CN110464349A - A kind of upper extremity exercise function score method based on hidden Semi-Markov Process - Google Patents

Abstract

The invention discloses an upper limb motor function scoring method based on a hidden semi-Markov model, which comprises the steps of: (1) selecting a standard motor function evaluation action; (2) collecting the upper limb pose and posture of the unaffected upper limb of the patient performing the motor function evaluation action and EMG data sets; (3) training a hidden semi-Markov model suitable for motor function assessment; (4) using the model in (3) for motor function assessment. The invention can evaluate the recovery degree of upper limb motor function of stroke patients with hemiplegia, and to a certain extent replace the method of empirical evaluation by rehabilitation physicians using the motor function evaluation scale, thereby reducing the work intensity of rehabilitation physicians and assisting physicians in their work. , to improve the efficiency of physicians.

Description

A scoring method for upper limb motor function based on hidden semi-Markov model

technical field

The invention belongs to the field of rehabilitation medicine, in particular to an upper limb motor function scoring method based on a hidden semi-Markov model.

Background technique

Stroke patients often have defects in the motor function of the upper limbs, and the clinical manifestations are hemiplegia symptoms such as muscle or limb flaccidity, spasticity, and slow movement. Due to the shortage of rehabilitation treatment resources in society, a large number of stroke disabled patients cannot receive timely and effective rehabilitation treatment. At the same time, treatment costs remain high. Therefore, it is of great significance to use appropriate computer methods to replace the therapist to effectively evaluate the motor function recovery status of hemiplegic patients to a certain extent.

Methods commonly used in clinical rehabilitation medicine to assess limb motor function include the Brunnstrom scale and the Fugel-Meyer scale. The Brunnstrom Scale divides the recovery of upper limb motor function in stages according to muscle tension and exercise ability, and divides it into 6 stages. In the first stage, the muscle tension level of patients is almost zero, complete flaccid paralysis, no response or extremely slow response, no Any exercise ability; in the second stage, the patient has spasm, combined reaction, the muscles of the affected limb can contract slightly, but the joint cannot move; in the third stage, the muscle spasm is severe, and joint movement and compensatory movement can be initiated; in the fourth stage, the muscle spasm At the beginning of the weakening, the affected limb can move away from common movement and separate movement; in the fifth stage, the degree of spasticity of the patient is further weakened, and the ability of separation movement is enhanced; in the sixth stage, the patient's affected limb basically recovers, close to the normal level, but the movement flexibility is slightly weaker than that of the healthy side . The Fugel-Meyer scale is a standardized assessment test for motor recovery in hemiplegic patients proposed by Axel Fugel-Meyer and colleagues. It is widely used together with the Brunnstrom scale for the clinical assessment of motor function. The Fugl-Meyer assessment scale was found to have excellent consistency, responsiveness and good accuracy.

Hidden Semi-Markov Model (HSMM) is an extended model of Hidden Markov Model (HMM). Compared with the hidden Markov model, the hidden semi-Markov model further considers the state residence probability distribution, and adds the residence time on the basis of the hidden Markov model. The hidden semi-Markov model has more advantages in solving practical problems. Good modeling ability and evaluation analysis ability. The hidden semi-Markov model λ is described by parameters (N, M, π, A, B, P), where N is the number of hidden states, M is the length of the observation sequence, the initial state probability distribution vector π, the state transition probability matrix A, State residence time distribution matrix P, observation value probability matrix B. The hidden semi-Markov model can be written as λ=(N,M,π,A,B,P).

Contents of the invention

Purpose of the invention: Aiming at the above problems, a method for scoring upper limb motor function based on hidden semi-Markov model is proposed

For realizing the purpose of the present invention, the technical solution adopted in the present invention is: its concrete steps are as follows:

(1) Planning standard upper limb motor function evaluation actions;

(2) Using the healthy limb of the patient to collect the upper limb pose and EMG data set of the standard upper limb motor function evaluation action;

(3) Using upper limb pose and EMG data sets to train a hidden semi-Markov model suitable for motor function assessment;

(4) Collect the upper limb posture and myoelectric data of the patient's affected limb performing standard motor function evaluation actions;

(5), utilize forward-to-backward algorithm in combination with the model described in step (3) to obtain the likelihood probability value of data in step (4);

(6), according to FMA calculation patient's uninjured side upper limb motor function score;

(7) According to step (6) and step (5), calculate the motor function score of the upper limb of the affected side of the patient.

