A ground equivalent distance conversion method, device, system and storage medium

Abstract

The application discloses a ground equivalent distance conversion method and device, a ground equivalent distance conversion system, a storage medium and a computer program product. The method comprises the following steps: acquiring original walking distance data of a subject collected by a non-ground walking test device, and individual characteristic parameters and gait characteristic parameters of the subject; inputting the original walking distance data, the individual characteristic parameters and the gait characteristic parameters into a pre-trained equivalent conversion model to obtain a ground equivalent walking distance; and matching the ground equivalent walking distance with a preset clinical function classification threshold value to output a corresponding clinical function classification result. According to the scheme, the systematic deviation between the non-ground walking test distance and the ground actual measurement value is corrected, the non-ground test result is directly connected with the ground walking test clinical classification standard, and the clinical application value and scene adaptability of the non-ground walking test device are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of sports function assessment and rehabilitation medical equipment technology, specifically relating to a ground equivalent distance conversion method, device, ground equivalent distance conversion system, storage medium and computer program product, and particularly to a ground equivalent distance conversion method, device, ground equivalent distance conversion system, storage medium and computer program product for the test results of a six-minute walking test treadmill. Background Technology

[0002] The six-minute walk test (6MWT) is a core submaximal exercise test for assessing exercise tolerance, cardiac function, and prognostic risk in patients with chronic cardiopulmonary diseases. The six-minute walk distance (6MWD) obtained from the test is a key quantitative basis for NYHA functional classification, rehabilitation efficacy evaluation, and prognostic risk stratification, and has irreplaceable application value in the clinical fields of cardiovascular, respiratory, and rehabilitation. According to standard operating procedures, the 6MWT must be completed in a flat, straight corridor of 30 meters in length. The subject walks at his maximum tolerable speed within a limited time, and the measured walk distance is used as the assessment basis.

[0003] However, the standard 6MWT ground-based walking test has strict requirements for site space, the testing process is greatly affected by environmental interference, and the reproducibility of results is easily affected by factors such as corridor length and turning points, making it difficult to widely adopt in primary healthcare institutions, bedside rehabilitation, and home rehabilitation scenarios. Meanwhile, various non-ground-based walking test devices, such as treadmills and walking trainers, are gradually becoming important alternatives to ground-based testing due to their small footprint, controllable speed, and ability to simultaneously collect monitoring parameters such as electrocardiogram and blood oxygenation, and their application in clinical rehabilitation and home monitoring scenarios is becoming increasingly widespread.

[0004] However, both clinical research and practice have shown that there are significant systematic differences between the results of walking tests performed on non-ground equipment and those measured on the ground. For example, on a treadmill, the 6MWD measured is generally shorter than the standard 6MWD measured on the ground due to factors such as exercise pattern, psychological factors, and gait adjustment, with the largest difference exceeding 30%. Therefore, the test results cannot be directly applied to the NYHA functional classification thresholds (300 meters, 375 meters, and 450 meters) corresponding to the 6MWD on the ground, otherwise it will lead to bias in cardiac function assessment and mislead clinical decision-making.

[0005] Existing technical solutions for gait testing devices mainly fall into two categories: one focuses on test data acquisition and report generation, using displacement sensing and physiological monitoring modules to automatically record walking distance and vital signs, but only outputs the device's raw test distance and lacks ground-equivalent conversion capabilities; the other focuses on optimizing the device's motion control, improving the testing experience through adaptive speed adjustment and gait guidance, but fails to address the clinical equivalence issue of "how much ground distance is equivalent to the device's test distance." Although some academic studies have confirmed a correlation between device test distance and ground distance, an industrially deployable, clinically accurate equivalent conversion model has not yet been developed, nor has a standardized mechanism for linking conversion results with clinical grading standards been established, and quantitative assessment methods for the reliability of conversion results are lacking.

[0006] In summary, there is currently a lack of a universal method for converting ground equivalent distances in this field. This makes it impossible to accurately convert the original test distances of various non-ground walking test devices into clinically equivalent ground distances, which in turn prevents direct integration with mature ground walking test clinical evaluation systems. Consequently, this directly limits the clinical application value and promotion scope of non-ground walking test devices.

[0007] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0008] The purpose of this invention is to provide a ground equivalent distance conversion method, device, system, storage medium, and computer program product to solve the problem that there is a systematic deviation between the test distance of non-ground walking test equipment and the actual ground measurement value in related solutions, which makes it impossible to directly connect the test results with the clinical grading standards of ground walking tests, thus restricting its clinical application value. This invention effectively corrects the systematic deviation between the test distance of non-ground walking test equipment and the actual ground measurement value, realizes the direct connection between the non-ground test results and the clinical grading standards of ground walking tests, and significantly improves the clinical application value and scenario adaptability of non-ground walking test equipment.

[0009] This invention provides a method for converting ground equivalent distances, comprising: acquiring raw walking distance data of subjects collected by non-ground walking test equipment, as well as individual characteristic parameters and gait characteristic parameters of the subjects; inputting the raw walking distance data, individual characteristic parameters and gait characteristic parameters into a pre-trained equivalent conversion model to obtain ground equivalent walking distances; matching a preset clinical functional grading threshold with the ground equivalent walking distances, and outputting the corresponding clinical functional grading results.

[0010] In some embodiments, the method further includes: acquiring at least two sets of raw walking distance data obtained from multiple tests of the same subject; performing learning effect correction processing on the at least two sets of raw walking distance data to obtain corrected walking distance data; and replacing the raw walking distance data with the corrected walking distance data and inputting it into the equivalent conversion model.

[0011] In some embodiments, the gait characteristic parameters include at least one of gait frequency, stride length, gait symmetry parameters, and gait frequency variation parameters; the individual characteristic parameters include at least one of age, sex, body mass index, and disease diagnosis classification.

[0012] In some implementations, when the individual characteristic parameters and / or gait characteristic parameters are missing, the equivalent conversion model outputs the equivalent walking distance on the ground based on the existing input parameters.

[0013] In some implementations, the method further includes: obtaining the preset prediction standard deviation corresponding to the equivalent conversion model, and the missing proportion of the current input parameters; using the ground equivalent walking distance as the center value, and combining the preset prediction standard deviation and the missing proportion of the input parameters, calculating and outputting the confidence interval range of the ground equivalent walking distance.

[0014] In some implementations, the equivalent conversion model is a supervised regression model trained based on paired training samples; the paired training samples are one-to-one corresponding samples consisting of the original walking distance data obtained by the same subject on non-ground walking test equipment and the measured ground walking distance data obtained in a ground walking environment.

[0015] In line with the above method, another aspect of the present invention provides a ground equivalent distance conversion device, comprising: an acquisition unit configured to acquire raw walking distance data of a subject collected by a non-ground walking test device, as well as the subject's individual characteristic parameters and gait characteristic parameters; an equivalent conversion unit configured to input the raw walking distance data, individual characteristic parameters, and gait characteristic parameters into a pre-trained equivalent conversion model to obtain the ground equivalent walking distance; and a graded output unit configured to match the ground equivalent walking distance with a preset clinical functional grading threshold and output the corresponding clinical functional grading result.

[0016] In conjunction with the above-mentioned device, the present invention further provides a ground equivalent distance conversion system, comprising: the ground equivalent distance conversion device described above.

[0017] In conjunction with the above method, the present invention further provides a storage medium comprising a stored program, wherein, when the program is executed, the device on which the storage medium is located executes the above-described ground equivalent distance conversion method.

[0018] In conjunction with the above method, the present invention further provides a computer program product comprising a computer program that, when processed and executed, implements the steps of the above-described ground equivalent distance conversion method.

