Doktora Tezi Savunma Sunusu

1
Ahmet YARDIMCIDepartment of Biomedical
Equipment Technology, TBMYOAkdeniz University,
Kampus, 07059 Antalya, Turkiyee-mailyardimci_at_akd
eniz.edu.trwebwww.ahmetyardimci.com
2
Stroke I

• Stroke is the most common neurologic disease that
  leads to death and disability in the elderly
  population. Every year, a signi?cant number of
  stroke patients survive and are left with
  signi?cant disabilities. Hemiparesis is the most
  common cause of disability after stroke,
  affecting 70 85 of all patients, and it has
  been estimated that 60 of all surviving stroke
  patients may require rehabilitation treatment. It
  is important to identify effective stroke
  rehabilitation strategies as the number of stroke
  survivors and medical costs increase.

3
Stroke II

• Different treatment strategies for the
  rehabilitation of hemiplegic patients are
  available today, such as conventional exercise
  programs, proprioceptive neuromuscular
  facilitation techniques, muscle strengthening and
  physical conditioning programs, neurophysiologic
  approaches, and functional electrical
  stimulation.
• Rehabilitation techniques have been more
  successful in restoring function in the lower
  limbs than in the upper limbs. The assesment of
  rehabilitation period is very important to find a
  correct method for treatment process. The aim of
  this study find a new way to assesment of
  hemiplegic patients gate.

4
Study
Aim of the study Classification of Hemiplegic
Patients

• Methods
• Fuzzy Logic ?
• Neuro Fuzzy (ANFIS) ?
• NN ?

Stage 1. Discrimination of subject
situation Healthy? Patient?
Stage 2. Classification of patients Brunnstrom
stages III, IV, V, VI
5
Software

• Matlab V6.5, Mathworks
• FuzzyTECH V5.54d Professional edition, Inform

6
Data gathering
Information about Measurements
Subject 7 healthy elderly subjects and 26
hemiplegic patients Parameter Waist
acceleration Condition Walking on
corridor Instruction With orthosis and/or cane
(hemiplegic patients) Sampling 1024Hz
7
Statistics about subjects
8
What we do to reach our aims?

• Find significant amplitude features of signals,
• Find symmetry features of signals,
• Decide to inputs of classification system,
• Decide to rule blocks,
• Find suitable rules for all conditions ( consult
  a specialist),
• Test the system with your own data,
• Test the system with blind approach (find test
  data which is not included the your own data),
• Turn back if the system response does not satisfy
  you,
• Check all steps again from 3 to 8.

9
Three orthogonal acceleration signals from a
normal healthy subject walking at a normal speed
Evans AL, Duncan G, Gilchrist W. Recording
accelerations in body movements. Med. Biol. Eng.
Comput., 1991, 29, 102-104
10
Description of accelerometer signal
Bussmann JBJ, Damen L, Stam HJ. Analysis and
decomposition of signals obtained by thigh-fixed
uni-axial accelerometry during normal
walking.Med.Biol.Eng.Comput.,2000, 38, 632-638
11
Temporal events in stroke hemiparesis
Features

• Walking speed
• Stride period
• Cadence
• Stride length
• Stance period
• Swing Period
• Stance/swing ratio
• Double support
• Stance symmetry
• Swing symmetry

Sandra JO, Richards C.Hemiparetic gait
following stroke. Part I Characteristics. Gait
Posture 4 (1996) 136-148
12
Temporal events in stroke hemiparesis
Features

• Walking speed
• Stride period
• Cadence
• Stride length
• Stance period
• Swing Period
• Stance/swing ratio
• Double support
• Stance symmetry
• Swing symmetry

All of them temporal gait variables!..
Measurement and analysis of those variables did
not further characterize the pathologic nature of
locomotion in hemiplegic patients. Because most
of the relevant temporal information in
hemiplegic gaits is included in the measurement
of walking speed.
13
Some measurable features of gait

