A system for analyzing the movement posture of an infant to assist in diagnosing cerebral palsy

Abstract

The application provides a baby motion posture analysis system for assisting in diagnosing cerebral palsy, and belongs to the technical field of medical measurement. The system comprises a video processing module, an image processing module, a data processing module, an analysis module, a discrimination module and a user graphical interface module, objectively evaluates the motion function in the early brain development of babies and infants, and assists doctors in diagnosing whether the babies and infants have cerebral palsy. The application has the following advantages: 1. The risk of cerebral palsy of a baby to be measured can be determined by using the whole body motion video of the baby, so that early treatment and rehabilitation of high-risk babies can be realized; 2. The analysis module is based on the statistical principle, and a more reliable prediction classification feature vector is obtained, so that the correlation between the motion characteristics of babies and neurodevelopmental abnormalities can be quantitatively studied; and 3. The discrimination module is based on the key point time window sequence decision threshold strategy and the symmetric decision tree algorithm, so that the accuracy and generalization of the prediction classification are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of medical measurement, and particularly relates to an infant motion posture analysis system for assisting in diagnosing cerebral palsy. BACKGROUND

[0002] Cerebral palsy, also known as cerebral paralysis, is a non-progressive brain damage syndrome caused by various reasons during the early brain development of infants from before birth to one month after birth. It mainly manifests as central motor disorders and abnormal posture, and can also be accompanied by mental retardation, epilepsy, sensory disorders, language disorders and mental behavior abnormalities, etc., and is one of the main diseases causing motor disability of children.

[0003] At present, early diagnosis of infant cerebral palsy mainly relies on GMFM gross motor function assessment table, head imaging (such as abnormality of nuclear magnetic resonance, CT or B-ultrasound), surface electromyography signal sEMG, gait motion characteristic parameter analysis, etc., combined with the experience of doctors to obtain the diagnosis result of cerebral palsy. By using the above scheme, cerebral palsy is generally diagnosed after three years of age, which delays the early intervention treatment of cerebral palsy. In addition, the signal-to-noise ratio of electromyography signal detection is low, and it is difficult to exclude interference; the judgment accuracy of GMFM scale is not enough, and it depends on subjective judgment of people; considering the factors such as radiation, head imaging is not suitable for infant detection.

[0004] Whole body motion assessment is an important means for identifying early infantile cerebral palsy and other neurodevelopmental disorders, which can provide a certain convincing diagnosis result for the health development of infants without affecting them. However, only relying on observers to perform the assessment work not only needs a long training mode, but also may cause misdiagnosis due to individual bias, so it has certain important practical significance to assist whole body motion assessment by using video, electromagnetic tracking and other auxiliary equipment. SUMMARY

[0005] Therefore, the purpose of the present application is to provide an infant motion posture analysis system for assisting in diagnosing cerebral palsy. In the early crawling stage of infants, whether the infant to be tested has the risk of cerebral palsy is judged based on the motion posture, so that early treatment and rehabilitation of high-risk infants can be more conveniently realized.

[0006] To achieve the above purpose, the present application provides an infant motion posture analysis system for assisting in diagnosing cerebral palsy, which comprises a video processing module, an image processing module, a data processing module, an analysis module, a discrimination module and a user graphical interface module; the video processing module transmits images to the image processing module, and the image processing module transmits data to the data processing module; the data processing module transmits data to the analysis module, the analysis module transmits data to the discrimination module, and the data processing module, the analysis module and the discrimination module all interact with the user graphical interface module.

