Metabonomics-radiomics prediction method for recurrence risk of chronic subdural hematoma

Abstract

The invention relates to the technical field of health risk prediction, in particular to a metabonomics-radiomics prediction method for chronic subdural hematoma recurrence risk, which comprises the following steps: acquiring a CT image and extracting edge gray fluctuation, constructing fluctuation parameters in combination with metabolome data, screening coordination characteristics to generate a risk combination, and predicting a chronic subdural hematoma recurrence risk. Feature pairs consistent in trend are extracted to form a collaborative channel, and a feature matrix is constructed to generate an input vector set; according to the method, disturbance features are extracted through a CT image edge gray level path, a cross-modal fluctuation trend comparison mechanism is established in combination with patient brain metabolism indexes, biological consistency between the features is enhanced, feature combinations with uncoordinated changes are eliminated, feature pairs with collaborative structure and function trends are screened, and a linkage path is constructed. The evolution relation from structural disturbance to metabolic response is reflected, channel data sorting and recombination improve the difference of input characteristics, the stability and accuracy of recurrence discrimination are enhanced, and the systematicness and interpretability of risk assessment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health risk prediction, in particular to a metabolomics-imagingomics prediction method for the recurrence risk of chronic subdural hematoma. BACKGROUND

[0002] The technical field of health risk prediction involves systematic analysis of the health status of individuals to determine the likelihood of developing a certain disease or health event in the future. The core of this technology includes the collection, analysis and modeling of multiple sources of data such as clinical information, biomarkers, omics data and medical imaging data. This technology usually uses data fusion analysis methods to model and quantify risk factors, and to achieve disease risk assessment and stratified management of target populations through the construction of risk prediction models. Traditional chronic subdural hematoma recurrence risk assessment refers to a method that determines the probability of recurrence based on clinical symptoms and postoperative imaging review results. This method usually relies on single-dimensional statistical analysis or discrimination based on imaging omics data such as the amount of bleeding, brain tissue displacement, and hematoma density in postoperative CT images combined with clinical information. However, it is difficult to comprehensively explain the biological mechanisms of recurrence, so traditional imaging interpretation standards and medical record statistics are often used for risk stratification and recurrence judgment, and it is difficult to effectively integrate metabolomics data and imaging omics data that reflect pathological state changes to construct a multi-omics joint risk prediction model.

[0003] In the assessment of chronic subdural hematoma recurrence risk, existing technologies often rely on postoperative imaging and some clinical indicators for assessment, and often use static features based on postoperative CT images for interpretation, such as hematoma volume, brain tissue displacement, or density. However, such features are often low in timeliness or result-oriented, making it difficult to reflect the dynamic evolution trend of the lesion edge and ignoring potential pathological activity clues. Clinical information is often not effectively linked to microscopic metabolic states, resulting in a one-sidedness in the assessment system in terms of etiology tracing and mechanism explanation. For example, when postoperative imaging shows that the hematoma has been absorbed and the tissue has not been significantly displaced, it is often determined to be a low recurrence risk, but in fact there may be a hidden recurrence risk due to the lack of recovery of cell activity in the edge region and local metabolic abnormalities. Such phenomena are difficult to identify through traditional statistical methods. In addition, existing methods often use single-omics data as the analysis object, making it difficult to integrate features and make consensus judgments between different dimensions of data, resulting in blind spots in risk assessment and a lack of recognition ability for heterogeneous samples, thereby reducing the generalization ability and accuracy of the risk prediction model. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art, and to provide a metabolomics-imagingomics prediction method for the recurrence risk of chronic subdural hematoma.

[0005] In order to achieve the above object, the present application adopts the following technical scheme: a metabolomics-imagingomics prediction method for recurrence risk of chronic subdural hematoma, comprising the following steps:

[0006] S1: obtaining CT images of chronic subdural hematoma patients, constructing transverse and longitudinal pixel sampling paths in the bleeding edge region, extracting gray value difference and identifying gray fluctuation region, obtaining edge disturbance multi-parameter description value based on fluctuation intensity and region span information;

[0007] S2: combining the patient number in the edge disturbance multi-parameter description value, extracting brain metabolism correlation indicators in the metabolomics data, constructing fluctuation feature parameter set according to the distribution range, change amplitude and dispersion degree of the indicators in the sample group, and arranging into a metabolomics fluctuation distribution list;

[0008] S3: combining the metabolomics fluctuation distribution list and edge disturbance multi-parameter description value, identifying the change trend of the features in the same patient, screening out feature combinations with inconsistent change direction or inconsistent change amplitude, and generating a recurrence risk joint feature combination item;

[0009] S4: based on the recurrence risk joint feature combination item, respectively measuring the trend change direction of each group of metabolic features and image features in the patient sequence, screening the feature pairs with consistent trend, integrating into a structure-function linkage path set, and generating a metabolomics imaging collaborative channel list.

[0010] As a further scheme of the present application, the edge disturbance multi-parameter description value includes gray fluctuation intensity, region span, fluctuation frequency, the metabolomics fluctuation distribution list includes brain metabolism index distribution range, change amplitude, dispersion degree, the recurrence risk joint feature combination item includes change direction consistency feature, change amplitude coordination feature, structure and function information integration feature, and the metabolomics imaging collaborative channel list includes structure-function linkage path number, feature pair trend consistency index, and channel response difference.

[0011] As a further scheme of the present application, the specific steps of S1 are:

[0012] S101: after obtaining the CT images of chronic subdural hematoma patients, based on the bleeding edge region in the image, constructing transverse and longitudinal pixel sampling paths, extracting the gray values of the pixel points on the path, reconstructing the gray value sequence in the difference direction, and obtaining the edge gray sampling matrix;

[0013] S102: according to the gray value sequence in the edge gray sampling matrix, identifying the section with prominent gray value fluctuation, extracting the continuous gray fluctuation region, recording the position span and intensity characteristics, integrating the gray fluctuation information in the difference direction, and obtaining the edge gray fluctuation interval set;

[0014] S103: Construct a plurality of directional gray scale disturbance feature indexes based on the region span and intensity characteristics in the edge gray scale fluctuation interval set, integrate the direction information to form a description value set, and obtain an edge disturbance multi-parameter description value.

