CN104834888A - Abnormal gait identification method capable of facilitating screening Parkinsonism - Google Patents

Abstract

The invention discloses an abnormal gait identification method capable of facilitating screening Parkinsonism. The method is characterized in that the gait plantar pressure characteristics can be extracted, and the modeling and the identification of the neural network of the gait system of the normal healthy people and the patients suffering from the Parkinsonism can be dynamically carried out; the constant neutral network can be established; a dynamic estimator can be built by using the constant neutral network, and based on the difference between the gait modes of the normal healthy people and the patients suffering from the Parkinsonism in the gait system dynamics, the abnormal gait caused by the Parkinsonism and the normal gait of the normal healthy people can be distinguished according to the minimum error principle, and the screening detection of the Parkinsonism can be facilitated. By arranging a pressure sensing floor system or wearing the special shoes provided with the pressure sensor insole, the plantar pressure characteristics can be acquired, and the abnormal gait caused by the Parkinsonism and the normal gait of the normal healthy people can be distinguished conveniently, simply, and non-invasively, and therefore the daily gait monitoring of the family members can be realized, and the screening detection of the Parkinsonism can be facilitated.

Description

A kind of abnormal gait recognition methods of auxiliary examination parkinsonism

Technical field

The present invention relates to a kind of abnormal gait recognition methods based on plantar pressure feature of auxiliary examination parkinsonism.

Background technology

The walking movement of people is an accurately process for complexity, and the dynamic interaction that its mode of motion is unified between feedback mechanism by central nervous system determined.Some infirmitiess of age and some neurogenic diseases all can cause said process generation problem.People in the process of walking, produces in central nervous system between muscle that the kinesitherapy nerve of motor mindedness and lower limb produce action and there is a signal transduction pathway, if this path exists motor message conductive obstruction, it directly characterizes is exactly the abnormal gait of people.Disease common in the elderly, as parkinsonism (Parkinson ' sDisease) etc., all can exception throw gait, therefore can carry out auxiliary examination and detection to nerve degenerative diseases such as parkinsonisms based on abnormal gait.Epidemiology shows, and the morbidity rate of parkinsonism is 15 ~ 3,28/,100,000 populations, over-65s prevalence about 1%, and the incidence of disease was 10 ~ 21,/10 ten thousand populations/years.Its clinical manifestation includes slow, the myotonia of moving, static tremor, abnormal gait, cognition/insanity, sleep-disorder, dysautonomia, sensory disturbance etc.The parkinsonism cause of disease and pathogenesis are not yet clear and definite, may be relevant with social factor, medicine factor, patients factors etc., carry out early screening detect significant to it.Although medically there is the technological means of a lot of checkout and diagnosis at present, as blood test, surface electromyography (Electromyography:EMG) signal analysis, CT, nuclear magnetic resonance, genetic test, lumbar puncture etc., but these means are usually cumbersome, and all with wound to a certain degree.

The dynamic characteristic of normal gait, as plantar pressure, joint moment etc. usually present complicated wave characteristic between step and step.And the gait of Parkinson's disease patients has marked difference in the change of plantar pressure with normal gait, the plantar pressure of left and right pin correspondence position presents obvious left-right asymmetry property.No matter be Parkinson's disease patients or healthy population, the dynamics of their gait all has complicated non-linear nature, and this is mainly because the nonlinear characteristic of human dynamics system.Important difference is there is between the dynamics of Parkinson's disease patients gait and healthy normal person.And how modeling is dynamically carried out to non-linear gait system, and distinguish based on the difference between this two classes crowd in gait system dynamics, detect parkinsonism with auxiliary examination, then lacking corresponding research, is also difficulties wherein.

The detection of abnormal gait can be regarded as identification and the identification problem of a dynamic mode in essence, and an one of difficult problem for dynamic pattern recognition inherently area of pattern recognition.On the basis of the Persistent Excitation characteristic research to radial basis function (RBF) neural network, C.Wang etc. propose and determine the theories of learning, comprising the identification of the dynamic mode produced Kind of Nonlinear Dynamical System, express and method for quickly identifying, namely by determining that study obtains the dynamic local of dynamic mode built-in system accurately neural net model establishing, time dependent dynamic mode with time the constant and mode of space distribution effectively express, the dynamics topological resemblance of dynamic mode inherence is utilized to provide similarity definition between dynamic mode further, and propose a set of new method of dynamic mode being carried out to identification fast.

