Recognition of Posture and Gait Disturbances in Patients with Normal Pressure Hydrocephalus Using a Posturography and Computer Dynography Systems

Assessment of stability and balance system consist in quantitatively measuring and analysing movements of the centre of foot pressure (COP). Position of COP steadily changes due to the so called postural sways, and of course due to voluntary moves. Enlarged sways, observed in normal pressure hydrocephalus, are not however specific and cannot give simple diagnosis. Postural balance can be impaired due to pathology in various organs including vestibular and cerebellar disorders and various forms of ataxia (Mohan at al. 2009), Parkinsonism (Bloem at al. 1995, Stolze at al. 2001, Jagielski at al. 2006), multiple sclerosis (Kessler at al. 2011) and even alcohol dependence (Wöber at al. 1998) and muscle fatigue or aging (Błaszczyk and Michalski 2006).


Introduction
There are great difficulties in clinical practice to differentiate between normal pressure hydrocephalus (NPH) and brain atrophy (Tans 1979, Galia at al. 2005. The consequences of inaccurate diagnosis are serious therefore we observe steady searching of new non-invasive or minimal-invasive diagnostic methods.
The purpose of this study is to quantify the characteristics of the postural sway and locomotion in NPH patients in two states: before and after shunt implantation and to compare posture and gait features among: NPH, brain atrophy patients and healthy persons.
Assessment of stability and balance system consist in quantitatively measuring and analysing movements of the centre of foot pressure (COP). Position of COP steadily changes due to the so called postural sways, and of course due to voluntary moves. Enlarged sways, observed in normal pressure hydrocephalus, are not however specific and cannot give simple diagnosis. Postural balance can be impaired due to pathology in various organs including vestibular and cerebellar disorders and various forms of ataxia (Mohan at al. 2009), Parkinsonism (Bloem at al. 1995, Stolze at al. 2001, Jagielski at al. 2006, multiple sclerosis (Kessler at al. 2011) and even alcohol dependence (Wöber at al. 1998) and muscle fatigue or aging (Błaszczyk and Michalski 2006).
Evaluation of gait relates to postural stability in standing upright position. The gait disturbance is probably the most prominent clinical feature of NPH and it is often the first NPH symptom to develop (Radvin 2008). Gait disturbances are part of so called Hakim triad (Hakim & Adams 1965). NPH gait disturbances are very characteristic and rely on shuffling manner of walking, without raising the feet as if they were glued to the floor. This kind of gait is called also magnetic. Gait disturbances are still not fully described quantitatively due to lack of reliable, specific parameters measuring most typical features of gait in NPH.
Some papers related to postural stability and gait evaluation in NPH have already been published by Szczepek and Czerwosz (Szczepek at al. 2008. The current study is trying to summarize some of our results.

Methods
Recently rapid development of precise methods of quantitative measurements of body position while standing or walking has been observed. Two techniques used by us in our investigations should be discussed here: 1. static posturography -measurement of body sways while standing on a force plate, 2. dynography -measurement of gait.
In both systems the resultant force -feet pressure acting on the horizontal surface (XY) is calculated on the basis of some number of pressure sensors. The most important is the point of application of this force. This point is called Centre of foot/feet Pressure (COP). In static conditions the COP point is a projection of the Centre of Gravity (COG) position -on the XY horizontal plane. COM and COG signals are highly correlated (Błaszczyk 2008). It has been documented that COG signal can be extracted from COP by low-pass filtering (Benda et al. 1994). The high frequency component comes in dynamic and realistic conditions from inertia forces that influence COP instantaneous location. Inertia forces arise from accelerations of the body while it is swaying or moving -losing and recovering balance (Newton's second law of motion).

