METHOD AND SYSTEM FOR GAIT ANALYSIS

Description

Technical field

[0001] The invention relates to a method and a system for gait analysis. This can be based in particular on a kinetic analysis. Technical background

[0002] Gait analysis examines a person's natural mode of locomotion, namely walking and running. In a kinetic analysis, this is typically done by measuring the ground reaction force, which is the reaction force of the ground to the force that the body transmits to the ground through the feet when striking the ground.

[0003] The qualitative and / or quantitative description of ground reaction forces is used in fields such as medicine, biomechanics, and robotics. In clinical applications, in particular, it can be used to identify stresses, risks, and pathological abnormalities or changes, especially those resulting from illnesses or injuries (acute and chronic). In sports medicine and clinical practice, gait also serves as an indicator of potential stress risks, as well as for assessing and clarifying existing impairments in patients and as a measure of therapy success.

[0004] A fundamental parameter in diagnostics is the temporal profile of the ground reaction force, which is usually classified visually and is therefore highly dependent on the experience and knowledge of the diagnostician. Due to the complexity of the ground reaction force profile and numerous influencing parameters, in addition to the diagnostician's experience, particular emphasis is placed on individual parameters of the ground reaction force. These include, among others, for the vertical ground reaction force, a maximum force, a time of maximum force, and a ground contact time, and for the horizontal ground reaction force, a maximum acceleration, maximum deceleration, net acceleration, etc. From these parameters, further parameters can be determined, such as momentum / impulse, symmetries, etc.Despite these characteristic parameters and values, it is still not possible to objectify the description of the soil reaction force curve, so misjudgments can occur.

[0005] One approach to better objectification can be found, for example, in Alaqtash, M., Sarkodie-Gyan, T., Yu, H., Fuentes, O., Brower, R. & Abdelgawad, A. (2011). Automatic classification of pathological gait patterns using ground reaction forces and machine learning algorithms. This approach classifies ground reaction forces of stroke patients based on characteristic parameters using a machine learning algorithm. A disadvantage of this approach is that the entire ground reaction force profile is not considered and analyzed, meaning that the method proposed for stroke patients is not transferable to other fields.

[0006] Thomas Ertelt: "Force Morphology of Human Leg Movement: Dissertation", 2008, found on the Internet: URL: https: / / www.researchgate.net / profile / Thomas_Ertelt / publication / 259704550_Kraftmorphologie_der_menschlichen_Beinbewegung / links / 58e22481aca272059ab13059 / Kraftmorphologie-der-menschlichen-Bewegung.pdf, describes a stationary system for jump analysis.

[0007] Gait analysis using sensors arranged in shoe soles is known from US2009 / 240171 A1 and US2016 / 324445 A1. Summary of the invention

[0008] The object of the invention is therefore to provide an improved method for gait analysis with the highest possible objectivity.

[0009] The problem is solved by the subject matter of the independent claims. Advantageous embodiments of the invention are described in the dependent claims, the description, and the accompanying figures.

[0010] A first aspect of the invention relates to a method for gait analysis, comprising the following steps: Obtaining, by means of an electronic evaluation device integrated into an insole for a shoe, data containing a ground reaction force curve from a force measuring device, wherein the force measuring device is formed by an insole for a shoe with a plurality of individual sensors; determining, by means of the evaluation device, a first approximation of the ground reaction force curve by means of at least one first approximation function defined by a set of coefficients; and determining, by means of the evaluation device, a second approximation of the ground reaction force curve by means of a second approximation function formed on the basis of the set of coefficients of the at least one first approximation function;and wherein at least one set of coefficients is grouped into a coefficient pool from which individual coefficients can be combined to form the second approximation function, providing a neural network configured to receive the coefficients as input and to generate class information as output.

[0011] In other words, a selected force segment, here the ground reaction force curve, is approximated in order to describe this measured curve as objectively as possible with the greatest possible agreement.

[0012] This configuration offers several advantages. It provides a way to objectify gait analysis and, in particular, to achieve a quantifiable description independent of a diagnostician. Uniform interventions and strategies can be derived from the gait analysis according to the invention, for example, for potential surgical procedures as well as therapeutic and rehabilitative measures.

