Neural network-based post-stroke patient pathologic gait recognition system and method

Abstract

The application relates to the technical field of action or behavior recognition, in particular to a post-stroke patient pathological gait recognition system and method based on a neural network. The method obtains support force related data and center of gravity data of feet on a diseased side and a healthy side of a target patient during walking, preliminarily divides a gait cycle of the walking process of the target patient according to the support force related data, obtains gait phase sequences of the diseased side and the healthy side, then combines the center of gravity data to determine the possibility of pseudo phases of each gait phase in the gait phase sequence of the diseased side, and updates the gait phase sequence of the diseased side; according to the gait phase sequence of the healthy side and the updated gait phase sequence of the diseased side and the support force related data of the feet on the diseased side and the healthy side, the neural network is used to recognize patient gait information. The application realizes fine gait cycle segmentation of the diseased side of the patient, and the accuracy of patient gait information recognition is improved.

Description

Technical Field

[0001] This invention relates to the field of action or behavior recognition technology, specifically to a pathological gait recognition system and method for post-stroke patients based on neural networks. Background Technology

[0002] Because stroke can impair motor control, sensory integration, and balance regulation within the nervous system, abnormal gait is a common sequela. Abnormal gait is mainly characterized by foot drop, foot inversion, toe plantar flexion, unaffected side tilting, hip external rotation, and knee hyperextension during walking, severely impacting patients' daily lives and motor abilities. Therefore, gait recognition in stroke patients can effectively monitor their neurological recovery.

[0003] In related technologies, gait recognition for stroke patients typically involves using an inertial measurement unit (IMU) to acquire abrupt changes in the z-axis acceleration of the foot's supporting force on the affected side. These acceleration changes are then used to segment the gait period on the affected side, and a neural network is employed to identify the patient's gait information based on these segmented gait periods. However, since stroke patients often exhibit unilateral motor dysfunction while the other side remains relatively normal, the affected side may experience irregular, intermittent ground contact during gait recognition due to insufficient muscle control. This means that some of the acceleration changes in supporting force acquired during gait recognition are not entirely due to gait period transitions, leading to low accuracy in traditional gait period segmentation and ultimately, inaccurate gait information recognition. Summary of the Invention

[0004] To address the aforementioned technical problem of inaccurate gait information recognition due to low accuracy in gait cycle segmentation, the present invention aims to provide a neural network-based pathological gait recognition system and method for post-stroke patients. The specific technical solution adopted is as follows:

[0005] In a first aspect, the present invention provides a method for pathological gait recognition in post-stroke patients based on neural networks, comprising the following steps:

[0006] Acquire data on the support force and center of gravity of the feet on the affected and unaffected sides during the walking process of the target patient;

[0007] Based on the support force data, the gait cycle of the target patient's walking process is initially divided to obtain the gait phase sequence of the target side. The target side is either the affected side or the healthy side of the patient. The gait phase sequence consists of several gait phases that alternate between swing phase and support phase.

[0008] Based on the numerical differences in the center of gravity data of the affected and healthy sides at each moment, the type of each gait phase in the gait phase sequence of the affected side, and the local numerical changes in the support force related data of the healthy side, the possibility of spurious phases in each gait phase in the gait phase sequence of the affected side is determined.

[0009] Based on the possibility of artifacts, the gait phase sequence of the affected side is updated to obtain the updated gait phase sequence of the affected side.

[0010] Based on the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side, as well as the support force related data of the feet on the affected and healthy sides, a neural network is used to identify the patient's gait information.

[0011] In conjunction with the first aspect mentioned above, in some possible implementation methods, the gait cycle of the target patient's walking process is initially divided to obtain the gait phase sequence on the target side, including:

[0012] Based on the numerical differences in the support force related data on the target side at each time point and its adjacent time points, determine the degree of data mutation on the target side at each time point;

[0013] Based on the degree of data mutation, several suspected support phase transition times are determined on the target side, where the degree of data mutation at the suspected support phase transition times is greater than a set mutation degree threshold.

[0014] Based on the suspected support phase transition time, the gait cycle of the target patient's walking process is divided to obtain the gait phase sequence on the target side.

[0015] In conjunction with the first aspect above, in some possible implementations, the support force-related data includes vertical support force acceleration data and support force angular velocity data perpendicular to the ground, determining the degree of data mutation on the target side at each moment, including:

[0016] The mean value of acceleration is obtained by determining the average value of the acceleration data of the support force on the target side of each time step, which is the number of adjacent time steps preceding each time step.

[0017] The mean angular velocity value is obtained by determining the average angular velocity value of the support force angular velocity data on the target side of several adjacent time points before each time point;

[0018] The degree of data mutation on the target side at each moment is determined based on the difference between the acceleration value in the support force acceleration data on the target side at each moment and the mean value of the acceleration value, and the difference between the angular velocity value in the support force angular velocity data on the target side at each moment and the mean value of the angular velocity value.

[0019] In conjunction with the first aspect mentioned above, among some possible implementation methods, determining the degree of data mutation on the target side at each time step includes:

[0020] The absolute value of the difference between the acceleration value in the support force acceleration data on the target side at each moment and the mean value of the acceleration values ​​is determined to obtain the first difference value;

[0021] The absolute value of the difference between the angular velocity value in the support force angular velocity data on the target side at each moment and the mean value of the angular velocity values ​​is determined to obtain the second difference value;

[0022] The normalized result of the product of the first difference value and the second difference value is determined as the degree of data mutation on the target side at each time step.

[0023] In conjunction with the first aspect mentioned above, among some possible implementations, determining the likelihood of pseudophases in each gait phase of the gait phase sequence on the affected side includes:

[0024] Based on the numerical differences in the center of gravity data between the affected and healthy sides at each moment, the degree to which the center of gravity is biased towards the affected side in each gait phase of the gait phase sequence at each moment is determined.