Posturography
The instantaneous COP xy position can be calculated on the basis of instantaneous values of p 1 (t), p 2 (t), p 3 (t), p 4 (t) forces measured on four corners of the square plate; d is a length of it's side. In case of our device: d= 40 cm. 1234 1234 1234 1234 d x(t) ( p (t) p (t) p (t) p (t)) /(p (t) p (t) p (t) p (t)) 2 d y(t) ( p (t) p (t) p (t) p (t)) /(p (t) p (t) p (t) p (t)) 2 To obtain the exact position of COP, the p n forces must be reduced by tare weights measured independently on each corner. The real force plate is shown on Figure 1; p n forces are pointed and orientation of xy plane is given by X and Y axes. All p n (t) values, and therefore x(t) and y(t) change in time. In practice we collect them in 0.01 seconds intervals (sampling frequency 100 Hz) in digital form with 12 bit accuracy. Data were low-pass filtered (15 Hz cut-off frequency). Trajectory can be observed on-line and off-line in an analogue way on a chart called posturogram or stabilogram.
A single measurement on a force plate takes usually 30-60 seconds. The resultant time had to be reduced due to some artefacts related to unsolicited activity of the patient such as his movement or speaking influencing the outcomes. Removing artefacts is still an unresolved problem in posturography due to questionable difference between unsolicited movement and essential balance restoration, especially in case of large sways. There are four sensors measuring forces on each corner: p1, p2, p3, p4. Y represents forward-backward, anterior-posterior sways, in sagittal plane, X represents left-right, mediolateral sways, in frontal plane.  implantation. The next two were measured after surgery; the last two belong to a healthy person. As one can see, there are big differences in the shapes of the trajectories, especially NPH patient before surgery demonstrates very large sways -both, with eyes open and closed. Sways of the NPH patient are very large both for EO and EC. After surgery sways are reduced but they are still larger than in a healthy person.

Posturography parameters
COP trajectory represents sways of an object standing in upright position for some time. There is some number of various metrics developed that evaluate average or typical "behaviour" of the curve in many aspects (Baratto at al. 2002, Raymakers at al. 2005.
The starting point of our analysis was to define the global parameters expressing the "size" of sways. We have taken into consideration: -R -average COP sway Radius, -A -Area of developed surface of COP trajectory, AS -Area Speed -L -Length of COP trajectory, V -Velocity.
An average Radius of sway is a simple average of distances between curve samples and (0,0) point on XY plane -coordinate origin. Actually all points of the curve have been shifted by (,) x y vector beforehand and thus (0,0) turns into the "centre of gravity" of all samples ( x denotes average value of all x i ). This simple way of calculating R has been applied by Czerwosz and Szczepek in their papers  and also Mraz at al. 2007, Bosek at al. 2005, Kubisz at al. 2011.
Calculation of developed area bases on all available samples. The developed surface consists of triangles created from every two consecutives COP positions sampled every 0.01 second and the coordinate origin (0,0). Let's compare this surface to wooden pencil shavings when sharpening a pencil. Figure 3 shows the idea of the calculation on a piece of COP curve. The area depends on the number of samples -thus it depends on measurement duration. It is easy to normalize it dividing the area by measurement duration. In this way we are obtaining Area Speed (AS) in mm 2 /sec.
The length of COP trajectory is calculated as a simple sum of consecutives segments between COP positions sampled every 0.01 second. Length after division by measurement duration becomes a Velocity (V) in mm/sec. Each parameter can relate to two measurement conditions: eyes either open (EO) or closed (EC). In order to express change in value of some parameters due to different conditions one can use simple difference (2). For computational reasons in advanced statistical analysis sum of the same parameters (3) has been introduced.
where X can be R, A, AS, L, or V.
A derived parameter -the vision index (4) has been developed on the basis of Radius, Area, Area speed, Length or Speed to express difference of chosen parameter in relation to its mean value. Index is an absolute, dimensionless number with theoretical range [-100%, 100%]. Zero means no difference. One should notice that I x is bigger for bigger D x and smaller S x and vice versa: I x gets smaller for smaller D x and bigger S x . A similar index related to the parameters measured in two different conditions have been introduced by Mraz as ICOP (Mraz at al. 2007).
where X can be R, A, AS, L, or V. EC means "eyes closed", EO -"eyes open ", while measurement has been performed. We have made use of vision indices related to radius, area and length. Notice that the index related to area is equal to the index related to area speed (I AS = I A ) and the index related to length is equal to the index related to velocity (I V = I L ).
Let's take sight index of radius as an example. Exact way of calculation I R is given below:

Dynography
Computerized dynography (Infotronic 2007) is a gait analysis system which consists of two soles containing sensors sensible to a foot pressure acting on the ground. These sensors measure the vertical ground reaction forces and their distribution during walking. It's a good alternative to much less quantitative gait scale -for measuring gait impairment of NPH patients (Boon at al. 2007). Another alternative is camera based system but then the only information that is provided are body parts positions and angles without any data related to forces. Walking on a treadmill gives just the speed of gait; to measure more gait features some extra instrumentation should be added. In this study we don't take advantage of forces explicitly, but our system (see below) uses them internally to determine gait phases. Other alternative method of gait evaluation can be performed on two or more joined force plates. This method limits gait to only few steps because of size. It can easily be used for gait analysis in small animals (Voss at al. 2007).
Special boot for dynography is presented in Figure 4 (on the left side). There are eight sensors inside each sole located as shown in the middle picture. Actually eight histograms are shown here on the sole, not sensors, but they are distributed just as sensors. Histogram height expresses instantaneous or average force acting on a sensor. The picture on the right shows how the point of application of the resultant ground reaction force is being displaced while stance phase of gait cycle. This point is the Centre of foot/feet Pressure (COP) and its position changes during each step. The overall load, the value of resultant ground reaction changes due to inertia forces as well. The displacement of COP presented in Figure 4 relates to single foot only. In this case COP position is calculated on the basis of eight sensors. Figure 5 presents so called gait-lines where successive COP positions of each step are drawn overlapped. Gait-lines of each foot are calculated independently. Pictures on the right side are averaged over all gait cycles of 20 seconds walking. Gait-line represents stance phase of gait from initial contact, while a heel touches the ground till toe off moment (Perry & Burnfield 2010). COP can also be calculated on the basis of 16 sensors enclosed in two soles -from both feet. Cyclogram arises if connecting successive COP positions of each step and drawing the lines overlapped. They are presented in Figure 6. Pictures on the right side are averaged over all gait cycles. During double support the COP position lies somewhere between the feet depending on the load ratio and its position changes from one side to another. During single support COP is located within single foot boundary.
Let there be N sensors. N = 8 or N = 16. Let i be the number of a sensor: 1 ≤ i ≤ N. Let the position of sensor i be (x i , y i ). Let F i (t) be the force at moment t acting on sensor i.
Coordinates of the COP at moment t can be calculated from equations (6) -see Jeleń at al. 2008.
Equations (6) are very similar to (1) related to force plate, only the number of sensors (N = 4) and its positions are different. The gait-lines and cyclograms are normalised and related to dimensionless foot length, therefore all distances and speed can be calculated only if the total distance that the person has walked has been entered. Data is sampled with the frequency equal to 100 Hz.

Examples of dynography measurements
Pictures in Figure 5 show examples of gait-lines; pictures in Figure 6 -cyclograms obtained for NPH patient before and after shunt implantation and for healthy person. Gait-lines and cyclograms show gait in compact way. The symmetry of gait and regularity of cycles can  easily be observed. One can notice shuffling gait of NPH patient (before surgery) -there is almost no single support phase -the patient only slightly rises his feet. The whole stance phase is consisting of double support.

Dynography parameters
After data collection CDG software recognizes gait phases and calculates gait parameters. Gait was described in our research as a duration of a single (SSUP) and double support (DSUP) and duration of a stance phase (STANCE). These measures relate to the left and right leg independently. The phases are shown in Figure 7.

Hardware and software usage
For posturographic measurements Pro-Med (Poland) force plate (Olton & Czerwosz 2006) was used by us. The software for data collection and analysis has been written by the author (Leszek Czerwosz).
For measurement of ground reaction forces in walking we used the Ultraflex Computer DynoGraphy system from Infotronic Company (Netherlands). Applications of CDG system in NPH are not aware. Explanations of usefulness of this system are provided by Bhargava at al. 2007, Majumdar at al. 2008, Jeleń at al. 2008.