[0013] The first aspect of the invention also provides that at least one set of coefficients is combined in a coefficient pool, from which individual coefficients are combined to form the second approximation function.

[0014] The more first approximation functions are determined, the more sets of coefficients are collected in the coefficient pool and become available for the second approximation. By selecting individual coefficients from the coefficient pool, the soil reaction force profile can be described exactly.

[0015] According to a further training method, a plurality of distinct first approximation functions can be determined, defined by their respective sets of coefficients. These first approximation functions can differ, for example, in only one of several coefficients. All coefficients can be grouped together in the coefficient pool described above. The ability to select from multiple sets of coefficients allows for a high degree of accuracy in the second approximation.

[0016] In another advanced approach, at least two first approximation functions can be selected iteratively. In each iteration, a sum function and its deviation from the ground reaction force curve are determined from these functions until the deviation reaches a predetermined termination criterion. The corresponding first approximation function for the initial approximation is then determined. For example, residuals—i.e., the square of the sum of the deviations—can be calculated from the sum function of the selected first approximation functions. Squaring allows small deviations to be mitigated and large deviations, such as deviations greater than 1, to be amplified. This selection or adjustment process can be repeated until a minimum of residuals is achieved and the function's performance reaches an R² of, for example, at least 0.8.The minimum of the residuals, the values ​​of R2, or a combination of both can serve as the termination criterion. The first approximation functions that meet this termination criterion can then be used to determine the second approximation. This allows a good first approximation to be obtained with comparatively little computational effort, providing a good starting point for the second approximation.

[0017] According to a further training method, the coefficients forming the second approximation function can be compared in advance with at least one predetermined dataset, which may contain a plurality of classified ground reaction force curves, and used to validate whether the coefficients enable the second approximation. In other words, the selected first approximation functions or coefficients can be validated in a kind of training set. For this purpose, existing ground reaction force curves are first distinguished from one another based on, for example, a visual analysis.

[0018] It may be known, for example through observation and assignment to a running style or similar, through findings from literature, etc., which pattern of ground reaction force curve corresponds to a particular running style. It can then be tested using the training set whether it is statistically possible to actually differentiate a predetermined class, e.g., a specific running style. For example, visually classified patterns can be assigned a known category, the differentiation of which is then tested in a training set.

[0019] Another advanced approach involves iteratively generating a second approximation function from the validated coefficients and assigning this second approximation to a class. In other words, information about the coefficient values ​​in relation to a classification group, such as running style, sport, pathology, etc., can be obtained from the dataset. These validated coefficients can then be used to achieve the best possible match with the ground reaction force profile being analyzed. This means that a recorded ground reaction force profile can be approximated based on the available coefficients. A termination criterion can be the minimization of the residual. It is now also known what the fundamental values ​​of the coefficients in each class are, so that the class can be determined for the ground reaction force profile with a corresponding probability.

[0020] According to further training, the first approximation function can be a normal distribution or a Gaussian function. These are easy to determine and have low complexity. Nevertheless, a good first approximation can be achieved with them.

[0021] Another approach proposes that the second approximation function can be a sum function of at least one first approximation function. This is easy to construct and allows, for example, using a residual as a termination criterion, an iterative approximation with low computational effort.

[0022] In a continuing education course, the set of coefficients can include a maximum value, a mean value, and / or a width of the first approximation function. In other words, each first approximation function can have three coefficients. This allows, for example, a complete definition of a normal distribution or a Gaussian function.

[0023] According to another advanced training, the number of determined first approximation functions can be set between 1 and 20, preferably between 1 and 15, and most preferably exactly 8. The first approximation functions can, for example, be defined by three coefficients, so that for two first approximation functions, 2 × 3 = 6 coefficients are obtained, for three first approximation functions, 3 × 3 = 9, and so on. It has been shown that for a clinical application of gait analysis, particularly good results for describing the ground reaction force curve can be achieved with eight first approximation functions, i.e., 8 × 3 = 24 coefficients. However, depending on the complexity, 15 or more first approximation functions can also be determined. Surprisingly, however, it has been shown that with eight first approximation functions, a good quality approximation can be achieved with comparatively little computational effort.