• Walking speed m/sn
• Cadence step/min
• Step length m
• Double step length m
• Step time difference
• (Mean step time Mean step length/ walking speed
  STD ?MSTL - MSTR? )
• Double step time difference
• (Double mean step time double mean step length/
  walking speed DSTD ?DMSTL - DMSTR? )

14
Some important notes from literature
Prior studies revealed that temporal variables of
hemiplegic gait, (walking speed and symmetry of
the swing phases) are significantly related to
motor recovery as classified according to defined
stages.
Hemiplegic patients, even those with good motor
recovery, by comparison all walked much more
slowly. Walking speed was related to the clinical
status of the patient, being progressively slower
as the motor deficit became more severe. Walking
speed is an important temporal variable of
hemiplegics gait, as reported by many
investigators.
There are several algorithms to compute step
times and quantifying symmetry. (Aminian et
al., Sadeghi et al., Robinson et al., Ganguli et
al., Vagenas et al.)
15
Symmetry and Laterality Quantification
Gait symmetry has been defined as a perfect
agreement between the actions of the lower limbs.
A way of categorizing different means of
determining whether or not symmetry and
laterality exist between the lower limbs using
indices and statistical analysis.
Sadeghi H, Allard P, Prince F, Labelle H,
Symmetry and Limb dominance in able-bodied gaita
review. Gait and Posture 12 (2000) 34-45
16
Symmetry and Laterality Quantification
In pathological gait, marked differences have
been noted between the affected and unaffected
limbs. Asymmetrical properties were reported for
34 gait variables in a group of 31 hemiplegic
subjects. The gait of hemiparetic patients was
characterized by slower velocity and more
asymmetry as they swayed more laterally on the
unaffected leg compared to healthy persons.
Sadeghi H, Allard P, Prince F, Labelle H,
Symmetry and Limb dominance in able-bodied gaita
review. Gait and Posture 12 (2000) 34-45
17
To determines asymmetries Symmetry Index (SI)
1
Robinson RO, Herzog W, Nigg BM. Use of force
platform variables to quantify the effects of
chiropractic manipulation on gait symmetry. J
Manipulative Physiol Ther 198710172-6
2
Ganguli S, Mizrahi J, Bose KS. Gait evaluation of
unilateral belowknee amputees fitted with
patellar-tendon-bearing prostheses. J Ind Med
Assoc 197463(8)256-9
R XR / XL
3
Vagenas G, Hoshizaki B. A multivariable analysis
of lower extremity kinematic asymmetry in
running. Int J Sports Biomech 1992 8(1)11-29
4
18
Lets see the differences between the SIs in a
sample problem
B
A
1
2
3
SI(A) -50 SI(B) -33
SI(A) 0,6 SI(B) 0,71
SI(A) -40 SI(B) -28
4
SI(A) 25 SI(B) 16
0 100
Symmetry
Asymmetry
19
Hemiplegic gate signals
Healthy ST6 ST5 ST4 ST3
Anteroposterior Acceleration signals
20
Feature of signal
Amplitude of signal A Step1 time S1 Step2
time S2 Slope of signal? SL Two steps
time T Absolute Step Difference ASD?S1-S2? R
ate of Step Difference RSDASD / T
21
Detection algorithm
After find the phl ptl , phr, ptr parameters
Duration of each gait cycle gc(i) phr(i1)
phr(i) 1? i ? N Left stance LS(i) pt1(i)
phi (i) right stance RS(i) ptr(i)
phr(i) Left double support LDS(i) ptl(i) -
phr(i) Right double support RDS ptr(i) -
phl(i)
Aminian K, Rezahhanlou K, Andres E, Fritsch C,
Leyvraz PF, Robert P. Temporal feature estimation
during walking using miniature accelerometers an
analysis of gait improvement after hip
arthroplasty. MedicalBiological Engineering
Computing, 1999 Vol.37,p.686-691
22
Anteroposterior Step times
Step time comparison
23
Anteroposterior signal ranges and mean values
indefinite
24
Anteroposterior signal ranges and mean values
25
Vertical acceleration signals
26
Lateral acceleration signals
27
Comparison of some features of vertical
acceleration signals (Range, max, min)
28
Comparison of some features of lateral
acceleration signals (Range, max, min)
29
Vertical and lateral acceleration signals mean
ranges
30
Fuzzy logic based classification
A N A L Y S I S
Temporal Features of Gait Signals
Symmetri Features of signals
Physiological Features of Subject
Amplitudes of signals
?
31
Preferred features of acceleration signals
32
System block diagram direct fuzzy classifier
33
System Structure
XAM
XAR
Signal Peak Features
81 rules
YVR
ZLR
Main Decision Rule Block
25 rules
XASI
9 rules
Signal Symmetry Features
XST
34
Fuzzy logic system diagram
35
Membership Functions of 1st Rule Block
MBF of XAM
MBF of XAR
MBF of ZLR
MBF of YVR
36
Membership Functions of 2nd and 3rd Rule Blocks
MBF of XASI
MBF of XAST
MBF of Classification
37
Rules
1st RB
81 rules
2nd RB
3rd RB
25 rules
38
Test (MoM)
39
Test Results
Healthy ST6 ST5 ST4 ST3
XAM,XAR,XASI,XAST,YVR,ZLR,Classification,__flags_
-0.2446,1.555,0.031,0.4183,1.211,1.0468,1,0 -0.157
,0.7305,0.1575,0.6081,1.0172,0.9236,1,0 -0.1128,0.
6394,0.101,0.6157,0.8952,0.6339,3,0 -0.047,0.5791,
0.2849,1.06,1.2105,0.7844,3,0 0.025,5846,0.5633,2.
39,0.8012,0.6479,5,0
Healthy Healthy ST3 ST3 ST5