Further, the step (1) specifically includes:

The patient sits with the waist straight and the arms hang down naturally;

(1.1), the initial posture is to place the palm of the hand on the knee of the lower limb on the opposite side, then slowly bend the elbow joint, and slowly lift and abduct the shoulder joint until the hand touches the ear on the same side;

(1.2), the initial posture is to place the palm of the hand on the knee of the lower limb on the same side, and then perform the movement of touching the back of the hand to the lumbosacral region;

(1.3), the initial posture is that the arms are naturally drooping, the shoulder joints are abducted 90 degrees, and the elbow joints are kept in a stretched state;

(1.4), the initial posture is that the arm is naturally drooping, the elbow joint is slowly flexed to 90 degrees and the ground level, and the forearm is pronated 90 degrees and supinated 90 degrees;

(1.5), the initial posture is that the arms are naturally drooping, the shoulder joints are slowly stretched forward 90 degrees to the level of the arms and the ground, and the elbow joints are kept extended;

(1.6), the initial posture is that the arms are naturally drooping, the shoulder joints are slowly stretched forward 45 degrees, while the elbow joints are kept stretched, the palms are fully extended, and the upper arms are pronated 90 degrees and supinated 90 degrees;

(1.7) The initial posture is that the arms are drooping, the palms are fully extended and the palms are facing forward, the upper arms are raised forward while the elbow joints are flexed and stretched to the inside of the palms to touch the ears on the same side;

(1.8) The initial posture is that the arms are drooping, and the upper arms are lifted forward to 180 degrees above the head, while the elbow joints are kept in a stretched state.

Further, in the step (2): collecting the posture and the surface electromyography amplitude signal of the associated muscles when the upper arm and the forearm of the patient's healthy side upper limb perform standard assessment exercises; specifically include:

(2.1), install attitude measurement sensors on the upper arm and forearm of the healthy side respectively;

(2.2), several surface electromyography sensors are installed on the uninjured upper limb movement-related muscles;

(2.3), the healthy limbs of the patient are evaluated according to all the standards in step (1), and the movement actions are respectively performed, and the posture data and surface EMG amplitude data of the limbs during the movement are collected and recorded;

(2.4), repeat the step (2.3) multiple times, and collect multiple sets of data to enhance the robustness of the model.

Further, the step (3) specifically includes:

(3.1), adding window to extract the original signal features of surface electromyography;

(3.2), the attitude data collected each time is combined with the myoelectric feature data to form an observation matrix;

(3.3), aligning multiple groups of observation matrix data to form an observation matrix set, ensuring that the rows and columns of each group of matrix in the observation matrix set are consistent;

(3.4), using the observation matrix set combined with the EM algorithm or the Baum-Welch algorithm to train the hidden semi-Markov model, and obtain a single action hidden semi-Markov model suitable for the patient's motor function assessment;

(3.5) For multiple different actions, use the observation matrix set corresponding to the action and repeat step (3.4).

Further, the step (4) is to collect the upper limb posture and myoelectric data of the patient's affected limb performing standard motor function evaluation actions, specifically including:

(4.1), install attitude measurement sensors on the upper arm and forearm of the affected side respectively;

(4.2), several surface electromyographic sensors are installed on the movement-related muscles of the upper limb of the affected side;

(4.3), the affected limbs of the patient are evaluated according to all the standards in step (1), and the movement actions are respectively performed, and the posture data and surface EMG amplitude data of the limbs during the movement are collected and recorded;

(4.4) Combining the collected limb posture data and myoelectric feature data to form an observation matrix for each standard evaluation action.

Further, the step (5) uses the forward-backward algorithm to calculate the likelihood probability value of the motor function data of the upper limb of the affected side relative to the motor function data of the upper limb of the healthy side in step (4) according to the model described in the step (3). It is: using the hidden semi-Markov model suitable for motor function evaluation, and an observation matrix composed of posture and myoelectric data of the upper limb motor function evaluation action performed by the affected limb, combined with a forward-backward algorithm, Obtain the likelihood probability value of the upper limb motor function data of the affected side relative to the upper limb motor function data of the healthy side, which includes:

(5.1), using the specific model λ=(N, M, π, A, B, P) obtained in step (3) as the parameters of the forward-backward algorithm, using each group of standards collected in step (4) Evaluate the action observation matrix as the current observation value of the forward-backward algorithm, and output the result of the forward-backward algorithm, that is, the likelihood probability value of the current observation value;

(5.2), all adopt (5.1) described method to obtain its likelihood probability value to each group of actions.