Summary of the invention

The object of the invention is to overcome the shortcoming of prior art and deficiency, a kind of abnormal gait recognition methods of auxiliary examination parkinsonism is provided, for differentiating that the abnormal gait of Parkinson's disease patients provides one more succinct examination detection method accurately.

The abnormal gait recognition methods of a kind of auxiliary examination parkinsonism of the present invention, comprises the following steps:

Step 1, gathered the plantar pressure characteristic of each Parkinson's disease patients and healthy normal person respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the some Parkinson's disease patients gathered and healthy normal person forms training set;

Step 2, the gait feature variable extracted according to step 1, dynamically carry out modeling to the unknown nonlinear gait system of normal person healthy in training set and Parkinson's disease patients, design RBF neural identifier, approaches the unknown dynamic local of gait system;

The foundation of step 3, constant value neural network:

According to determining the theories of learning, neuron along the RBF neural of gait system features track meets persistent excitation condition, its weight convergence is to optimal value, get the average of weights in a period of time after weight convergence as learning training result, and utilize these results to set up constant value neural network, the gait system dynamic acquired is gained knowledge with the storage of the form of constant value neural network weight, forms a training gait pattern storehouse;

Step 4, gathered the plantar pressure characteristic of each Parkinson's disease patients to be measured respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the Parkinson's disease patients some to be measured of collection forms test set;

Step 5, classification and Detection:

Constant value neural network is utilized to build one group of dynamic estimator, non-linear gait system dynamic corresponding to healthy normal person in the training gait pattern storehouse that step 2 and step 3 learn and Parkinson's disease patients is gained knowledge to be embedded in dynamic estimator, the gait feature data of Parkinson's disease patients to be detected and this are organized dynamic estimator and does difference, form one group of error in classification, detect according to the abnormal gait of least error principle by Parkinson's disease patients to be detected, realize the auxiliary detection to parkinsonism.

Further, in step 1 and 4, adopt gait analysis database PhysioNet, the plantar pressure characteristic of each 8 symmetric positions of human body left and right pin is gathered by the pressure transducer being placed on experimenter's sole, choose the plantar pressure characteristic of the most obvious about the two groups pin symmetric positions of wherein difference, form one group of plantar pressure characteristic variable.

Further, in step 2, unknown nonlinear gait system dynamic modeling is as follows:

$\stackrel{\·}{x}=F(x;p)+v(x;p)$

Wherein, x=[x ₁..., x _n] ^t∈ R ⁿbe the plantar pressure characteristic variable that step 1 is extracted, p is the normal parameter value of system, and n is the dimension of plantar pressure characteristic variable;

F (x; P)=[f ₁(x; P) ..., f _n(x; P)] ^tbe smooth and the Nonlinear Dynamic state variable of the unknown, the gait system representing healthy normal person and Parkinson's disease patients is dynamic, v (x; P)=[v ₁(x; P) ..., v _n(x; P)] ^tbe modeling indeterminate, the two is merged into and it is dynamic to be defined as general nonlinearity gait system;

(2) design neural network identifier and be used for identification:

Adopt dynamic RBF neural network constructing neural network identifier, dynamic RBF neural network identifier form is as follows: wherein it is the plantar pressure characteristic variable chosen; A=diag [a ₁..., a _n] be diagonal matrix, a _ibe the constant of design, meet 0 < | a _i| < 1, be dynamic RBF neural network, the general nonlinearity gait system being used for approaching the unknown is dynamic s (x)=[S ₁(|| X-ξ ₁|| ..., S _n(|| X-ξ _n||] ^tbe Gaussian radial basis function, N > 1 is neural network number of network nodes, ξ _ineuronal center point, RBF neural weights adjustment rule as follows:

${\stackrel{\&CenterDot;}{\hat{W}}}_i=-{\mathrm{\&Gamma;}}_{i}S\left(x\right){\tilde{x}}_i-{\mathrm{\&sigma;}}_i{\mathrm{\&Gamma;}}_i{\hat{W}}_i,i=1,...,n,$

