for lumbar spinal
  stenosis
• Not dependent on users knowledge and experience
• Accuracy from 75 80
• Good enough for screening and initial diagnosis
• Suitable for general practitioners

46
Do You Know ?

• Giraffes and human have SEVEN vertebrae in their
  necks