[0024] A second aspect of the invention relates to a gait analysis system comprising a force measuring device configured to detect a ground reaction force profile, wherein the force measuring device is formed by an insole for a shoe with a plurality of individual sensors, an electronic evaluation device integrated into an insole for a shoe and configured to approximate the detected ground reaction force profile by at least a first approximation function defined by a set of coefficients, and to further approximate the ground reaction force profile by a second approximation function formed on the basis of the set of coefficients;and wherein the at least one set of coefficients is grouped into a coefficient pool from which individual coefficients can be combined to form the second approximation function, further comprising at least one neural network configured to receive the coefficients as input and to generate class information as output.

[0025] In a further development, for example, 8 to 25, preferably 12 to 20, more preferably 14 to 18, and particularly preferably 15 to 16, individual sensors can be distributed over a planar extent of the insole. The individual sensors can provide individual signals, in particular in the form of a deflection and / or peak, as a result of a force acting upon them, wherein the sum of the individual signals yields an overall signal of the ground reaction force. This overall signal can be approximated, for example, by a combination of coefficients, as explained in the method described above. Alternatively to the combination of coefficients, it is possible to determine only the deflection and / or peak, which can be approximately comparable to the amplitude of a Gaussian function or Gaussian curve, and a time point, in particular an absolute or relative time point, which can be approximately comparable to the position of a Gaussian function or Gaussian curve.The width of the distribution can be determined by calculating the standard deviation over a time function. This yields, for example, three coefficients from which an approximation can be derived. Due to the reduced amount of data, this determination can be performed directly by an electronic evaluation unit that can be integrated into the insole or form a single assembly with it.

[0026] This creates a way to objectify gait analysis and, in particular, to achieve a quantifiable description independent of a diagnostician.

[0027] A further development of the invention provides that the gait analysis system may also include a database unit configured to assign a combination of coefficients forming the second approximation function to a predetermined gait behavior and / or gait abnormality. For example, the database unit may comprise the predetermined data set described above and be configured to validate and / or classify the coefficients based on the data set, in particular by the evaluation unit.

[0028] According to the invention, the gait analysis system further comprises at least one neural network configured to receive at least the coefficients as input and to generate class information as output. In a further development, the termination criterion can additionally represent a further input. The neural network can be multilayered and / or convolutional, with the coefficients being fed to an input layer and the class information, e.g., running style, sport, pathology, or similar, being output by the output layer. The neural network can, for example, comprise a Bayesian Regularization Artificial Neural Network (BRANN). This offers high robustness with regard to training and / or validation, and the neural network can preferably take the termination criterion into account for the approximation. The output can be a specific sport, or, in a medical setting, possibly the specific pathology.the stage of impairment. The neural network can also recognize gradual differences between individual coefficients and patterns, even in very complex ground reaction force profiles, and can therefore be advantageous compared to statistical methods such as discriminant or cluster analyses.

[0029] A third aspect of the invention relates to a program element which, when executed by means of a processor of an electronic evaluation device, causes the evaluation device integrated into an insole for a shoe to perform the following steps: Receiving data containing a ground reaction force curve from a force measuring device, wherein the force measuring device is formed by an insole for a shoe with a plurality of individual sensors; determining a first approximation of the ground reaction force curve by at least one first approximation function defined by a set of coefficients; and determining a second approximation of the ground reaction force curve by a second approximation function formed on the basis of the set of coefficients of the first approximation function; and wherein the at least one set of coefficients is summarized in a coefficient pool from which individual coefficients can be combined to form the second approximation function; providing a neural network configured to receive the coefficients as input and to generate class information as output.

[0030] The resulting advantages are described above for the procedure and the gait analysis system.

