Long-time-history motion test system

Abstract

A long-time-history motion test system comprises a data acquisition unit, a data fitting unit, a principal component analysis unit and an index inspection unit. The data acquisition unit is used for recording the amplitude and area of the muscle composite action potential to obtain an original LET data set; the data fitting unit is used for performing data fitting on each piece of LET data in the original LET data set according to a fourth-order B-spline basis function to obtain a function type data set formed by LET functions; the principal component analysis unit is used for performing principal component analysis on each LET function in the function type data set, and obtaining a pre-diagnosis index according to the principal component analysis; and the index inspection unit is used for constructing a periodic paralysis diagnosis model according to the pre-diagnosis indexes, performing logistic regression analysis on the diagnosis model, and extracting diagnosis indexes from the pre-diagnosis indexes. According to the invention, more useful indexes can be extracted for combined application, the sensitivity is further improved on the premise of ensuring the specificity, and the periodic paralysis can be diagnosed earlier and more accurately.

Description

Technical Field

[0001] The invention belongs to the technical field of medical equipment, and in particular relates to a long-term exercise test system. Background Art

[0002] Primary periodic paralysis is a group of skeletal muscle ion channel diseases caused by mutations in calcium, sodium or potassium channel genes, characterized by paroxysmal limb paralysis, including hypokalemic periodic paralysis, hyperkalemic periodic paralysis, Anderson-Tavel syndrome and thyrotoxic hypokalemic periodic paralysis. Attacks can last from hours to days, seriously affecting the patient's psychology, work and life. Long-term exercise test is an electrophysiological method for specific diagnosis of periodic paralysis. Long-term exercise test is to let the subject exert the maximum force on the abductor digiti minimi muscle, causing fatigue of the abductor digiti minimi muscle, resulting in characteristic changes in the muscle compound action potential (CMAP). However, the existing long-term exercise test only focuses on the percentage of maximum amplitude decrease in diagnostic indicators and diagnostic criteria, but does not focus on the dynamic change characteristics of the CMAP amplitude during the whole process, resulting in too single test indicators. This leads to the poor accuracy of this diagnostic criterion. Therefore, how to develop a new long-term exercise test system that can further improve the sensitivity while ensuring specificity, and then achieve earlier and more accurate diagnosis of periodic paralysis, is a direction that technicians in this field need to study. Summary of the invention

[0003] The purpose of the present invention is to provide a long-term exercise test system, which can extract more effective diagnostic indicators for primary periodic paralysis under the premise of ensuring specificity, so as to achieve an earlier and more accurate diagnosis of periodic paralysis.

[0004] The invention discloses a long-term exercise test system, which comprises:

[0005] A data acquisition unit, wherein the data acquisition unit is used to record the amplitude and area of ​​the muscle compound action potential and obtain an original LET data set;

[0006] A data fitting unit, the data fitting unit is used to perform data fitting on each LET data in the original LET data set according to a fourth-order B-spline basis function to obtain a function-type data set composed of LET functions;

[0007] A principal component analysis unit, the principal component analysis unit is used to perform principal component analysis on each LET function in the functional data set, and obtain a pre-diagnosis index according to the principal component analysis;

[0008] An index verification unit is used to construct a diagnostic model for periodic paralysis according to the pre-diagnosis index, and to perform logistic regression analysis on the diagnostic model to extract diagnostic indexes.

[0009] Preferably, the data acquisition unit is also used to process missing data in the original LET data by linear interpolation to obtain an original LET data set whose completeness meets the data fitting requirements.

[0010] Preferably, the data fitting unit is used to perform data fitting on each LET data in the original LET data set according to a fourth-order B-spline basis function.

[0011] Preferably, the data fitting unit is used to perform the data fitting through the following steps:

[0012] Define the nodes of the basis function according to the n observation time points (Timeline) in the LET data set, and establish the corresponding B-spline basis function group φ(t)=(φ1(t),...,φ K (t))′;

[0013] Using the regularization method for functional data, treat the fitted functional data x(t)=c′φ(t),

[0014] Total curvature is the penalty functional; where D 2 (x) represents the second-order derivative function of x(t);

[0015] By minimizing the penalized residual sum of squares Solve for the optimal coefficient vector c;

[0016] The generalized cross validation method is used to obtain the best value of the smoothing parameter, and its GCV index is defined as:

[0017]

[0018] Here, SSE stands for the residual sum of squares.