Wherein, i represents that n ties up the i-th dimension variable in plantar pressure characteristic variable, state error, Γ _i=Γ _i ^t> 0, σ _i> 0 is the regulating parameter regulating rule, the weights of dynamic RBF neural network initial value

Constant value neural network is set up in step 3 specific practice:

According to determining the theories of learning, neuron along the RBF neural of gait system features track meets persistent excitation condition, its weight convergence, to optimal value, is got the average of weights in a period of time after weight convergence as learning training result, and is utilized these results to set up constant value neural network the gait system dynamic acquired is gained knowledge with the storage of the form of constant value neural network weight, forms a training gait pattern storehouse; Described constant value neural network weight levy by as shown in the formula sublist: wherein, [t _a, t _b] represent constant value neural network weight completing to a time period after the transient process that its optimal value restrains, make like this can by constant value neural network carry out local accurately to approach:

wherein, ε _i2it is approximate error;

Classification and Detection in step 5, specific as follows:

According to the constant value neural network weight of healthy normal person and Parkinson's disease patients in training gait pattern storehouse construct one group of dynamic estimator, be expressed as follows:

${{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k=-b_i({{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k-x_{\mathrm{ti}})+{{\stackrel{\&OverBar;}{W}}_i}^{k^T}S\left(x_{\mathrm{ti}}\right),i=1,...,n,k=1,...,M,$

Wherein, χ _ifor the state of dynamic estimator, b _ifor dynamic estimator parameter, x _tifor the gait feature data of Parkinson's disease patients to be detected in test set, k represents the kth training mode in M training mode, and M is the pattern total amount in training gait pattern storehouse;

By the gait feature data x of Parkinson's disease patients to be detected in test set _tiorganize dynamic estimator with this and do difference, obtain following classification and Detection error system:

Wherein, be state estimation error, calculate average L ₁norm is as follows:

${\left|\right|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(t\right)\left|\right|}_1=\frac{1}{T_c}{\&Integral;}_{i-T_c}^t\left|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(\mathrm{\&tau;}\right)\right|\mathrm{d\&tau;},t\&GreaterEqual;T_c,$

Wherein, T _crepresent gait cycle;

Classification and Detection strategy is as follows: if there is a finite time t ^s, s ∈ 1 ..., k} and a certain i ∈ 1 ..., n}, makes to all t > t ^sset up, then the abnormal gait pattern of the Parkinson's disease patients to be detected occurred is classified and detects.

The adjustment rule of described RBF neural weights is according to Li Yapuluofu stability theorem and determines the theories of learning to design, and makes state error and weights estimation all bounded, and exponential convergence, wherein, the weight convergence of RBF neural has two kinds of situations:

The first situation: the neuron returning the RBF neural of track along gait feature meets persistent excitation condition, and its weight convergence is in the small neighbourhood of optimal value;

The second situation: the neuron not excited target and not being conditioned returning the RBF neural of track away from gait feature, its weights are approximately zero.

Can be represented by following formula general nonlinearity gait system dynamic local accurate modeling:

Wherein, ε _i1it is approximate error; Here local accurate modeling refers to that the internal dynamic away from track is not then approached by RBF neural approaching the built-in system dynamic trajectory along gait feature data.

The present invention will determine the theories of learning in conjunction with plantar pressure feature application in dynamically carrying out local accurate modeling and identification to the non-linear gait system of healthy population and Parkinson's disease patients, the gait dynamics that learns gain knowledge and to store with the form of constant value neural network weight, the difference between healthy population and Parkinson's disease patients in gait system dynamics is utilized to classify, with auxiliary detection parkinsonism.

The present invention is by setting pressure induction floor system or the special shoes dressing band pressure transducer shoe-pad, obtain plantar pressure feature, can convenient and simple, the normal gait of non-invasively distinguishing abnormal gait caused by parkinsonism and general health crowd, realize the daily gait monitoring of kinsfolk and the auxiliary examination detection of parkinsonism.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is left and right pin plantar pressure sensor distributing position schematic diagram in the embodiment of the present invention;

Fig. 3 (a1) is Parkinson's disease patients left foot plantar pressure feature L used in the embodiment of the present invention ₃schematic diagram;