[0019] The present invention acquires raw walking distance data of subjects collected by a non-ground walking test device, as well as the subjects' individual characteristic parameters and gait characteristic parameters. These multi-source parameters are then input into a pre-trained equivalent conversion model to calculate the equivalent ground walking distance, which has the same clinical reference value as the standard ground walking test. Finally, based on the equivalent distance and a preset clinical functional grading threshold, the corresponding clinical functional grading result is output. This effectively corrects the systematic deviation between the non-ground walking test distance and the measured ground value, achieving direct alignment between the non-ground test results and the clinical grading standards of the ground walking test, significantly improving the clinical application value and scenario adaptability of the non-ground walking test device.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an embodiment of the ground equivalent distance conversion method of the present invention;

[0023] Figure 2 This is a schematic diagram of a structure of an embodiment of the ground equivalent distance conversion device of the present invention;

[0024] Figure 3 System block diagram;

[0025] Figure 4 This is a system flowchart.

[0026] Referring to the accompanying drawings, the reference numerals in the embodiments of the present invention are as follows:

[0027] 101 - Acquisition unit; 102 - Equivalent conversion unit; 103 - Hierarchical output unit. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0029] According to embodiments of the present invention, a method for converting ground equivalent distances is provided, such as... Figure 1 The flowchart of an embodiment of the method of the present invention is shown. The ground equivalent distance conversion method may include steps S110 to S130.

[0030] In step S110, the raw walking distance data of the subject collected by the non-ground walking test device, as well as the subject's individual characteristic parameters and gait characteristic parameters, are obtained.

[0031] Non-ground walking testing equipment refers to testing and training devices such as treadmills, lower limb rehabilitation robots, and walking training devices that do not require subjects to walk independently on real ground but instead achieve walking movements through the support of the device. These devices typically have a small footprint, controllable motion parameters, and can simultaneously collect multi-dimensional physiological and motor data. Raw walking distance data refers to the total walking distance directly collected by the device after the subject completes the entire test on the non-ground walking testing equipment. This value has not undergone equivalent correction and exhibits a systematic deviation from the measured walking distance in a real-ground environment. Individual characteristic parameters refer to parameters reflecting the subject's basic physiological state and disease background, and are fundamental individual factors affecting walking ability and equivalent distance conversion results. Gait characteristic parameters refer to parameters reflecting the characteristics of the subject's walking movement pattern, reflecting the rhythm of movement, lower limb functional balance, and stability during walking, and are key kinematic characteristics for correcting equivalent distance.

[0032] Current technologies typically use only the raw walking distance as a single variable, performing a simple conversion using a fixed proportional coefficient, resulting in extremely low conversion accuracy. Deviations in the equivalent walking distance stem not only from differences in the movement patterns of the devices themselves but are also highly correlated with individual factors such as the subject's age, physical condition, disease type, and degree of gait abnormality. Introducing individual characteristics and gait features as multi-dimensional inputs can cover the impact of individual differences on the equivalent distance, improving conversion accuracy.

[0033] Specifically, the raw total walking distance of the subject after completing the entire test is read from the data acquisition system of the non-ground walking test equipment. At the same time, the subject's basic physiological information, disease diagnosis information, and gait feature information extracted during the test are collected. The three types of information are used as the input data source for subsequent conversion calculations.

[0034] In some embodiments, the gait characteristic parameters include at least one of gait frequency, stride length, gait symmetry parameters, and gait frequency variation parameters; the individual characteristic parameters include at least one of age, sex, body mass index, and disease diagnosis classification.

[0035] Cadence frequency (CFM) refers to the number of steps taken per unit of time during walking, measured in steps per minute. Stride length refers to the average distance covered by each step, measured in meters. Gait symmetry parameter is a quantitative indicator used to measure the balance of walking performance between the left and right lower limbs. Its value typically ranges from 0 to 1; a value closer to 1 indicates a more balanced gait. This parameter reflects the impact of unilateral limb dysfunction, abnormal neural control, and unilateral pain on walking. Cadence frequency variability parameter is a quantitative indicator used to measure the degree of fluctuation in cadence frequency during walking. It is obtained by statistically analyzing the dispersion of local cadence frequencies in segments. A higher value indicates poorer cadence stability. This parameter reflects the impact of walking balance ability, fatigue level, and neurodegenerative changes on walking stability.

[0036] Gait feature parameters characterize the subject's walking pattern from four dimensions: movement rhythm, stride length, left-right balance, and walking stability. This allows the equivalent conversion model to identify gait abnormalities and the impact of gait adjustments on distance, correcting conversion biases caused by simple distance input. These parameters can all be extracted from existing displacement time-series data from non-ground walking testing equipment, eliminating the need for additional pressure sensors, inertial sensors, or camera equipment. This significantly improves the feature dimensions and conversion accuracy of the conversion model without increasing hardware costs.

[0037] Body Mass Index (BMI) is a quantitative index calculated based on a subject's height and weight. It is a core indicator reflecting body shape and physical load. The higher the BMI, the greater the physical load during walking, and the more significant the impact on walking endurance and distance. Disease classification refers to the classification of a subject according to their corresponding disease type. Different diseases affect walking ability through different pathways. For example, cardiopulmonary diseases limit exercise endurance, while neurological diseases alter gait patterns. Disease classification is a key basis for adapting conversion rules to different pathological populations.

[0038] Individual characteristic parameters cover three core influencing factors: physiological baseline, body morphology, and pathological background. This allows the equivalent conversion model to adapt to differentiated conversion rules for different populations, avoiding fixed-coefficient conversion errors and improving conversion accuracy across different populations. In particular, the introduction of disease diagnosis classification allows the model to establish specific conversion logics for the gait characteristics of different diseases. For example, the endurance decline patterns of people with cardiopulmonary diseases and the gait abnormalities of people with neurological diseases can be differentiated through diagnostic classification, further improving conversion accuracy in clinical scenarios.

[0039] In step S120, the original walking distance data, individual characteristic parameters, and gait characteristic parameters are input into a pre-trained equivalent conversion model to obtain the equivalent walking distance on the ground.

[0040] The equivalent conversion model establishes a mapping relationship between non-ground walking characteristics and ground walking distance, enabling equivalent distance conversion. The ground equivalent walking distance is a value obtained after conversion by the model. This value is clinically equivalent to the measured distance obtained by the subject in completing a walking test of the same intensity on real ground, and can be directly applied to the clinical assessment criteria of ground walking tests.

[0041] The force patterns and movement regulation patterns of the human body in non-ground walking environments differ inherently from those in real-ground walking. There is a systematic bias between the walking performance of subjects on non-ground devices and their performance on the ground. Directly using the raw distances leads to severely inaccurate assessment results, rendering them unusable for clinical decision-making. By learning the complex mapping relationship between multi-dimensional features and real-ground distances through pre-trained models, the inherent biases introduced by non-ground devices can be systematically corrected, yielding equivalent distances with the same clinical reference value as ground-based measurements.

[0042] Specifically, after standardizing and encoding the collected multi-source parameters, they are input into a pre-trained equivalent conversion model. The model processes the input parameters based on the nonlinear mapping between the multi-dimensional features learned during the training phase and the actual walking distance on the ground, and finally outputs the corresponding equivalent walking distance on the ground.

[0043] In some embodiments, the method further includes: acquiring at least two sets of raw walking distance data obtained from multiple tests of the same subject; performing learning effect correction processing on the at least two sets of raw walking distance data to obtain corrected walking distance data; and replacing the raw walking distance data with the corrected walking distance data and inputting it into the equivalent conversion model.

[0044] The learning effect refers to the phenomenon where, when the same subject repeatedly performs a walking test, the walking distance increases with each test due to increasing familiarity with the testing procedure, equipment movement patterns, and operation methods. This increase does not represent an improvement in the subject's actual exercise endurance, cardiopulmonary function, or lower limb function; rather, it is a systematic bias in the testing process, and is typically more pronounced on non-ground walking equipment than on ground.