[0007] Video processing module: split the baby video into images every 10 frames, with a time interval of Δt;

[0008] Image processing module: feature extraction on the images split by the video processing module, extracting the plane coordinates of each feature joint of the baby every Δt time, and performing Gaussian filtering on the split images;

[0009] The baby feature joints include: nose, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, right hip, right knee, right ankle, left hip, left knee, left ankle, right eye, left eye, right ear, and left ear 18 joints;

[0010] Data processing module: data processing of baby motion feature information from time and frequency domains;

[0011] In the time domain, the distance change between the pixel coordinates of the joints of the baby in the previous frame and the pixel coordinates of the joints in the next frame is calculated, thereby obtaining the velocity sequence and acceleration sequence of each feature joint of the baby;

[0012]

[0013]

[0014] In the frequency domain, the energy spectrum density of the baby's motion is calculated;

[0015] Analysis module:

[0016] I. Dimensionality reduction processing for feature data: for the feature data of the 18 feature joints of the baby, PCA principal component analysis is used for dimensionality reduction processing, and the retention amount of the dimensionally reduced information is evaluated;

[0017] II. Correlation analysis: correlation analysis from the size and curve trend of the motion feature data, comparing the correlation of the standard video and the sample video, using the correlation coefficient to represent the correlation degree of the sample video and the standard video;

[0018] Discrimination module: provided with a baby neural development discrimination model based on CatBoost and a time window threshold decision loss function, used to assist in the diagnosis of baby cerebral palsy.

[0019] Further, the data processing module calculates the energy spectrum density of the baby's motion in the frequency domain as follows:

[0020] 1). Let the energy of a motion signal s(t) be E, and the energy of the motion signal be determined by equation (3):

[0021]

[0022] 2) The Fourier transform of the motion signal s(t), i.e. the spectral density S(f), is known from the Parseval theorem:

[0023]

[0024] where |S(f)|2is the energy spectral density, representing the signal energy in a frequency band of width f at frequency f, or the signal energy per unit frequency band. 2

[0025] Further, the analysis module is used for the following process of dimension reduction:

[0026] S1. Assume that the characteristic joint of the baby involves p joints, and use X1, X2, …, Xp to represent the speed or acceleration of each joint every 12 frames, respectively, each of which represents the speed or acceleration of the corresponding joint every 12 frames. p

[0027]

[0028] S2. Calculate the covariance matrix of the normalized sample;

[0029] First, calculate the covariance of the normalized sample:

[0030]

[0031] The sample correlation coefficient matrix is obtained as:

[0032]

[0033] S3. Calculate the eigenvalues and eigenvectors of R to obtain the eigenvalues and eigenvectors of the baby's limbs;

[0034] Eigenvalues: λ1≥ λ2≥ … ≥ λp≥ 0; p

[0035] Eigenvectors:

[0036] The motion characteristics of the baby's whole body are represented by p principal components Y, each of which is a linear combination of p joints;

[0037]

[0038] S4. Calculate the contribution rate and cumulative contribution rate of the principal component Y of each joint position;

[0039] The contribution rate is a normalized representation of the eigenvalue, and the cumulative contribution rate is a superposition of the contribution rate, and the calculation formula is as follows:

[0040]

[0041] ​​​

[0042] Further, the correlation analysis includes correlation analysis of the size of the infant motion speed and acceleration and the neural development, grey correlation analysis.

[0043] Further, the correlation analysis of the infant motion speed and acceleration and the neural development has the following process:

[0044] The distance metric shown in formula (12) is used to measure the gap between the sample and the normal infant in the motion feature data, d 1i The smaller the distance is, the smaller the motion data value difference between the sample video and the standard video is, indicating that the individual difference is small.

[0045]

[0046] Further, the process of the grey correlation analysis is as follows:

[0047] First, calibrate the mother sequence and the subsequence, and define the correlation coefficient of each index in the subsequence and the mother sequence:

[0048]

[0049] In the formula, a represents the two-pole minimum difference of |x0(k)-x i (k)|; b represents the two-pole maximum difference of |x0(k)-x i (k)|; ρ1 is the resolution coefficient, and ρ1=0.5 is taken.

[0050] Calculate the grey correlation degree:

[0051]

[0052] The greater the grey correlation degree is, the closer the curves are, and the greater the correlation degree between the corresponding sequences is, and vice versa.

[0053] Further, the process of the discriminant module in assisting the diagnosis of cerebral palsy is as follows:

[0054] ①. Extract the motion posture features of the initial layer, and perform prediction classification based on the first ten important joints of the initial layer. The subsequent skeleton feature layer processes 18 joint information, and a 5s time window sequence is constructed accordingly.