Fig. 3 (a2) is Parkinson's disease patients left foot plantar pressure feature L used in the embodiment of the present invention ₆schematic diagram;

Fig. 3 (a3) is Parkinson's disease patients right crus of diaphragm plantar pressure feature R used in the embodiment of the present invention ₃schematic diagram;

Fig. 3 (a4) is Parkinson's disease patients right crus of diaphragm plantar pressure feature R used in the embodiment of the present invention ₆schematic diagram;

Fig. 3 (b1) is healthy normal person's left foot plantar pressure feature L used in the embodiment of the present invention ₃schematic diagram;

Fig. 3 (b2) is healthy normal person's left foot plantar pressure feature L used in the embodiment of the present invention ₆schematic diagram;

Fig. 3 (b3) is healthy normal person's right crus of diaphragm plantar pressure feature R used in the embodiment of the present invention ₃schematic diagram;

Fig. 3 (b4) is healthy normal person's right crus of diaphragm plantar pressure feature R used in the embodiment of the present invention ₆schematic diagram;

Fig. 4 is the topological structure simplified schematic diagram of the RBF neural adopted in the embodiment of the present invention;

Fig. 5 is the convergence situation of RBF neural weights in the embodiment of the present invention.

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail.

Embodiment

As shown in Figure 1, the abnormal gait recognition methods of a kind of auxiliary examination parkinsonism of the present invention, comprises the following steps:

Step 1, gathered the plantar pressure characteristic of each Parkinson's disease patients and healthy normal person respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the some Parkinson's disease patients gathered and healthy normal person forms training set, and wherein plantar pressure characteristic variable leaching process is as follows:

The gait data storehouse that the present invention adopts is the online gait analysis database PhysioNet provided by people such as Hausdorff

( http:// www.physionet.org/physiobank/database/gaitpdb), its gait signal is that the pressure transducer by being placed on experimenter's sole obtains, and the sample frequency of signal acquiring system is 100Hz.This gait data storehouse provides following gait feature parameter at present: the plantar pressure of each 8 symmetric positions of human body left and right pin as shown in Figure 2, is L respectively ₁, L ₂, L ₃, L ₄, L ₅, L ₆, L ₇, L ₈and R ₁, R ₂, R ₃, R ₄, R ₅, R ₆, R ₇, R ₈, unit: newton (N).Found through experiments, Parkinson's disease patients L ₃, R ₃and L ₆, R ₆the plantar pressure of position is compared to the plantar pressure of other positions, and the difference existed between the plantar pressure at healthy normal person's same position place is the most obvious, therefore left and right pin L ₃, R ₃and L ₆, R ₆the plantar pressure most of these two groups of symmetric positions is representative.Choose the plantar pressure L of these two groups of symmetric positions ₃, R ₃and L ₆, R ₆, form one group of plantar pressure characteristic variable: x=[L ₃, R ₃, L ₆, R ₆] ^t, to reduce intrinsic dimensionality and calculated amount.Process of the test one has 93 Parkinson's disease patients, comprises 58 male sex and 35 women, 66.3 years old mean age; And 73 normal Healthy Peoples of gait, comprise 40 male sex and 33 women, 63.7 years old mean age.As shown in Fig. 3 (a1), Fig. 3 (a2), Fig. 3 (a3), Fig. 3 (a4) and Fig. 3 (b1), Fig. 3 (b2), Fig. 3 (b3), Fig. 3 (b4), the difference schematic diagram between Parkinson's disease patients and healthy normal person in plantar pressure feature respectively, wherein: Parkinson's disease patients left foot plantar pressure feature L ₃as shown in Fig. 3 (a1), Parkinson's disease patients left foot plantar pressure feature L ₆as shown in Fig. 3 (a2), Parkinson's disease patients right crus of diaphragm plantar pressure feature R ₃as shown in Fig. 3 (a3), Parkinson's disease patients right crus of diaphragm plantar pressure feature R ₆as shown in Fig. 3 (a4); Healthy normal person's left foot plantar pressure feature L ₃as shown in Fig. 3 (b1), healthy normal person's left foot plantar pressure feature L ₆as shown in Fig. 3 (b2), healthy normal person's right crus of diaphragm plantar pressure feature R ₃as shown in Fig. 3 (b3), healthy normal person's right crus of diaphragm plantar pressure feature R ₆as shown in Fig. 3 (b4).