[0045] By fitting and correcting the learning effect on multiple test data of the same subject, systematic biases caused by the subject's increased proficiency with the equipment and testing procedures can be effectively eliminated, avoiding misjudging increased proficiency as improved motor ability. This improves the accuracy of single equivalent conversion results and ensures longitudinal comparability of results at different test numbers and follow-up points. It is particularly suitable for clinical scenarios that require multiple test comparisons, such as rehabilitation efficacy evaluation and long-term functional follow-up, further improving the clinical applicability of the ground equivalent distance conversion method.

[0046] Specifically, raw walking distance data from multiple sets of walking tests completed by the same subject under the same testing specifications were collected. The data were arranged in the order of the tests, with each set of data corresponding to one complete test process. Based on the distance increment change pattern of multiple tests, a learning curve conforming to saturation characteristics was fitted, and the two core parameters of learning effect strength and learning rate were estimated. Then, combined with the raw distance of the last test, the steady-state walking distance after the subject fully adapted to the device was calculated as the corrected walking distance data. The corrected steady-state walking distance data replaced the original raw walking distance data and was input into the equivalent conversion model along with individual characteristic parameters and gait characteristic parameters.

[0047] For example, the same subject completes three six-minute walking tests on a treadmill, with sufficient rest intervals between each test. The original walking distances of the three tests increase progressively, with the increments gradually decreasing, exhibiting typical learning effect characteristics. By fitting the three sets of distance data, the intensity of the subject's learning effect and the learning rate are estimated, and then the steady-state walking distance after the subject has fully adapted to the treadmill is calculated as the corrected data. This corrected walking distance data replaces the original distance data of the third test, and is combined with individual characteristics and gait characteristic parameters and input into the equivalent conversion model.

[0048] If the same subject completes only one walking test, and the requirement of at least two sets of data for correction is not met, then the learning effect correction process will not be performed, and the original walking distance data from the single test will be directly input into the equivalent conversion model.

[0049] In some implementations, when the individual characteristic parameters and / or gait characteristic parameters are missing, the equivalent conversion model outputs the equivalent walking distance on the ground based on the existing input parameters.

[0050] Some low-end devices lack the ability to collect gait parameters, failing to provide complete gait features. In some scenarios, subjects fail to complete the collection of all basic information, resulting in missing individual characteristic parameters. Most existing conversion methods require all input parameters to be complete, typically using zeros or averages to manually fill in missing values. This method introduces fictitious data, directly leading to systematic biases or even complete distortion in the conversion results. In contrast, the equivalent conversion model possesses missing value inference capabilities. During training, it learns the mapping patterns under different combinations of missing parameters, automatically matching the optimal computational path for missing parameters. It eliminates the need for manually filled-in fictitious values, avoiding estimation errors caused by manual filling and ensuring the conversion process runs normally under varying data completeness. It is not rendered unusable due to the absence of individual parameters, significantly improving its applicability and flexibility.

[0051] Specifically, before inputting parameters into the equivalent conversion model, the validity of all input parameters is verified. Each individual feature parameter and gait feature parameter is checked for valid values, and all missing parameter items are marked. Input parameters containing missing items are then fed into the equivalent conversion model. The model, relying on its own missing value perception logic, performs feature calculations on the valid parameters, automatically adapting to scenarios with missing parameters, and finally outputs the corresponding equivalent walking distance on the ground. This capability can be achieved using a tree model with sparse perception splitting characteristics. The model learns the optimal computation path under different parameter missing states during the training phase, eliminating the need for additional value completion calculations.

[0052] In some implementations, the equivalent conversion model is a supervised regression model trained based on paired training samples; the paired training samples are one-to-one corresponding samples consisting of the original walking distance data obtained by the same subject on non-ground walking test equipment and the measured ground walking distance data obtained in a ground walking environment.

[0053] Supervised regression models refer to regression algorithms within a supervised learning framework. These models use labeled input features and corresponding ground truth outputs as training data, learning the mapping relationship between input features and continuous output values. Ultimately, they can output corresponding continuous numerical results based on new input features, adapting to the prediction needs of continuous values ​​such as equivalent walking distances. Paired training samples refer to paired data used for model training. Each set of samples comes from the same subject and includes input data obtained from non-ground walking tests and ground truth output data obtained from ground walking tests. The two sets of data correspond one-to-one, forming a complete training sample.

[0054] The essence of ground equivalent distance conversion is to establish a nonlinear mapping from non-ground walking features to ground walking distances. This is a continuous numerical prediction task, and supervised regression architecture is the optimal algorithmic paradigm for this task. Compared to empirical conversion methods with fixed ratios, supervised regression models can learn the complex nonlinear relationship between multi-dimensional input features and ground distances, significantly improving conversion accuracy. Simultaneously, the regression model outputs continuous distance values, flexibly adapting to different versions of clinical grading thresholds without being limited by adjustments to grading rules, making its adaptability far superior to classification models that directly output grading results. Training with paired data from the same subject eliminates the interference of individual baseline differences, allowing the model to accurately learn the equivalent correction rules from non-ground test results to the true ground distance, ensuring the accuracy of the conversion results from a data perspective.

[0055] Specifically, during the training data collection phase, the same subject completed both a non-ground walking test and a standard ground walking test under similar physical conditions and test periods. Raw walking distance data from the non-ground device and measured walking distance data from the standard ground environment were collected. These two sets of data were then paired into a single training sample. All training samples followed the same pairing rule to ensure accurate correspondence between input and ground truth. During the model training phase, the prepared paired training samples were divided into training, validation, and test sets. Non-ground walking-related parameters from the samples were used as input features, and measured ground walking distances were used as labeled ground truths to train and optimize the supervised regression model. Through iterative learning, the model continuously reduced the error between predicted distance and actual ground distance, ultimately obtaining a stable equivalent conversion model.

[0056] In step S130, a preset clinical functional grading threshold is matched based on the equivalent walking distance on the ground, and the corresponding clinical functional grading result is output.

[0057] Clinical functional grading thresholds refer to standardized distance cutoff values ​​established based on standard ground walking tests to classify human motor function levels. They serve as the basis for clinical assessment of functional status, judgment of efficacy, and prognosis.

[0058] By directly linking equivalent distances to standardized grading thresholds, the results of off-ground testing can be directly applied to clinical functional assessment, efficacy evaluation, and prognostic stratification, achieving seamless integration between off-ground testing and existing clinical assessment systems.

[0059] Specifically, the equivalent walking distance on the ground output by the model is compared with a pre-stored standardized clinical functional grading threshold table to determine the subject's corresponding motor function level and output the final grading conclusion.

[0060] In some implementations, the method further includes: obtaining the preset prediction standard deviation corresponding to the equivalent conversion model, and the missing proportion of the current input parameters; using the ground equivalent walking distance as the center value, and combining the preset prediction standard deviation and the missing proportion of the input parameters, calculating and outputting the confidence interval range of the ground equivalent walking distance.

[0061] The preset prediction standard deviation refers to the inherent baseline error value obtained from the prediction residual statistics based on independent test samples after the equivalent conversion model has been trained. It reflects the baseline conversion accuracy of the model itself and is a fixed statistic that is updated synchronously with the model version. The smaller the value, the higher the inherent conversion accuracy of the model. The missing proportion refers to the ratio of the number of missing and invalid input parameters to the total number of input parameters in this conversion process, reflecting the completeness of the input information. A higher missing proportion indicates less sufficient input information and greater uncertainty in the conversion result. The confidence interval range refers to the numerical range calculated based on the equivalent walking distance on the ground, combined with the inherent error of the model and the degree of input missingness. It represents the reasonable fluctuation range of the equivalent walking distance on the ground under a preset confidence level. A narrower interval indicates higher reliability, while a wider interval indicates greater uncertainty.