Statistical methods
Wilcoxon matched pairs test for intragroup and Mann-Whitney U test for between-group comparisons were performed in our studies ). Non-parametric ANOVA Kruskal-Wallis test (Kruskal & Wallis 1952) was used while comparing more cohorts and then Bonferroni correction (Dunn 1961) for post-hoc comparisons.
There is much confusion related to the application of corrections if performing post-hoc comparisons, especially if testing schema is a priori established. It is hard to accept the fact that the result of one statistical test, i.e. while comparing statistically the NPH and CONTROL groups, can be influenced by other tests, while collecting some extra data, for example ataxia group and by performing other comparisons. We will of course change the level of significance required for rejection the null hypothesis to p<0.016 in case of three group analysis that we perform in this study. And we are introducing pattern recognition methods for differentiation of groups to avoid purely academic arguments. www.intechopen.com

Recognition of Posture and Gait Disturbances in Patients with Normal Pressure Hydrocephalus Using a Posturography and Computer Dynography Systems 197
All the nonparametric analyses were conducted using Origin software (OriginLab Corporation) and PASW Statistics (IBM-SPSS Statistics).
We applied pattern recognition algorithms -two advanced statistical methods: Dicriminant Analysis (DA) and k-Nearest Neighbour method (K-NN) (Devijver andKittler 1982, Duda at al. 2001). DA calculations were performed by means of PASW Statistics. For K-NN calculations (Jóźwik at al. 2011) a computer programme developed by Adam Jóźwik was used.
The K-NN classifier is a pattern recognition algorithm for recognizing classes of objects.
Objects are just vectors of features which values can be measured in patients. Thus an object represent a patient in n-dimensional space, where n is a number of features -measured parameters for each patient. The k-NN algorithm requires a reference set of objects with known class membership. Class means the same as patient group. Any new object, from outside the reference set, is assigned to the class most frequently represented among its k nearest neighbours, searched in the reference set. The leave-one-out method is used to experimentally establish the best value of k giving minimal misclassification rate.
The DA and K-NN methods differ absolutely because DA is a strictly parametric method related closely to the analysis of variance (ANOVA) and produces, among other, linear combination of features (parameters or variables). There is a fundamental assumption that all independent variables have to be normally distributed. In our case -there is no proof for normal distributions, therefore DA outcomes may not be reliable. For the K-NN method it is not important whether distributions are normal.
Pattern recognition methods have already been used in relation to some gait parameters (Bertrani at al. 1999). They should not be mixed with the pattern recognition of gait (Maduko). Discriminant analysis and neural networks were used for gait classification (Kaczmarczyk at al. 2009).

Material
After ethical approval by a local Ethics Committee, posturographic and dynographic recordings were taken from NPH patients and from healthy volunteers. Patients with brain atrophy were also recorded to obtain a comparison group to test the power of calculated parameters and statistical methods in differentiation NPG and atrophy (Tans 1979, Galia at al. 2005. NPH diagnosis was based on the following criteria 1. Enlargement of brain ventricles seen on CT or MR -Evans' ratio above 0.3 (Evans 1942), 2. Neurological symptoms (Hakim triad -minimum two of three symptoms), 3. Mean intracranial pressure ≥ 10 cmH 2 O, 4. Resorption resistance R ≥ 11mmHg/ml/min.
Infusion test , Czosnyka at al. 1988) is performed on the basis of cerebrospinal intracranial fluid pressure measurement with simultaneous infusion of physiological saline in L4, L5, and S1 regions. Infusion test seems to be the most important and limitative qualification for shunt implantation.
In all cases balance disturbances and impairment of gait was observed. www.intechopen.com

Hydrocephalus 198
The ATROPHY group has been formed according to the following inclusion criteria: 1. Enlargement of brain ventricles seen on CT or MR (Evans' ratio above 0.3), 2. Both subcortical and cortical atrophy, 3. No characteristic neurological symptoms, 4. Mean intracranial pressure < 10 cmH 2 O, 5. Resorption resistance R < 11 mmHg/ml/min.
Balance disturbances and some impairment of gait were observed in atrophy patients.
Posturography and dynography evaluations were performed in NPH cases before shunt implantation and shortly after the surgery (within seven days).

Results
A number of results have been obtained in the posture and gait study. Posturography and dynography results will be reported separately because so far no joined analysis has been made.

Simultaneous comparison of three groups -posturography
To compare three groups: NPH BEFORE, ATROPHY, and CONTROL non-parametric ANOVA Kruskal-Wallis test was used. The results are in Similar three-group analysis was done for NPH AFTER, ATROPHY, and CONTROL. These groups differ significantly for parameters provided in Table 2. There are no significant differences for D R , D A , and D L differences as well as for the vision indices: I R , I A , I L . This is quite obvious if you compare left and right columns in each pair of columns in CONTROL, ATROPHY, and NPH AFTER groups in Figure 8 -in the upper column chart. Only one chart -the radius chart is presented here. The lower column chart shows vision indices. Indeed CONTROL, ATROPHY, and NPH AFTER columns are very similar.