[0055] ②. According to the correlation degree of the features obtained from each frame image in the 5s time window sequence and the neural development abnormalities, the threshold is used for judgment and division: when the proportion of abnormal features detected is more than 50%, it is judged as an object with potential neural abnormal disease; when the proportion is less than 50%, it is judged as a low-risk object.

[0056] ③. Further divide the proportion of abnormal features into 75% and 25%, and combine the GMS evaluation, to make a supplementary evaluation and prediction on the monotony GMS mode and the confusion GMS mode of the mild abnormality existing in the observation of the infant in the writhing movement stage;

[0057] Monotony GMS: there are certain features different from normal infants, but the abnormal features account for 25% to 50%, and the abnormal features have no reference value for clinical diagnosis;

[0058] Confusion GMS: the features are not obvious during detection, and obvious abnormalities develop after several weeks, and the proportion of abnormal features is 50% to 75%.

[0059] Advantages of the present application:

[0060] 1. The infant motion posture analysis system for assisting in diagnosing cerebral palsy provided by the present application is provided with a video processing module, an image processing module, a data processing module, an analysis module, a discrimination module and a user graphical interface module, can assist in diagnosing through the whole body motion video of an infant, and can judge whether the infant to be detected has a risk of cerebral palsy, so that early treatment and rehabilitation of high-risk infants can be more conveniently realized.

[0061] 2. The infant motion posture analysis system for assisting in diagnosing cerebral palsy provided by the present application, the analysis module obtains a prediction classification feature vector with certain reliability through principal component analysis, variance analysis and grey correlation calculation based on the statistical principle, and can quantitatively study the correlation between infant motion features and neural development abnormalities; the discrimination module solves the problems of gradient deviation and prediction deviation of typical prediction classification algorithms in cerebral palsy prediction based on a key point time window sequence decision threshold strategy and a symmetric decision tree algorithm, and improves the accuracy and generalization of prediction classification.

[0062] Other advantages, objects and features of the present application will be set forth in part in the following specification, and in part will become apparent to those skilled in the art from the examination of the following text, or can be learned from the practice of the present application. The objects and other advantages of the present application can be realized and obtained by the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 It is a schematic diagram of the relationship between the system parts;

[0064] Figure 2 It is a flowchart of system detection;

[0065] Figure 3 It is the variance percentage of the first n components. DETAILED DESCRIPTION

[0066] In order to make the technical solutions, advantages and objectives of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without any inventive effort fall within the protection scope of the present application.

[0067] As shown in Figure 1 The present application provides an infant motion posture analysis system for assisting diagnosis of cerebral palsy, comprising a video processing module, an image processing module, a data processing module, an analysis module, a discrimination module and a user graphical interface module; the video processing module transmits images to the image processing module, and the image processing module transmits data to the data processing module; the data processing module transmits data to the analysis module, the analysis module transmits data to the discrimination module, and the data processing module, the analysis module and the discrimination module all interact with the user graphical interface module;

[0068] Video processing module: split the infant video into images every 10 frames, with a time interval of Δt=0.48s;

[0069] Image processing module: extract the plane coordinates of each feature joint of the infant every Δt time, and perform Gaussian filtering on the split images;

[0070] The feature joints of the infant include 18 joints: nose, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, right hip, right knee, right ankle, left hip, left knee, left ankle, right eye, left eye, right ear and left ear;

[0071] Data processing module: perform data processing on the infant motion feature information from the time domain and the frequency domain;

[0072] a. In the time domain, calculate the distance change of the pixel coordinates of the joints of the infant in the previous frame and the pixel coordinates of the joints in the next frame, thereby obtaining the velocity sequence and the acceleration sequence of each feature joint of the infant;

[0073]

[0074]

[0075] b. In the frequency domain, calculate the energy spectrum density of the infant motion;

[0076] The flow of calculating the energy spectrum density of the infant motion is as follows:

[0077] 1). Let the energy of a motion signal s(t) be E, and the energy of the motion signal is determined by formula (3):

[0078]

[0079] 2). The Fourier transform of the motion signal s(t), i.e. the spectral density S(f), is known from the Parseval theorem:

[0080]

[0081] where |S(f)|2is the energy spectral density, representing the signal energy in a frequency band of width f at frequency f, or the signal energy per unit frequency band. 2

[0082] Analysis module:

[0083] I. Dimensionality reduction processing for feature data: for the feature data of 18 feature joints of the infant, PCA principal component analysis is used for dimensionality reduction processing, and the retention amount of the dimensionality reduced information is evaluated, and it is included in the classification vector for judging the neural development of the infant;

[0084] The process of the analysis module for dimensionality reduction processing is as follows:

[0085] S1. Assume that the feature joints of the infant involve p joints, denoted as X1, X2, …, Xp, respectively, and each list represents the speed or acceleration of every 12 frames of motion of the corresponding joint; p

[0086]

[0087] S2. Calculate the covariance matrix of the standardized sample;

[0088] First, calculate the covariance of the standardized sample:

[0089]

[0090] The sample correlation coefficient matrix is obtained as:

[0091]

[0092] S3. Calculate the eigenvalues and eigenvectors of R to obtain the eigenvalues and eigenvectors of the infant's limbs;

[0093] Eigenvalues: λ1≥ λ2≥ … ≥ λp≥ 0; p

[0094] Eigenvectors:

[0095] The motion features of the infant's whole body are represented by p principal components Y, and each principal component is a linear combination of p joints;

[0096] ​​​

[0097] S4. Calculate the contribution rate and cumulative contribution rate of each joint position principal component Y;

[0098] The eigenvalue is an index indicating whether the principal component is comprehensive or not. The larger the eigenvalue, the better the principal component can summarize the original index data. The contribution rate is the normalized embodiment of the eigenvalue, and the cumulative contribution rate is the superposition of the contribution rate. The calculation formula is as follows:

[0099]

[0100]

[0101] II. Correlation analysis: correlation analysis from the size and curve trend of the motion characteristic data, and comparison of the correlation between the standard video and the sample video, the correlation coefficient represents the correlation degree between the sample video and the standard video;

[0102] The correlation analysis includes the correlation analysis of the size of infant motion speed and acceleration and neural development, and the grey correlation analysis.

[0103] The process of the correlation analysis of infant motion speed and acceleration and neural development is as follows:

[0104] The distance measure shown in formula (12) is used to measure the gap between the sample and the normal infant in the motion characteristic data, d 1i The smaller the distance, the smaller the difference between the motion data values of the sample video and the standard video, indicating that the difference between individuals is small.

[0105]

[0106] The process of the grey correlation analysis is as follows:

[0107]

[0108] In the formula, a represents the two-pole minimum difference of |x0(k)-x i (k)|; b represents the two-pole maximum difference of |x0(k)-x i (k)|; ρ1 is the resolution coefficient, and ρ1=0.5.

[0109] Calculate the grey correlation degree:

[0110]

[0111] The larger the grey correlation degree, the closer the curves, and the greater the correlation degree between the corresponding sequences, and vice versa.

[0112] The discrimination module is provided with a baby neural development discrimination model established based on CatBoost and a time window threshold decision loss function, and is used for assisting in diagnosis of cerebral palsy of the baby.

[0113] The process of the discrimination module assisting in diagnosis of cerebral palsy is as follows:

[0114] ①. The motion posture features of the initial layer are extracted, and the first ten important joints in the initial layer are used for prediction and classification. The subsequent skeleton feature layer processes 18 joint information, and a 5s time window sequence is constructed accordingly.

[0115] ②. According to the correlation degree of the features obtained from each frame of image in the 5s time window sequence and the neural development abnormality, the threshold is used for judgment and division: when the proportion of abnormal features detected is more than 50%, it is judged as an object with potential neural abnormal disease; when the proportion is less than 50%, it is judged as a low-risk object.

[0116] ③. Further, the proportion of abnormal features is divided into two standards of 75% and 25%, and the monotony GMS mode and the confusion GMS model of the mild abnormality appearing in the observed baby in the twisting motion stage are supplemented and evaluated.