[0019] Preferably, the principal component analysis unit is used to obtain the pre-diagnosis index through the following steps:

[0020] Seek several mutually orthogonal weight functions ξ m (t), the square of its quadratic norm ||ξ m || 2 = 1, and the inner product of the function The sum of squares reaches the maximum;

[0021] Extracting the kth principal component is equivalent to solving the optimization problem

[0022]

[0023] Solving the above problem, we get m (t) in,

[0024] ξ m (t) is the mth characteristic function;

[0025] is the mth eigenvalue;

[0026] f m，i For data x i (t) is the mth principal component score.

[0027] Preferably, the principal component analysis unit is further used to obtain a pre-diagnosis indicator by describing the derivative function of each LET function.

[0028] Preferably, the indicator verification unit is used to verify whether each variable of the pre-diagnosis indicator obeys a normal distribution according to a normal QQ plot:

[0029] A logistic regression model is established by taking each non-empty subset of the set of variables of the pre-diagnosis index as an independent variable.

[0030]

[0031] And calculate the AIC value of the logistic regression model;

[0032] The variables with insignificant index X2 in the whole model are eliminated from the independent variables to obtain the optimized independent variables.

[0033] Preferably, the indicator verification unit is further used to draw an empirical ROC curve of a logistic regression model of the set of optimized independent variables and existing indicators.

[0034] Compared with the prior art, the present invention can pay more comprehensive attention to the dynamic changes of CMAP amplitude in the entire exercise test, extract more useful indicators for combined application, further improve sensitivity while ensuring specificity, and diagnose periodic paralysis earlier and more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the system module of the present invention.

[0036] Figure 2 In the present invention, log 10 Schematic diagram of the settlement results of calculating GCV within the range of (λ)∈[-3, 3].

[0037] Figure 3 This is a graph of the LET amplitude curve data extracted and fitted.

[0038] Figure 4 A graph showing the first principal component extracted from the fitted LET amplitude curve data.

[0039] Figure 5 A graph showing the extraction of the second principal component from the fitted LET amplitude curve data.

[0040] Figure 6 A graph showing the third principal component extracted from the fitted LET amplitude curve data.

[0041] Figure 7 Box plot of the first principal component scores from the extracted fitted LET amplitude curve data.

[0042] Figure 8 Box plot of the scores for the second principal component in the extracted fitted LET amplitude curve data.

[0043] Fig. 9 Box plot of the third principal component scores from the extracted fitted LET amplitude curve data.

[0044] Fig.10 This is the derivative function graph of the LET amplitude curve data set.

[0045] Fig.11 Plot of the first principal component of the derivative of the LET amplitude curve.

[0046] Fig.12 Plot of the second principal component of the derivative of the LET amplitude curve.

[0047] Fig.13 Plot of the third principal component of the derivative of the LET amplitude curve.

[0048] Fig.14 Box plot of the first principal component score of the derivative of the LET amplitude curve.

[0049] Fig.15 Box plot of the second principal component scores for the derivative of the LET amplitude curve.

[0050] Fig.16 Box plot of the third principal component score for the derivative of the LET amplitude curve.

[0051] Fig.17 is the normal QQ plot of variable X1.

[0052] Fig.18 This is the normal QQ plot of variable X2.

[0053] Fig.19 This is the normal QQ plot of variable X3.

[0054] Fig. 20 It is the empirical ROC curve of the logistic regression model for the existing indicators.

[0055] Fig.21 is the empirical ROC curve of the logistic regression model for {X1, X3}.

[0056] In the figure, the corresponding reference numerals of the components are as follows:

[0057] 100. Data collection unit; 200. Data fitting unit; 300. Principal component analysis unit; 400. Index inspection unit. DETAILED DESCRIPTION

[0058] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0059] Example 1, please refer to Figure 1-21 :

[0060] A long-term motion test system comprises: a data acquisition unit, a data fitting unit, a principal component analysis unit and an index inspection unit.