Comparison of groups in pairs -posturography
All collected posturographic data related to the Radius are exhibited in Figure 8. There are four groups here; each one consists of EO and EC measurements. Individual values of the Radius in EO and EC measurements are shown overlapped to respective columns. For comparisons among three groups in pairs see Tables 3, 4, 5, 7, and 8. Nonparametric Mann-Whitney U test with Bonferroni correction was used.
One can notice that NPH BEFORE patients reach largest sways, both with EO and EC (see Table 3 and 4 respectively). CONTROL group exhibits the smallest sways, both with EO and EC. The same effect can be observed on sums of corresponding parameters (S X , where X can be R, AS, or V, see Table 5). All tested differences of X EO , X EC , and S X parameters are significant but only V EC and S V parameters do not differentiate NPH BEFORE and ATROHY pair of groups (Tables 4 and 5).

Effect of shunt implantation on posturography parameters
One can see in Figure 9 that radius of sways measured in both conditions: EO and EC in a NPH group before shunt implantation treatment exceeded corresponding values measured in the same patients after surgery.
There is a full set of parameters included in Table 6 -average values, standard deviation and significance level of nonparametric Wilcoxon paired test. The NPH BEFORE and NPH AFTER groups differ in relation to almost all parameters. The most powerful is radius with EO (R EO ) -see also Figure 9.
Sways with eyes opened before shunting is much bigger than after surgery.

Impact of vision on postural sway characteristics
We introduced EC-EO differences and vision index for explaining in various groups the effect of eye opening or closure on sways. Table 7 shows D X and Table 8 -I X (where X can be R, AS, or V). The interpretation of differences D X can be skipped here.
Most interesting features are the vision indices. They differentiate NPH BEFORE group from NPH AFTER and from any other group. Table 8 shows three comparisons in pairs of groups. The groups NPH BEFORE -ATROPHY significantly differ as well as the NPH BEFORE -CONTROL groups do.
A direct comparison of EO and EC parameters in four groups independently is provided in Table 9. Sways with EO and EC seem to be equal in NPH BEFORE group only. In any other group there are significant changes of sways related to the eyes closure. Enlargement of sways is a normal phenomenon after closing eyes, but not in NPH patients in acute state.
Sways of NPH patients do not depend on the sight, they seem to be the same in EO and EC conditions. This observation is the most important result of this study.

Two dimension analysis -posturography
Much more insight into the data collected is provided in the two dimensional graph. There is four-group scattergram in Figure 10. One can see a separation of NPH BEFORE cases and the centroid from all other groups. A separation is statistically significant -see tables 5 and 8 for NPH BEFORE -CONTROL and NPH BEFORE -ATROPHY data.  Table 9. Comparison of R, AS, and V parameters measured with EO and EC in CONTROL, ATROPHY, NPH BEFORE and NPH AFTER groups separately. Fig. 10. Scattergram presents all cases in four groups: NPH BEFORE, ATROPHY, CONTROL, and NPH AFTER in I R (vision index related to radius) versus S R (sum of R EO and R EC ) coordination system. Centroids of the groups are shown with standard deviations. Notice that ATROPHY and NPH AFTER group centroids do overlap.
Two dimensional statistical analysis was performed by means of methods described in the following section. www.intechopen.com

Recognition of Posture and Gait Disturbances in Patients
with Normal Pressure Hydrocephalus Using a Posturography and Computer Dynography Systems 205