[0025] Based on the degree of center of gravity tilt towards the affected side at each moment in the gait phase sequence of the affected side, the type of each gait phase in the gait phase sequence of the affected side, and the magnitude of the numerical fluctuation of the support force acceleration data of each gait phase in the gait phase sequence of the affected side at each moment in the healthy side, the probability of spurious phases in each gait phase in the gait phase sequence of the affected side is determined.

[0026] In conjunction with the first aspect above, in some possible implementations, the center of gravity data includes shoulder longitudinal acceleration data and shoulder vertical acceleration data, determining the degree of center of gravity tilt towards the affected side at each moment in the gait phase sequence of the affected side, including:

[0027] The acceleration value in the longitudinal acceleration data of the shoulder on the target side at each moment is determined as the first acceleration value;

[0028] The acceleration value in the vertical shoulder acceleration data on the target side at each moment is determined as the second acceleration value;

[0029] The four-quadrant arctangent value of the ratio of the first acceleration value to the second acceleration value is used as the shoulder roll angle of the target side at each moment;

[0030] The difference in shoulder roll angle between the affected and healthy sides at each moment is determined to obtain the roll angle difference value;

[0031] The result of positively correlated mapping of the roll angle difference is used as the degree of center of gravity tilt towards the affected side, thereby obtaining the degree of center of gravity tilt towards the affected side for each gait phase in the gait phase sequence of the affected side at each time step.

[0032] In conjunction with the first aspect mentioned above, among some possible implementations, determining the likelihood of pseudophases in each gait phase of the gait phase sequence on the affected side includes:

[0033] Determine the average degree of center of gravity tilt towards the affected side for each gait phase in the gait phase sequence of the affected side at all times, and obtain the average degree of center of gravity tilt towards the affected side;

[0034] Determine the variance of the acceleration values ​​in the support force acceleration data of the healthy side at each moment of each gait phase in the gait phase sequence of the affected side, and obtain the acceleration value variance;

[0035] Determine the product of the mean degree of center of gravity tilt towards the affected side and the variance of the acceleration value, and determine the spurious phase probability of each gait phase in the gait phase sequence of the affected side based on the product value and the type of each gait phase in the gait phase sequence of the affected side.

[0036] In conjunction with the first aspect above, in some possible implementations, the probability of pseudo-phases in each gait phase of the gait phase sequence on the affected side is determined based on the product value and the type of each gait phase in the gait phase sequence on the affected side, including:

[0037] If the gait phase in the gait phase sequence of the affected side is a swing phase, the normalized result obtained by normalizing the product value is used as the spurious phase probability of each gait phase in the gait phase sequence of the affected side.

[0038] If the gait phase in the gait phase sequence of the affected side is the support phase, the difference between the set value and the normalized result obtained by normalizing the product value is taken as the pseudo-phase probability of each gait phase in the gait phase sequence of the affected side.

[0039] In conjunction with the first aspect mentioned above, in some possible implementations, the gait phase sequence of the affected side is updated to obtain an updated gait phase sequence of the affected side, including:

[0040] Based on the probability of artifacts, the pseudo-gait phases in the gait phase sequence of the affected side are identified, and the artifact probability of the pseudo-gait phase is greater than a set probability threshold.

[0041] The gait phases in the gait phase sequence of the affected side, excluding the pseudo-gait phases, are taken as non-pseudo-gait phases, and all pseudo-gait phases are merged into the previous non-pseudo-gait phase, thus obtaining the updated gait phase sequence of the affected side.

[0042] Secondly, the present invention also provides a neural network-based pathological gait recognition system for post-stroke patients, including a memory and a processor. The memory is used to store executable computer program code, and the processor is used to call and run the executable computer program code from the memory, causing the system to perform the neural network-based pathological gait recognition method for post-stroke patients as described in the first aspect or any possible implementation thereof.

[0043] Thirdly, the present invention also provides a computer program product comprising: computer program code, which, when run on a computer, causes the computer to execute the neural network-based pathological gait recognition method for post-stroke patients as described in the first aspect or any possible implementation thereof.

[0044] Fourthly, the present invention also provides a computer-readable storage medium storing computer program code that, when executed on a computer, causes the computer to perform the neural network-based pathological gait recognition method for post-stroke patients as described in the first aspect or any possible implementation thereof.

[0045] The present invention has the following beneficial effects: First, by acquiring support force-related data and center of gravity data of the feet on the affected and healthy sides during the walking process of a target patient, the present invention first performs a preliminary division of the gait cycle based on the support force-related data, obtaining gait phase sequences for the affected and healthy sides of the target patient. Then, based on the numerical differences in the center of gravity data of the affected and healthy sides at each moment, the type of each gait phase in the gait phase sequence of the affected side, and the local numerical changes in the support force-related data of each gait phase at each moment in the gait phase sequence of the healthy side, the probability of pseudo-phases in each gait phase in the gait phase sequence of the affected side is determined. Based on the probability of pseudo-phases, the gait phase sequence of the affected side is updated, obtaining an updated gait phase sequence for the affected side. Finally, based on the gait phase sequence of the healthy side, the updated gait phase sequence of the affected side, and the support force-related data of the feet on the affected and healthy sides, a neural network is used to identify the patient's gait information. This invention achieves fine gait cycle segmentation of the patient's affected side by determining the probability of pseudophases in each gait phase sequence on the affected side and updating the gait phase sequence on the affected side based on the probability of pseudophases, thereby improving the accuracy of gait information recognition for the patient. Attached Figure Description

[0046] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the steps of a neural network-based method for identifying pathological gait in post-stroke patients according to an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the gait cycle distribution on the right side of a patient according to an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the pseudo-support phase and pseudo-oscillation phase that appear on the affected side of the target patient in an embodiment of the present invention after a certain oscillation phase;

[0050] Figure 4 This is a schematic diagram of the structure of a neural network-based pathological gait recognition system for post-stroke patients according to an embodiment of the present invention. Detailed Implementation

[0051] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings.