[0062] Specifically, the baseline prediction error value corresponding to the currently used equivalent conversion model is retrieved. This value is obtained from the statistical analysis of the residuals of independent test samples during the model training phase. Simultaneously, the number of missing or invalid parameters among all input parameters in this conversion is counted, and the proportion of missing parameters to all input parameters is calculated to obtain the input missing ratio. Using the equivalent walking distance output by the equivalent conversion model as the central reference point of the interval, and combining it with the preset statistical quantile coefficient, the inherent standard deviation of the model and the amplification caused by the input missing ratio are superimposed to obtain the half-width value of the interval. Then, the upper and lower limits of the interval are calculated to form a complete confidence interval range. Finally, the confidence interval, equivalent walking distance, and clinical grading results are output synchronously. For example, if a subject with unilateral hemiplegia has partially missing gait parameters and significantly abnormal gait symmetry, resulting in a high input parameter missing ratio, the calculated confidence interval width will be significantly increased. The interval will simultaneously cover the threshold range of two adjacent clinical functional grading levels, and the output results will synchronously reflect the cross-level status of the interval, indicating to clinicians that the grading result is uncertain and needs to be comprehensively judged in conjunction with clinical examination and other test results.

[0063] By quantifying the uncertainty range of equivalent conversion results, the reliability of the conversion results is visualized intuitively. It can adapt to application scenarios with different input completeness, accurately reflect the credibility of the results, and reduce the risk of misjudgment in clinical decision-making through interval cross-level prompts. It effectively improves clinical safety and decision reference value, making the output of ground equivalent distance conversion results more rigorous and more in line with clinical use logic.

[0064] This invention achieves precise conversion between off-ground walking test distances and ground-based equivalent distances through a multi-dimensional parameter fusion equivalent conversion model, and directly interfaces with the standardized clinical grading system, fundamentally solving the core problem of incompatibility between off-ground walking test results and ground-based clinical assessment standards. It retains the advantages of off-ground equipment—small footprint, flexible deployment, and controllable parameters—while giving off-ground test results consistent clinical application value with standard ground-based tests. It is widely adaptable to various off-ground walking test and training devices, possessing strong versatility and clinical applicability.

[0065] Figure 3The system diagram illustrates a process where data acquisition begins with treadmill hardware. Through multiple functional modules, the system processes data step-by-step, ultimately outputting a standardized assessment report directly applicable to clinical decision-making. Specifically, the treadmill displacement sensing subsystem provides the raw data source for the entire process, acquiring real-time position x(t) and velocity v(t) time-series data of the subject during walking using displacement sensing devices. This data is then transmitted down to the gait acquisition module, which uses signal processing algorithms such as peak detection to extract four core gait parameters—step frequency f, stride length s, gait symmetry index SI, and step frequency variation coefficient CV_f—from the continuous position sequence. This extraction of gait features requires no additional pressure sensors or inertial measurement units; it relies solely on displacement data. The data then enters the learning effect elimination module. Based on multiple test data from the same subject, a saturation-type learning curve is fitted, automatically correcting for distance deviations caused by the subject's increased familiarity with the equipment and testing procedures. The corrected treadmill walking distance D_t' after eliminating the learning effect is output. The equivalent conversion model is the core computing unit of the system, supporting various algorithm architectures such as XGBoost, Generalized Additive Model (GAM), and linear regression. Its input includes the corrected treadmill distance, all gait parameters, and the subject's basic physiological and pathological parameters. After computation by a pre-trained supervised regression model, it outputs the equivalent six-minute ground walking distance D_g, while also incorporating the model's inherent prediction standard deviation and the completeness of the input parameters. The module calculates and outputs the corresponding confidence interval [D_g-, D_g+] to quantify the uncertainty of the result. This module also has bidirectional data storage interaction with the data management module; it stores the calculation results in the data management module and can also call stored historical data and samples to support model iteration and optimization. After the calculated equivalent distance and confidence interval are passed to the NYHA equivalent grading module, this module calls the NYHA functional grading threshold table, which is maintained independently of the model, to match the equivalent ground distance to the corresponding functional level. If the confidence interval crosses the grading threshold, it will simultaneously mark the grading uncertainty, following the safety guidelines for clinical scenarios. The design principle is as follows: the graded results are then fed down to the report generation module, which integrates all information from the entire chain, including raw test data, gait parameters, equivalent distance, confidence intervals, and cardiac function levels, to generate a standardized clinical report that conforms to clinical guidelines. The final report is output to clinicians for diagnostic and treatment reference, and also supports integration with the hospital's electronic medical record system to achieve seamless integration of test results and clinical treatment processes. The data management module, as a supporting unit of the system, adopts a de-identified and encrypted compliant design, and is responsible for the storage, management, and traceability of data throughout the entire process, while providing a data foundation for the model's sample accumulation and continuous optimization.

[0066] Figure 4The system flowchart includes: S0: Subject preparation, including informed consent, basic information entry, baseline assessment, and contraindication screening to ensure test compliance and safety; S1: Initiating the six-minute treadmill walking test, collecting real-time location time-series data x(t) and speed time-series data v(t) of the subjects throughout the test, serving as the raw motion data source for subsequent gait analysis and distance calculation; S2: Simultaneously collecting multi-dimensional physiological monitoring parameters of the subjects, such as heart rate, blood oxygen saturation, blood pressure, and electrocardiogram, for both safety monitoring during the test and as data collection. S3: Based on the aforementioned collected displacement time-series data, extract four core gait parameters—step frequency, stride length, gait symmetry index, and step frequency variation coefficient—using signal processing algorithms such as peak detection; S4: Determine whether this is multiple tests conducted by the same subject: If determined to be multiple tests, proceed to S5, the learning effect elimination step. By fitting the learning effect coefficient λ and learning rate μ to the distance sequence of multiple tests, correct the distance deviation caused by the improvement in proficiency, output the steady-state treadmill walking distance D_t' after eliminating the learning effect, and then... The distance is then sent to the equivalent conversion stage; if it is determined to be a single test, the correction step is skipped, and the original treadmill distance data is directly sent to the S6 equivalent conversion stage; S6: Using a pre-trained equivalent conversion model, the treadmill walking distance, gait parameters, and individual characteristic parameters of the subject are fused to complete the calculation, outputting the equivalent ground walking distance D_g and simultaneously generating a preliminary confidence interval; S7: Confidence assessment is performed, combining the model's inherent prediction standard deviation and the missing proportion of the input parameters to determine the upper and lower limits of the final equivalent distance confidence interval, quantifying the uncertainty of the result. If the clinical grading threshold is exceeded, a grading prompt will be simultaneously displayed; S8: Based on the equivalent ground walking distance, match the independently maintained NYHA functional grading threshold and output the corresponding functional grading results of grade I / II / III / IV; S9: Integrate all information such as the original test data, gait parameters, equivalent distance results, confidence intervals, and functional grading throughout the process to generate a standardized test report that conforms to clinical norms; S10: Output the report to clinicians for reference in diagnosis and treatment decisions, and also support integration with the hospital's electronic medical record system to achieve seamless integration of test data and clinical diagnosis and treatment processes.

[0067] In some embodiments, a regression algorithm architecture under a supervised learning framework is adopted. The overall construction process includes, in sequence: task definition, training dataset construction, feature engineering, model selection and parameter configuration, training execution, performance verification, interpretability implementation and long-term iteration mechanism, and finally forms a computational model that can stably output the equivalent walking distance on the ground.