[0015] As a further scheme of the present application, the specific steps of S2 are:

[0016] S201: Obtain the patient number in the edge disturbance multi-parameter description value, call the number corresponding record in the metabolomics data, extract the brain metabolism correlation index, construct a patient metabolism feature parameter set, and generate a brain metabolism feature parameter set value;

[0017] S202: Based on the brain metabolism feature parameter set value, detect the distribution state of the index in the sample group, analyze the offset degree between the original value and the distribution range, construct an offset degree set corresponding to the brain metabolism index, and obtain a brain metabolism index offset sequence value;

[0018] S203: According to the brain metabolism index offset sequence value, identify the fluctuation amplitude level corresponding to the index, determine the fluctuation interval of the index combined with the change in the group sample, and integrate to establish a metabolism fluctuation distribution list.

[0019] As a further scheme of the present application, the specific steps of S3 are:

[0020] S301: Based on the metabolism fluctuation distribution list, extract the fluctuation value and distribution characteristics of the metabolite, and combine with the disturbance coefficient and direction in the edge disturbance multi-parameter description value, screen the feature set with consistent change direction and coordinated amplitude, and generate a fluctuation-disturbance consistent feature trend value;

[0021] S302: Call the fluctuation-disturbance consistent feature trend value, identify the items with fluctuation abnormalities or direction mutations in the feature combination, judge according to the amplitude difference threshold and direction offset threshold, eliminate combinations that do not meet the conditions, and obtain a double offset coordination feature coefficient value;

[0022] S303: According to the double offset coordination feature coefficient value, the feature combination that meets the coordination standard is summarized, the disturbance consistency and fluctuation amplitude are normalized, the combination items are integrated according to the patient dimension, and a recurrence risk joint feature combination item is generated.

[0023] As a further scheme of the present application, the fluctuation value of the metabolite refers to the change degree of the concentration or expression level of the metabolite under different states (such as disease state and healthy state, or different time points, different tissue samples);

[0024] The distribution characteristics of the metabolite describe the statistical characteristics of the metabolite fluctuation, including skewness, kurtosis, normality, and multimodality.

[0025] The amplitude difference threshold refers to a limit value of whether the difference in numerical value between two variables (such as metabolic fluctuation trend and disturbance trend) is acceptable when screening the feature combination;

[0026] The direction offset threshold refers to the tolerance degree of direction deviation when judging whether two trends are "directionally consistent". The direction can be upward (positive), downward (negative), or stable (no obvious change).

[0027] As a further scheme of the present application, the specific steps of S4 are:

[0028] S401: Based on the relapse risk joint feature combination item, extract the distribution trend of each group of metabolic features and image features in the patient sequence, judge whether the change direction in the increasing number direction is consistent, assign a label to the combination item with the same trend, and generate a trend direction consistency marker value;

[0029] S402: Call the trend direction consistency marker value, screen the feature combination items with the same label, extract the corresponding number and merge, remove the repeated combination items with inconsistent numbers, and obtain a stable combination item number set;

[0030] S403: According to the stable combination item number set, map the metabolic features and image features to the original matrix, calculate the aggregation degree of the feature combination, classify the combination paths with aggregated numbers and consistent trends, and obtain a metabolic image collaborative channel list.

[0031] As a further scheme of the present application, the method further comprises:

[0032] S5: Based on the metabolic image collaborative channel list, extract the channel response data from the relapse and non-relapse samples, sort and reorganize according to the difference degree, construct a feature matrix, and generate a relapse risk feature joint input vector set;

[0033] The relapse risk feature joint input vector set includes channel response sorting features, reorganized feature matrix, and model input feature vector.

[0034] As a further scheme of the present application, the specific steps of S5 are:

[0035] S501: Based on the metabolic image collaborative channel list, extract the response signal of the channel in the relapse sample and the non-relapse sample, call the response values of the two types of samples under the channel, calculate the response difference degree between the channels, and generate a channel response difference matrix;

[0036] S502: According to the channel response difference matrix, the channel index of the response difference is screened, the number sequence in the original channel list is extracted, the original signal matrix is called, and the recombined channel response matrix is generated after adjusting the channel dimension sequence;

[0037] S503: Based on the recombined channel response matrix, all channel response values are extracted according to the sample index, an input structure with channel as a feature is constructed, a unified two-dimensional data format is integrated to generate a recurrent risk feature joint input vector set.

[0038] As a further scheme of the application, the recurrent sample refers to a sample labeled as having a recurrence event in the study, such as a patient or lesion that has relapsed in medical image analysis;

[0039] The response signal of the channel generally refers to the signal intensity, characteristic value or statistical response recorded by the channel or region in metabolic imaging (such as PET, fMRI);

[0040] The original signal matrix refers to the signal response data set collected initially before metabolic image analysis of the sample, which has not been reordered or screened.

[0041] Compared with the prior art, the application has the advantages and positive effects that:

[0042] In the application, the fluctuation trend comparison mechanism between cross-modal data is established by extracting disturbance features through the CT image edge gray path and combining the corresponding brain metabolism indicators of the patient, the biological consistency between features is enhanced, the feature combination with inconsistent change direction is eliminated, the feature pairs with coordinated structure and function trends are screened, the biologically representative linkage path is constructed, the evolution relationship from structural disturbance to metabolic response is reflected, the ordering and reordering of channel data realize the high differentiation of input features, effectively improve the stability and accuracy of recurrence discrimination, make up for the recognition blind area caused by separation of information dimension, and enhance the systematization and explainability of risk assessment. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The figure is a step flowchart of the application;

[0044] Figure 2 The figure is a step flowchart of S1 of the application;

[0045] Figure 3 The figure is a step flowchart of S2 of the application;

[0046] Figure 4 The figure is a step flowchart of S3 of the application;

[0047] Figure 5 The figure is a step flowchart of S4 of the application;

[0048] Figure 6 A step flow chart for S5 of the present application. DETAILED DESCRIPTION

[0049] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0050] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0051] Please refer to Figure 1 , a metabolomics-imagingomics prediction method for the risk of recurrence of chronic subdural hematoma, comprising the following steps:

[0052] S1: Obtain the CT image of a patient with chronic subdural hematoma, construct a transverse and longitudinal pixel sampling path in the bleeding edge region, extract the gray value difference and identify the gray fluctuation region, and obtain the edge disturbance multi-parameter description value based on the fluctuation intensity and region span information;

[0053] Explanation of imagingomics data acquisition method: including CT image preprocessing method (histogram equalization, edge detection algorithm), sampling path parameters (length, interval), gray fluctuation identification rule (difference threshold, span threshold);

