1 Computer Aided Diagnosis System for Lumbar Spinal Stenosis Using X-ray Images Soontharee Koompairojn Kien A. Hua School of EECS University of Central.

1

Computer Aided Diagnosis System forComputer Aided Diagnosis System forLumbar Spinal Stenosis Lumbar Spinal Stenosis

Using X-ray ImagesUsing X-ray Images

Soontharee KoompairojnKien A. Hua

School of EECSUniversity of Central

Florida

Chutima Bhadrakom

Department of RadiologyThai Nakarin Hospital

Thailand

Outline

Background

Methodology Classifiers Construction Automatic diagnosis

Prototype

Experimental Studies

Conclusions2

Our Back

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

3

Spine

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

4

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

5

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

6

HoleHolesline up Tough

outershell

Jelly-likenucleus

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

7

Link between Brain & Body

8

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

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

9

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

10

More “white”matter

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

11

chronic pain, and loss of control over

some functions because communication with the brain is interrupted

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 sitting12

We focus on Lumbar Spine Stenosis

13

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

14

We Try to Detect These Conditions

Disc Space Narrowing

Abnormal Bony Growth (Posterior osteophytes)

Abnormality of FacetJoint (Posterior Apophyseal Arthropathy)

Vertibral Slippage (Spondylolisthesis)

15

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

16

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

17

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

18

Spondylolisthesis

A Vertebra is slipping off another

19

Summary

Disc Space Narrowing – bulging and herinated discs

Posterior osteophytes – bone spurs

Posterior Apophyseal Arthropathy – abnormal growth on facet joints

Spondylolisthesis – vertebral slippage

20

We detect these conditions using X ray

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

21

Imaging Technology

1. X-RAYS: These show (1) disc narrowing, (2) bone spurs (osteophytes), and (3) vertebrae slipping off another (spondylo-listhesis)

2. 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

3. 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.

22

Features

23

B: Mid vertebral height

B

A: Anterior vertebral height

A

C: Posterior vertebral height

C G,H: Anteroposterior (A-P) width of usual spinal canal

H

G

I,J: Anteroposterior (A-P) width of unusual spinal canal

I

JD,E,F: Intervertebral disc space height D E F

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

25

Boundary point

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

26

Initial position After 1 iteration After 2 iteration Convergence

Face modelBuilt from

400 images

Apply AAM to our Environment

1. A radiologist manually labels boundary points of training images

2. Apply the AAM technique to build a lumbar model (with boundary points)

3. Apply the lumbar model to determine the boundary points of the image under investigation

4. Measure the distances among the boundary points to obtain the feature values

27

Spine X-ray image

28

Result from AAM

posterior osteophyte(bone spur)

apophyseal arthopathy(growth on facet joint)

29

spondylolisthesis(vertebral slippage)

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

),...,|(* 1

#

1di

conditions

i

xxCpP Max

If P* > threshold spinal stenosis

Naïve Bayes Classifier (1)

• Prior Probability: Prior probabilities are based on previous experience

31

60

40

objects ofnumber Total

objectsGreen ofNumber GREENfor y probabilitPrior

60

20

objects ofnumber Total

objects RedofNumber for REDy probabilitPrior


• Likelihood: Likelyhood of X given Red/Green

32

40

1

casesGREEN ofnumber Total

of vicinity theinGREEN ofNumber GREEN given of Likelihood

XX

20

3

cases REDofnumber Total

of vicinity thein REDofNumber REDgiven of Likelihood

XX

X


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

33

GREENgiven of Likelihood GREEN ofy probabilitPrior

GREEN being ofy probabilitPosterior

X

X

Percentage of Green population

Percentage of Green inthe neighborhood X


34

60

1

40

1

6

4 GREENgiven of Likelihood GREEN ofy probabilitPrior

GREEN being ofy probabilitPosterior

X

X

20

1

20

3

6

2 REDgiven of d Likelihoo REDofy probabilitPrior

REDbeing ofy probabilitPosterior

X

X

We classify X as RED

Multiple Independent Variables

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

35

CxxxCxxxC jdijdijppp |,...,,,...,,|

22

Posterior probability of class membership, i.e., the probability that X belongs to Cj

Likelihood

d

kjkjdij CxCxxxC ppp

12

|,...,,|

Conditional probability of independentVariables are statistically independent Likelihood

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

36

d

kjkjdij CxCxxxC ppp

12

|,...,,|

)|()|( Max#

1

XpXp CC i

classes

im

X belongs to Cm

Automatic Stenosis Diagnosis

• Probability that X belongs to Cj

• Using Bayes’ rule above, we diagnose a new case X as follows:

37

d

kjkjdij CxCxxxC ppp

12

|,...,,|

)|()|( Max#

1

XpXp CC i

conditions

im

If p(Cm|X) > threshold spinal stenosis

System Architecture

38

FeatureExtraction

Training &learning process

Feature Vectors

Training interface

User interface

Imagesegmentation

Classification

FeatureExtraction

Result

X-ray training cases

New X-ray case

Classifier

Classifiers constructionAutomatic diagnosis

GUI for Classifier Construction

39

The user interface for managing training images and building lumbar spine classifiers

GUI for Stenosis Diagnosis

40

The user interface for submitting X-ray images for analysis of spinal conditions

Data Set for Experiments

41

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 databasecollected by the second National Health and Nutrition Examination Survey

Spinal ConditionsIntervertebral Disc Level

L2-L3 L3-L4 L4-L5 Total

Posterior Osteophyte 5 2 4 11

Posterior Apophyseal Arthorphathy 7 13 20 40

Disc Space Narrowing 30 33 35 98

Spondylooisthesis 1 0 1 2

Spinal Stenosis 12 15 24 51

Average Percentage of correct prediction of training images

42



Posterior Osteophyte 100.0 98.6 100.0 99.5

Posterior Apophyseal Arthorphathy 97.1 82.9 80.0 86.7

Disc Space Narrowing 84.3 87.1 80.0 83.8

Spondylooisthesis 100.0 100.0 100.0 100.0

Spinal Stenosis 100.0 95.7 97.1 97.6

Average Percentage of Correct Prediction of test images

43







Spinal Stenosis 79.7 75.0 68.8 74.5

Average Percentage of correct prediction using perfect labels

44

Better labeling improves performance







Spinal Stenosis 93.8 87.5 75.0 85.4

Conclusions

A fully automatic CAD system for lumbar spinal stenosis

Not dependent on user’s knowledge and experience

Accuracy from 75 – 80%

Good enough for screening and initial diagnosis

Suitable for general practitioners

45

Do You Know ?

Giraffes and human have SEVEN vertebrae in their necks

46

1 Computer Aided Diagnosis System for Lumbar Spinal Stenosis Using X-ray Images Soontharee Koompairojn Kien A. Hua School of EECS University of Central.

Documents

spinal nerves spinal

spinal column spinal

spinal canal

shorter spinal cord

cases of spinal stenosis

neck lumbar spinal stenosis

spine slide

pair of spinal nerves