[0068] The first step in model building is defining the task boundaries. This establishes a non-linear mapping between the multi-dimensional features of the off-ground walking test and the standard six-minute walking distance on the ground, outputting a continuous, equivalent walking distance value. A regression task was chosen over a classification task for two reasons: firstly, clinical assessment requires meter-level precision in distance measurements, and classification outputs would lose this precision, failing to meet the refined requirements of efficacy assessment and follow-up comparisons; secondly, it decouples the clinical grading thresholds from model computation. When subsequent clinical guidelines update grading standards, only the threshold configuration needs adjustment, without retraining the model, resulting in greater adaptability.

[0069] The training dataset is a fundamental step in model building. The training dataset strictly utilizes paired samples from the same subject on the same day. This means the same subject, under similar physical conditions, completes both a six-minute treadmill walk test and a six-minute ground walk test. Treadmill-based feature data is used as input, and the measured ground walking distance is used as the ground truth, creating a one-to-one paired training dataset. This eliminates interference from individual baseline differences, allowing the model to accurately learn and correct for systematic biases not caused by the ground equipment itself. The training dataset covers healthy controls, individuals with cardiovascular disease, and those with chronic respiratory diseases, ranging in age from 20 to 50 years, and includes subjects of different genders and body mass indexes, ensuring the model's generalization ability across diverse populations. All samples are divided into training, validation, and test sets at a ratio of 70%, 15%, and 15%, respectively. The training set is used to fit the model's core parameters, the validation set is used for hyperparameter tuning and early stop control, and the test set is used to finally evaluate the model's generalization performance. Test set data is not visible throughout the training process to ensure the objectivity of performance evaluation.

[0070] Feature engineering is the core step in improving model accuracy. The original input to the model consists of 9 features, namely, treadmill walking distance after learning effect correction, cadence, stride length, gait symmetry index, cadence variation coefficient, age, gender, body mass index, and disease diagnosis classification, covering four core influencing factors: athletic ability, gait pattern, physiological baseline, and pathological background. In the feature encoding stage, the 7 continuous features are normalized using Z-scores, and standardization is performed based on the mean and standard deviation of each feature from the training set, eliminating the influence of different units on model operation. Gender is a binary feature and its numerical input is directly retained. Disease diagnosis classification is expanded into a 5-dimensional feature using one-hot encoding. The final actual input dimension of the model is 13 dimensions. To address the common problem of missing data in clinical scenarios, the model natively supports missing value inference, using non-numeric labels for missing features without the need for padding with 0s or population means. This capability is achieved through a missing value-aware splitting mechanism. During the training phase, for each splitting node, the information gain of the splitting direction is calculated for both feature presence and feature absence. The direction with the greater gain is selected as the default computation path for the missing feature. If the corresponding feature is missing during inference, the computation is completed along the default path, avoiding the introduction of fictitious data errors by manual padding and achieving graceful degradation under different data completeness scenarios.

[0071] Model selection and core architecture design are crucial decision-making steps in the construction process. Suitable supervised regression algorithms include four categories: multiple linear regression, generalized additive models, extreme gradient boosting trees, and multilayer perceptron neural networks. After comprehensive comparison, the extreme gradient boosting tree model demonstrated the best overall performance across five dimensions: conversion accuracy, inference speed, missing value support, interpretability, and deployment size, making it the primary chosen architecture. This model belongs to the gradient boosting tree architecture in ensemble learning, employing an additive model approach where multiple shallow decision trees jointly complete the prediction. Each newly generated decision tree specifically corrects the cumulative prediction error of all previous trees, gradually reducing the overall prediction residual through residual learning and gradient boosting mechanisms. Compared to traditional gradient boosting decision trees, this model introduces a second-order Taylor expansion to optimize the loss function, improving training convergence speed and fitting accuracy; it incorporates L1 and L2 regularization terms to control model complexity and explicitly reduce the risk of overfitting; it achieves feature-level parallel training and pre-sorting optimization, significantly improving training efficiency; and it natively supports the aforementioned missing value-aware splitting mechanism, adapting to the characteristics of clinical data.

[0072] Model hyperparameter configuration and training execution are the execution phases that translate the algorithm into a usable model. Recommended hyperparameter configurations are: 200 decision trees, maximum tree depth of 6, learning rate of 0.05, sample subsampling ratio of 0.8, feature subsampling ratio of 0.8, minimum child node weight of 5, L1 regularization coefficient of 0.1, and L2 regularization coefficient of 1.0. The loss function used is the mean absolute error (MAE), whose unit is consistent with the walking distance unit, making it more intuitive for clinical interpretation and less sensitive to outliers, thus reducing the interference of extreme scenarios such as subjects stopping mid-training on the model. An early stopping mechanism is employed during training, with 20 early stopping rounds. If the prediction error on the validation set does not improve after 20 consecutive training rounds, training is terminated early to ensure fitting accuracy while avoiding overfitting. The overall training objective is to minimize the mean absolute error between the predicted values ​​and the ground truth, while controlling model complexity and balancing accuracy and generalization ability.

[0073] The model performance validation system serves as the acceptance phase of the model development process, validating model performance across two dimensions: regression accuracy and clinical accuracy. For regression accuracy, mean absolute error (MAO), root mean square error (RMSE), coefficient of determination (CDO), and Pearson correlation coefficient (CCR) are used as core evaluation metrics. On the independent test set, the extreme gradient boosting tree model achieves a MAO of approximately 15 meters, a RMSE of approximately 19 meters, and a CDO of 0.97, meeting the accuracy requirements for clinical applications. For clinical accuracy, the core evaluation metrics are the consistency rate of cardiac function classification, the weighted Coencapsulation coefficient, and the rate of class crossing. After matching the equivalent distance of the model output to the NYHA classification threshold, the consistency rate between the cardiac function classification and the clinician's judgment reaches 95%, and the rate of class crossing the confidence interval is less than 5%, meeting the accuracy and safety requirements of clinical assessment.

[0074] Interpretability is a crucial supplementary element for adapting models to clinical scenarios. The model is equipped with an additive interpretation system for Shapley values, which decomposes the prediction result of a single sample into the contribution of each input feature dimension. This clarifies the direction and magnitude of each feature's influence on the final equivalent distance, while also statistically analyzing the global feature importance and quantifying the contribution percentage of each input dimension to the overall prediction. This system provides a basis for clinicians to interpret results and supports model validation and regulatory compliance, thus compensating for the lack of clinical interpretability in black-box models.

[0075] A long-term iteration mechanism is designed as a guarantee for the entire lifecycle operation and maintenance of the model. The model is equipped with an online update mechanism with four types of retraining trigger conditions: triggering at fixed time periods, triggering when the cumulative number of newly added paired samples reaches a preset number, triggering when the online prediction error continuously exceeds a threshold, triggering when a new diagnostic group is added, or triggering when a partner hospital or equipment model is selected. Model updates follow a standardized process, completing data preparation, model retraining, performance evaluation, canary release, and full release in sequence. Each version corresponds to a unique version number, supporting historical version rollback and fault rollback. At the same time, a data closed-loop system is built. The input data and prediction results of each inference are stored synchronously. Ground-based measured values ​​can be fed back through the hospital's electronic medical record system, automatically accumulating to form new paired training samples, continuously expanding the training dataset, and supporting continuous iterative optimization of model accuracy. The model construction process is equipped with full-process boundary handling rules. Training is rejected when the number of training samples is lower than a preset threshold. Prediction is rejected and supplementary data is prompted when a new diagnostic group is not covered. Data cleaning and retry are automatically performed when numerical anomalies occur during model training. If the model file is corrupted, it supports rollback to the previous stable version, ensuring the operational stability of the model throughout its entire lifecycle.