[0052] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0053] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0054] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0055] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0056] Although operations or steps are described in a specific order in the accompanying drawings in the embodiments of the present invention, this should not be construed as requiring these operations or steps to be performed in the specific order or serial order shown, or requiring all of the shown operations or steps to be performed to obtain the desired result. In the embodiments of the present invention, these operations or steps may be performed serially; they may be performed in parallel; or a portion of these operations or steps may be performed.

[0057] Furthermore, it is understood that the data involved in the technical solutions of this invention (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains, and all parameters or indicators in the formulas involved in this invention are normalized values ​​that have eliminated the influence of dimensions.

[0058] The following will describe in detail, with reference to the accompanying drawings, the neural network-based pathological gait recognition system and method for post-stroke patients provided by the embodiments of the present invention.

[0059] Figure 1 This diagram illustrates the basic flowchart of a neural network-based pathological gait recognition method for post-stroke patients provided by an embodiment of the present invention. Figure 1 As shown, the method specifically includes the following steps:

[0060] Step S100: Obtain relevant data on the support force and center of gravity of the feet on the affected and healthy sides of the target patient during walking.

[0061] Specifically, triaxial inertial measurement units (IMUs) are installed on the bottom of the left and right arches of the target patient, and also worn on both sides of the patient's shoulders. All triaxial IMUs have the same spatial axis orientation: the x-axis points horizontally in the direction the patient is walking forward; the y-axis points vertically in the direction perpendicular to the x-axis and towards the affected side of the target patient; and the z-axis points vertically upwards in the direction perpendicular to the horizontal plane. It should be understood that all triaxial IMUs must be installed properly to avoid interfering with the target patient's normal walking.

[0062] A walking test is conducted on the target patient. During the patient's walking process, walking test data is continuously collected over a set time period using all three-axis inertial measurement units (IMUs). The walking test data includes acceleration and angular velocity data along three axes, and it is recorded whether the patient takes their first step with the affected or unaffected leg. The data collection frequency of each IMU is kept consistent and the data is collected synchronously. In one specific implementation, the set time period is 1 minute, and the data collection frequency of the IMUs is 100Hz.

[0063] Data collected by triaxial inertial measurement units (IMUs) installed on the bottom of the left and right arches of the target patient can be obtained to acquire data on the support force of the foot on the affected and unaffected sides during walking. This support force data includes vertical support force acceleration data and support force angular velocity data perpendicular to the ground. Data collected by triaxial IMUs installed on both sides of the patient's shoulders can be obtained to acquire data on the center of gravity manifestation on the affected and unaffected sides during walking. This center of gravity manifestation data includes longitudinal and vertical shoulder acceleration data. The affected side refers to the side of the target patient affected by the disease, and the unaffected side refers to the side of the target patient that is healthy.

[0064] Step S200: Based on the support force related data, the gait cycle of the target patient's walking process is initially divided to obtain the gait phase sequence of the target side. The target side is either the affected side or the healthy side of the patient. The gait phase sequence consists of several gait phases that alternate between swing phase and support phase.

[0065] Specifically, based on the support force data of the feet on the affected and unaffected sides during the target patient's walking process, the gait cycle is initially divided, thus obtaining the gait phase sequence of the target patient on the affected and unaffected sides. The main principle of this initial gait cycle division based on support force data is to segment the gait cycle by observing the abrupt changes in acceleration and angular velocity perpendicular to the ground after the foot touches the ground. Because the movement suddenly stops at the moment the heel touches the ground on one side, the acceleration and angular velocity change abruptly and approach zero. Therefore, the abrupt changes in acceleration and angular velocity perpendicular to the ground can be used to determine whether the foot has touched the ground. Under normal gait conditions, when one side touches the ground again, the time elapsed from the last touch to the second touch constitutes one gait cycle. One gait cycle consists of one swing phase and one support phase. Figure 2 This diagram illustrates the gait cycle distribution on the right side of the patient. Figure 2 It can be seen that this gait cycle (right walking cycle) consists of a support phase (right standing phase) and a swing phase (right stepping phase). When the patient's right side enters the swing phase, the patient's left side enters the support phase.

[0066] Compared to the normal gait process, the uncoordinated gait of stroke patients does not have a relatively regular gait cycle. The affected side may exhibit compensatory brief support during the support phase of the healthy side, such as intermittent ground tapping or slow circling. Traditional methods of using the characteristic that the acceleration and angular velocity of the foot support force after ground contact tend to approach 0 to divide the gait cycle of the affected side will have a large error. Therefore, further analysis of the patient's motion data is needed to correct the gait phase sequence of the affected side of the target patient in order to achieve precise gait cycle segmentation of the affected side.

[0067] Step S300: Based on the numerical differences in the center of gravity data of the affected and healthy sides at each moment, the type of each gait phase in the gait phase sequence of the affected side, and the local numerical changes in the support force related data of the healthy side, determine the spurious phase probability of each gait phase in the gait phase sequence of the affected side.

[0068] Specifically, the above-mentioned preliminary segmentation of the gait cycle of the target patient's walking process was achieved based on the support force data during walking, resulting in a gait phase sequence for the target side, in which the swing phase and support phase alternate. However, due to irregular and intermittent ground contact such as stumbling on the affected side during walking, corresponding artifacts may occur, leading to an inaccurate preliminary segmentation of the gait phase sequence for the affected side. Figure 3 As shown, the affected side of the target patient exhibited a pseudo-support phase and a pseudo-oscillation phase after a certain swing phase.

[0069] Because the movement patterns on the affected side of the patient are not very regular, the artifacts appearing on the affected side of the target patient are usually not periodic, but only exist within certain periods. However, these parts that are considered artifacts are mainly due to insufficient support and weak control on the affected side, resulting in intermittent ground contact. In this case, the patient's center of gravity is generally located on the healthy side, and the vertical acceleration curve of the support force on the healthy side usually does not fluctuate much in this range. Therefore, for each gait phase in the gait phase sequence of the target patient's affected side, the probability that each gait phase in the gait phase sequence is an artifact can be calculated by using the degree to which the center of gravity is tilted towards the healthy side and the degree of fluctuation of the support force acceleration data corresponding to the healthy side during that gait phase. Based on this probability, the gait phase sequence of the affected side is updated, and finally a more accurate gait phase sequence for the affected side is obtained.