Further, the step (6) calculates the motor function score of the healthy side upper limb of the patient according to the FMA: the treating physician reviews the recorded video tape of the patient's healthy side upper limb performing the described motor function evaluation action and fills in a Likert questionnaire to evaluate the patient The uninjured upper limb performed the performance during the motor function evaluation described above, and the FMA was used to calculate the motor function score of the uninjured limb:

(6.1), collecting the upper limbs of the healthy side to perform each standard evaluation action movement posture and associated muscle electromyographic data;

(6.2), utilize step (4.4) method, obtain the observation matrix of single action;

(6.3), using the forward-backward algorithm in combination with the trained model in step (3) to output the likelihood probability value of the uninjured upper limb movement as the maximum likelihood probability value;

(6.4), utilize the likelihood probability value that step (5.2) method draws and the maximum likelihood probability value that obtains in step (6.3) to ask for full marks 100 points, the lowest evaluation sports score of 0 points;

(6.5). For a plurality of different evaluation actions, the evaluation score of each evaluation action can be obtained by repeating steps (6.1) to (6.4).

The step (7) uses step (6) and step (5) to calculate the motor function score of the affected upper limb of the patient as follows: first, for the same rehabilitation evaluation action, according to step (5), calculate the motor function data of the affected upper limb of all patients Relative to the likelihood probability value of the upper limb motor function data of the healthy side, and through normalization processing, the normalized likelihood probability value of each patient's affected side for each specific rehabilitation evaluation action is obtained; secondly, according to the steps (6 ) score of the healthy side of the patient, to obtain the highest and lowest scores of the healthy side of the patient for each rehabilitation assessment action; finally, the likelihood probability value obtained by normalization and the highest and lowest scores of the healthy side of the patient for each rehabilitation assessment action , to obtain the motor function score of the affected side of the patient on the rehabilitation evaluation action.

Beneficial effects of the present invention: the present invention can evaluate the degree of recovery of the motor function of the upper limbs of stroke hemiplegia patients, and to a certain extent replace the method of empirical evaluation by rehabilitation physicians using the motor function evaluation scale, thereby reducing the workload of rehabilitation physicians. Intensity, assisting the work of the treating physician, and improving the efficiency of the physician's work.

Description of drawings

Fig. 1 is the block diagram of implementation process of the present invention;

Fig. 2 is a scheme diagram of acquisition of upper limb motion posture in the present invention;

Fig. 3 is an installation plan diagram of the surface electromyography sensor of the muscle associated with upper limb movement in the present invention.

Detailed ways

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

First, select 8 standard evaluation actions involving the upper limb shoulder joint and elbow joint movement function, and then install posture sensors and surface electromyography sensors on the healthy limbs of the patient, and guide the patients to perform the 8 standard evaluation actions respectively. The electrical sensors record the physiological data during exercise. In order to ensure the robustness of the system, each standard evaluation action is repeated multiple times to collect multiple sets of exercise physiological data; then the data of the posture sensor and the surface electromyography sensor are extracted during the exercise of the healthy limb. The eigenvalues are used to form an observation matrix set, and the hidden semi-Markov model is trained using the EM algorithm or the Baum-Welch algorithm; the posture sensor and the surface electromyography sensor are installed on the affected limb of the patient again, and the patient is guided to perform the evaluation actions according to the five standards , the attitude sensor and the surface electromyography sensor respectively record the physiological data during exercise, and extract the data feature values of the attitude sensor and the surface electromyography sensor during the movement of the affected limb to form an observation matrix; finally, use the forward-backward algorithm and the trained The hidden semi-Markov model outputs the likelihood probability value based on the observation matrix of the affected limb, and calculates the assessment movement score.