k-NN classification of posturografic parameters
All data was analysed by means of two methods: Discrimination Analysis (DA) and k nearest neighbours (K-NN) method. Both methods give similar results for most of classifications.
Feature is a single value of multidimensional vector assigned to any object/case/patient. Groups of patients are called classes within the "classification world". There are 15 features used here as follows: R EC , R EO , AS EC , AS EO , V EC , V EO , S R , S AS , S V , D R , D AS , D V , I R , I AS , and I V , exactly as parameters.
The analysis was performed in several sections. At the beginning every feature was used, one at a time. Then features were grouped into pairs according to the templates: (X EO and X EC ), (sums S X and differences D X ), (sums S X and vision indices I X ), where X can be R, AS, or V. Features related to R, AS and V were paired separately. Then clusters consisting of six features were applied.
We classified groups/classes in two following schemas -like in traditional analysis: 1) NPH BEFORE, ATROPHY and CONTROL 2) NPH BEFORE and NPH AFTER. Briefing of the classification results is presented below. It is interesting that there are three pairs of features with equal classification power. They relate to the analysis in two dimensions. Notice that (S AS ,I AS ) and (S AS ,D AS ) can be calculated from (AS EO ,AS EC ) using formulas (2),(3), and (4). These dimensions are not completely independent because, for example in the CONTROL group the features R EO and R EC are correlated (R = 0.73); in other groups the correlation is not as high. An example of twodimensional approach is shown in Figure 10.

Single feature applied individually, classes: NPH BEFORE and NPH AFTER:
Six-feature cluster without and with feature selection, classes: NPH BEFORE and NPH AFTER: Multi-feature classification by K-NN method was also performed with the following features: S R , S AS , S V , I R , I AS , and I V . Classification without feature selection (all features forced to enter the analysis) resulted in 89.2% of correctly classified cases. Classification with automatic selection of "best" features resulted in 94.6%. The selected features: S R and I AS .
Single feature applied individually, classes: NPH BEFORE, ATROPHY, and CONTROL: Three-group classification (NPH BEFORE, ATROPHY and CONTROL) using single features cannot be good both in DA and K-NN methods. The best results reached AS EO -77.4% of correct classified cases.

Two-feature pairs, classes: NPH BEFORE, ATROPHY, and CONTROL:
There were nine DA/K-NN classifications here, the best are:

Analysis of gait parameters
Evaluation of gait is performed by means of five gait parameters: time of single supper (T SSUP ), time of double support (T DSUP ), time of stance (T STANCE ), length of single support (D SSUP ), and length of double support (D DSUP ). Initially ten parameters were involved; they were related to the left and right leg. Average values with standard deviations were plotted in Figure 11. Values of the parameters related to the left and right leg were compared statistically by means of nonparametric Wilcoxon paired test in each group separately. There was no difference found between legs. Therefore the outcomes from the left and right leg were combined -cases are not patients now but legs. Direct comparison of gait parameters between BEFORE and AFTER states in NPH are given in Table 12. Because NPH BEFORE and NPH AFTER are paired groups, the Wilcoxon test was used to compare the gait. Differences of all parameters are statistically significant. Figure 12 shows average values with standard deviations as well as individual values of T SSUP , T DSUP , T STANCE in NPH patients before and after surgery.
After surgery gait of NPH patients resembles gait of patients with atrophy. Table 13 shows that there is no statistical significant difference between NPH AFTER and ATROPHY groups. www.intechopen.com

k-NN classification of dynografic parameters
Dynographic parameters were analysed by means of k-NN method in 2008 ). Now the calculations should be repeated with larger number of subjects. Usage of pattern recognition methods can give chance for better evaluation of multi-parameter data.

Discussion and conclusions
There are three main theses that come out from this study: 1. Sways in NPH patients before shunting is much bigger than after surgery. This relates mostly to eyes open (EO) condition. 2. Sways of NPH patients do not depends on the sight, they seem to be the same in EO and EC conditions. 3. NPH and atrophy imbalances differ when evaluate by means of more than one parameter (feature). Pattern recognition methods should be used.
There are many proofs of strong relationship between vision and postural control. Vision has a greater influence on standing postural control, resulting in greater sway when individuals are presented with erroneous or conflicting visual cues (Redfern at al. 2001). In impairment of the vestibular system and possibly other sensing systems and probably in cerebellar ataxias, vision can help in balance recovery. In NPH we do not observe any improvement, there is some probability that vision can disturb maintenance of the balance. Interestingly the vision index was slightly negative in NPH BEFORE group, meaning that with eyes closed sways are smaller. Blomsterwall (Blomsterwall at al. 2000) found that healthy individual had a 29% better postural function with open eyes while NPH patients only improved their balance score by 18% with open eye. The impact of vision should be studied farther due to this discrepancy.