[0070] Step S400: Update the gait phase sequence of the affected side based on the possibility of artifacts to obtain the updated gait phase sequence of the affected side.

[0071] The above steps determine the spurious phase probability of each gait phase in the gait phase sequence of the affected side of the target patient. When the spurious phase probability is large, it indicates that the corresponding gait phase is a spurious phase caused by the patient's intermittent ground contact on the affected side. Therefore, it can be incorporated into the previous gait phase, thereby updating the gait phase sequence of the affected side and finally obtaining the accurate gait phase sequence of the affected side.

[0072] Step S500: Based on the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side, as well as the support force related data of the feet on the affected and healthy sides, the patient's gait information is identified using a neural network.

[0073] The above steps involve precisely segmenting the gait cycles of the target patient's healthy and affected sides, ultimately obtaining the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side. Based on these gait phase sequences, along with support force data from the feet on both sides, a deep learning model is used to identify the target patient's gait information, thus ultimately obtaining the target patient's gait data. In one specific implementation, the basic framework of the deep learning model can be the OpenGait open-source model. When using this OpenGait open-source model to identify the target patient's gait information, the deep learning model first needs to be pre-trained. During training, following the same method used to acquire support force-related data (including support force-related data in all three axes) obtained from the triaxial inertial measurement units installed on the bottom of the left and right arches during the target patient's walking process, as well as the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side of the target patient, a large amount of historical data on support force-related data (including support force-related data in all three axes), gait phase sequences of the healthy side and the updated gait phase sequence of the affected side, was acquired. Each historical patient's historical data was used as a training sample, and all training samples constituted a training sample set. Each training sample in the training sample set was labeled with gait information corresponding to the historical patient. The gait information included the type of gait, which could specifically include normal gait, foot drop, foot inversion, toe plantar flexion, healthy side tilt, hip external rotation, knee hyperextension, etc. The deep learning model was trained using the labeled training sample set to obtain a trained deep learning model. Since the process of labeling each training sample in the training sample set and the specific implementation process of training the deep learning model using the training sample set are both existing technologies, they will not be elaborated here. The gait phase sequence of the healthy side and the updated gait phase sequence of the affected side of the target patient, as well as the support force related data of the feet on the affected and healthy sides (including support force related data in the three-axis directions), are obtained above and input into the trained deep learning model. The deep learning model then outputs the gait information of the target patient, thereby realizing the gait information recognition of the target patient.

[0074] The neural network-based pathological gait recognition method for post-stroke patients provided in this embodiment preliminarily divides the gait cycle of the target patient's walking process based on the support force data of the feet on the affected and healthy sides during the walking process, thereby obtaining the gait phase sequences of the affected and healthy sides of the target patient. Then, by determining the probability of false phases of each gait phase in the gait phase sequence of the affected side based on the numerical differences in the center of gravity data of the affected and healthy sides at each moment, the type of each gait phase in the gait phase sequence of the affected side, and the local numerical changes in the support force data of the healthy side, the gait phase sequence of the affected side is updated based on the probability of false phases, thereby obtaining the updated gait phase sequence of the affected side, thus achieving fine gait cycle segmentation of the affected side of the target patient. Finally, based on the gait phase sequence of the healthy side, the updated gait phase sequence of the affected side, and the support force data of the feet on the affected and healthy sides, the neural network is used to identify the patient's gait information, effectively improving the accuracy of patient gait information recognition.

[0075] Furthermore, in one possible implementation, step S200 above involves a preliminary division of the gait cycle during the target patient's walking process to obtain a gait phase sequence on the target side, including:

[0076] Step S201: Determine the degree of data mutation on the target side at each time step based on the numerical differences in the support force related data on the target side at each time step and the adjacent time steps before it.

[0077] Specifically, the data on the support force of the patient's foot on both the affected and unaffected sides during walking include vertical (z-axis) support force acceleration and angular velocity data. These data represent the changes in the patient's foot support force relative to the ground. Since the magnitudes of the acceleration and angular velocity perpendicular to the ground change instantaneously when the patient's foot touches the ground and forms support force, or loses support and transitions from a supported to a swinging state, the degree of change in vertical data before and after a given moment can be calculated, allowing for a preliminary assessment of whether each moment represents a potential support transition point.

[0078] Furthermore, in one possible implementation, the degree of data mutation on the target side at each time step is determined, including:

[0079] Step S2011: Determine the average value of the acceleration values ​​in the support force acceleration data on the target side for each time step, and obtain the average value of the acceleration values.

[0080] Specifically, at the moment of transition in the gait phase, the data will undergo a sudden change, and this change is relative to multiple moments before that moment. Therefore, it is necessary to calculate the difference between each moment on both sides of the target patient and the previous local data to determine the degree of data change at each moment.

[0081] Therefore, obtaining the front view at each moment during the target patient's walking process is crucial. In a specific implementation, for each adjacent time interval, set =10, and determine the mean value of the acceleration values ​​in the support force acceleration data on the target side at these several adjacent moments, and record this mean value as the mean acceleration value. It should be understood that for the first few moments in the target patient's walking process, there may be less than 10 adjacent moments before a certain moment. In this case, the mean acceleration value is directly calculated based on the existing adjacent moments. However, for the first moment, since there are no adjacent moments before it, its corresponding mean acceleration value is not needed.

[0082] Step S2012: Determine the average value of the angular velocity in the support force angular velocity data on the target side for each time step, and obtain the average value of the angular velocity.

[0083] Similarly, obtain the previous time step at each moment. The mean value of the acceleration values ​​in the support force acceleration data on the target side at adjacent time points is recorded as the mean value of the angular velocity.

[0084] Step S2013: Determine the degree of data mutation on the target side at each moment based on the difference between the acceleration value in the support force acceleration data on the target side at each moment and the average value of the acceleration value, and the difference between the angular velocity value in the support force angular velocity data on the target side at each moment and the average value of the angular velocity value.