[0054] S2: Combine the patient number in the edge disturbance multi-parameter description value, extract the brain metabolism related indicators in the metabolomics data, and construct the fluctuation feature parameter set according to the distribution range, change amplitude and dispersion degree of the indicators in the sample group, and arrange it into a metabolomics fluctuation distribution list;

[0055] Source and acquisition method of metabolomics data:

[0056] Experimental steps of PET for glucose metabolic rate, MRS method for detecting lactic acid / NAA / ATP, and explanation of data standardization processing;

[0057] S3: Combine the metabolic fluctuation distribution list with the edge disturbance multi-parameter description value, identify the change trend of the characteristics in the same patient, screen out the characteristic combination with inconsistent change direction or inconsistent change amplitude, and generate a recurrence risk joint characteristic combination item after integration;

[0058] S4: Based on the recurrence risk joint characteristic combination item, measure the trend direction of each group of metabolic characteristics and image characteristics in the patient sequence respectively, select the feature pairs with consistent trend, integrate them into a structure-function linkage path set, and output a metabolic image coordination channel list;

[0059] S5: Based on the metabolic image coordination channel list, extract the channel response data from the recurrence and non-recurrence samples, sort and reorganize them according to the difference degree, construct a feature matrix for model input, and output a recurrence risk feature joint input vector set.

[0060] The edge disturbance multi-parameter description value includes gray fluctuation intensity, region span, and fluctuation frequency. The metabolic fluctuation distribution list includes brain metabolism index distribution range, change amplitude, and dispersion degree. The recurrence risk joint characteristic combination item includes change direction consistency feature, change amplitude coordination feature, and structure and function information integration feature. The metabolic image coordination channel list includes structure-function linkage path number, feature pair trend consistency index, and channel response difference degree. The recurrence risk feature joint input vector set includes channel response sorting feature, reorganized feature matrix, and model input feature vector.

[0061] Please refer to Figure 2 , the specific steps of S1 are as follows:

[0062] S101: After obtaining the CT image of the patient with chronic subdural hematoma, based on the bleeding edge region in the image, construct horizontal and vertical pixel sampling paths, extract the gray value of the pixel points on the path, reconstruct the gray value sequence in the difference direction, and obtain the edge gray sampling matrix;

[0063] The imageomics data acquisition method is specifically described as follows:

[0064] CT image acquisition conditions (scanning parameters such as slice thickness, resolution);

[0065] Image preprocessing and path sampling details;

[0066] The imageomics data used comes from the postoperative routine head CT scan of the patient. The recommended acquisition parameters are: slice thickness 5mm, pixel matrix 512x512, tube voltage 120kV, and tube current automatic adjustment mode. The original DICOM image obtained needs to be uniformly converted into NIfTI format for processing.

[0067] Before entering the gray path sampling, the image needs to be preprocessed, including:

[0068] Adaptive histogram equalization (AHE) is used to enhance the edge contrast;

[0069] Sobel or Canny operator is used to extract the hematoma boundary;

[0070] Based on the extracted edge coordinates, define the horizontal and vertical pixel paths, the path length is set to 64 pixels, and the starting point interval is 5 pixels;

[0071] During the path sampling process, the pixel gray value on each path is stored in the matrix in the original order, which is used for subsequent identification and calculation of gray level fluctuation characteristics;

[0072] After obtaining the CT image of chronic subdural hematoma, first, image preprocessing operation is needed, adaptive histogram equalization method is used to enhance the edge contrast of the original image, then edge detection algorithm such as Sobel or Canny method is used to identify the bleeding profile area, and the edge pixel point coordinates are obtained. Select the starting sampling point at the edge profile, extend 64 pixel units in the horizontal and vertical directions respectively to construct the path, and the path step is 1 pixel. The horizontal path is defined as fixed vertical coordinate and horizontal coordinate increasing; the vertical path is fixed horizontal coordinate and vertical coordinate increasing. Set multiple sampling path starting points at an interval of 5 pixels in the edge area, and sequentially collect the gray values on each path. The sampling operation can extract the gray level by traversing the coordinates on the image matrix, and two gray sequences are formed in the horizontal and vertical directions respectively. For example, if the starting point is horizontal coordinate 120 and vertical coordinate 200, 64 points are collected to the right to obtain the sequence such as 52, 55, 58, and downward sampling obtains the sequence such as 48, 50, 51. Each path corresponds to two direction gray sequences. Taking 10 paths sampled from each image as an example, a gray sampling matrix containing 20 sequences in horizontal and vertical directions is finally formed, with the matrix dimension of 2 rows and 64 columns, and the number of 10 groups.

[0073] S102: According to the gray value sequence in the edge gray sampling matrix, identify the section with prominent gray value variation, extract the continuous gray fluctuation area, record the position span and intensity characteristics, integrate the gray fluctuation information in different directions, and obtain the edge gray fluctuation interval set;

[0074] In the acquired gray scale sampling matrix, the change of each path sequence needs to be analyzed, the difference between adjacent gray values is calculated, if the gray difference between adjacent points is continuously more than 10 and the continuous section is greater than or equal to 3 pixel units, it is marked as a mutation section, and the judgment standard adopts the absolute value change judgment method. If there are multiple mutation sections on a path that meet the conditions, the starting position, length and intensity are recorded respectively, wherein the span refers to the number of pixels from the first change point to the last point, and the intensity is the difference between the maximum gray value and the minimum gray value in the section. For example, in a certain path sequence, the gray value changes from 56 to 74 at positions 20 to 27, the span is 8, and the intensity is 18. The direction attribute, horizontal direction or vertical direction, path number, starting point, length, intensity and other information of each path are recorded and stored in the gray scale fluctuation interval set in turn. Assuming that the horizontal path with path number 3 in the image has a mutation from point 25 to point 32, the length is 8 and the intensity is 17, it is recorded as “horizontal, path 3, starting point 25, length 8, intensity 17”. Repeat the traversal of all paths to form a complete edge gray scale fluctuation interval set for subsequent analysis.

[0075] S103: Based on the region span and intensity characteristics in the edge gray scale fluctuation interval set, a plurality of directional gray scale disturbance feature indexes are constructed, the direction information is integrated to form a description value set, and an edge disturbance multi-parameter description value is obtained.