As shown in Figure 1, 1, a kind of upper limb motor function scoring method based on hidden semi-Markov model, comprises the following steps:

(1) Planning the upper limb motor function assessment action;

(2) According to the upper limb motor function evaluation action, collect the posture and EMG data of the healthy side of the patient's upper limb;

(3) Use the uninjured upper limb pose and EMG dataset to train a hidden semi-Markov model suitable for motor function assessment;

(4) Collect the posture and myoelectric data of the patient's affected upper limb performing motor function evaluation actions;

(5), according to the model described in step (3), utilize the forward-backward algorithm to calculate the likelihood probability value of the motor function data of the affected upper limb relative to the upper limb motor function data of the healthy side in the step (4);

(6), according to FMA calculation patient's uninjured side upper limb motor function score;

(7) According to step (6) and step (5), calculate the motor function score of the upper limb of the affected side of the patient.

The step (1) upper limb motor function evaluation action consists of the following evaluation actions:

(2.1), forearm pronation and supination movement;

(2.2), joint movement of flexor muscles;

(2.3), touch the lumbar spine movement;

(2.4), shoulder flexion 90° movement;

(2.5), arm pronation and supination movement;

(2.6), 90° movement of shoulder joint abduction;

(2.7), shoulder flexion 90°-180° movement;

(2.8), flex the shoulder joint until the palm of the upper limb touches the ear on the same side.

The step (2) is to evaluate the action according to the upper limb motor function, and collect the pose and EMG data of the healthy side of the patient's upper limb as follows: install the attitude sensor on the upper arm and forearm of the healthy side of the upper limb, and collect the data of the patient performing the upper limb movement. The movement posture of the limbs during the function evaluation action; the surface electromyography sensor is installed on the muscle group associated with the movement, and the surface electromyography signal of the affected limb is collected when performing the upper limb motor function evaluation action; in order to ensure the robustness of the system , the posture and EMG data of the same group of upper limb movement evaluation actions can be collected several times.

The step (3) utilizes the posture and posture of the patient's healthy upper limb and the EMG data set to train the hidden semi-Markov model suitable for motor function evaluation, including: evaluating the movement of the upper limb motor function, using the collected patient's healthy limb motion posture and Using the surface EMG data, construct a multi-dimensional hidden semi-Markov model explicit state observation matrix; for the explicit state observation matrix, use EM algorithm or Baum-Welch algorithm to train a hidden semi-Markov model suitable for a single evaluation action.

The step (4) collecting the pose and myoelectric data of the patient's affected upper limb performing the motor function evaluation action is as follows: the posture sensor is installed on the upper arm and the forearm of the patient's affected limb, and the affected limb is collected while performing the upper limb The movement posture angle during the movement assessment action; the surface electromyography sensor is installed on the muscle group of the affected limb associated with the movement assessment, and the surface electromyography signal of the affected limb is collected when performing the upper limb movement assessment action.

According to the model described in step (3), the step (5) uses the forward-backward algorithm to calculate the likelihood value of the upper limb motor function data of the affected side relative to the upper limb motor function data of the healthy side in step (4): using The hidden semi-Markov model suitable for motor function evaluation, and the observation matrix composed of posture and myoelectric data of the affected limb performing the upper limb motor function evaluation action, combined with the forward-backward algorithm, obtain the affected side The likelihood probability value of upper limb motor function data relative to uninjured upper limb motor function data.

The step (6) calculates the motor function score of the patient's healthy side upper limb according to the FMA: the treating physician reviews the recording videotape of the patient's healthy side upper limb performing the described motor function evaluation action and fills in a Likert questionnaire to evaluate the patient's healthy side upper limb Performance during the motor function assessment was performed as described, and the FMA was used to calculate the motor function score of the healthy limb.

The step (7) uses step (6) and step (5) to calculate the motor function score of the affected upper limb of the patient as follows: first, for the same rehabilitation evaluation action, according to step (5), calculate the motor function data of the affected upper limb of all patients Relative to the likelihood probability value of the upper limb motor function data of the healthy side, and through normalization processing, the normalized likelihood probability value of each patient's affected side for each specific rehabilitation evaluation action is obtained; secondly, according to the steps (6 ) score of the healthy side of the patient, to obtain the highest and lowest scores of the healthy side of the patient for each rehabilitation assessment action; finally, the likelihood probability value obtained by normalization and the highest and lowest scores of the healthy side of the patient for each rehabilitation assessment action , to obtain the motor function score of the affected side of the patient on the rehabilitation evaluation action.

