Justin Lange MA

muscleblouseΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

92 εμφανίσεις

Justin Lange M.A.
, Falk Zaumseil M.A.

Pressure distribution measurement at the diabetic
foot


Potentialities of multivariate analysis to
detect risky unroll patterns

World Congress


ORTHOPAEDIE + REHA
-
TECHNIK 2012

Exhibition Centre Leipzig, 18 May 2012

Justin Lange M.A.

Department Human
Locomotion

|
justin.lange@hsw.tu
-
chemnitz.de

2

Mobile systems for long
-
term pressure measurement
to detect harmful pressure distribution

Appropriate

sensor devices


Appropriate

bedding

/
attaching

of

sensors


Appropriate

data

processing



to

detect

harmfull

pressure
?


risky

unroll

pattern
?


Peripheral neuropathy is a frequent complication of diabetes shown
to affect gait (
Cavanagh

et al. 1993).



Patients with diabetes frequently exhibit a conservative gait
strategy where there is slower walking speed, wider base of
support, and prolonged double support time (
Petrofsky

et al. 2005).



These changes affect the foot

ground interface and are reflected in
plantar pressure parameters (
Wrobel

et al. 2010).

Pressure and ulceration


A
multifactorial

problem

3

Pressure and ulceration


A
multifactorial

problem

4

Conservative gait strategy

Advanced
Glycosylation

end products (AGEs)

affect the soft tissues of
the foot. (Skin, tendons,
joints, bones, collagen,
fat pads)

Limited
joint

mobility

Plantar
Pressure

Distribution /
Plantar
Pressure

Parameters

Consequences of Diabetes
and Neuropathy

Impact

Putting diabetic patients at risk of foot
ulcerations

Pressure and ulceration


A
multifactorial

problem

5

Plantar
pressure

/
unroll

pattern

Peak Pressures

Relative
Load

Distribution

Pressure
-
Time
-

Integrals

Shear

Forces

Pressure

Duration &
Frequency

Spatial

Peak
Pressure

Distribution

Mean

Pressure

Multivariate

statistical

methods

with

regard

to

plantar

pressure

distribution

parameters

could

be

useful

to

detect

critical

„unroll

patterns“

and

pressure

distri
-
bution
.

Pressure

evokes

foot

ulcerations

in

a

multifactorial

way

(
Wrobel

et

al
.

2010
)
.

Plantar

pressure

is

affected

multifactorial

by

gait

patterns,

footwear,

tissue

characteristics,

foot

deformities,

etc
.

(
Wrobel

et

al
.

2010
)
.

Plantar
pressure

/
unroll

pattern

6


Multivariate approaches to classify plantar pressure distribution data
into unroll pattern via cluster
-
analysis


(De Cock et al. 2006; Hughes et al. 1991)



Artificial neural networks to predict gait parameters and gait
patterns based on several characteristic variables (e.g. plantar
pressure)


(
Breit

and Whalen, 1997; Barton et al. 2007;
Rajendra

et al. 2007)

Multivariate approaches to classify pressure distribution data

Pattern recognition

Cluster analysis

Data mining / Classification

Artificial neural networks

Classification / Prediction / Learning

Benefit of pattern recognition for prevention of foot ulceration

7

Diverse pressure

data of
risk
groups

(e.g. diabetics with neuropathy
& risk of ulcerations
)

Diverse pressure data of
risk
-
free groups


(e.g. healthy subjects, subjects
with effective therapeutic
footwear)

Classification of unroll patterns with the aid
of e.g. Cluster
-
Analysis, Neural Networks

Plantar pressure
during daily gait
activity

(
PrinDASA



System)

Classification of measured
data to corresponding
group

“Intervention” in
case of risk
-
like
unroll pattern


60 subjects



40 individuals obese



20 of
them

had
diabetes and
polyneuropathy



Plantar pressure data was captured by a capacitive
pressure distribution platform during walking

Example: neural network to classify roll over patterns

8

Disease

N (
sex
)

Age

Height

Weight

BMI

Healthy

20
(12w|8m)

52,4
yrs


=12,67

170 cm


=10,2

67,8 kg


=10,04

23,2 kg/m²


=1,77

Obese

20
(12w|8m)

50,7
yrs


㴸ⰱ4

172 捭


㴹ⰵ

128ⰸ kg


㴲4ⰵ1

43ⰲ kg⽭²


㴸ⰳ8

佢敳e


P潬祮敵r潰athy

㈰2
⠱2睼w洩

53ⰷ
祲s


㴱1ⰵ4

172 捭


㴸ⰹ

130ⰵ kg


㴳4ⰵ9

44ⰰ kg⽭²


㴱1ⰲ1

9



Anatomical sub
-
areas were identified as
described by Maiwald et al. (2008)



Peak pressure (
pmax
) and relative force
time integral (
rfti
) were calculated for 6
sub
-
areas:



Heel,


Midfoot
,


Lateral
-
, middle
-
, medial
-
forefoot,


Hallux



Analysis of variance (
Kruskal

Wallis
-
Test)
&
pairwise

test (Mann
-
Whiteney

U)

Maiwald et al. 2008

Example:
neural network

to classify roll over patterns

10


‘Multilayer
perceptron


neural
network based on
rfti

and
pmax

data
of the sub
-
areas to map roll over
pattern of the groups



Two hidden layers



75% of subjects in training group



25% of subjects in holdout group

Example: neural network to classify roll over patterns

11

Example: neural network to classify roll over patterns


No significant differences between obese and obese diabetics with
neuropathy in any of the pressure parameters and foot regions



But 92.9% of subjects disease were correct predicted in the training
group and 79,3% in the holdout group!