[0076] Based on the extracted gray scale fluctuation interval set, the direction, span and intensity in each section are classified and valued. The span is divided into short section 1 to 4, medium section 5 to 8 and long section 9 and above, and is valued as 1, 2 and 3 respectively. The intensity is divided into low disturbance 0 to 10, medium disturbance 11 to 20 and high disturbance 21 and above, and is valued as 1, 2 and 3 respectively. The horizontal direction attribute is defined as 1, and the vertical direction attribute is defined as 2. The disturbance value of each fluctuation section is obtained by multiplying the corresponding values of the three items, and the average value, maximum value, minimum value and standard deviation of all disturbance values are counted in each direction. If 6 fluctuation sections are detected in the horizontal direction of the image, the corresponding disturbance values are 6, 12, 6, 18, 6 and 12, then the horizontal average value is 10; in the vertical direction, the corresponding disturbance values of 4 sections are 12, 12, 6 and 18, and the vertical average value is 12. In addition, the maximum value is 18, the minimum value is 6, and the standard deviation is about 4.2. The above statistical items are integrated to form a complete disturbance description value set, including the average value of each direction, the maximum and minimum values of the total set and the standard deviation of the fluctuation degree, which constitutes a description index set for quantitatively identifying the edge disturbance characteristics.

[0077] Please refer to Figure 3 , the specific steps of S2 are:

[0078] S201: Obtain the patient number indicated in the edge disturbance multi-parameter description value, call the corresponding record in the metabonomics data, extract the brain metabolism correlation index, construct the patient metabolism characteristic parameter set, and generate the brain metabolism characteristic parameter set value;

[0079] The metabonomics data acquisition method is specifically described as follows:

[0080] PET scanning protocol (FDG injection dose, acquisition time, calculation method);

[0081] MRS detection conditions (1H-MRS spectrum acquisition, calibration method);

[0082] The metabolic indicators are mainly derived from positron emission tomography (PET) and magnetic resonance spectrum (MRS) data;

[0083] 18F-FDG PET:

[0084] The patient is intravenously injected with 18F-FDG (recommended dose 5 MBq / kg), and the scan is performed after waiting for 45 minutes in a resting state;

[0085] The scan adopts a three-dimensional acquisition mode, and the acquisition time per bed is 10 minutes;

[0086] The glucose metabolic rate (μmol·100g -1 ·min -1 ) is calculated by Patlak graph analysis method;

[0087] 1 H-MRS (proton magnetic resonance spectrum):

[0088] Under the 3.0T MRI scanner, the lesion and the corresponding region on the opposite side are selected, and the PRESS sequence is collected (TE=35 ms, TR=1500 ms, voxel size 20×20×20 mm 3 );

[0089] The concentrations of lactic acid, N-acetylaspartate (NAA), gamma-aminobutyric acid (GABA) and glutamic acid are converted by peak area integration;

[0090] The result unit is mmol / L;

[0091] 31 P-MRS or high performance liquid chromatography (HPLC):

[0092] Used for detecting cell energy metabolism related indicators (ATP, choline, etc.), unit mmol / L;

[0093] Data standardization processing:

[0094] All metabolic indicators need to be standardized by Z-score (subtract the population mean and divide by the standard deviation);

[0095] Generate a patient's individual metabolic feature vector (e.g. 6-dimensional vector: glutamate, lactate, NAA, GABA, glutamine, choline) as input for subsequent feature extraction and trend comparison;

[0096] The patient number referred to in the edge disturbance multi-parameter description value needs to be parsed by extracting the patient identification in the encoding field from the composite structure and converting it to a unified format to identify the patient involved. Then, the metabolic detection record corresponding to the number in the metabolome database is called. During the call of the metabolic data, the correspondence between the metabolic sample number and the patient number is first retrieved, and then the data record with the closest detection date to the specified observation window is filtered. The brain metabolism-related indicators contained in the metabolome data usually include glutamate, lactate, NAA, GABA, glutamine, choline, etc. When extracting these indicators, their detection concentration values are read in a fixed order. The obtained data are the concentration levels of each metabolite, such as glutamate 2.3 mM, NAA 1.8 mM, GABA 0.7 mM. The brain metabolism feature parameter set needs to be normalized. Each data needs to be normalized based on the mean and standard deviation of the population sample. After processing, each metabolite forms a standardized value and constitutes a six-dimensional vector, such as 1.2, 0.8, 1.0, 0.9, 1.1, 0.7. Finally, the metabolic vector is recorded as the feature parameter set value, and the corresponding patient number is bound to generate a set of brain metabolism parameters specific to the patient.

[0097] S202: Based on the brain metabolism feature parameter set value, the distribution state of the indicators in the sample group is detected, the deviation degree between the original value and the distribution range is analyzed, the deviation degree set corresponding to the brain metabolism indicators is constructed, and the brain metabolism indicator deviation sequence value is obtained.

[0098] The specific calculation formula for analyzing the deviation degree between the original value and the distribution range is:

[0099]

[0100] Calculate the brain metabolism indicator deviation feature value to obtain the brain metabolism indicator deviation sequence value.

[0101] Wherein, represents the deviation degree of the i-th item in the m-th brain metabolism indicator in the current sample, represents the original value of the i-th brain metabolism indicator in the m-th class in the current sample, represents the mean value of the i-th brain metabolism indicator in the m-th class in the sample group, and N represents the total number of samples in the sample group, T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators, T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators, T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators, T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators,

[0102] T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators,

[0103] T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators,

[0104] T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators, i (m) T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators,

[0105] T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators, T represents the original value of the brain metabolism of the ith item in the jth sample of the mth type in the sample group, and T represents the total number of brain metabolism indicators,

[0106] N is the number of samples, and in this example, the FDG-PET database of 37 healthy volunteers is taken;

[0107] T is the total number of brain metabolism indicator categories, and in this example, 3 categories are taken, including glucose metabolic rate, lactic acid concentration, and intracellular ATP concentration;

[0108] T is the original value of the current sample in the kth indicator, wherein the lactic acid concentration is obtained by collecting the spectrum peak intensity of 1H-MRS and converted by combining the calibration factor, and the unit is mmol per liter, and the ATP concentration is measured by 31P-MRS or tissue homogenate high performance liquid chromatography, and the unit is mmol per liter;

[0109] T is the original value of the current sample in the kth indicator, wherein the lactic acid concentration is obtained by collecting the spectrum peak intensity of 1H-MRS and converted by combining the calibration factor, and the unit is mmol per liter, and the ATP concentration is measured by 31P-MRS or tissue homogenate high performance liquid chromatography, and the unit is mmol per liter;

[0110] T is the original value of the current sample in the kth indicator, wherein the lactic acid concentration is obtained by collecting the spectrum peak intensity of 1H-MRS and converted by combining the calibration factor, and the unit is mmol per liter, and the ATP concentration is measured by 31P-MRS or tissue homogenate high performance liquid chromatography, and the unit is mmol per liter;