Step 2, the gait feature variable extracted according to step 1, dynamically carry out modeling to the unknown nonlinear gait system of normal person healthy in training set and Parkinson's disease patients, design RBF neural identifier realizes accurately approaching the unknown dynamic local of gait system:

(1) unknown nonlinear gait system dynamic modeling is as follows:

$\stackrel{\&CenterDot;}{x}=F(x;p)+v(x;p)$

Wherein, x=[x ₁..., x _n] ^t∈ R ⁿbe the plantar pressure characteristic variable that step 1 is extracted, p is the normal parameter value of system, and n is the dimension of plantar pressure characteristic variable;

F (x; P)=[f ₁(x; P) ..., f _n(x; P)] ^tbe smooth and the Nonlinear Dynamic state variable of the unknown, the gait system representing healthy normal person and Parkinson's disease patients is dynamic, v (x; P)=[v ₁(x; P) ..., v _n(x; P)] ^tmodeling indeterminate, due to modeling indeterminate v (x; P) with the dynamic F (x of gait system; P) cannot decoupling zero mutually, therefore the two is merged into one and it is dynamic to be defined as general nonlinearity gait system;

(2) design neural network identifier and be used for identification:

Adopt dynamic RBF neural network constructing neural network identifier, as shown in Figure 4, dynamic RBF neural network identifier form is as follows for the topological structure sketch dynamically learnt non-linear gait system:

$\stackrel{\&CenterDot;}{\hat{x}}=-A(\hat{x}-x)+{\hat{W}}^{T}S\left(x\right),$

Wherein the state of neural network identifier, the plantar pressure characteristic variable namely chosen; A=diag [a ₁..., a _n] be diagonal matrix, a _ibe the constant of design, meet 0 < | a _i| < 1, such as, get 0.5; be dynamic RBF neural network, the general nonlinearity gait system being used for approaching the unknown is dynamic s (x)=[S ₁(|| X-ξ ₁|| ..., S _n(|| X-ξ _n||] ^tbe Gaussian radial basis function, N > 1 is neural network number of network nodes, such as, get N=83521, ξ _ibe neuronal center point, neuron is evenly distributed within region [-1,1] × [-1,1] × [-1,1] × [-1,1], and width gets 0.15; All plantar pressure characteristics are normalized to [-1,1] interval; RBF neural weights adjustment rule as follows:

${\stackrel{\&CenterDot;}{\hat{W}}}_i=-{\mathrm{\&Gamma;}}_{i}S\left(x\right){\tilde{x}}_i-{\mathrm{\&sigma;}}_i{\mathrm{\&Gamma;}}_i{\hat{W}}_i,i=1,...,n,$

Wherein, i represents that n ties up the i-th dimension variable in plantar pressure characteristic variable, state error, Γ _i=Γ _i ^t> 0, σ _i> 0 is the regulating parameter regulating rule, the weights of dynamic RBF neural network initial value

Can be represented by following formula general nonlinearity gait system dynamic local accurate modeling:

Wherein, ε _i1it is approximate error; Here local accurate modeling refers to that the internal dynamic away from track is not then approached by RBF neural approaching the built-in system dynamic trajectory along gait feature data;

The adjustment rule of above-mentioned RBF neural weights is according to Li Yapuluofu stability theorem and determines the theories of learning to design, make all bounded the exponential convergences of state error and weights estimation, wherein the weight convergence of RBF neural has two kinds of situations: the neuron along the RBF neural of gait feature data regression track meets persistent excitation condition, and its weight convergence is in the small neighbourhood of optimal value; Away from the neuron not excited target and not being conditioned of the RBF neural of gait feature data regression track, its weights are approximately zero.