[0117] Monotony GMS: There are certain characteristics different from normal babies, but still have no reference value for clinical diagnosis, and the proportion of abnormal features is 25%-50%;

[0118] Confusion GMS: The feature is not obvious during detection, and obvious abnormality develops after several weeks, and the proportion of abnormal features is 50%-75%.

[0119] Example 1

[0120] In this embodiment, 148 pieces of baby whole body motion videos are obtained from a hospital, each of which has been detected by a professional doctor through GMA, and each has a length of about 3 minutes. Among them, 61 pieces of baby motion videos of normal neural development and 87 pieces of baby motion videos of abnormal neural development. One piece of baby motion video data of normal neural development is selected from the 148 pieces of videos as standard data, and the remaining 147 pieces are used as sample data. As shown in Figure 2 The above baby whole body motion video is imported into the baby motion posture analysis system for assisting in diagnosis of cerebral palsy of the present application, and detection and analysis are performed, and the specific process is as follows:

[0121] 1). Collecting and picture preprocessing of baby spontaneous motion video.

[0122] Video recording standard: The room temperature is maintained at 26-28℃, the light is soft and the environment is quiet. The small infant is supine on a single-color bed, wearing a tight cotton vest or exposed, without a pillow, holding the shooting device at a distance of 1.2-1.5m high and 45° angle from the side of the infant's feet to shoot the awake state; the shooting position is fixed to avoid camera shaking and ghosting. The infant's whole body state needs to be recorded, avoiding crying, belching, sucking the pacifier and other disturbances.

[0123] Picture preprocessing: import the video into the system, and split the infant video every 10 frames into images by the video processing module, with a time interval of Δt=0.48s.

[0124] 2). Feature extraction is performed on the images split by the video processing module by the image processing module, and the plane coordinates of each feature joint of the infant are extracted every Δt time, and the split images are subjected to Gaussian filtering.

[0125] 3). Data processing module is used to process the infant motion feature information from time domain and frequency domain.

[0126] In the time domain, the distance change (represented by the Euclidean distance) of the pixel coordinates of the joint nodes of the infant in the last frame and the pixel coordinates of the joint nodes in the next frame is calculated, and the velocity sequence and acceleration sequence of each feature joint of the infant are obtained. In the frequency domain, set the energy of a motion signal s(t) as E, and the Fourier transform of the motion signal s(t) is obtained to get the energy spectrum density of the infant motion.

[0127] 4). Principal component analysis

[0128] The 18 extracted motion features are processed by the analysis module for dimension reduction, and the PCA algorithm function is used for calculation. By obtaining different n_components parameters, the information retention after dimension reduction is compared and evaluated. Figure 3 The percentage of variance accounted for by the first b components can be seen from Figure 3 It can be seen that the contribution of the first 10 principal components has reached more than 90%, and the principal components after the 10th point have less impact on the whole. Therefore, the first 10 are extracted as new sample inputs and are classified.

[0129] 5). Coefficient of variation analysis

[0130] The coefficient of variation can reflect the dispersion degree of the infant motion data, and the greater the dispersion degree, the greater the volatility, which is a key factor affecting the infant's neural development and should also be included in the classification vector for judging the infant's neural development. The definition of the coefficient of variation is as follows: where σ i is the standard deviation of the motion feature data of the i-th group of principal components of the infant within a certain time, μ iThe average value of the motion characteristic data of the i-th group of principal component sequence of the infant in a certain time.

[0131] 6. The correlation analysis is performed by the analysis module from the size and curve trend of the motion characteristic data.

[0132] By comparing the correlation of the standard video and the sample video, the correlation coefficient obtained can represent the correlation degree of the sample video and the standard video, and it is included in the classification vector for judging the neural development of the infant. Distance measurement is used to measure the gap between the sample and the normal infant in the motion characteristic data, and the smaller the distance, the smaller the difference between the motion data values of the sample video and the standard video, indicating that the difference between individuals is small. The grey correlation analysis is used to infer the correlation degree from the similarity of the geometric curve.