Concrete operation of the present invention comprises the following steps:

(1) Selection of standard motor function assessment actions:

(1.1) The initial posture is to place the palm of the hand on the knee of the lower limb on the opposite side, then slowly bend the elbow joint, and slowly lift and abduct the shoulder joint until the hand touches the ear on the same side;

(1.2), the initial posture is to place the palm of the hand on the knee of the lower limb on the same side, and then perform the movement of touching the back of the hand to the lumbosacral region;

(1.3), the initial posture is that the arms are naturally drooping, the shoulder joints are abducted 90 degrees, and the elbow joints are kept in a stretched state;

(1.4), the initial posture is that the arm is naturally drooping, the elbow joint is slowly flexed to 90 degrees and the ground level, and the forearm is pronated 90 degrees and supinated 90 degrees;

(1.5), the initial posture is that the arms are naturally drooping, the shoulder joints are slowly stretched forward 90 degrees to the level of the arms and the ground, and the elbow joints are kept extended;

(1.6), the initial posture is that the arms are naturally drooping, the shoulder joints are slowly stretched forward 45 degrees, while the elbow joints are kept stretched, the palms are fully extended, and the upper arms are pronated 90 degrees and supinated 90 degrees;

(1.7) The initial posture is that the arms are drooping, the palms are fully extended and the palms are facing forward, the upper arms are raised forward while the elbow joints are flexed and stretched to the inside of the palms to touch the ears on the same side;

(1.8) The initial posture is that the arms are drooping, and the upper arms are lifted forward to 180 degrees above the head, while the elbow joints are kept in a stretched state.

(2) Collection of standard exercise data on the healthy side of the upper limbs of hemiplegic patients:

(2.1) Keep the patient's healthy upper limb in a natural drooping state;

(2.2), install two attitude sensors, respectively on the upper arm and the forearm;

(2.3), install 5 surface electromyographic sensors, and install 1 sensor on the surface of 5 muscles related to exercise: biceps brachii, deltoid, deltoid-scapula, pronator, and supinator;

(2.4), initialize and clear the sensor;

(2.5), instructing the patient to perform the 8 standard motion evaluation actions described in step (1) respectively using the upper limb of the healthy side, and the posture sensor and the surface myoelectric sensor collect the motion posture data and the original amplitude data of the myoelectricity respectively;

(2.6), repeat the steps (2.4) and (2.5) for a total of K times in sequence, and collect and record the data of K groups of different standard motion evaluation actions.

(3) Construct a hidden semi-Markov model suitable for patient motion assessment:

(3.1), each group of each group of each myoelectric original amplitude value data collected in step (2.5) is carried out to add window respectively to seek eigenvalue, obtains eigenvalue sequence, gets suitable window sliding value to guarantee each eigenvalue obtained The sequence length is consistent with the attitude angle data sequence length;

(a), in particular, extract the mean value, root mean square value, first-order difference standard deviation, variance, first-order difference mean value, average power frequency, standard deviation, zero-crossing points for each group of each myoelectric raw amplitude data, There are 12 eigenvalue sequences including absolute value slope, integral EMG value, first-order difference median value, and median frequency. The eigenvalue matrix O _kEi (m _k × 12) of a single original EMG signal, where k is the number of acquisitions, 1≤k≤K, i is the serial number of the movement-related muscles 1≤i≤4, and m _k is the kth time The length of the acquired sequence;

(b), the two attitude sensors collect the movement attitude quaternion data of the big arm and the forearm respectively, and obtain a single movement attitude quaternion matrix O _kQj (m _k × 4), where j is the serial number of the two attitude sensors 1 ≤j≤2;

(c), the signal eigenvalue matrix O _kEi (m _k × 12) collected by 5 surface electromyography sensors forms the observation matrix O _kE (m _k × 60) of the myoelectric characteristic signal collected in a single time, O _kE = (O _kE1 ,O _kE2 ,O _kE3 ,O _kE4 ,O _kE5 );

(d), the quaternion matrix O _kQj (m _k × 4) collected by two attitude sensors constitutes the observation matrix O _kQ (m _k × 8) of the myoelectric characteristic signal acquired in a single acquisition, O _kQ = (O _kQ1 , O _kQ2 );