[0031] A fourth aspect of the invention relates to a computer-readable medium on which the program element described above is stored. Brief description of the characters

[0032] Preferred embodiments of the invention are explained below with reference to the accompanying figures. These show: Figure 1 shows a system according to the invention for gait analysis, which is suitable for carrying out a method according to the invention for gait analysis, Figure 2 shows an exemplary ground reaction force curve with its first and second approximations, Figure 3 shows an exemplary first approximation function defined by three coefficients, and Figure 4 shows a flowchart of a method according to the invention for gait analysis.

[0033] The figures are merely schematic representations and serve only to illustrate the invention. Identical or equivalent elements are consistently identified by the same reference numerals. Preferred embodiments of the invention

[0034] Figure 1 shows a schematic structure of a gait analysis system 100, which is particularly suitable for showing stresses, risks and pathological abnormalities or changes, especially as a result of diseases or injuries.

[0035] The gait analysis system 100 initially comprises a force measuring device 110 for the metrological recording of a ground reaction force over a period of time, i.e., a ground reaction force profile 111. The force measuring device 110 is exemplified here as a force plate and provides electronic data containing the ground reaction force profile 111. In some embodiments, however, the force measuring device 110 can also be formed by an insole for a shoe, which may have a plurality of individual sensors. For example, 10 to 25 individual sensors can be distributed over the area of ​​the insole, each providing an individual signal, e.g., in the form of a deflection and / or peak. Furthermore, the gait analysis system 100 has an electronic evaluation unit 130, which includes a processor 131 and a memory 132 and is used to obtain the data from the force measuring device 110, e.g.,via a data interface. For example, the electronic evaluation unit 130 can also be integrated into an insole for a shoe. Program instructions or a program element 133 are stored in the memory 132, which can be executed by the processor 131 and in which at least one artificial intelligence module and / or a neural network with an input layer, one or more intermediate layers, and an output layer is implemented. Preferably, the neural network can comprise or be a Bayesian Regularization Artificial Neural Network (BRANN). The evaluation unit 130 also has a database unit 134 or is connected to one for data exchange. At least one predetermined data set 135 with comparative data for evaluating or classifying the ground reaction force profile 111 is available in the database unit 134, so that an evaluation can be carried out as automatically as possible.

[0036] Based on Figure 2 The following section describes an exemplary operation of the gait analysis system 100, as shown in a diagram illustrating an exemplary ground reaction force curve 111 supplied as data by the force measuring device 110. In the diagram, the vertical axis represents the ground reaction force, which is therefore labeled GRF. The horizontal axis of the diagram represents time, which is therefore labeled t.

[0037] First, the data obtained from the force measuring device 110 are normalized to the body weight of the person whose gait is to be analyzed. This can be done automatically using weight data, for example, or by input via a user interface or similar.

[0038] Then, using the evaluation unit 130, a first approximation of the ground reaction force curve 111 is determined, for which several first approximation functions 140 are selected. In the Figure 2 In the illustrated embodiment, exactly eight first approximation functions 140 have been selected for the clinical gait analysis. The first approximation functions 140 represent normal distributions or Gaussian functions, differ from one another, and are defined by a set of coefficients 141, 142, 143 that represent a maximum value, a position, e.g., relative to an x-axis, in particular e.g., the position of a maximum value and / or a mean value, and a width of the respective first approximation function 140 (see Figure 1). Figure 3The first approximation functions 140 are selected iteratively by determining a sum function and its deviation from the ground reaction force curve 111 in each iteration step until the deviation reaches a predetermined termination criterion, at which point the corresponding first approximation function for the first approximation is determined. In this embodiment, residuals, i.e., the square of the sum of the deviations, are determined from the sum function of the first approximation functions 140. If these deviations of a first approximation function 140 are too large, another is selected. This is preferably repeated iteratively until a minimum of the residuals is obtained and the function's performance reaches an R2 of, for example, at least 0.8. Preferably, the combination of the minimum of the residuals and the values ​​of the R2 constitutes the termination criterion for the iterative process.Once the first approximation functions 140 for the first approximation have been found, the coefficients 141, 142, 143 of all first approximation functions 140 are stored in a coefficient pool 144 (see . Figure 1 ) collected and / or summarized, whereby in this embodiment 8 · 3 = 24 coefficients are contained in the coefficient pool 144.