[0111] The above mean and variance are calculated with the same source data, and the variance is obtained by formula The square root is used for normalization to ensure consistent dimension;

[0112] Numerical integration procedure:

[0113] Given N = 37;

[0114] m is the glucose metabolic rate, current sample V i (1) = 33.2 μmol-100g -1 ·min -1 , sample mean μ i (1) = 38.0 μmol-100g -1 ·min -1 , sample standard deviation σ1= 4.2 μmol-100g -1 ·min -1 , from σ1 2 = (1 / N)Σ(V j (1) - μ i (1) ) 2 we get Σ(V j (1) - μ i (1) ) 2 = N- σ1 2 = 37 x (4.2) 2 = 37 x 17.64 = 652.68;

[0115] First term calculation:

[0116]

[0117] Taking the absolute value, we get 1.1429;

[0118] Second term summation over T = 3 and averaging:

[0119] k = 1 glucose metabolic rate, from above |-1.1429| = 1.1429;

[0120] k = 2 lactate concentration, current sample V i (2) = 0.90 mmol-L -1 , mean μ i (2) = 0.70 mmol-L -1 , σ2= 0.20 mmol-L -1 , Σ(V j (2) - μ i (2) ) 2 = N- σ2 2= 37 x 0.04 = 1.48;

[0121] k = 3 ATP concentration, current sample V i (3) = 2.30 mmol L -1 , mean μ i (3) = 2.70 mmol L -1 , σ3= 0.30 mmol L -1 , Σ(V j (3) - μ i (3) ) 2 = 37 x 0.09 = 3.33;

[0122]

[0123] Sum and average:

[0124]

[0125] Overall Δ i (1) :

[0126]

[0127] The results show that the current sample's comprehensive deviation measure on indicators such as glucose metabolic rate is close to zero, indicating that the overall deviation is small under the combined action of cross-class standardization deviation and the class, and the value as the corresponding position element in the brain metabolic indicator deviation sequence participates in sequence construction and subsequent sorting or clustering processing;

[0128] The formula constructs the deviation measure Δ i (m) , where the first part is the standardized deviation of the current sample on the mth indicator in the ith class, reflecting its deviation from the group mean, and the difference between the original value and the sample group mean is used as the numerator, and the square root of the sample variance is used as the denominator for normalization. The use of the square root ensures that the denominator dimension is consistent with the numerator, while improving the control ability of extreme values, forming a dimensionless deviation value; the second part of the formula is the absolute value of the standardized deviation of the current sample on the ith indicator in all T classes and its group mean, and then the sum is averaged, which reflects the overall fluctuation level of the sample in the horizontal indicator class dimension. By adding the first part, the fusion measure of longitudinal (single class) and horizontal (cross-class) deviation intensity is realized, and the absolute value operation eliminates the directional interference of positive and negative deviation, retaining only the deviation amplitude, so that Δ i (m)The individual and the whole are relative, and the requirements of dimension consistency and mathematical rigor are met.

[0129] The brain metabolism index deviation characteristic value is used to quantify the deviation degree of a single sample relative to the reference distribution of the healthy population under a specific brain metabolism index. This value is obtained by combining the normalized difference between the current index value and the sample group mean value and standard deviation, and superimposing the average result of the relative fluctuation amplitude of the index in different categories of brain metabolism systems. It comprehensively reflects the metabolic deviation state of the sample in the vertical dimension (the abnormality degree of the individual to the group) and the horizontal dimension (the consistency or imbalance trend across the system). The larger the value, the more significant the abnormality in the brain metabolism pattern. It has the function of characterizing pathological changes, metabolic imbalance or early abnormality screening. As a single point element in the deviation sequence value, it provides basic quantitative support for subsequent sequence analysis, clustering classification, pathological prediction, etc.

[0130] S203: According to the brain metabolism index deviation sequence value, the fluctuation amplitude grade corresponding to the index is identified, the fluctuation range of the index is determined by combining the variation in the group sample, and the metabolic fluctuation distribution list is established after integration;

[0131] After the brain metabolism index deviation sequence value is formed, the fluctuation amplitude grade of each index needs to be determined. The judgment standard is based on the relationship between the absolute value of the deviation and the standard deviation of the group. If the deviation value exceeds 1.5 times the standard deviation, it is identified as high fluctuation; if it is between 0.5 and 1.5 times, it is identified as medium fluctuation; and if it is less than 0.5 times, it is low fluctuation. If the glutamate deviation value is -1.8 times the standard deviation, it is classified as high fluctuation. When analyzing the fluctuation of this index in the group, the proportion relationship between the number of samples belonging to the same fluctuation grade and the total number of samples is calculated. For example, among 100 samples, 30 are high fluctuation, forming a high fluctuation proportion of 30%. The fluctuation range of this index is defined as -2.5 to -1.5. After processing all the metabolic indexes in this way, the metabolic fluctuation distribution list is integrated. Each item in the list includes the index name, the original value, the deviation degree, the fluctuation grade, the group variation range, and the corresponding sample proportion. For example, NAA is 1.8 mM, the deviation is medium, the fluctuation grade is medium, and the proportion of samples in which this item is in the medium grade is 45%. The GABA deviation is low, the fluctuation is low, and the distribution proportion is 55%. In this way, a complete set of individual metabolic fluctuation characteristics is obtained, which serves as the basic data for subsequent analysis and comparison.

[0132] Please refer to Figure 4 , and the specific steps of S3 are:

[0133] S301: Based on the metabolic fluctuation distribution list, the fluctuation value and distribution characteristics of the metabolites are extracted, and the fluctuation coefficient and direction in the edge disturbance multi-parameter description value are combined. The feature set with consistent change direction and coordinated amplitude is selected, and the fluctuation-disturbance consistent feature trend value is generated.