(e), as described in step (3.1), when performing sliding windowing on the original EMG signal, it is necessary to ensure that the length of the feature sequence is consistent with the length of the quaternion sequence output by the attitude sensor, and both are m _k , therefore, a single original EMG signal The eigenvalue matrix O _kE of is consistent with the length of a single motion posture data matrix O _kQ , forming the observation value matrix O _k of a single observation;

(f), repeat steps (a) to (e) for a total of K times of motion, and obtain the training matrix set O={O ₁ ,...,O _K } of the hidden semi-Markov model

(3.2) Due to the inconsistency of the sequence length in each collection, the length of each observation sequence M={m ₁ ,m ₂ ,…,m _K } is different, and the DTW algorithm is used for O={O ₁ ,…,O Each matrix in _K } performs an alignment operation;

(3.3), use the K-means clustering algorithm to cluster the observation matrix set O, take G cluster centers, and obtain the mean matrix μ, covariance matrix U, and weight matrix W of each cluster;

(3.4), initialize the hidden semi-Markov model parameters:

(a), initial state probability distribution vector π=norm(rand(N×1));

(b), state transition probability matrix A=norm(rand(N×N));

(c), the observation value probability matrix B is initialized with a multivariate Gaussian mixture model;

(d), the state residence time distribution matrix P is initialized with a single Gaussian model.

In the above formula, norm(...) is a normalization function, and rand(...) is a random number generation function;

(3.5), using the hidden semi-Markov model λ=(N, M, π, A, B, P) initialized in step (3.4) combined with EM algorithm or Baum-Welch algorithm to set the threshold iteratively reevaluate the hidden semi-Markov Cove model The optimal hidden semi-Markov model suitable for motor function assessment of patients was obtained.

(4) Collect the upper limb posture and myoelectric data of the patient's affected upper limb performing standard motor function evaluation actions:

(4.1) Keep the patient's affected upper limb in a natural drooping state;

(4.2), install two attitude sensors, respectively on the upper arm and the forearm;

(4.3), install 5 surface electromyography sensors, and install 1 on the surface of 5 muscles related to exercise: biceps, deltoid, deltoid-scapula, pronator, and supinator;

(4.4), initialize and clear the sensor;

(4.5) Instruct the patient to use the upper limb of the affected side to perform the 8 standard motion evaluation actions described in step (1), and the posture sensor and the surface electromyography sensor collect the motion posture data and the original amplitude data of the myoelectricity respectively;

(4.6), construct the observation value matrix X of the affected limb:

(a), in particular, extract the mean value, root mean square value, first-order difference standard deviation, variance, first-order difference mean value, average power frequency, standard deviation, zero-crossing points for each group of each myoelectric raw amplitude data, There are 12 eigenvalue sequences including absolute value slope, integral EMG value, first-order difference median value, and median frequency; the eigenvalue matrix X _Ei (m _x ×12) that forms the original EMG signal of the affected limb, i is the motion correlation The serial number of the muscle is 1≤i≤5, and m _x is the length of the collected sequence;

(b), the two attitude sensors collect the movement attitude quaternion data of the upper arm and the forearm respectively, and obtain the movement attitude quaternion matrix X _Qj (m _x × 4) of the affected limb, where j is the quaternion data of the two attitude sensors Serial number 1≤j≤2;

(c), the signal eigenvalue matrix X _Ei (m _x × 12) collected by five surface electromyography sensors forms the myoelectric characteristic signal observation matrix X _E (m _x × 60) of the affected limb, X _E = (X _E1 ,X _E2 ,X _E3 ,X _E4 ,X _E5 );

(d), the quaternion matrix X _Qj (m _x × 4) collected by two attitude sensors constitutes the observation matrix X _Q (m _x × 8) of the myoelectric characteristic signal acquired in a single acquisition, X _Q = (X _Q1 , X _Q2 );

(e), as described in step (3.1), when performing sliding windowing on the original EMG signal of the affected limb, it is necessary to ensure that the length of the feature sequence is consistent with the length of the quaternion sequence output by the attitude sensor, and both are m _x , therefore, the affected side The eigenvalue matrix X _E of the limb myoelectric signal has the same length as the motion posture data matrix X _Q of the affected limb, forming the observed value matrix X=(X _Q , X _E ) of the affected limb.