[0061] The data acquisition unit includes an electrode patch for recording the amplitude and area of ​​the muscle compound action potential to obtain an original LET data set; the data fitting unit is signal-connected to the data acquisition unit, and the data fitting unit is used to perform data fitting on each LET data in the original LET data set according to a fourth-order B-spline basis function to obtain a functional data set composed of LET functions; the principal component analysis unit is used to perform principal component analysis on each LET function in the functional data set to obtain pre-diagnostic indicators based on the principal component analysis; the indicator verification unit is used to construct a diagnostic model for periodic paralysis based on the pre-diagnostic indicators, and use logistic regression analysis to extract diagnostic indicators for the diagnostic model.

[0062] The working process is as follows:

[0063] The first step is to obtain the original LET dataset:

[0064] The patient lies down, the abductor digiti minimi muscle is selected as the testing muscle, the skin surface is cleaned, and an electrode patch is attached to the testing muscle.

[0065] Baseline muscle compound action potential (CMAP) amplitude and area were recorded.

[0066] Ask the patient to abduct the digiti minimi muscle with maximum force for 45 seconds, then rest for 15 seconds. Repeat 5 times. Record the CMAP amplitude and area every one minute during the exercise.

[0067] The flexor digiti minimi muscles were then allowed to rest for 60 minutes. The CMAP amplitude and area were recorded every 1 minute for the first 5 minutes of rest, every 2 minutes for 5-15 minutes, and every 3 minutes for 15-60 minutes.

[0068] The original LET data sets of 31 patients were obtained as shown in Table 1

[0069]

[0070]

[0071] Table 1

[0072] Step 2: Functional Data Analysis

[0073] In view of the possible missing data at some observation time points in the original LET data set, preferably, the missing values ​​are first processed by linear interpolation to obtain a complete LET data set that can be considered to be more in line with the actual situation.

[0074] Select the fourth-order B-spline basis function to fit the LET data, and define the nodes of the basis function according to the n=31 observation time points (Timeline) in the LET data set.

[0075] Establish the corresponding B-spline basis function group φ(t)=(φ1(t),...,φ K The number of basis functions is determined by the sum of their order and the number of internal nodes, so the basis function group here contains K = 4 + 31-2 = 33 basis functions.

[0076] The present invention adopts a regularization method for function-type data as a fitting method. The regularization method is to impose a certain penalty on a certain measure of the complexity of the function-type data to ensure the smoothness of the function and avoid overfitting.

[0077] For the function data to be fitted x(t) = c′φ(t),

[0078] Total curvature As the penalty functional,

[0079] Where D 2 (x) represents the second-order derivative function of x(t), and minimizes the residual sum of squares with penalty Solve for the optimal coefficient vector c. The hyperparameter λ ≥ 0 in the above formula is called the smoothing parameter, which measures the weight of the imposed penalty relative to the goodness of fit (residual sum of squares). In practice, its optimal value needs to be selected.

[0080] Using generalized cross validation (GCV) as the selection method, it can be shown that the fitted values ​​obtained by the regularization method There is a linear relationship between the response variable y and

[0081]

[0082] Where n×K matrix Φ=(φ k (t j )) contains the value of each basis function at each node, a K-order symmetric matrix is the penalty matrix. φ，λ It is called the smoothing matrix, where df(λ) = tr(S φ，λ ) is called the fitting degrees of freedom.

[0083] The GCV indicator of the regularization method is defined as

[0084]

[0085] where SSE represents the residual sum of squares (without penalty). In general, near the minimum value of GCV, its value increases with the logarithmic smoothing parameter log 10 (λ) changes very little, so by visually comparing log 10 The GCV corresponding to different values ​​of (λ) is selected to minimize it, which is an acceptable λ.

[0086] In the log 10 GCV is calculated within the range of (λ)∈[-3, 3], and the calculation results are as follows Figure 2 As shown, the optimal λ=10 0.6 .

[0087] The LET amplitude data is fitted according to the optimal smoothing parameter to obtain a functional data set. The fitting results are as follows: Figure 3 As shown in the figure, each curve represents the test data of one subject.