ST4

40
Tests
1
2
41
Classification Accuracy Assesment
42
Classification Accuracy Assesment
43
PMCC (Pearson product-moment correlation
coefficient) results

• The Pearson coefficient is a statistic which
  estimates the correlation of the two given
  random variables. The linear equation that best
  describes the relationship between X and Y can be
  found by linear regression.
• This equation can be used to "predict" the value
  of one measurement from knowledge of the other.
  That is, for each value of X the equation
  calculates a value which is the best estimate of
  the values of Y corresponding the specific value.
  We denote this predicted variable by Y'.

Correlation coefficient is 0.85
44
Statistical Results
Successful discrimination rate of Patients
1
Successful discrimination rate of Healthy
Subjects
100
100
2
Successful classification rate of hemiplegic
patients
ST6? 66 ST5? 66 ST4? 66 ST3? 46
Good results for discrimination of subjects as
healthy and patient!
Low success for classification of ST3 patients.
This study has shown that it is possible to
discriminate subjects as healthy or patients with
fuzzy logic approach. But successful
classification of patients, due to the unstable
behaviors of signals, is rather difficult than
discrimination of subjects for fuzzy approach.
45
Future works

• Check the wrong results to find whether a failure
  in system.
• Check all the rules.
• If necessary do some fine arrangements on rules
  and membership functions.
• Expand the system by adding new inputs.

46
Future works
XAM
XAR
Signal Peak Features
81 rules
YVR
ZLR
Main Decision Rule Block
?
XASI
9 rules
Signal Symmetry Features
XST
Age
Height
Subjects Physiological Features
?
Weight
Gender
47
Future works

• Examine the literature on detection algorithms.
• Develop an algorithm for detect precise moments
  and compute temporal parameters.
• Make new measurements with using footswitch
  equipped shoe.
• Try the neuro-fuzzy methods to produce membership
  functions and rule base from the data records.
• Compare results with prior studies.

48
Thank you for your attention!