(5) Collect the upper limb posture and myoelectric data of the patient's affected upper limb performing standard motor function evaluation actions:

(5.1), utilize the concrete model that obtains in step (3.5) As a parameter of the forward-backward algorithm, use the standard evaluation action observation matrix X of each group of affected limbs collected in step (4.6) as the current observation value of the forward-backward algorithm, and output the result of the forward-backward algorithm , that is, the likelihood value l of the current observation value;

(5.2) Step (5.1) is performed on the 8 different actions described in step (1), and the likelihood values L=(l ₁ ,l ₂ ,...,l ₈ ) of 8 observations are obtained.

(6) According to the likelihood probability value L=(l ₁ ,l ₂ ,…,l ₈ ) in (5), calculate the motor score of the affected limb:

(6.1), for a single standard evaluation action, collect the posture and associated muscle electromyographic data of the exercise performed by the uninjured upper limb;

(6.2), for a single standard evaluation action, extract the myoelectric feature data, and combine it with the attitude sensor data to obtain the observation matrix of a single action;

(6.3), for a single standard evaluation action, use the forward-backward algorithm and the model corresponding to the observation mean calculation action in step (6.2) The likelihood probability value of is taken as the maximum likelihood probability value l _max ;

(6.5), the likelihood probability value L=(l ₁ ,l ₂ ,...,l ₈ ) obtained by the method of step (5.2) and the maximum likelihood probability value L _max of each action obtained in step (6.3) =( l _max1 ,l _max2 ,…,l _max8 ) Calculate the total score of the evaluation exercise with a full score of 100 and a minimum score of 0. The calculation method is shown in the following formula:

As shown in Figure 2, the upper limb motor function evaluation method based on the hidden semi-Markov model of the present invention includes the following contents:

(1) Two attitude sensors are respectively installed on the inner side of the upper arm and the inner side of the forearm of the patient's upper limbs. When installing, pay attention to keep the original posture of the two attitude sensors consistent;

(2) When the arm is naturally drooping, initialize the reset sensor, and then keep the patient's upper limbs still;

(3) Guide the patient to perform standard motion assessment actions, and the attitude sensor simultaneously records the attitude quaternion of the real-time movement of the upper arm and the forearm;

(4) When the motion ends, the sensor data recording is ended, and the motion posture quaternion sequence of the big arm and the small arm is obtained.

As shown in Figure 3, the upper limb motor function evaluation method based on the hidden semi-Markov model of the present invention includes the following contents:

(1), 5 surface electromyographic sensors are respectively installed on the surface of the biceps brachii, deltoid, deltoid-scapula, pronator muscle group, and supinator muscle group of the upper limbs of the patient. The skin should be clean and dry during installation;

(2) Guide the patient to perform standard motion assessment actions, and 5 surface electromyographic sensors simultaneously record 5 groups of original electromyographic signals of the real-time movement of the upper arm and forearm;

(3) At the end of the exercise, the sensor data recording is ended, and the original EMG signal sequence of the biceps, deltoid, deltoid-scapula, pronator muscle group, and supinator muscle group is obtained;

(4) Windowing is performed on the original signal of each muscle, and an appropriate window length and frame shift are selected to ensure that the length of the attitude data sequence is consistent with the length of the feature value sequence.

The invention can evaluate the recovery degree of upper limb motor function of stroke patients with hemiplegia, and to a certain extent replace the method of empirical evaluation by rehabilitation physicians using the motor function evaluation scale, thereby reducing the work intensity of rehabilitation physicians and assisting physicians in their work. , to improve the efficiency of physicians.

Claims (8)

1. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process, it is characterised in that: its appraisal procedure packet Include following steps:

(1), Planning Standard upper extremity exercise functional assessment acts；

(2), the upper limb pose and myoelectricity data set of limb acquisition standard upper extremity exercise functional assessment movement are good for using patient；

(3), it is suitable for the hidden Semi-Markov Process of motor function assessment using upper limb pose and the training of myoelectricity data set；

(4), acquisition patient's suffering limb executes the upper limb pose and myoelectricity data of standard movement functional assessment movement；

(5), the likelihood probability value of data in step (4) is obtained using model described in forward-backward algorithm algorithm combination step (3)；

(6), patient's healthy side upper limb motor function scores are calculated according to FMA；

(7), according to step (6) and step (5), patient's ipsilateral upper limb motor function scores are calculated.

2. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process according to claim 1, special Sign is: step (1) the standard upper extremity exercise functional assessment movement includes following locomotion evaluation movement composition:

(2.1), upper limb palm touches ipsilateral ear movement；

(2.2), upper limb the back of the hand touches lumbosacral region movement；

(2.3), upper limb joint flexion and extension；

(2.4), upper limb joint rotary motion.

3. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process according to claim 1 or 2, Be characterized in that: the step (2) is good for the upper limb pose and myoelectricity of limb acquisition standard upper extremity exercise functional assessment movement using patient Data set are as follows: attitude transducer is installed on patient and is good in the large arm and forearm of limb, acquisition patient is executing the standard fortune Limb motion attitude angle when dynamic assessment movement；Surface myoelectric sensor is installed on muscle associated with movement and is acquired Surface electromyogram signal of the patient when executing the standard movement assessment movement；For the robustness for guaranteeing system, same group of mark The pose and myoelectricity data of quasi-moving assessment movement can acquire once or for several times.

4. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process according to claim 1 or 3, Be characterized in that: the step (3) is suitable for hidden half Ma Er of motor function assessment using upper limb pose and the training of myoelectricity data set Section's husband's model includes: that the patient of acquisition is good for the one or more groups of athletic postures and surface myoelectric letter that limb executes criterion evaluation movement The multidimensional data of number composition forms the aobvious state observation matrix of hidden Semi-Markov Process；Aobvious state observation matrix combination EM is calculated Method or the training of Baum-Welch algorithm obtain the hidden Semi-Markov Process for being suitable for single assessment movement.

5. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process according to claim 1 or 2, Be characterized in that: step (4) acquisition patient's suffering limb executes the upper limb pose and myoelectricity data of standard movement functional assessment movement Are as follows: attitude transducer is installed in the large arm and forearm of patient's suffering limb, acquisition patient's suffering limb is executing the standard movement Limb motion attitude angle when assessment acts；Surface myoelectric sensor is installed on the muscle associated with movement of suffering limb Acquire surface electromyogram signal of patient's suffering limb when executing the standard movement assessment movement.

6. a kind of according to claim 1, upper extremity exercise function score method based on hidden Semi-Markov Process described in 3,4, It is characterized by: the step (5) is general using the likelihood that forward-backward algorithm algorithm combines model described in (3) to obtain data in (4) Rate value are as follows: execute standard using patient's suffering limb of the hidden Semi-Markov Process for being suitable for motor function assessment and acquisition The posture of athletic performance and the observing matrix of myoelectricity data composition are assessed, obtains likelihood probability value in conjunction with Forward-backward algorithm.

7. a kind of upper extremity exercise function score method based on hidden Semi-Markov Process according to claim 1, special Sign is: the step (6) calculates patient's healthy side upper limb motor function scores according to FMA are as follows: treating physician examines that patient is good for side Upper limb executes the record video-tape of the motor function assessment movement and fills in Li Kete questionnaire survey, is good for side to assess patient Upper limb executes the performance during the motor function assessment movement, and calculates strong limb motor function scores with FMA.

8. a kind of according to claim 1, upper extremity exercise function score method based on hidden Semi-Markov Process described in 4,7, It is characterized by: the step (7) calculates patient's ipsilateral upper limb motor function scores using step (6) and step (5) are as follows: first First, same rehabilitation assessment is acted, according to step (5), calculates all patient's ipsilateral upper limb motor function data relative to strong side The likelihood probability numerical value of upper extremity exercise performance data, and by normalized, every patient Ipsilateral is got to specific each The normalization likelihood probability numerical value of rehabilitation assessment movement；Secondly, being good for side scoring according to step (6) patient, obtains patient and be good for side pair The highest and lowest score of each rehabilitation assessment movement；Finally, the likelihood probability numerical value obtained using normalized and patient The highest and lowest score that strong side acts each rehabilitation assessment, get the motor function that patient's Ipsilateral acts rehabilitation assessment Scoring.