Such as within a period of time, weight convergence is to constant value (optimal value), and as shown in Figure 5, the neuronic weights near system trajectory meet part persistent excitation condition to the convergence situation of its learning phase neural network weight, thus converge to its optimal value; And very little and be conditioned hardly away from the degree of the neuron excited target of system trajectory, remain essentially in the small neighbourhood of zero;

Step 3, set up constant value neural network

Constant value neural network constant and space distribution when being, that is: effective information is only stored on the neuron of the built-in system dynamic trajectory of gait feature data, and does not have storage information away from the neuron of track, constant value neural network only approach the internal dynamic along gait feature data space track, the internal dynamic away from track is not approached; Therefore, according to determining the theories of learning, the neuron along the RBF neural of gait system features track meets persistent excitation condition, and its weight convergence is to optimal value, get the average of weights in a period of time after weight convergence as learning training result, and utilize these results to set up constant value neural network the gait system dynamic acquired is gained knowledge with the storage of the form of constant value neural network weight, forms a training gait pattern storehouse; Described constant value neural network weight levy by as shown in the formula sublist: wherein, [t _a, t _b] represent constant value neural network weight completing to a time period after the transient process that its optimal value restrains, make like this can by constant value neural network carry out local accurately to approach:

wherein, ε _i2it is approximate error;

Step 4, gathered the plantar pressure characteristic of each Parkinson's disease patients to be measured respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the Parkinson's disease patients some to be measured of collection forms test set;

Step 5, utilize constant value neural network build one group of dynamic estimator, the gait feature data of Parkinson's disease patients to be detected in test set and this are organized dynamic estimator and does difference, form one group of error in classification, detect according to the abnormal gait Accurate classification of least error principle by Parkinson's disease patients to be detected, realize the auxiliary detection to parkinsonism, concrete steps are as follows:

(1) according to the dynamic RBF neural identification result of general nonlinearity gait system of healthy normal person and Parkinson's disease patients in training gait pattern storehouse, i.e. constant value neural network weight construct one group of dynamic estimator, by step 2 and 3 study to healthy normal persons and the gait system dynamic of Parkinson's disease patients gain knowledge and be embedded in dynamic estimator, be expressed as follows:

${{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k=-b_i({{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k-x_{\mathrm{ti}})+{{\stackrel{\&OverBar;}{W}}_i}^{k^T}S\left(x_{\mathrm{ti}}\right),i=1,...,n,k=1,...,M,$

Wherein, χ _ifor the state of dynamic estimator, b _ifor dynamic estimator parameter, x _tifor the gait feature data of Parkinson's disease patients to be detected in test set, k represents the kth training mode in M training mode, M is the pattern total amount in training gait pattern storehouse, the plantar pressure characteristic sequence that healthy normal person and Parkinson's disease patients extract in walking process each time just forms a pattern, in process of the test, subjects has been walked how many times, and the corresponding plantar pressure characteristic sequence extracted just constitutes how many patterns;

(2) by the gait feature data x of Parkinson's disease patients to be detected in test set _tiorganize dynamic estimator with this and do difference, obtain following classification and Detection error system:

Wherein, be state estimation error, calculate average L ₁norm is as follows:

${\left|\right|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(t\right)\left|\right|}_1=\frac{1}{T_c}{\&Integral;}_{t-T_c}^t\left|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(\mathrm{\&tau;}\right)\right|\mathrm{d\&tau;},t\&GreaterEqual;T_c,$

Wherein, T _crepresent gait cycle;

(3) if in test set Parkinson's disease patients to be detected gait pattern similar in appearance to training gait pattern s (s ∈ 1 ..., k}), then embed the constant value RBF neural in dynamic estimator s the knowledge learned can be remembered fast and accurately approaching gait dynamics is provided; Therefore, corresponding error in all errors in become minimum, based on least error principle, the abnormal gait of this Parkinson's disease patients to be detected can be detected by Fast Classification, and classification and Detection strategy is as follows:

If there is a finite time t ^s, s ∈ 1 ..., k} and a certain i ∈ 1 ..., n}, makes to all t > t ^sset up, then the abnormal gait pattern of the Parkinson's disease patients to be detected occurred can be classified and detect, and realizes the auxiliary detection to parkinsonism.

The performance index such as sensitivity (Sensitivity), specificity (Specificity) and accuracy (Accuracy) are utilized to assess classification and Detection result, being calculated as follows of these indexs:

$\mathrm{Sensitivity}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}\&times;100\%,$

$\mathrm{Specificity}=\frac{\mathrm{TN}}{\mathrm{TN}+\mathrm{FP}}\&times;100\%,$

$\mathrm{Accuracy}=\frac{\mathrm{TP}+\mathrm{TN}}{\mathrm{TP}+\mathrm{TN}+\mathrm{FN}+\mathrm{FP}}\&times;100\%,$

Wherein, TP represents real positive sample, and TN represents real negative sample, and FP represents false positive sample, and FN represents false negative sample.