[0134] Based on the metabolic fluctuation distribution list, first, the concentration data of each metabolite at multiple sampling time points or different pathological states is classified, the change amplitude between the maximum and minimum is identified, the mean value and fluctuation range in the whole cycle are counted, the fluctuation characteristic curve is constructed by distribution trend fitting, the average value and fluctuation range are labeled, the disturbance coefficient in the edge disturbance multi-parameter description is extracted, which can be usually calculated according to the difference ratio before and after the corresponding physiological or pathological index changes, and the direction is set to positive or negative according to the rising and falling trend of the index value, then the fluctuation direction of the metabolite is compared with the disturbance direction, the combination items with the same direction are screened out, then the difference degree between the fluctuation amplitude of the metabolite and the disturbance coefficient is calculated, the coordination degree is evaluated by the relative ratio between the two difference values, if the difference value between the two is less than the set standard, for example, within 20%, it is determined that the coordination is good, if the coordination degree is higher than the specified threshold, for example, more than 80%, the metabolite and the disturbance factor form a qualified combination item, finally, the trend values of all qualified combination items in each sample are summarized, and the average value is integrated respectively, and the trend characteristic numerical value with the same fluctuation and disturbance direction is obtained, which can be used for the next step of screening.

[0135] S302: Call fluctuation-disturbance consistent feature trend value, identify the items with fluctuation abnormality or direction mutation in the feature combination, judge according to the amplitude difference value threshold and direction offset threshold, remove the combinations that do not meet the conditions, and obtain the double offset coordination feature coefficient value;

[0136] Call the fluctuation and disturbance consistent feature trend value obtained, and detect whether there is obvious deviation in each combination of metabolites and disturbance factors in turn. First, the fluctuation amplitude of the metabolite during sampling and the disturbance intensity corresponding to the disturbance factor are calculated respectively, and the numerical difference between the two is recorded. If the difference exceeds the specified threshold, for example, 0.2, it is marked as amplitude deviation. Then, it is judged whether the signs of the change directions of the two are consistent. If one is positive and the other is negative, it means that the direction has mutated. Such cases will be marked as direction offset. Set the offset judgment reference value, such as 0.5. When any deviation in the combination item exceeds the above standard, it is removed. Only the combinations that are coordinated in both amplitude and direction are retained. Then, a coefficient value is given to each combination that meets the conditions according to its coordination performance. The closer the value is to 1, the higher the coordination degree. The coefficient values of all combinations are uniformly archived, and the coordination combinations and their average coordination degree in each sample are recorded according to the samples. Finally, the double offset coordination feature coefficient of each sample is calculated, which is used as the input basis for subsequent summary.

[0137] S303: According to the double offset coordination feature coefficient value, summarize the feature combinations that meet the coordination standard, normalize the disturbance consistency and fluctuation amplitude, integrate the combination items according to the patient dimension, and generate the recurrence risk joint feature combination item;

[0138] According to the dual offset coordination feature coefficient, the combination items meeting the set standard are screened, for example, the items with a coefficient higher than 0.85 are taken as qualified sets, then the disturbance consistency performance of each qualified combination is uniformly scaled, and a unified standard is used to map the disturbance performances of different intensities to the same numerical interval, and similarly, the fluctuation amplitude of the metabolites is also relatively converted, for example, converted into percentage values based on the maximum fluctuation amplitude, then for a single individual sample, all the qualified combination items contained therein are summarized, the number of occurrences of each item is taken as the frequency parameter, then the disturbance performance and the fluctuation amplitude result of each combination item are multiplied to form a weight value, the frequency and the weight of the combination item are multiplied to obtain the total value of the feature combination of each sample, which is taken as a numerical index representing the recurrence risk, forming structured data for subsequent input source for hierarchical classification judgment.

[0139] Please refer to Figure 5 , the specific steps of S4 are:

[0140] S401: Based on the recurrence risk joint feature combination item, the distribution trend of each group of metabolic features and image features in the patient sequence is extracted, and it is judged whether the change direction in the numbering increasing direction is consistent, and the combination item with the same trend direction is labeled to generate a trend direction consistency label value;

[0141] Based on the determination of the recurrence risk joint feature combination item, the distribution trend of each group of metabolic features and image features needs to be extracted in the patient number sequence. It is assumed that the patient number increases from P1 to P100, the metabolic features can be selected such as glucose concentration, lactic acid accumulation rate, etc., the image features can include lesion density change, edge irregularity score, morphological distortion index, etc., the values of each group of features are arranged in sequence to form a sequence curve, and the original sequence is smoothed by using a five-point moving average method, and then the increment value between adjacent numbers is calculated to identify the change direction. If the value of a metabolic feature increases from 3.1 to 5.0 in P1 to P10, and the image feature also shows a similar increasing trend, then the trend direction of the two is consistent, the change direction determination method can be that a positive difference value indicates an increase, and a negative difference value indicates a decrease. If the metabolic and image trends in the combination are in the same direction, the combination is marked as consistent, and a label value of 1 is assigned, otherwise 0. This operation needs to be performed for each combination item, and a trend consistency label column is generated. The label value is used as a basis for subsequent screening.

[0142] S402: Call the trend direction consistency label value, screen the feature combination items with the same trend, extract the corresponding numbers and merge them, remove the repeated combination items with inconsistent numbers, and obtain a stable combination item number set;

[0143] Combination items are screened according to the generated trend consistency label value, combination items with inconsistent trend directions are removed, only feature combinations with a label value of 1 are retained, and the associated number pairs are extracted. Assuming that a combination item is M21 and I03, the involved patient numbers are P5, P7, P10, etc., an initial number set is constructed, all numbers are aggregated according to the order of appearance and de-duplicated, duplicate items are excluded using set processing logic, then the frequency of each number in all combinations is counted, and the screening threshold is set to a frequency of no less than 2 times. This value can be obtained by statistical analysis of the number frequency distribution. Assuming that the average frequency is 3.6 and the standard deviation is 1.1, the lower limit is about 2 as the effective combination judgment standard, numbers with a frequency lower than this value are excluded, and finally the number set of stable combination items with a qualified frequency is retained to provide a number basis for the subsequent feature mapping.

[0144] S403: According to the stable combination item number set, map the metabolic features and image features to the original matrix, calculate the aggregation degree of the feature combination, classify the combination paths with consistent trends, and obtain the metabolic image collaborative channel list;

[0145] The specific calculation formula for calculating the aggregation degree of the feature combination is:

[0146]

[0147] Calculate the aggregation strength index, classify the combination paths with consistent trends, and obtain the metabolic image collaborative channel list.

[0148] Wherein, represents the aggregation strength index of the i-th feature combination, n i represents the number of metabolic or image feature pairs contained in the i-th group, M ij represents the normalized value of the j-th metabolic feature in the i-th group, represents the average value of all metabolic features in the i-th group, represents the average value of metabolic features in the k-th combination path, represents the average value of the average values of metabolic features in all combination paths, I ij represents the normalized value of the j-th image feature in the i-th group, represents the average value of the image features in the i-th group, represents the average value of image features in the k-th combination path, represents the average value of the average values of image features in all combination paths, K represents the total number of combination paths.