[0088] The third step is principal component analysis of functional data.

[0089] By extracting the part of the functional data that can reflect the differences between the data to the greatest extent, it plays a role in reducing the dimension and presenting the data characteristics.

[0090] Seek several mutually orthogonal weight functions ξ m (t), the square of its quadratic norm ||ξ m || 2 = 1, and the inner product of the function The sum of squares reaches a maximum.

[0091] Extracting the kth principal component is equivalent to solving the optimization problem

[0092]

[0093] Solving the above problem, we get m (t) are called the mth eigenfunction and eigenvalue respectively, and the contribution rate of the eigenvalue is similar to the definition in multivariate principal component analysis. m，i Called data x i The mth principal component score of (t) is a quantitative reflection of the functional data characteristics.

[0094] Extract the curve graph of the first three principal components of the fitted LET amplitude curve data, and the result is as follows Figure 4-6 shown.

[0095] The total contribution rate of the first three principal components is 95.7%, which can reflect most of the information in the data set. In order to explore the distribution of each principal component score, a box plot of the three principal component scores was drawn according to the normal and hemp groups, as shown in the figure below: Figure 5 In each frame of the principal component analysis result graph, the solid line represents the mean function of all data, and all principal component scores are zero; the two curves drawn by "+" and "-" represent the changing trend of the mean function when the principal component score increases or decreases slightly from zero.

[0096] The joint analysis of the principal component analysis results and the box plot of the principal component scores helps to explain the data characteristics reflected by each principal component. The first principal component describes the rising trend of the LET amplitude curve during the exercise period and the overall decline during the rest period. The lower the principal component score, the more the curve shows the characteristics similar to the Zhouma patient group "rapid rise during exercise and sharp decline during rest", and vice versa, it tends to be stable near the baseline, which is consistent with the rule that the first principal component score of the Zhouma group is significantly lower than that of the normal group. The second principal component mainly reflects the overall level of the curve before resting for 30 minutes, and the principal component score is positively correlated with it. The second principal component score of the normal group is slightly lower than that of the Zhouma patient group, indicating that the LET amplitude curve of normal people can drop to a relatively stable level earlier than that of Zhouma patients. The third principal component describes the extreme characteristics of some curves, including the dramatic fluctuations or stable state during exercise and before to mid-rest. The variance of the third principal component score of the Zhouma patient group is greater than that of the normal group, which reflects the abnormal trend of the LET amplitude curve of Zhouma patients from a certain perspective.

[0097] Through the above scheme, on the one hand, the analysis of the distribution of principal component scores helps to reveal both the actual meaning of the principal components and the intuitive characteristics of the data; on the other hand, unsupervised cluster analysis of the principal component scores can test the performance of the principal components when they participate in diagnosis independently, thereby preliminarily evaluating the feasibility of using the indicators derived from the principal component analysis for actual diagnosis.

[0098] The fourth step is derivative function analysis:

[0099] The LET amplitude curve is a functional data describing the change of LET amplitude over time. Its derivative function reflects the speed of amplitude change. The derivative function of the LET amplitude curve data set is as follows: Fig.10 shown.

[0100] The principal component analysis and K-means clustering of the derivative function of the LET amplitude curve were performed in a similar way to the study of the LET amplitude curve. The curve graphs of the three principal components of the derivative function of the LET amplitude curve and the box plot of the principal component scores are shown in the figure below: Figure 11-16 shown.

[0101] From the principal component of the derivative function, it can be seen that the main characteristics of the LET amplitude curve of patients with peripheral anesthesia in terms of the speed of change include two points: one is the peak from the beginning of exercise to the beginning of rest (the derivative drops sharply from a high positive value to a negative value), and the other is the continuous rapid decline during the rest period of 7-33 minutes (the derivative continues to be at a low level). Compared with the principal component score of the curve itself, the specificity of predicting the group corresponding to the LET amplitude curve based on the principal component score of the derivative function is lower, but the sensitivity is the same. Therefore, possible diagnostic indicators can be proposed based on the characteristics of the curve change speed.