[0039] Based on the coefficients 141, 142, 143 of the first approximation functions 140, which are contained in the coefficient pool 144, a second approximation of the ground reaction force curve 111 is determined by a second approximation function 150. This is done by feeding the coefficients 141, 142, 143 to the input layer of the neural network of the evaluation unit 130. The second approximation function 150 is preferably a sum function of the first approximation functions and is formed from some or all of the coefficients 141, 142, 143 of the coefficient pool 144. Figure 2The second approximation function 150 is shown as a dashed line. A comparison with the ground reaction force curve 111 shows that the second approximation describes it extremely precisely, as they are almost identical. Optionally, the coefficients 141, 142, and 143 of the coefficient pool 144, which are used to form the second approximation function 150, are compared beforehand with the dataset 135, which contains a number of already classified comparative ground reaction force curves. This comparison determines whether it is statistically possible to actually distinguish a predetermined class, e.g., a specific running style. This validates the coefficients 141, 142, and 143 of the coefficient pool 144 in advance. Then, from the coefficients 141, 142, 143 validated on the basis of the data set 135, the second approximation function 150 is formed in an iterative procedure and the second approximation is assigned to a class.The coefficients 141, 142, and 143 are combined repeatedly until a termination criterion is reached, which might, for example, consist of minimizing a residual. The resulting approximate ground reaction force profile 111 can exhibit values ​​for the coefficients 141, 142, and 143, each of which can be assigned to a specific class. The output layer of the neural network then provides a description and classification of the ground reaction force profile 111. This result allows for an objective and valid gait analysis.

[0040] Figure 3 For better illustration, an exemplary first approximation function is shown, which is defined by the three coefficients, namely the maximum value 141, the position 142 and the width 143.

[0041] Based on Figure 4A flowchart is shown, and a method for gait analysis according to the invention is now described. First, in step S1, the data containing the ground reaction force profile 111 are obtained from the force measuring device 110. Then, in step S2, a first approximation of the ground reaction force profile 111 is determined by one or more of the first approximation functions 140, which are defined by the coefficients 141, 142, 143. In step S3, a second approximation of the ground reaction force profile 111 is determined by the second approximation function 150, which is formed on the basis of the coefficients 141, 142, 143 of the one or more first approximation functions 140. In an optional step S4, the coefficients 141, 142, 143 are compared with at least one predetermined data set 135, which contains a plurality of classified ground reaction force profiles.This comparison validates whether coefficients 141, 142, and 143 allow for a second approximation. Optionally, coefficients 141, 142, and 143 are then assigned to a predetermined gait pattern, gait abnormality, and / or other foot-ground interaction by comparison with data set 135.

[0042] Although the present invention has been described above with reference to preferred embodiments, it is not limited to these embodiments but can be modified in many ways. In particular, the present invention can be altered or modified in many ways without deviating from the core of the invention.

[0043] It should also be noted that "comprehensive" and "include" do not exclude any other elements or steps, and "a" or "an" does not exclude a multitude.

[0044] Furthermore, it should be noted that features or steps described with reference to one of the above embodiments may also be used in combination with other features or steps from other embodiments described above. Reference numerals in the claims are not to be considered limitations. Reference symbol list

[0045] 100 Gait analysis system 110 Force measuring device 111 Ground reaction force curve 130 Evaluation device 131 Processor 132 Memory 133 Program element 134 Database device 135 Predefined data set 140 First approximation function 141 First coefficient (e.g., maximum value) 142 Second coefficient (e.g., position) 143 Third coefficient (e.g., width) 144 Coefficient pool 150 Second approximation function

Claims

1. A method for gait analysis, comprising: - obtaining (S1), by an electronic evaluation device (130) integrated in an insole for a shoe, data including a ground reaction force curve (111) from a force measuring device (110), wherein the force measuring device (110) is formed by an insole for a shoe having a plurality of single sensors; - determining (S2), by the evaluation device (130), a first approximation of the ground reaction force curve (111) by at least a first approximation function (140) defined by a set of coefficients (141, 142, 143), and - determining (S3), by the evaluation device (130), a second approximation of the ground reaction force curve (111) by a second approximation function (150) formed on the basis of the set of coefficients (141, 142, 143) of the at least one first approximation function (140), and - wherein the at least one set of coefficients (141, 142, 143) is combined into a coefficient pool (144) from which individual coefficients (141, 142, 143) can be combined to form the second approximation function (150), - providing a neural network which is configured to receive the coefficients (141, 142, 143) as an input variable and to generate a class information as an output variable.