[0076] The technical solution of this embodiment collects the raw walking distance data of the subjects output by the non-ground walking test device, and simultaneously obtains the subjects' individual characteristic parameters and gait characteristic parameters. All input parameters are fed into a pre-trained equivalent conversion model to calculate the ground walking distance with clinical equivalence. Then, the corresponding classification conclusion is output by comparing it with the preset clinical functional classification threshold. This can effectively correct the systematic deviation between the non-ground device test distance and the ground measured value, realize the seamless connection between the non-ground test results and the ground clinical classification system, and greatly expand the clinical application scenarios and application value of the non-ground walking test device.

[0077] According to embodiments of the present invention, a ground equivalent distance conversion device corresponding to the ground equivalent distance conversion method is also provided. See also Figure 2 The schematic diagram shown is a structural diagram of an embodiment of the device of the present invention. The ground equivalent distance conversion device may include: an acquisition unit 101, an equivalent conversion unit 102, and a hierarchical output unit 103.

[0078] The acquisition unit 101 is configured to acquire raw walking distance data of the subject collected by the non-ground walking test device, as well as the subject's individual characteristic parameters and gait characteristic parameters.

[0079] Non-ground walking testing equipment refers to testing and training devices such as treadmills, lower limb rehabilitation robots, and walking training devices that do not require subjects to walk independently on real ground but instead achieve walking movements through the support of the device. These devices typically have a small footprint, controllable motion parameters, and can simultaneously collect multi-dimensional physiological and motor data. Raw walking distance data refers to the total walking distance directly collected by the device after the subject completes the entire test on the non-ground walking testing equipment. This value has not undergone equivalent correction and exhibits a systematic deviation from the measured walking distance in a real-ground environment. Individual characteristic parameters refer to parameters reflecting the subject's basic physiological state and disease background, and are fundamental individual factors affecting walking ability and equivalent distance conversion results. Gait characteristic parameters refer to parameters reflecting the characteristics of the subject's walking movement pattern, reflecting the rhythm of movement, lower limb functional balance, and stability during walking, and are key kinematic characteristics for correcting equivalent distance.

[0080] Current technologies typically use only the raw walking distance as a single variable, performing a simple conversion using a fixed proportional coefficient, resulting in extremely low conversion accuracy. Deviations in the equivalent walking distance stem not only from differences in the movement patterns of the devices themselves but are also highly correlated with individual factors such as the subject's age, physical condition, disease type, and degree of gait abnormality. Introducing individual characteristics and gait features as multi-dimensional inputs can cover the impact of individual differences on the equivalent distance, improving conversion accuracy.

[0081] Specifically, the raw total walking distance of the subject after completing the entire test is read from the data acquisition system of the non-ground walking test equipment. At the same time, the subject's basic physiological information, disease diagnosis information, and gait feature information extracted during the test are collected. The three types of information are used as the input data source for subsequent conversion calculations.

[0082] In some embodiments, the gait characteristic parameters include at least one of gait frequency, stride length, gait symmetry parameters, and gait frequency variation parameters; the individual characteristic parameters include at least one of age, sex, body mass index, and disease diagnosis classification.

[0083] Cadence frequency (CFM) refers to the number of steps taken per unit of time during walking, measured in steps per minute. Stride length refers to the average distance covered by each step, measured in meters. Gait symmetry parameter is a quantitative indicator used to measure the balance of walking performance between the left and right lower limbs. Its value typically ranges from 0 to 1; a value closer to 1 indicates a more balanced gait. This parameter reflects the impact of unilateral limb dysfunction, abnormal neural control, and unilateral pain on walking. Cadence frequency variability parameter is a quantitative indicator used to measure the degree of fluctuation in cadence frequency during walking. It is obtained by statistically analyzing the dispersion of local cadence frequencies in segments. A higher value indicates poorer cadence stability. This parameter reflects the impact of walking balance ability, fatigue level, and neurodegenerative changes on walking stability.

[0084] Gait feature parameters characterize the subject's walking pattern from four dimensions: movement rhythm, stride length, left-right balance, and walking stability. This allows the equivalent conversion model to identify gait abnormalities and the impact of gait adjustments on distance, correcting conversion biases caused by simple distance input. These parameters can all be extracted from existing displacement time-series data from non-ground walking testing equipment, eliminating the need for additional pressure sensors, inertial sensors, or camera equipment. This significantly improves the feature dimensions and conversion accuracy of the conversion model without increasing hardware costs.

[0085] Body Mass Index (BMI) is a quantitative index calculated based on a subject's height and weight. It is a core indicator reflecting body shape and physical load. The higher the BMI, the greater the physical load during walking, and the more significant the impact on walking endurance and distance. Disease classification refers to the classification of a subject according to their corresponding disease type. Different diseases affect walking ability through different pathways. For example, cardiopulmonary diseases limit exercise endurance, while neurological diseases alter gait patterns. Disease classification is a key basis for adapting conversion rules to different pathological populations.

[0086] Individual characteristic parameters cover three core influencing factors: physiological baseline, body morphology, and pathological background. This allows the equivalent conversion model to adapt to differentiated conversion rules for different populations, avoiding fixed-coefficient conversion errors and improving conversion accuracy across different populations. In particular, the introduction of disease diagnosis classification allows the model to establish specific conversion logics for the gait characteristics of different diseases. For example, the endurance decline patterns of people with cardiopulmonary diseases and the gait abnormalities of people with neurological diseases can be differentiated through diagnostic classification, further improving conversion accuracy in clinical scenarios.

[0087] The equivalent conversion unit 102 is configured to input the original walking distance data, individual feature parameters and gait feature parameters into a pre-trained equivalent conversion model to obtain the equivalent walking distance on the ground.

[0088] The equivalent conversion model establishes a mapping relationship between non-ground walking characteristics and ground walking distance, enabling equivalent distance conversion. The ground equivalent walking distance is a value obtained after conversion by the model. This value is clinically equivalent to the measured distance obtained by the subject in completing a walking test of the same intensity on real ground, and can be directly applied to the clinical assessment criteria of ground walking tests.

[0089] The force patterns and movement regulation patterns of the human body in non-ground walking environments differ inherently from those in real-ground walking. There is a systematic bias between the walking performance of subjects on non-ground devices and their performance on the ground. Directly using the raw distances leads to severely inaccurate assessment results, rendering them unusable for clinical decision-making. By learning the complex mapping relationship between multi-dimensional features and real-ground distances through pre-trained models, the inherent biases introduced by non-ground devices can be systematically corrected, yielding equivalent distances with the same clinical reference value as ground-based measurements.

[0090] Specifically, after standardizing and encoding the collected multi-source parameters, they are input into a pre-trained equivalent conversion model. The model processes the input parameters based on the nonlinear mapping between the multi-dimensional features learned during the training phase and the actual walking distance on the ground, and finally outputs the corresponding equivalent walking distance on the ground.

[0091] In some embodiments, the equivalent conversion unit 102 is further configured to: acquire at least two sets of original walking distance data obtained by the same subject through multiple tests; perform learning effect correction processing on the at least two sets of original walking distance data to obtain corrected walking distance data; and input the corrected walking distance data into the equivalent conversion model instead of the original walking distance data.

[0092] The learning effect refers to the phenomenon where, when the same subject repeatedly performs a walking test, the walking distance increases with each test due to increasing familiarity with the testing procedure, equipment movement patterns, and operation methods. This increase does not represent an improvement in the subject's actual exercise endurance, cardiopulmonary function, or lower limb function; rather, it is a systematic bias in the testing process, and is typically more pronounced on non-ground walking equipment than on ground.

[0093] By fitting and correcting the learning effect on multiple test data of the same subject, systematic biases caused by the subject's increased proficiency with the equipment and testing procedures can be effectively eliminated, avoiding misjudging increased proficiency as improved motor ability. This improves the accuracy of single equivalent conversion results and ensures longitudinal comparability of results at different test numbers and follow-up points. It is particularly suitable for clinical scenarios that require multiple test comparisons, such as rehabilitation efficacy evaluation and long-term functional follow-up, further improving the clinical applicability of the ground equivalent distance conversion method.