[0149] Take the i-th group containing n_i=10 metabolic-image feature pairs as an example, M ij , I ij are obtained by monitoring the filling value mode.

[0150]

[0151] K=5 combined paths, each path is monitored by different samples to obtain the average value between paths.

[0152]

[0153] Enter numerical examples:

[0154] Assume that M is monitored ij The cumulative deviation of the 10 items ∑(M ij _i) = 0.2, then:

[0155]

[0156] If ∑_{k=1}^5|_k-| = 0.15,

[0157]

[0158] The metabolic part is 0.02+0.1732=0.1932.

[0159] The image part is the same, if ∑(I_{ij}-_i) = -0.1, then:

[0160]

[0161] If ∑|_k-| = 0.08, then:

[0162]

[0163] The image part is 0.01+0.1265=0.1365.

[0164] Overall, enter the previous results:

[0165]

[0166] The results show that the numerical value represents the aggregation strength index of the feature combination group. By corresponding to the step results, the numerical value is used for classification and consistent combination path; the aggregation degree is further filtered by the value size to determine the synergistic channel group;

[0167] The operation logic of the formula is based on the distribution deviation within each of the metabolic feature group and the image feature group and the trend stability across paths for segmented aggregation analysis. First, the absolute value of the average of the sum of the differences between each metabolic feature within the group and the average value in the group is calculated to measure the concentration of the features within the group. Then, the square root mean of the deviation amplitude of the metabolic feature mean in all paths relative to the overall path metabolic mean is calculated to reflect the overall fluctuation trend across paths. The sum of the two parts is used to integrate local consistency and global trend consistency. The image feature part performs the same operation with the same structure to ensure that there is no dimensional interference in the collaborative calculation of metabolism and images. Finally, the metabolic and image results are added to represent the comprehensive aggregation effect of the feature combination in the metabolic channel and image channel dimensions, thereby forming a unified index to support the subsequent path clustering screening mechanism.

[0168] The aggregation strength index is used to quantify the overall consistency of metabolic features and image features in a combination. Its specific meaning is to measure the concentration of the features within the group and the trend stability across the combination paths. The internal concentration reflects the tightness of the feature values around the group mean, and the trend stability reflects the consistency of the mean values of the same type combination in different paths. The smaller the index value, the stronger the aggregation and trend consistency of the feature combination in the metabolic channel and image channel dimensions, which can be considered as a candidate channel with good metabolic and image collaboration, thereby serving as a quantitative basis for subsequent channel screening, path classification and collaborative channel construction.

[0169] Please refer to Figure 6 The specific steps of S5 are as follows:

[0170] S501: Based on the metabolic and image collaborative channel list, the response signals of the channels in the recurrent samples and non-recurrent samples are extracted, the response values of the two types of samples under the channel are called, the response difference degree between the channels is calculated, and a channel response difference matrix is generated.

[0171] Based on the metabolic image collaborative channel list, the channel index and number information contained in the list need to be imported first, for example, there are 100 channels numbered from T1 to T100. Then, the metabolic image data of the recurrence group and the non-recurrence group are called respectively, and are organized into a three-dimensional structure according to the sample quantity, channel quantity and time or space dimension. If the sample quantity of each group is 20 and 25 respectively, and each channel contains 10 time point response values, all response data of the corresponding channel need to be extracted. In order to ensure the comparability of the channel response values, the channel values need to be standardized. The mean value in the channel can be used to reduce and divide by the standard deviation to convert to a unified range. After completing the preprocessing, the response difference of each channel in the two types of samples is evaluated by the way of two-sample difference comparison. For example, the response mean of the 57th channel in the two groups of samples is compared, and the difference degree is calculated combined with the response fluctuation amplitude and the sample quantity. When the difference score exceeds the preset standard, such as the statistical significance threshold is set to 0.05, it is determined that there is a difference. The difference degree is summarized according to the channel dimension to form a vector structure containing the scores of all channels, for example, the difference value of channel 3 is 0.12, the difference value of channel 57 is 3.42, and the difference value of channel 92 is 0.78, etc. That is, a difference degree list is formed, which can be used as the basis for subsequent screening.

[0172] S502: According to the channel response difference degree matrix, the channel index with response difference is screened, the number order in the original channel list is extracted, the original signal matrix is called, and the recombined channel response matrix is generated after adjusting the channel dimension order;

[0173] According to the channel response difference degree score obtained in the last stage, all channel difference values are sorted, and channels that meet the difference standard are selected, for example, the top 30 channels are selected, and the numbers may include channel 3, channel 7, channel 15, channel 22, etc. Then, the original response matrix data is called, and the channel dimension is reordered, with the high difference channels at the front, and the order arranged according to the number in the original channel. This operation ensures that the channel sorting is adjusted according to the channel difference degree, rather than arranged according to the initial channel number. When executed, the sample and time point or space point dimensions remain unchanged, and only the channel position order is adjusted. For example, the original channel arrangement is T1 to T100, and the high difference channels selected are T3, T7, T15, T22, etc., so the channel order is recombined as: T3, T7, T15, T22, and the rest of the channels. The adjusted channel order is applied to the entire data matrix to output a new three-dimensional data structure, with the sample dimension and time dimension remaining unchanged, and only the channel dimension being rearranged according to the difference degree. For example, 45 samples are processed, and the channel arrangement order of each sample is updated after recombination to form a unified format matrix for the next step of structure input.

[0174] S503: Based on the reordering channel response matrix, all channel response values are extracted by sample index to construct the input structure of channel as a feature, integrate into a unified two-dimensional data format, and generate a recurrent risk feature joint input vector set;

[0175] On the basis of obtaining the channel reordered matrix, all channel response values need to be extracted in turn according to the sample index. Each sample contains several channels and multiple time point response data of each channel, which can be aggregated by time point average, maximum or standard deviation, etc. to form a channel level value. For example, channel 3 in each sample contains 10 time point response values, and the average value is calculated to obtain the feature value of channel 3. Repeat the process to generate a channel feature vector corresponding to each sample. If the number of channels is 100, the vector length of each sample is 100. By traversing all samples, a two-dimensional matrix structure can be formed, with rows representing samples and columns representing channel feature values. To improve the feature expression ability, low response values can also be filtered out, for example, set the response lower threshold to 1.2, and when the maximum response value of a channel is lower than the threshold, the channel response is set to 0. In addition, the channel difference degree can be converted into a weight value to participate in feature construction, for example, the channel with higher difference degree is given greater weight, and all channel difference degrees are normalized as a weight vector. The weighted feature is obtained by multiplying the channel feature value and the corresponding weight. The final generated structure is a two-dimensional input data matrix with sample number rows and channel number columns, which is used for subsequent modeling analysis and other operations.