2. Method of claim 1, wherein a plurality of mutually different first approximation functions (140) defined by the respective set of coefficients (141, 142, 143) are determined.

3. Method according to any one of the preceding claims, wherein at least two first approximate functions (140) are iteratively selected, from which a sum function and its deviation from the ground reaction force curve (111) are determined in a respective iteration step until the deviation reaches a predetermined termination criterion and the corresponding first approximate function (140) for the first approximation is determined therefrom.

4. Method according to any one of the preceding claims, wherein the coefficients (141, 142, 143) forming the second approximation function (150) are compared (S4) in advance with at least one predetermined data set (135) comprising a plurality of classified ground reaction force trajectories and validated therefrom whether the coefficients (141, 142, 143) enable the second approximation.

5. Method according to claim 4, wherein the second approximation function (150) is iteratively formed from the validated coefficients (141, 142, 143) and the second approximation is assigned to a class.

6. Method according to any one of the preceding claims, wherein the first approximation function (140) is a normal distribution or Gaussian function.

7. Method according to any one of the preceding claims, wherein the second approximation function (150) is a sum function of the at least one first approximation function (140).

8. Method according to any one of the preceding claims, wherein a number of the determined first approximation function (140) is set between 1 and 20, preferably between 1 and 15, more preferably to exactly 8.

9. A Gait analysis system (100), comprising - a force measuring device (110) configured to detect a ground reaction force (111), wherein the force measuring device (110) is formed by an insole for a shoe having a plurality of single sensors, - an electronic evaluation device (130), integrated in an insole for a shoe and configured to: - approximate the detected ground reaction force curve (111) by at least a first approximation function (140) defined by a set of coefficients (141, 142, 143), and - further approximate the ground reaction force curve (111) by a second approximation function (150) formed on the basis of the set of coefficients (141, 142, 143) - wherein the at least one set of coefficients (141, 142, 143) is combined into a coefficient pool (144) from which individual coefficients (141, 142, 143) can be combined to form the second approximation function (150), - further comprising a neural network which is configured to receive the coefficients (141, 142, 143) as an input variable and to generate a class information as an output variable.

10. Gait analysis system (100) according to claim 9, further comprising a database (134) adapted to associate a combination of coefficients (141, 142, 143) forming the second approximation function (150) with a predetermined gait, a gait abnormality, and / or another foot-ground interaction.

11. Gait analysis system (100) according to claim 10, wherein the database (134) is integrated into an insole for a shoe.

12. A program element (132) which, when executed by means of a processor (131) of an electronic evaluation device (130), causes the evaluation device (130) integrated into an insole for a shoe to perform the following steps: - obtaining data including a ground reaction force curve (111) from a force measuring device (110), wherein the force measuring device (110) is formed by an insole for a shoe having a plurality of single sensors, - determining a first approximation of the ground reaction force curve (111) by at least a first approximation function (140) defined by a set of coefficients (141, 142, 143), and - determining a second approximation of the ground reaction force curve (111) by a second approximation function (150) formed on the basis of the set of coefficients (141, 142, 143) of the first approximation function (140), and - wherein the at least one set of coefficients (141, 142, 143) is combined into a coefficient pool (144) from which individual coefficients (141, 142, 143) can be combined with each other to form the second approximation function (150), - providing a neural network which is configured to receive the coefficients (141, 142, 143) as an input variable and to generate a class information as an output variable.

13. A computer readable medium on which is stored a program element (132) according to claim 12.