The use of multivariate treatment of pressure data for
the prevention of foot ulceration

12


Feeding the ‘pressure measurement system’ with various / relevant
plantar pressure data of healthy and affected people


Classification of unroll patterns and pressure distribution images using
cluster analysis / neural networks


Continuous review of pressure data during the day using a suitable
measuring system based on suitable sensors (
PrinDASA
) and lasting
reconciliation with risk
-
like pressure patterns (from clusters / neural
networks)


Feedback as soon as a risk
-
like pressure distribution / unroll pattern is
achieved






(interesting picture / deleting background )


www.tu
-
chemnitz.de/PrinDASA

Thanks for your attention

Pressure and ulceration


A
multifactorial

problem

14

Plantar
Pressure

Distribution /
Plantar
Pressure

Parameters

Conservative gait strategy

Advanced
Glycosylation

end
products (AGEs)

affect the soft tissues of the
foot. (Skin, tendons, joints,
bones, collagen, fat pads)

Limited
joint

mobility

Initially, skin thickness decreases and skin hardness
increases; tendons thicken; muscles atrophy

and exhibit activation delays; bones become less
dense; joints have limited mobility; and fat pads are
less thick,

demonstrate fibrotic atrophy, migrate distally, and may be
stiffer

The use of multivariate treatment of pressure data for
the prevention of foot ulceration

15


‘Feeding’ the system with various / relevant plantar pressure data of
healthy and affected people



Classification of unroll patterns and pressure distribution images using
cluster analysis / neural networks



Continuous review of pressure data during the day using a suitable
measuring system based on suitable sensors (
PrinDASA
) and lasting
reconciliation with risk
-
like pressure patterns (from clusters / neural
networks)



Feedback as soon as a risk
-
like pressure distribution / unroll pattern is
achieved





Problemstellung
:
Multifaktorielle

Betrachtung

von
Druck

und
Ulzeration
:


Ulzerationen

durch

Vielzahl

an
Druckparametern

bedingt



Interpretation of multiple gait signal interactions and quantitative
comparisons of gait waveforms are

identified

as important data analysis topics in need of further research.


Pressure and ulceration


A
multifactorial

problem

16



Only 38% of ulcer locations matched the peak pressure location. Peak
pressure location actually changed in 59% of patients over the time.


(
Veves

et al.,
Diabetologia
. 1992)



The duration and frequency of stress is severer in foot ulceration
development than intensity of peak pressure.


(
Baites
,
Clin

Biomech
. 2003)



Patients with risk of diabetic foot ulcerations had significantly higher
forefoot peak pressure and pressure time integrals than controls.


(
Stess

et al., Diabetes Care. 1997)


Pressure and ulceration


A
multifactorial

problem

17

Pressure and ulceration


A
multifactorial

problem

19

Developement

of

Foot
Ulceration

Peak Pressures

Relative
Load

Distribution

Pressure
-
Time
-

Integrals

Shear

Forces

Shear

Forces

Spatial

Peak
Pressure

Distribution

Mean

Pressure

Multivariate

statistical

methods

with

regard

to

plantar

pressure

ditribution

parameters

could

be

usefull

to

detect

critical


unroll

pattern


and

pressure

distribution
,


Pressure

evokes

foot

ulcerations

in

a

multifactorial

way
.

Plantar

pressure

is

affected

multifactorial

by

gait

patterns
,

footwear
,

tissue

characteristics
,

foot

deformities
,

etc
..

Multivariate approaches to classify pressure distribution data

20

Pressure

data of diseased subjects

(e.g. Diabetics with neuropathy &
risk of ulcerations
)

Pressure data of healthy subjects

Classification of unroll patterns with the aid of
e.g. Cluster
-
Analysis, Neuronal Networks

Classification of measured data to
corresponding group

Measured plantar
pressure during daily
gait activity

“Intervention” when
measured unroll
pattern correspond to
risk
-
like unroll pattern

21

Heel

Midfoot

Lateral Forefoot

Middle Forefoot

Medial Forefoot

Hallux

Figure
:

Means

&

Confidence

Intervals

of

for

Each

Cluster
´
s

Anatomical

Sub
-
Areas

(
rfti

&

pmax
)

Example: Cluster analysis to classify roll over patterns



Diabetic patients developing foot ulcers seem to have less cumulative
plantar stress than those that do not develop foot ulcers.


(
Maluf

& Mueller,
Clin

Biomech
. 2003)



No specific peak pressure threshold that predicts the development of
ulcerations.


(Armstrong et al., J Foot Ankle Surg. 1998;
Najafi

et al., Gait Posture. 2010)



Pressure and ulceration


A
multifactorial

problem

22

Lott

DJ, Zou D, Mueller MJ.
Pressure

gradient

and

subsurface

shear

stress on
the

neuropathic

forefoot
.
Clin

Biomech

(Bristol,
Avon
) 2008; 23: 342

348.

23

D

DP

H

O

OD

ODP

Figure
:

Cumulative

Percentage

Distribution

of

Diseases

in

Cluster

Example: Cluster analysis to classify roll over patterns