[0085] Specifically, when the acceleration value of the target side at each moment differs significantly from the average acceleration value of the preceding few adjacent moments, and the angular velocity value at each moment also differs significantly from the average angular velocity value of the preceding few adjacent moments, it indicates that the corresponding moment is more likely to be a moment of sudden change in support force, and the corresponding value of the degree of data change will be greater.

[0086] In one specific implementation, determining the degree of data mutation on the target side at each moment includes: determining the absolute value of the difference between the acceleration value in the support force acceleration data on the target side at each moment and the mean value of the acceleration values, to obtain a first difference value; determining the absolute value of the difference between the angular velocity value in the support force angular velocity data on the target side at each moment and the mean value of the angular velocity values, to obtain a second difference value; and determining the normalized result of the product of the first difference value and the second difference value as the degree of data mutation on the target side at each moment. The degree of data mutation on the target side at each moment is then calculated using the following formula: ;

[0087] In the formula: Indicates the target side at the first The degree of data mutation at each moment; and These represent the target side at the 1st... The acceleration and angular velocity values ​​at each instant; , These represent the target side at the 1st... A moment before The average acceleration and angular velocity values ​​at consecutive time points. Indicates the first difference value; Indicates the second difference value; This represents a normalization function used to normalize values ​​to the range (0,1). This normalization function can be... function.

[0088] Using the above method, by considering the difference between the acceleration value of the target side at each moment and the average acceleration value of several adjacent moments, as well as the difference between the angular velocity value of the target side at each moment and the average angular velocity value of several adjacent moments, the degree of data mutation of the target side at each moment can be accurately determined.

[0089] Step S202: Based on the degree of data mutation, determine several suspected support phase transition times on the target side, wherein the degree of data mutation at the suspected support phase transition times is greater than a set mutation degree threshold.

[0090] Specifically, a threshold for the degree of mutation is preset. In one implementation, the threshold value is set to 0.3. The degree of mutation in the target side data at each moment is compared with this threshold. When the degree of mutation exceeds the threshold, it indicates that the vertical acceleration and angular velocity data at that moment have fluctuated significantly, possibly indicating a shift in the support phase. This is considered a suspected moment of support phase transition, and the moment corresponding to this degree of mutation is determined as the suspected moment of support phase transition. Thus, several suspected moments of support phase transition on the target side during the patient's walking process can be obtained.

[0091] Step S203: Based on the suspected support phase transition time, the gait cycle of the target patient's walking process is divided to obtain the gait phase sequence of the target side.

[0092] Specifically, since the gait test on the target patient records whether the first step was taken by the affected or unaffected lower limb, the basal cycle of the target side can be divided based on the suspected support phase transition time on the target side. For example, if the record shows the target patient's right lower limb taking the first step, with the right side swinging first and the left side stationary, then the time from the start of the right side stepping to the first suspected support phase transition time is the swing phase, the time from the first to the second suspected support phase transition time is the support phase, the time from the second to the third suspected support phase transition time is the swing phase, and so on. Conversely, the time from the start of the left side stepping to the first suspected support phase transition time is the support phase, the time from the first to the second suspected support phase transition time is the swing phase, the time from the second to the third suspected support phase transition time is the support phase, and so on.

[0093] The above method determines the degree of data mutation on the target side at each moment by analyzing the numerical differences in the support force data on the target side at each moment and the adjacent moments before it, and determines several suspected support phase transition moments on the target side based on the degree of data mutation, thereby achieving preliminary gait cycle segmentation of the target patient's walking process on the target side.

[0094] Furthermore, in one possible implementation, determining the likelihood of pseudophases in each gait phase of the gait phase sequence on the affected side in step S300 above includes:

[0095] Step S301: Based on the numerical differences in the center of gravity data between the affected and healthy sides at each moment, determine the degree to which the center of gravity is biased towards the affected side at each moment in the gait phase sequence of the affected side.

[0096] Specifically, the gait phase sequences of the target patient on the healthy side and the affected side were obtained through the above step S200. Since the affected side of the target patient will have intermittent ground contact phenomenon while the healthy side usually does not, the obtained gait phase sequence of the healthy side is accurate, while the gait phase sequence of the affected side is not accurate enough and needs to be further corrected.

[0097] During the normal alternation of swing and stance phases, the patient's center of gravity also alternates along the left or right side of the body. When one side is in the swing phase, the center of gravity shifts to the other side, which is usually the stance phase. Since the preliminary gait phase sequence for the affected side of the target patient has been obtained, the degree to which the center of gravity shifts towards the affected side for each gait phase is assessed. By evaluating the consistency between the segmented gait phases and the shift of the center of gravity towards the affected side, the likelihood of the corresponding gait phase being a spurious phase can be determined. For example, if the affected side is in the stance phase, but the center of gravity is significantly shifted towards the healthy side, and the stability acceleration data for the healthy side shows relatively small fluctuations, then the stance phase is more likely to be a spurious phase. Similarly, if the affected side is in the swing phase, but the center of gravity is significantly shifted towards the affected side, and the stability acceleration data for the healthy side shows relatively large fluctuations, then the phase is more likely to be a spurious phase.

[0098] Furthermore, in one possible implementation, the center of gravity data includes shoulder longitudinal acceleration data and shoulder vertical acceleration data. Step S301, which determines the degree of center of gravity tilt towards the affected side in each gait phase at each moment in the gait phase sequence of the affected side, includes:

[0099] Step S3011: Determine the acceleration value in the longitudinal acceleration data of the shoulder on the target side at each moment as the first acceleration value.

[0100] Step S3012: Determine the acceleration value in the vertical acceleration data of the shoulder on the target side at each moment as the second acceleration value.

[0101] Step S3013: Determine the four-quadrant arctangent value of the ratio of the first acceleration value to the second acceleration value as the shoulder roll angle of the target side at each moment.