[0133] The specific process is as follows:

[0134] Firstly, the sequence to be analyzed is determined, and since the quantitative values corresponding to different experimental result indicators are different in this case, we analyze the relationship between two elements. Secondly, the correlation coefficient of each indicator in the sub-sequence and the mother sequence is defined. The mean value of each sequence is calculated by preprocessing the mother sequence and the sub-sequence, and then the indicators in the sequence are divided by the mean value to de-dimension and simplify the calculation by reducing the variable range. Finally, the grey correlation degree is calculated. The greater the grey correlation degree, the closer the curve, and the greater the correlation degree between the corresponding sequences, and vice versa.

[0135] 7. The infant neural development discrimination model based on CatBoost and the time window threshold decision loss function are used to assist in the diagnosis of infant cerebral palsy.

[0136] The integrated learning algorithm stacks multiple decision tree models (weak learners) layer by layer to improve the total model to a strong learner, and runs in a serial manner. The prediction results of multiple weak learners are added to obtain the final prediction result of the model. At the same time, the time window threshold decision loss function is added, and the final result is shown in Table 1.

[0137] Table 1: Prediction classification index

[0138]

[0139] From Table 1, it can be seen that the whole body motion evaluation system based on OpenPose and decision tree proposed in the present application achieves an accuracy of 93.02% and a recall rate of 93.02%, and an average detection accuracy value of 91.95%. In summary, the infant motion posture analysis system for assisting in the diagnosis of cerebral palsy proposed in the present application is feasible and effective, and can assist in the diagnosis of related diseases through whole body motion.

[0140] Finally, it is to be explained that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions, and all should be covered in the protection scope of the present application.

Claims

1. An infant motion posture analysis system for assisting diagnosis of cerebral palsy, characterized by: It comprises a video processing module, an image processing module, a data processing module, an analysis module, a discrimination module and a user graphical interface module; the video processing module transmits images to the image processing module, the image processing module transmits data to the data processing module; the data processing module transmits data to the analysis module, the analysis module transmits data to the discrimination module, and the data processing module, the analysis module and the discrimination module all interact with the user graphical interface module; The video processing module: split the baby video into images every 10 frames, with a time interval of Δt; The image processing module: extracts the feature points of the baby every Δt time, and performs Gaussian filtering on the split images; The baby feature points include: nose, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, right hip, right knee, right ankle, left hip, left knee, left ankle, right eye, left eye, right ear and left ear 18 joints; The data processing module: processes the baby's motion feature information from the time domain and the frequency domain; In the time domain, the distance change between the pixel coordinates of the joint nodes of the baby in the previous frame and the pixel coordinates of the joint nodes in the next frame is calculated, so as to obtain the speed sequence and acceleration sequence of each feature joint node of the baby; Where Δs is the distance between the pixel coordinates of the adjacent two frames of the baby joint nodes; Δt is the time interval between adjacent frames, which is used to describe the speed of the joint node position change; Δv is the change amount of the speed of the adjacent two frames; In the frequency domain, the energy spectrum density of the baby's motion is calculated; The data processing module calculates the energy spectrum density of the baby's motion in the frequency domain as follows: 1). Let the energy of a motion signal s(t) be E, and the energy of the motion signal be determined by formula (3): 2). The Fourier transform of the motion signal s(t) is S(f), and according to the Bessel theorem, we have: where |S(f)| 2 is the energy spectral density, representing the signal energy in a frequency band of width f at frequency f, or the signal energy per unit frequency band; Analysis module: I. Dimensionality reduction processing for feature data: for the feature data of the 18 feature joints of the baby, PCA principal component analysis is used for dimensionality reduction processing, and the retention amount of the dimensionality reduced information is evaluated; II. Correlation analysis: correlation analysis is performed from the size and curve trend of the motion feature data, and the correlation between the standard video and the sample video is compared, and the correlation coefficient is used to represent the correlation degree between the sample video and the standard video; The analysis module for dimensionality reduction processing has the following process: S1. Assume that the infant's characteristic joints involve a total of p joints, denoted by X1, X2, ..., X... p Each column represents the velocity or acceleration of the corresponding joint every 12 frames. wherein X ' is all motion data recorded at different times for all feature joints; n represents a certain feature joint; and m represents a certain time node. S2. Calculate the covariance matrix of the standardized sample; First, calculate the covariance of the standardized sample: where r ij is the covariance matrix of the normalized samples; X ki is the motion data recorded at a certain time for a certain joint node; The sample correlation coefficient matrix is obtained as follows: S3. Calculate the eigenvalues and eigenvectors of R to obtain the eigenvalues and eigenvectors of the baby's limbs; Eigenvalues: λ1≥ λ2≥... ≥ λ p ≥ 0; Feature vector: The motion feature of the baby's whole body is represented by p principal components Y, and each principal component is a linear combination of p joints; wherein, u pp represents the coefficient of the pth joint in the pth principal component; u 1p represents the coefficient of the pth joint in the 1st principal component; u p1 represents the coefficient of the 1st joint in the pth principal component; S4. Calculate the contribution rate and cumulative contribution rate of each joint part principal component Y; The contribution rate is the normalization of the eigenvalue, and the cumulative contribution rate is the superposition of the contribution rate, and the calculation formula is as follows: where λ i represents the eigenvalue of the i-th limb of the infant; is the sum of all limb eigenvalues; The discrimination module: there is an infant neural development discrimination model based on CatBoost and a time window threshold decision loss function, which is used to assist the diagnosis of infantile cerebral palsy.