[0094] Specifically, raw walking distance data from multiple sets of walking tests completed by the same subject under the same testing specifications were collected. The data were arranged in the order of the tests, with each set of data corresponding to one complete test process. Based on the distance increment change pattern of multiple tests, a learning curve conforming to saturation characteristics was fitted, and the two core parameters of learning effect strength and learning rate were estimated. Then, combined with the raw distance of the last test, the steady-state walking distance after the subject fully adapted to the device was calculated as the corrected walking distance data. The corrected steady-state walking distance data replaced the original raw walking distance data and was input into the equivalent conversion model along with individual characteristic parameters and gait characteristic parameters.

[0095] In some implementations, the equivalent conversion unit 102 is further configured to: when the individual characteristic parameters and / or gait characteristic parameters are missing, the equivalent conversion model outputs the equivalent walking distance on the ground based on the existing input parameters.

[0096] Some low-end devices lack the ability to collect gait parameters, failing to provide complete gait features. In some scenarios, subjects fail to complete the collection of all basic information, resulting in missing individual characteristic parameters. Most existing conversion methods require all input parameters to be complete, typically using zeros or averages to manually fill in missing values. This method introduces fictitious data, directly leading to systematic biases or even complete distortion in the conversion results. In contrast, the equivalent conversion model possesses missing value inference capabilities. During training, it learns the mapping patterns under different combinations of missing parameters, automatically matching the optimal computational path for missing parameters. It eliminates the need for manually filled-in fictitious values, avoiding estimation errors caused by manual filling and ensuring the conversion process runs normally under varying data completeness. It is not rendered unusable due to the absence of individual parameters, significantly improving its applicability and flexibility.

[0097] Specifically, before inputting parameters into the equivalent conversion model, the validity of all input parameters is verified. Each individual feature parameter and gait feature parameter is checked for valid values, and all missing parameter items are marked. Input parameters containing missing items are then fed into the equivalent conversion model. The model, relying on its own missing value perception logic, performs feature calculations on the valid parameters, automatically adapting to scenarios with missing parameters, and finally outputs the corresponding equivalent walking distance on the ground. This capability can be achieved using a tree model with sparse perception splitting characteristics. The model learns the optimal computation path under different parameter missing states during the training phase, eliminating the need for additional value completion calculations.

[0098] In some implementations, the equivalent conversion model is a supervised regression model trained based on paired training samples; the paired training samples are one-to-one corresponding samples consisting of the original walking distance data obtained by the same subject on non-ground walking test equipment and the measured ground walking distance data obtained in a ground walking environment.

[0099] Supervised regression models refer to regression algorithms within a supervised learning framework. These models use labeled input features and corresponding ground truth outputs as training data, learning the mapping relationship between input features and continuous output values. Ultimately, they can output corresponding continuous numerical results based on new input features, adapting to the prediction needs of continuous values ​​such as equivalent walking distances. Paired training samples refer to paired data used for model training. Each set of samples comes from the same subject and includes input data obtained from non-ground walking tests and ground truth output data obtained from ground walking tests. The two sets of data correspond one-to-one, forming a complete training sample.

[0100] The essence of ground equivalent distance conversion is to establish a nonlinear mapping from non-ground walking features to ground walking distances. This is a continuous numerical prediction task, and supervised regression architecture is the optimal algorithmic paradigm for this task. Compared to empirical conversion methods with fixed ratios, supervised regression models can learn the complex nonlinear relationship between multi-dimensional input features and ground distances, significantly improving conversion accuracy. Simultaneously, the regression model outputs continuous distance values, flexibly adapting to different versions of clinical grading thresholds without being limited by adjustments to grading rules, making its adaptability far superior to classification models that directly output grading results. Training with paired data from the same subject eliminates the interference of individual baseline differences, allowing the model to accurately learn the equivalent correction rules from non-ground test results to the true ground distance, ensuring the accuracy of the conversion results from a data perspective.

[0101] Specifically, during the training data collection phase, the same subject completed both a non-ground walking test and a standard ground walking test under similar physical conditions and test periods. Raw walking distance data from the non-ground device and measured walking distance data from the standard ground environment were collected. These two sets of data were then paired into a single training sample. All training samples followed the same pairing rule to ensure accurate correspondence between input and ground truth. During the model training phase, the prepared paired training samples were divided into training, validation, and test sets. Non-ground walking-related parameters from the samples were used as input features, and measured ground walking distances were used as labeled ground truths to train and optimize the supervised regression model. Through iterative learning, the model continuously reduced the error between predicted distance and actual ground distance, ultimately obtaining a stable equivalent conversion model.

[0102] The graded output unit 103 is configured to match a preset clinical functional graded threshold with the ground equivalent walking distance and output the corresponding clinical functional graded result.

[0103] Clinical functional grading thresholds refer to standardized distance cutoff values ​​established based on standard ground walking tests to classify human motor function levels. They serve as the basis for clinical assessment of functional status, judgment of efficacy, and prognosis.

[0104] By directly linking equivalent distances to standardized grading thresholds, the results of off-ground testing can be directly applied to clinical functional assessment, efficacy evaluation, and prognostic stratification, achieving seamless integration between off-ground testing and existing clinical assessment systems.

[0105] Specifically, the equivalent walking distance on the ground output by the model is compared with a pre-stored standardized clinical functional grading threshold table to determine the subject's corresponding motor function level and output the final grading conclusion.

[0106] In some implementations, the equivalent conversion unit 102 is further configured to: obtain the preset prediction standard deviation corresponding to the equivalent conversion model, and the missing proportion of the current input parameters; and calculate and output the confidence interval range of the ground equivalent walking distance with the ground equivalent walking distance as the center value, combined with the preset prediction standard deviation and the missing proportion of the input parameters.

[0107] The preset prediction standard deviation refers to the inherent baseline error value obtained from the prediction residual statistics based on independent test samples after the equivalent conversion model has been trained. It reflects the baseline conversion accuracy of the model itself and is a fixed statistic that is updated synchronously with the model version. The smaller the value, the higher the inherent conversion accuracy of the model. The missing proportion refers to the ratio of the number of missing and invalid input parameters to the total number of input parameters in this conversion process, reflecting the completeness of the input information. A higher missing proportion indicates less sufficient input information and greater uncertainty in the conversion result. The confidence interval range refers to the numerical range calculated based on the equivalent walking distance on the ground, combined with the inherent error of the model and the degree of input missingness. It represents the reasonable fluctuation range of the equivalent walking distance on the ground under a preset confidence level. A narrower interval indicates higher reliability, while a wider interval indicates greater uncertainty.

[0108] Specifically, the baseline prediction error value corresponding to the currently used equivalent conversion model is retrieved. This value is obtained from the statistical analysis of the residuals of independent test samples during the model training phase. Simultaneously, the number of missing or invalid parameters among all input parameters in this conversion is counted, and the proportion of missing parameters to all input parameters is calculated to obtain the input missing ratio. Using the equivalent walking distance output by the equivalent conversion model as the central reference point of the interval, and combining it with the preset statistical quantile coefficient, the inherent standard deviation of the model and the amplification caused by the input missing ratio are superimposed to obtain the half-width value of the interval. Then, the upper and lower limits of the interval are calculated to form a complete confidence interval range. Finally, the confidence interval, equivalent walking distance, and clinical grading results are output synchronously. For example, if a subject with unilateral hemiplegia has partially missing gait parameters and significantly abnormal gait symmetry, resulting in a high input parameter missing ratio, the calculated confidence interval width will be significantly increased. The interval will simultaneously cover the threshold range of two adjacent clinical functional grading levels, and the output results will synchronously reflect the cross-level status of the interval, indicating to clinicians that the grading result is uncertain and needs to be comprehensively judged in conjunction with clinical examination and other test results.