[0102] Step 5: Propose diagnostic indicators:

[0103] Based on the conclusions drawn from the principal component analysis of the LET amplitude curve and its derivative function, the following three possible indicators for diagnosing periodic paralysis are proposed:

[0104] X1: average level of LET amplitude during resting period of 40-60 min;

[0105] X2: the decrease in LET amplitude during the rest period of 7-30 min (average rate of decrease);

[0106] X3: The difference between the change in LET amplitude at the beginning of rest (5 minutes of exercise to 1 minute of rest) and the change in LET amplitude during the middle of exercise (2-3 minutes of exercise).

[0107] In addition, indicators in the literature:

[0108] X01: The percentage of decrease in the minimum CMAP amplitude after exercise and the maximum amplitude during exercise

[0109] X02: The percentage of decrease in the minimum CMAP amplitude after exercise and the maximum CMAP amplitude after exercise

[0110] X03: The percentage decrease of the minimum CMAP amplitude after exercise compared with the baseline amplitude.

[0111] Step 6: Conduct statistical tests on diagnostic indicators:

[0112] X1, X2, X3: All are continuous variables. The normal QQ plot is used to test whether each variable obeys the normal distribution. Figure 16-18 Shown are normal QQ plots of the three continuous variables.

[0113] The points in the QQ graph are roughly distributed near a straight line, so it can be considered that X1, X2, and X3 obey a normal distribution. A two-sided t test was performed on each indicator, and the test results are shown in Table 2 below.

[0114] index normal patient t p-value <![CDATA[X1]]> 83.53±17.32 51.74±11.95 7.2727 <0.0001 <![CDATA[X2]]> -15.76±16.60 -44.34±23.14 4.4468 0.0001 <![CDATA[X3]]> 0.56±9.51 -7.32±7.34 3.1171 0.0034

[0115] Table 2

[0116] The differences in indicators X1, X2, and X3 between the normal group and the patients with periodic paralysis were significant, indicating that the three indicators were significantly associated with whether the patient suffered from periodic paralysis and could be used to construct a diagnostic model for periodic paralysis.

[0117] Step 6: Logistic regression of diagnostic indicators

[0118] To evaluate the performance of different indicators and their combinations in diagnosing periodic paralysis, a logistic regression model was established with group as the response variable (0 for normal and 1 for diseased) and each non-empty subset of {X1, X2, X3} as the independent variable.

[0119]

[0120] And calculate its AIC value. According to the AIC criterion, the model with X1 and X3 as independent variables is the optimal model. The regression analysis results are shown in Table 3 below.

[0121] Subset of independent variables AIC <![CDATA[X1]]> 32.203 <![CDATA[X2]]> 46.227 <![CDATA[X3]]> 54.720 <![CDATA[X1,X2]]> 33.692 <![CDATA[X1,X3]]> 27.435 <![CDATA[X2,X3]]> 41.107 <![CDATA[X1,X2,X3]]> 28.028

[0122] Table 3

[0123] Table 4 below shows the logistic regression model with X1 and X3 as independent variables.

[0124] Independent Variable β SE(β) z p-value <![CDATA[X1]]> -0.1742 0.0600 -2.905 0.0037 <![CDATA[X3]]> -0.1907 0.0904 -2.109 0.0349

[0125] Table 4

[0126] Table 5 below shows the full model of logistic regression.

[0127] Independent Variable β SE(β) z p-value <![CDATA[X1]]> -0.2495 0.1028 -2.428 0.0152 <![CDATA[X2]]> 0.0496 0.0444 1.117 0.2641 <![CDATA[X3]]> -0.2186 0.1021 -2.142 0.0322

[0128] Table 5

[0129] Based on the above analysis, it can be seen that the indicator X2 in the full model is not significant (p value>0.05). After removing it from the independent variables, all independent variables are significant and the quality of the model is improved.

[0130] Step 7: ROC curve analysis of logistic regression model

[0131] Draw the empirical ROC curves of the logistic regression model for {X1, X3} and existing indicators respectively, such as Figure 20-21 As shown. Figure 20-21 It can be seen that the accuracy of diagnosing periodic paralysis using indicators X1 and X3 at the same time is higher than the accuracy of diagnosing using the current indicators.