In the present embodiment: TP=92, TN=73, FN=1, FP=0.

Following table is the classification and Detection result table of Parkinson's disease patients and healthy population:

Performance index	Result (%)
		Sensitivity	98.92
Specificity	100
		Accuracy	99.40

Above-described embodiment is the present invention's preferably embodiment; but embodiments of the present invention are not restricted to the described embodiments; change, the modification done under other any does not deviate from Spirit Essence of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (5)

1. an abnormal gait recognition methods for auxiliary examination parkinsonism, is characterized in that comprising following steps:

Step 1, gathered the plantar pressure characteristic of each Parkinson's disease patients and healthy normal person respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the some Parkinson's disease patients gathered and healthy normal person forms training set;

Step 2, the gait feature variable extracted according to step 1, dynamically carry out modeling to the unknown nonlinear gait system of normal person healthy in training set and Parkinson's disease patients, design RBF neural identifier, approaches the unknown dynamic local of gait system;

The foundation of step 3, constant value neural network:

According to determining the theories of learning, neuron along the RBF neural of gait system features track meets persistent excitation condition, its weight convergence is to optimal value, get the average of weights in a period of time after weight convergence as learning training result, and utilize these results to set up constant value neural network, the gait system dynamic acquired is gained knowledge with the storage of the form of constant value neural network weight, forms a training gait pattern storehouse;

Step 4, gathered the plantar pressure characteristic of each Parkinson's disease patients to be measured respectively by plantar pressure sensor, form one group of gait feature variable, the plantar pressure characteristic of the Parkinson's disease patients some to be measured of collection forms test set;

Step 5, classification and Detection:

Constant value neural network is utilized to build one group of dynamic estimator, non-linear gait system dynamic corresponding to healthy normal person in the training gait pattern storehouse that step 2 and step 3 learn and Parkinson's disease patients is gained knowledge to be embedded in dynamic estimator, the gait feature data of Parkinson's disease patients to be detected and this are organized dynamic estimator and does difference, form one group of error in classification, detect according to the abnormal gait of least error principle by Parkinson's disease patients to be detected, realize the auxiliary detection to parkinsonism.

2. the abnormal gait recognition methods of auxiliary examination parkinsonism according to claim 1, it is characterized in that, in step 1 and 4, adopt gait analysis database PhysioNet, the plantar pressure characteristic of each 8 symmetric positions of human body left and right pin is gathered by the pressure transducer being placed on experimenter's sole, choose the plantar pressure characteristic of the most obvious about the two groups pin symmetric positions of wherein difference, form one group of plantar pressure characteristic variable.

3. the abnormal gait recognition methods of auxiliary examination parkinsonism according to claim 1, is characterized in that: in step 2, and unknown nonlinear gait system dynamic modeling is as follows:

$\stackrel{.}{x}=F(x;p)+v(x;p)$

Wherein, x=[x ₁..., x _n] ^t∈ R ⁿbe the plantar pressure characteristic variable that step 1 is extracted, p is the normal parameter value of system, and n is the dimension of plantar pressure characteristic variable;

F (x; P)=[f ₁(x; P) ..., f _n(x; P)] ^tbe smooth and the Nonlinear Dynamic state variable of the unknown, the gait system representing healthy normal person and Parkinson's disease patients is dynamic, v (x; P)=[v ₁(x; P) ..., v _n(x; P)] ^tbe modeling indeterminate, the two is merged into and it is dynamic to be defined as general nonlinearity gait system;

(2) design neural network identifier and be used for identification:

Adopt dynamic RBF neural network constructing neural network identifier, dynamic RBF neural network identifier form is as follows: $\stackrel{\stackrel{.}{^}}{x}=-A(\hat{x}-x)+{\hat{W}}^{T}S\left(x\right),$ Wherein $\hat{x}=[{\hat{x}}_1,...,{\hat{x}}_n]$ It is the plantar pressure characteristic variable chosen; A=diag [a ₁..., a _n] be diagonal matrix, a _ibe the constant of design, meet 0 < | a _i| < 1, be dynamic RBF neural network, the general nonlinearity gait system being used for approaching the unknown is dynamic s (x)=[S ₁(‖ X-ξ ₁‖ ..., S _n(‖ X-ξ _n‖] ^tbe Gaussian radial basis function, N > 1 is neural network number of network nodes, ξ _ineuronal center point, RBF neural weights adjustment rule as follows:

${\stackrel{\stackrel{.}{^}}{W}}_i=-{\mathrm{\&Gamma;}}_{i}S\left(x\right){\tilde{x}}_i-{\mathrm{\&sigma;}}_i{\mathrm{\&Gamma;}}_i{\hat{W}}_i,i=1,...,n,$

Wherein, i represents that n ties up the i-th dimension variable in plantar pressure characteristic variable, state error, Γ _i=Γ _i ^t> 0, σ _i> 0 is the regulating parameter regulating rule, the weights of dynamic RBF neural network initial value

Constant value neural network is set up in step 3 specific practice:

According to determining the theories of learning, neuron along the RBF neural of gait system features track meets persistent excitation condition, its weight convergence, to optimal value, is got the average of weights in a period of time after weight convergence as learning training result, and is utilized these results to set up constant value neural network the gait system dynamic acquired is gained knowledge with the storage of the form of constant value neural network weight, forms a training gait pattern storehouse; Described constant value neural network weight levy by as shown in the formula sublist: wherein, [t _a, t _b] represent constant value neural network weight completing to a time period after the transient process that its optimal value restrains, make like this can by constant value neural network carry out local accurately to approach:

wherein, ε _i2it is approximate error;

Classification and Detection in step 5, specific as follows:

According to the constant value neural network weight of healthy normal person and Parkinson's disease patients in training gait pattern storehouse construct one group of dynamic estimator, be expressed as follows:

${{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k=-b_i({{\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_i}^k-x_{\mathrm{ti}})+{{{\stackrel{\&OverBar;}{W}}_i}^k}^{T}S\left(x_{\mathrm{ti}}\right),i=1,...,n,k=1,...,M,$

Wherein, χ _ifor the state of dynamic estimator, b _ifor dynamic estimator parameter, x _tifor the gait feature data of Parkinson's disease patients to be detected in test set, k represents the kth training mode in M training mode, and M is the pattern total amount in training gait pattern storehouse;

By the gait feature data x of Parkinson's disease patients to be detected in test set _tiorganize dynamic estimator with this and do difference, obtain following classification and Detection error system:

Wherein, be state estimation error, calculate average L ₁norm is as follows:

${\left|\right|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(t\right)\left|\right|}_1=\frac{1}{T_c}{\&Integral;}_{t-T_c}^t\left|{{\widetilde{\mathrm{\&chi;}}}_i}^k\left(\mathrm{\&tau;}\right)\right|\mathrm{d\&tau;},t\&GreaterEqual;T_c,$

Wherein, T _crepresent gait cycle;

Classification and Detection strategy is as follows: if there is a finite time t ^s, s ∈ 1 ..., k} and a certain i ∈ 1 ..., n}, makes to all t > t ^sset up, then the abnormal gait pattern of the Parkinson's disease patients to be detected occurred is classified and detects.

4. the abnormal gait recognition methods of auxiliary examination parkinsonism according to claim 3, it is characterized in that: the adjustment rule of described RBF neural weights is according to Li Yapuluofu stability theorem and determines the theories of learning to design, make state error and weights estimation all bounded, and exponential convergence, wherein, the weight convergence of RBF neural has two kinds of situations:

The first situation: the neuron returning the RBF neural of track along gait feature meets persistent excitation condition, and its weight convergence is in the small neighbourhood of optimal value;

The second situation: the neuron not excited target and not being conditioned returning the RBF neural of track away from gait feature, its weights are approximately zero.

5. the abnormal gait recognition methods of auxiliary examination parkinsonism according to claim 3, is characterized in that: can be represented by following formula general nonlinearity gait system dynamic local accurate modeling:

Wherein, ε _i1it is approximate error; Here local accurate modeling refers to that the internal dynamic away from track is not then approached by RBF neural approaching the built-in system dynamic trajectory along gait feature data.