[0109] By quantifying the uncertainty range of equivalent conversion results, the reliability of the conversion results is visualized intuitively. It can adapt to application scenarios with different input completeness, accurately reflect the credibility of the results, and reduce the risk of misjudgment in clinical decision-making through interval cross-level prompts. It effectively improves clinical safety and decision reference value, making the output of ground equivalent distance conversion results more rigorous and more in line with clinical use logic.

[0110] This invention achieves precise conversion between off-ground walking test distances and ground-based equivalent distances through a multi-dimensional parameter fusion equivalent conversion model, and directly interfaces with the standardized clinical grading system, fundamentally solving the core problem of incompatibility between off-ground walking test results and ground-based clinical assessment standards. It retains the advantages of off-ground equipment—small footprint, flexible deployment, and controllable parameters—while giving off-ground test results consistent clinical application value with standard ground-based tests. It is widely adaptable to various off-ground walking test and training devices, possessing strong versatility and clinical applicability.

[0111] Since the processing and functions implemented by the device in this embodiment are basically the same as the embodiments, principles and examples of the aforementioned methods, any details not covered in the description of this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be repeated here.

[0112] By employing the technical solution of this invention, the original walking distance, individual physiological and pathological parameters of the subject, and gait characteristic parameters collected by the non-ground walking test device are obtained. The equivalent walking distance on the ground is derived through a pre-trained equivalent conversion model, and then the grading result is output by matching the standardized clinical functional grading threshold. This can eliminate the inherent distance bias caused by the non-ground device, allowing the non-ground test results to be directly adapted to the existing clinical grading system, effectively improving the clinical reference value and applicability of the non-ground walking test.

[0113] According to an embodiment of the present invention, a ground equivalent distance conversion system corresponding to a ground equivalent distance conversion device is also provided. This ground equivalent distance conversion system may include the ground equivalent distance conversion device described above.

[0114] Since the processing and functions implemented by the ground equivalent distance conversion system in this embodiment are basically the same as those of the aforementioned device embodiments, principles and examples, any details not covered in this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be repeated here.

[0115] The technical solution of this invention acquires three types of data: the original walking distance, the individual characteristics of the subject, and the gait characteristics collected by the non-ground walking test device. These data are then input into a pre-trained equivalent conversion model to obtain the equivalent walking distance on the ground. Finally, the preset clinical functional grading threshold is matched to output the grading result. This can systematically correct the distance deviation between the non-ground test and the actual ground measurement, and achieve direct connection between the non-ground test and the ground clinical grading standard, significantly improving the clinical application value and universality of the non-ground walking test device.

[0116] According to an embodiment of the present invention, a storage medium corresponding to the ground equivalent distance conversion method is also provided. The storage medium includes a stored program, wherein the program controls the device where the storage medium is located to execute the ground equivalent distance conversion method described above when it is running.

[0117] Since the processing and functions implemented by the storage medium in this embodiment are basically the same as the embodiments, principles and examples of the aforementioned methods, any details not covered in this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be repeated here.

[0118] The technical solution of this invention integrates three types of multi-source data: raw walking distance collected by non-ground walking test equipment, basic information of the subject's individual, and walking gait characteristics. These data are then input into a pre-trained equivalent conversion model to complete the distance equivalence conversion, resulting in a walking distance with the same reference value as ground-based measurements. The corresponding clinical functional grading threshold is then matched to output a grading conclusion, which can effectively correct the systematic distance error of non-ground walking tests. This allows non-ground test results to directly adapt to ground-based clinical grading standards, significantly improving the clinical practical value of non-ground walking test equipment.

[0119] According to an embodiment of the present invention, a computer program product corresponding to the ground equivalent distance conversion method is also provided. The computer program product includes a computer program that, when processed and executed, implements the steps of the above-described ground equivalent distance conversion method.

[0120] Since the processing and functions implemented by the computer program product in this embodiment are basically corresponding to the embodiments, principles and examples of the aforementioned methods, any details not covered in the description of this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be repeated here.

[0121] The technical solution of this invention extracts the raw walking distance data of the subjects from the non-ground walking test device, simultaneously collects the subjects' individual characteristics and gait characteristics, and feeds them into a pre-trained equivalent conversion model to calculate the equivalent walking distance on the ground. Then, based on the preset clinical functional grading threshold, the corresponding grading result is output. This can correct the inherent deviation between the non-ground test and the actual walking distance on the ground, realize the direct adaptation of the non-ground test to the existing ground clinical grading system, and effectively expand the clinical application scenarios and reference value of the non-ground walking test device.

[0122] In summary, it is readily understood by those skilled in the art that, without conflict, the aforementioned advantageous methods can be freely combined and superimposed.

[0123] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for converting equivalent ground distances, characterized in that, include: Acquire raw walking distance data of subjects collected by non-ground walking testing equipment, as well as individual characteristic parameters and gait characteristic parameters of subjects; The original walking distance data, individual characteristic parameters, and gait characteristic parameters are input into a pre-trained equivalent conversion model to obtain the equivalent walking distance on the ground. Based on the equivalent walking distance on the ground, a preset clinical functional grading threshold is matched, and the corresponding clinical functional grading result is output.

2. The ground equivalent distance conversion method according to claim 1, characterized in that, Also includes: Obtain at least two sets of raw walking distance data from the same subject through multiple tests; The learning effect correction process is applied to the at least two sets of original walking distance data to obtain corrected walking distance data; The corrected walking distance data is used to replace the original walking distance data and is input into the equivalent conversion model.

3. The ground equivalent distance conversion method according to claim 1, characterized in that, The gait characteristic parameters include at least one of gait frequency, stride length, gait symmetry parameters, and gait frequency variation parameters; the individual characteristic parameters include at least one of age, sex, body mass index, and disease diagnosis classification.

4. The ground equivalent distance conversion method according to claim 1 or 3, characterized in that, When the individual characteristic parameters and / or gait characteristic parameters are missing, the equivalent conversion model outputs the equivalent walking distance on the ground based on the existing input parameters.

5. The ground equivalent distance conversion method according to any one of claims 1 to 3, characterized in that, Also includes: Obtain the preset prediction standard deviation corresponding to the equivalent conversion model, as well as the missing proportion of the input parameters; Using the equivalent walking distance on the ground as the center value, and combining the preset prediction standard deviation with the missing proportion of input parameters, the confidence interval range of the equivalent walking distance on the ground is calculated and output.

6. The ground equivalent distance conversion method according to claim 1, characterized in that, The equivalent conversion model is a supervised regression model trained based on paired training samples; the paired training samples are one-to-one corresponding samples consisting of the original walking distance data obtained by the same subject on non-ground walking test equipment and the measured ground walking distance data obtained in the ground walking environment.

7. A ground equivalent distance conversion device, characterized in that, include: The acquisition unit is configured to acquire raw walking distance data of the subject collected by the non-ground walking test device, as well as the subject's individual characteristic parameters and gait characteristic parameters; The equivalent conversion unit is configured to input the original walking distance data, individual feature parameters and gait feature parameters into a pre-trained equivalent conversion model to obtain the equivalent walking distance on the ground. The graded output unit is configured to match a preset clinical functional graded threshold with the equivalent walking distance on the ground and output the corresponding clinical functional graded result.

8. A ground equivalent distance conversion system, characterized in that, include: The ground equivalent distance conversion device as described in claim 7.

9. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the ground equivalent distance conversion method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the ground equivalent distance conversion method as described in any one of claims 1 to 6.

Images