[0102] Specifically, the target patient's center of gravity can be represented by the roll angle of the triaxial inertial measurement unit (TIM) of their shoulder. Since the roll angle reflects the angle between the shoulder's y-axis and the initial xoy plane, when the body's center of gravity is tilted to one side, the roll angle on that side is larger, and the roll angle on the other side is smaller. Generally, the roll angles on both sides are positive and negative, and there is a significant difference between them. Therefore, when the roll angle on the affected side of the target patient is larger than the roll angle on the healthy side, it indicates that the body's center of gravity is closer to the affected side.

[0103] Therefore, based on the obtained vertical and longitudinal shoulder acceleration data of the target patient's shoulder on the target side, the shoulder roll angle on the target side at each moment is calculated using the following formula:

[0104] ;

[0105] In the formula: Indicates the target side at the first Shoulder roll angle at any given moment; Indicates the first The acceleration value in the longitudinal acceleration data of the shoulder on the target side at the nth moment, i.e., the nth acceleration value. The acceleration component in the y-axis direction of the shoulder on the target side at a given moment, also known as the first acceleration value; Indicates the first The acceleration value in the vertical acceleration data of the shoulder on the target side at the nth moment, i.e., the nth The acceleration component in the z-axis direction of the shoulder on the target side at a given moment is also called the second acceleration value; This represents the arctangent function.

[0106] Using the method described above, the shoulder roll angles of the affected and unaffected sides of the target patient at each moment can be determined. It should be understood that, as another possible implementation, the shoulder roll angles of the affected and unaffected sides of the target patient at each moment can also be directly obtained using relevant equipment.

[0107] Step S3014: Determine the difference in shoulder roll angle between the affected side and the healthy side at each moment to obtain the roll angle difference value.

[0108] Step S3015: The result of positive correlation mapping of the roll angle difference is used as the degree of center of gravity tilt towards the affected side, thereby obtaining the degree of center of gravity tilt towards the affected side for each gait phase in the gait phase sequence of the affected side at each moment.

[0109] Specifically, based on the difference in shoulder roll angle between the affected and healthy sides at each moment, the degree of weight shift towards the affected side in each gait phase of the gait phase sequence at each moment is calculated using the following formula:

[0110] ;

[0111] In the formula: The gait phase sequence representing the affected side is the first... The gait phase at the ... The degree to which the center of gravity shifts to the affected side at any given moment; Indicates the affected side The gait phase sequence of the first The gait phase at the ... Shoulder roll angle at any given moment; Indicates the healthy side The gait phase sequence of the first The gait phase at the ... Shoulder roll angle at any given moment; The gait phase sequence representing the affected side is the first... The gait phase at the ... The difference in roll angle at each moment; This represents an exponential function with the natural constant e as the base.

[0112] Using the above method, the degree to which the center of gravity is tilted toward the affected side in each gait phase sequence of the target patient at each moment can be determined. This degree of tilt reflects the extent to which the patient's body center is tilted toward the affected side at each moment in each gait phase on the affected side. The larger the value of the degree of tilt, the higher the degree to which the patient's body center is tilted toward the affected side.

[0113] Step S302: Based on the degree of center of gravity tilt towards the affected side at each moment in the gait phase sequence of the affected side, the type of each gait phase in the gait phase sequence of the affected side, and the magnitude of the numerical fluctuation in the support force acceleration data of each gait phase in the gait phase sequence of the affected side at each moment in the healthy side, determine the possibility of pseudo-phases of each gait phase in the gait phase sequence of the affected side.

[0114] Specifically, based on the degree of weight shift towards the affected side in each gait phase of the target patient's affected side at various times, combined with the type of each gait phase in the gait phase sequence of the affected side, and the magnitude of the numerical fluctuation in the support force acceleration data of each gait phase on the healthy side at various times, the likelihood of each gait phase being a spurious phase in the gait phase sequence of the affected side can be determined. For example, if a certain gait phase on the affected side is a swing phase, and the value of the degree of weight shift towards the affected side in this gait phase is relatively high at various times, and the numerical change in the support force acceleration data of the healthy side on the patient's healthy side at various times in this gait phase is relatively large, then it indicates that this gait phase on the affected side is more likely to be a spurious phase, and the corresponding spurious phase probability is greater.

[0115] Furthermore, in one possible implementation, step S302 above, determining the likelihood of pseudo-phases in each gait phase of the gait phase sequence on the affected side, includes:

[0116] Step S3021: Determine the average value of the degree of center of gravity tilting towards the affected side in each gait phase at all times in the gait phase sequence of the affected side, and obtain the average value of the degree of center of gravity tilting towards the affected side.

[0117] Specifically, for any one of the gait phase sequences on the affected side of the target patient... The gait phase is determined to be the first... The mean value of the degree of center of gravity tilt towards the affected side is obtained by averaging the degree of center of gravity tilt towards the affected side at all times of each gait phase. This mean value of the degree of center of gravity tilt towards the affected side reflects the overall trend of the gait phase. The degree to which the patient's body center is biased towards the affected side during each gait phase.

[0118] Step S3022: Determine the variance of the acceleration values ​​in the support force acceleration data of the healthy side at each moment of each gait phase in the gait phase sequence of the affected side, and obtain the acceleration value variance.

[0119] Specifically, for any one of the gait phase sequences on the healthy side of the target patient... The gait phase is determined to be the first... The variance of the acceleration values ​​at each moment of each gait phase in the support force acceleration data on the healthy side is used to obtain the acceleration value variance. The larger the value of this acceleration value variance, the more significant the acceleration value variance. The greater the fluctuation in the support force acceleration data of the healthy side during the time period corresponding to each gait phase, the greater the volatility.

[0120] Step S3023: Determine the product of the mean value of the center of gravity tilt towards the affected side and the variance of the acceleration value, and determine the spurious phase probability of each gait phase in the gait phase sequence of the affected side based on the product value and the type of each gait phase in the gait phase sequence of the affected side.