2. The infant motion posture analysis system for assisting diagnosis of cerebral palsy according to claim 1, wherein: The correlation analysis includes correlation analysis of infant motion speed and acceleration and neurodevelopment, and grey correlation analysis.

3. The infant motion posture analysis system for assisting diagnosis of cerebral palsy according to claim 2, wherein The process of the correlation analysis of infant motion speed and acceleration and neurodevelopment is as follows: The distance metric shown in equation (12) is used to measure the gap between the sample and normal infants in the motion feature data, d 1i The smaller the distance, the smaller the difference between the motion data values of the sample video and the standard video, indicating that the difference between individuals is small. wherein x 1j is the motion feature data of a certain joint of a normal infant; x ij is the motion feature data of a certain joint of the ith sample video.

4. The infant motion posture analysis system for assisting diagnosis of cerebral palsy according to claim 3, wherein The process of the grey correlation analysis is as follows: First, calibrate the mother sequence and the subsequence, and define the correlation coefficient of each index in the subsequence and the mother sequence: wherein a represents the two-pole minimum difference of |x0(k)-x i (k) |, b represents the two-pole maximum difference of |x0(k)-x i (k) |; ρ1 is a resolution coefficient, and ρ1 = 0.5; k represents a time node sequence. Calculate the grey correlation degree: The greater the grey correlation degree is, the closer the curves are, and the greater the correlation degree between the corresponding sequences is, and vice versa.

5. The infant motion posture analysis system for assisting diagnosis of cerebral palsy according to claim 4, characterized in that: The process of the discriminant module in assisting diagnosis of cerebral palsy is as follows: ①. Extract the motion posture features of the initial layer, and perform prediction classification based on the first ten important nodes of the initial layer, and then process the 18 node information of the subsequent skeleton feature layer, and construct a 5s time window sequence accordingly; ②. According to the correlation degree of the features obtained from each frame of image in the 5s time window sequence and the neurodevelopmental abnormalities, the threshold is used for judgment and division: when the proportion of abnormal features detected is more than 50%, it is judged as an object with potential neuro-abnormal disease; when the proportion is less than 50%, it is judged as a low-risk object; ③. Further divide the proportion of abnormal features into two standards of 75% and 25%, and combine the GMS evaluation, and perform supplementary evaluation and prediction on the monotony GMS mode and the confusion GMS mode of the mild abnormalities of the observed infants in the twisting motion stage; Monotony GMS: There are certain characteristics different from normal infants, but the abnormal characteristics have no reference value for clinical diagnosis, and the proportion of abnormal characteristics is 25%-50%; Confusion GMS: The characteristics are not obvious during detection, and obvious abnormalities develop after several weeks, and the proportion of abnormal characteristics is 50%-75%.

Images