[0132] The combined diagnosis of X1 and X3 had a sensitivity of 93.1%, a specificity of 94.1% and an AUC of 0.972.

[0133] Conclusion: Among the possible diagnostic indicators X1, X2, and X3 of periodic paralysis, X1 and X3 can be used as independent predictors of periodic paralysis, and the combined diagnosis of the two is more accurate than that of the current indicators. X1 is the average level of LET amplitude after a long period of rest, and its significance is similar to that of the current indicators, but the latter only takes data from one observation time point, and its variance is larger than that of the former, so its performance is slightly lower than that of the former. On the other hand, X3 considers the change in LET amplitude between two adjacent time points, and approximately depicts the difference in the speed of change of LET amplitude at two more important moments, namely, the beginning of rest and the middle of exercise, which is a quantitative reflection of the fluctuation of LET amplitude during exercise. Although X3 only focuses on the LET amplitude before and after exercise and is not suitable for direct diagnosis, since other indicators do not fully consider the information of this period, when X3 and X1 are involved in diagnosis together, better diagnostic results can be obtained compared with using X1 or other existing indicators alone.

[0134] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they still fall within the protection scope of the present invention.

Claims

1. A long-term exercise test system, characterized in that: include: A data acquisition unit, wherein the data acquisition unit is used to record the amplitude and area of ​​the muscle compound action potential and obtain an original LET data set; A data fitting unit, the data fitting unit is used to perform data fitting on each LET data in the original LET data set according to a fourth-order B-spline basis function to obtain a function-type data set composed of LET functions; A principal component analysis unit, the principal component analysis unit is used to perform principal component analysis on each LET function in the functional data set, and obtain a pre-diagnosis index according to the principal component analysis; An index verification unit is used to construct a diagnostic model for periodic paralysis according to the pre-diagnosis index, and to perform logistic regression analysis on the diagnostic model to extract diagnostic indexes.

2. The test system according to claim 1, characterized in that: The data acquisition unit is also used to process data missing in the original LET data through linear interpolation method to obtain the original LET data set whose completeness meets the data fitting requirements.

3. The test system according to claim 2, characterized in that: The data fitting unit is used to perform data fitting on each LET data in the original LET data set according to a fourth-order B-spline basis function.

4. The test system according to claim 3, characterized in that: The data fitting unit is used to perform the data fitting through the following steps: Define the nodes of the basis function according to the n observation time points (Timeline) in the LET data set, and establish the corresponding B-spline basis function group φ(t)=(φ1(t),...,φ K (t))′; Using the regularization method for functional data, treat the fitted functional data x(t)=c′φ(t), Total curvature is the penalty functional; where D 2 (x) represents the second-order derivative function of x(t); By minimizing the penalized residual sum of squares Solve for the optimal coefficient vector c; The generalized cross validation method is used to obtain the best value of the smoothing parameter, and its GCV index is defined as: Here, SSE stands for the residual sum of squares.

5. The test system according to claim 4, characterized in that: The principal component analysis unit is used to obtain the pre-diagnosis index through the following steps: Seek several mutually orthogonal weight functions ξ m (t), the square of its quadratic norm ||ξ m || 2 = 1, and the inner product of the function The sum of squares reaches the maximum; Extracting the kth principal component is equivalent to solving the optimization problem Solving the above problem, we get m (t) in, ξ m (t) is the mth characteristic function; is the mth eigenvalue; f m，i For data x i (t) is the mth principal component score.

6. The test system according to claim 5, characterized in that: The principal component analysis unit is also used to obtain a pre-diagnosis indicator by describing the derivative function of each LET function.

7. The test system according to claim 6, characterized in that: The indicator testing unit is used to test whether each variable of the pre-diagnosis indicator obeys normal distribution according to the normal QQ graph: A logistic regression model is established by taking each non-empty subset of the set of variables of the pre-diagnosis index as an independent variable. And calculate the AIC value of the logistic regression model; The variables with insignificant index X2 in the whole model are eliminated from the independent variables to obtain the optimized independent variables.

8. The test system according to claim 7, characterized in that: The indicator verification unit is also used to draw an empirical ROC curve of the logistic regression model of the set of optimized independent variables and existing indicators.

Images