[0121] Specifically, if the target patient's affected side... The first gait phase is the swing phase. When the center of gravity is closer to the affected side and the support force acceleration data of the healthy side fluctuates significantly, it indicates that the first gait phase on the affected side is the swing phase. The first gait phase is more likely to be a false phase; if the affected side's first gait phase is... The first gait phase is the support phase. When the center of gravity is close to the healthy side and the support force acceleration data of the healthy side fluctuates little, it indicates that the affected side is in the first gait phase. The gait phase is more likely to be an artifact.

[0122] Furthermore, in one possible implementation, the above step S3023, determining the spurious phase probability of each gait phase in the gait phase sequence of the affected side, includes: if the gait phase in the gait phase sequence of the affected side is a swing phase, the normalized result obtained by normalizing the product value is used as the spurious phase probability of each gait phase in the gait phase sequence of the affected side; if the gait phase in the gait phase sequence of the affected side is a support phase, the difference between the set value and the normalized result obtained by normalizing the product value is used as the spurious phase probability of each gait phase in the gait phase sequence of the affected side.

[0123] In one specific implementation, the probability of pseudophases in each gait phase of the gait phase sequence on the affected side is calculated using the following formula:

[0124] ;

[0125] In the formula: The gait phase sequence representing the affected side is the first... The possibility of an apex in a gait phase; The gait phase sequence representing the affected side is the first... Mean value of the degree of center of gravity tilt towards the affected side in each gait phase; The gait phase sequence representing the affected side is the first... The variance of the acceleration values ​​in the gait phase, i.e., the variance of the acceleration values ​​in the gait phase sequence of the affected side. The variance of all acceleration data corresponding to the support force acceleration data on the healthy side at each moment of each gait phase; This refers to the set of sway phases in the gait phase sequence of the affected side. This refers to the set of support phases in the gait phase sequence on the affected side. This represents a normalization function used to normalize values ​​to the range (0,1). This normalization function can be... Function; the value 1 indicates the set value.

[0126] By determining the mean of the degree of center of gravity tilting towards the affected side and the variance of the acceleration values ​​for each gait phase in the gait phase sequence of the affected side, and combining this with the type of each gait phase in the gait phase sequence of the affected side, the probability of each gait phase in the gait phase sequence of the affected side being a pseudo-phase can be accurately determined, so as to reflect the likelihood that each gait phase in the gait phase sequence of the affected side is a pseudo-phase.

[0127] In one possible implementation, step S400 involves updating the gait phase sequence of the affected side based on the likelihood of artifacts, to obtain an updated gait phase sequence for the affected side, including:

[0128] Step S401: Based on the probability of artifacts, determine the pseudo-gait phases in the gait phase sequence of the affected side. The probability of artifacts in the pseudo-gait phases is greater than a set probability threshold.

[0129] Specifically, a pre-set probability threshold is configured, and in one implementation, this threshold is set to 0.8. The probability of a pseudo-gait phase in each gait phase sequence on the affected side is compared with this pre-set probability threshold. When the probability of a pseudo-gait phase is greater than the pre-set probability threshold, the corresponding gait phase is identified as a pseudo-gait phase. This allows for the identification of each pseudo-gait phase in the gait phase sequence on the affected side.

[0130] Step S402: Take the gait phases in the gait phase sequence of the affected side, excluding the pseudo-gait phases, as non-pseudo-gait phases, and merge all pseudo-gait phases into the previous non-pseudo-gait phase to obtain the updated gait phase sequence of the affected side.

[0131] Specifically, gait phases in the gait phase sequence on the affected side, excluding pseudo-gait phases, are considered non-pseudo-gait phases. For each pseudo-gait phase in the gait phase sequence on the affected side, it is merged into the nearest preceding non-pseudo-gait phase. The resulting gait phase sequence is then used as the updated gait phase sequence. Simultaneously, combining the step records from walking tests performed on the target patient, the specific type of each gait phase in the updated gait phase sequence is determined using the same method as described above for determining the type of each gait phase in the target side's gait phase sequence. For example... Figure 3 As shown, the pseudo-swing phase and pseudo-support phase of the affected side in the figure are merged into the previous swing phase, thus finally obtaining a more accurate updated gait phase sequence of the affected side. Among them, for the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side of the target patient, if both sides are in the support phase at the same time period, it means that the time period is a bilateral support phase, that is, both lower limbs of the target patient are in the support state.

[0132] Based on the same inventive concept, embodiments of the present invention also provide a pathological gait recognition system for post-stroke patients based on neural networks, such as... Figure 4 As shown, the system includes: a memory 401, a processor 402, and computer program code 403 stored in the memory 401 and running on the processor 402, wherein when the processor 402 executes the computer program code 403, the system can execute any of the neural network-based pathological gait recognition methods for post-stroke patients described above.

[0133] In this embodiment of the invention, the system can be divided into functional modules according to the above method example. For example, each module can correspond to a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0134] Based on the same inventive concept, embodiments of the present invention also provide a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute any of the aforementioned neural network-based pathological gait recognition methods for post-stroke patients.

[0135] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing computer program code, which, when run on a computer, causes the computer to execute any of the aforementioned neural network-based pathological gait recognition methods for post-stroke patients.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for recognizing post-stroke patient pathological gait based on neural network, characterized in that, The method comprises the following steps: Obtaining support force related data and center of gravity data of the foot of the target patient on the affected side and the healthy side during walking; Preliminarily dividing the gait cycle of the walking process of the target patient according to the support force related data, and obtaining a gait phase sequence of the target side, the target side being any one of the affected side and the healthy side of the patient, the gait phase sequence being composed of a plurality of gait phases alternately being swing phases and support phases; Determining the possibility of pseudo phases of each gait phase in the gait phase sequence of the affected side according to the numerical difference between the center of gravity data of the affected side and the healthy side at each moment, the type of each gait phase in the gait phase sequence of the affected side, and the local numerical change in the support force related data of the healthy side; Updating the gait phase sequence of the affected side according to the possibility of pseudo phases, and obtaining an updated gait phase sequence of the affected side; Identifying the gait information of the patient by using a neural network according to the gait phase sequence of the healthy side and the updated gait phase sequence of the affected side, and the support force related data of the foot of the affected side and the healthy side.

2. The neural network-based post-stroke patient pathology gait recognition method according to claim 1, characterized in that, The method for preliminarily dividing the gait cycle of the walking process of the target patient and obtaining the gait phase sequence of the target side comprises the following steps: Determining the data mutation degree of the target side at each moment according to the numerical difference between the support force related data of the target side at each moment and the adjacent moment in front of the each moment; Determining a plurality of suspected support phase transition moments of the target side according to the data mutation degree, the data mutation degree of the suspected support phase transition moment being greater than a set mutation degree threshold; Dividing the gait cycle of the walking process of the target patient according to the suspected support phase transition moment, and obtaining the gait phase sequence of the target side.

3. The neural network-based post-stroke patient pathology gait recognition method according to claim 2, characterized in that, The support force related data comprises vertical support force acceleration data and support force angular velocity data perpendicular to the ground, and the method for determining the data mutation degree of the target side at each moment comprises the following steps: Determining the mean value of the acceleration numerical value in the support force acceleration data of the target side at each moment and the adjacent moment in front of the each moment, and obtaining an acceleration numerical value mean value; Determining the mean value of the angular velocity numerical value in the support force angular velocity data of the target side at each moment and the adjacent moment in front of the each moment, and obtaining an angular velocity numerical value mean value; Determining the data mutation degree of the target side at each moment according to the difference between the acceleration numerical value in the support force acceleration data of the target side at each moment and the acceleration numerical value mean value, and the difference between the angular velocity numerical value in the support force angular velocity data of the target side at each moment and the angular velocity numerical value mean value.

4. The neural network-based post-stroke patient pathology gait recognition method according to claim 3, characterized in that, The method for determining the data mutation degree of the target side at each moment comprises the following steps: Determining the absolute value of the difference between the acceleration numerical value in the support force acceleration data of the target side at each moment and the acceleration numerical value mean value, and obtaining a first difference value; Determining the absolute value of the difference between the angular velocity numerical value in the support force angular velocity data of the target side at each moment and the angular velocity numerical value mean value, and obtaining a second difference value; Determining the normalization result of the product of the first difference value and the second difference value as the data mutation degree of the target side at each moment.

5. The neural network-based post-stroke patient pathology gait recognition method according to claim 3, characterized in that, The method for determining the possibility of pseudo phases of each gait phase in the gait phase sequence of the affected side comprises the following steps: According to the value difference in the center of gravity embodiment data of the affected side and the healthy side at each time, the degree of the center of gravity of each gait phase in the gait phase sequence of the affected side at each time is determined to tend to the affected side; According to the degree of the center of gravity of each gait phase in the gait phase sequence of the affected side at each time, the type of each gait phase in the gait phase sequence of the affected side, and the value fluctuation size of the acceleration data of the support force of the healthy side at each time of each gait phase in the gait phase sequence of the affected side, the possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side is determined.

6. The neural network-based post-stroke patient pathology gait recognition method according to claim 5, characterized in that, The center of gravity embodiment data includes shoulder longitudinal acceleration data and shoulder vertical acceleration data, and the degree of the center of gravity of each gait phase in the gait phase sequence of the affected side at each time is determined to tend to the affected side, including: The acceleration value in the shoulder longitudinal acceleration data of the target side at each time is determined as the first acceleration value; The acceleration value in the shoulder vertical acceleration data of the target side at each time is determined as the second acceleration value; The four-quadrant arctangent value of the ratio of the first acceleration value to the second acceleration value is determined as the shoulder roll angle of the target side at each time; The difference value of the shoulder roll angles of the affected side and the healthy side at each time is obtained as the roll angle difference value; The processing result of the roll angle difference value after positive correlation mapping processing is taken as the degree of the center of gravity tending to the affected side, so as to obtain the degree of the center of gravity of each gait phase in the gait phase sequence of the affected side at each time tending to the affected side.

7. The neural network-based post-stroke patient pathology gait recognition method according to claim 5, characterized in that, The possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side is determined, including: The average value of the degree of the center of gravity tending to the affected side of each gait phase in the gait phase sequence of the affected side at all times is determined as the average degree of the center of gravity tending to the affected side; The variance of the acceleration value in the acceleration data of the support force of the healthy side at each time of each gait phase in the gait phase sequence of the affected side is determined as the acceleration value variance; The product value of the average degree of the center of gravity tending to the affected side and the acceleration value variance is determined, and according to the product value and the type of each gait phase in the gait phase sequence of the affected side, the possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side is determined.

8. The neural network-based post-stroke patient pathology gait recognition method according to claim 7, characterized in that, According to the product value and the type of each gait phase in the gait phase sequence of the affected side, the possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side is determined, including: If the gait phase in the gait phase sequence of the affected side is a swing phase, the normalization result obtained by normalizing the product value is taken as the possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side; If the gait phase in the gait phase sequence of the affected side is a support phase, the difference value between a set value and the normalization result obtained by normalizing the product value is taken as the possibility of pseudo-phase of each gait phase in the gait phase sequence of the affected side.

9. The neural network-based post-stroke patient pathology gait recognition method according to claim 1, characterized in that, The gait phase sequence of the affected side is updated to obtain an updated gait phase sequence of the affected side, including: According to the possibility of pseudo-phase, the pseudo gait phase in the gait phase sequence of the affected side is determined, and the possibility of pseudo-phase of the pseudo gait phase is greater than a set possibility threshold value; The gait phase in the gait phase sequence of the affected side except the pseudo gait phase is taken as a non-pseudo gait phase, and all the pseudo gait phases are combined into the last non-pseudo gait phase, so as to obtain the updated gait phase sequence of the affected side.

10. A neural network based post-stroke patient pathologic gait recognition system characterized in that, The computer program product comprises a memory, a processor, and a computer program code stored in the memory and executable on the processor, and the processor executes the computer program code to perform the neural network-based post-stroke patient pathological gait recognition method according to any one of claims 1-9.

Images