[0176] The above is only a preferred embodiment of the present application, and does not limit the form of the present application. Any skilled person in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the present application.

Claims

1. A metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma, characterized in that, Includes the following steps: S1: Acquire CT images of patients with chronic subdural hematoma, construct horizontal and vertical pixel sampling paths in the hemorrhage edge region, extract gray value differences and identify gray value fluctuation regions, and obtain multi-parameter description values ​​of edge perturbation based on fluctuation intensity and region span information. S2: Combining the patient number indicated in the multi-parameter description value of the edge perturbation, extract brain metabolism-related indicators from the metabolomics data. Based on the distribution range, variation amplitude and dispersion of the indicators in the sample group, construct a set of fluctuation feature parameters for metabolic features and organize them into a list of metabolic fluctuation distributions. S3: Combining the metabolic fluctuation distribution list with the edge perturbation multi-parameter description value, identify the changing trend of features in the same patient, screen out feature combinations with inconsistent changing directions or inconsistent changing magnitudes, and generate joint feature combination items for recurrence risk. S4: Based on the combined features of recurrence risk, measure the trend of each group of metabolic features and imaging features in the patient sequence, screen feature pairs with similar trends, integrate them into a set of structural-functional linkage paths, and generate a list of metabolic-imaging collaborative channels.

2. The metabolomics-radiomics prediction method for the risk of recurrence of chronic subdural hematoma according to claim 1, characterized in that, The edge perturbation multi-parameter descriptive values ​​include grayscale fluctuation intensity, region span, and fluctuation frequency. The metabolic fluctuation distribution list includes the distribution range, variation amplitude, and dispersion of brain metabolic indicators. The recurrence risk joint feature combination items include the consistency feature of change direction, the coordination feature of change amplitude, and the integration feature of structural and functional information. The metabolic imaging collaborative channel list includes the number of structure-function linkage paths, the feature-to-trend consistency index, and the channel response difference.

3. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: After acquiring CT images of patients with chronic subdural hematoma, construct horizontal and vertical pixel sampling paths based on the bleeding edge region in the image, extract the gray values ​​of pixels on the path, reconstruct the gray value sequence in the difference direction, and obtain the edge gray sampling matrix. S102: Based on the gray value sequence in the edge gray value sampling matrix, identify the segments with prominent gray value changes, extract continuous gray value fluctuation areas, record the location span and intensity characteristics, integrate the gray value fluctuation information in different directions, and obtain the edge gray value fluctuation interval set. S103: Based on the regional span and intensity characteristics of the edge gray-level fluctuation interval, construct multiple sets of directional gray-level disturbance feature indicators, integrate directional information to form a set of descriptive values, and obtain multi-parameter descriptive values ​​of edge disturbance.

4. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Obtain the patient number from the edge perturbation multi-parameter description value, call the record corresponding to the number in the metabolomics data, extract brain metabolism-related indicators, construct a patient metabolic feature parameter set, and generate brain metabolic feature parameter set values. S202: Based on the set of brain metabolism feature parameters, detect the distribution of the index in the sample group, analyze the degree of deviation between the original value and the distribution range, construct the set of deviation degrees corresponding to the brain metabolism index, and obtain the brain metabolism index deviation sequence value. S203: Based on the brain metabolic index offset sequence value, identify the fluctuation level corresponding to the index, combine it with the changes in the population sample, determine the fluctuation range of the index, and establish a metabolic fluctuation distribution list after integration.

5. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the metabolic fluctuation distribution list, extract the fluctuation value and distribution characteristics of metabolic substances, and combine them with the perturbation coefficient and direction in the edge perturbation multi-parameter description value to screen the feature set with consistent change direction and coordinated amplitude, and generate fluctuation-perturbation consistent feature trend value. S302: Call the fluctuation-disturbance consistent feature trend value, identify items with abnormal fluctuations or sudden changes in direction in the feature combination, judge according to the amplitude difference threshold and the direction offset threshold, remove combinations that do not meet the conditions, and obtain the dual offset coordination feature coefficient value. S303: Based on the dual offset coordination feature coefficient values, summarize the feature combinations that meet the coordination criteria, normalize the perturbation consistency and fluctuation amplitude, integrate the combination items according to the patient dimension, and generate the recurrence risk joint feature combination item.

6. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 5, characterized in that, The fluctuation value of the metabolite refers to the degree of change in the concentration or expression level of the metabolite under different states (such as disease state and healthy state, or different time points, or between different tissue samples). The distribution characteristics of the metabolites describe the statistical properties of metabolite fluctuations, including: skewness, kurtosis, normality, and multimodality. The amplitude difference threshold refers to a limit value for comparing whether the numerical difference between two variables (such as metabolic fluctuation trend and perturbation trend) is acceptable when screening feature combinations. The directional offset threshold refers to the tolerance for directional deviation when determining whether two trends are "in the same direction". The direction can be upward (positive), downward (negative), or stable (no significant change).

7. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Based on the combined features of recurrence risk, extract the distribution trend of each group of metabolic features and imaging features in the patient sequence, determine whether the direction of change in the direction of increasing number is consistent, assign a label to the combined items with the same trend, and generate a trend direction consistency label value. S402: Call the trend direction consistency marker value, filter feature combination items with the same label, extract the corresponding number and merge them, remove duplicate combination items with inconsistent numbers, and obtain a stable combination item number set; S403: Based on the set of stable combination item numbers, map metabolic features and image features to the original matrix, calculate the degree of clustering of feature combinations, classify the combination paths with clustered numbers and consistent trends, and obtain a list of metabolic image collaborative channels.

8. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 1, characterized in that, The method further includes: S5: Based on the metabolic imaging collaborative channel list, extract channel response data from relapse and non-relapse samples, sort and recombine them according to the degree of difference, construct a feature matrix, and generate a joint input vector set of relapse risk features; The joint input vector set of relapse risk features includes channel response ranking features, recombination feature matrix, and model input feature vector.

9. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 8, characterized in that, The specific steps of S5 are as follows: S501: Based on the metabolic imaging collaborative channel list, extract the response signals of the channels in the relapsed and non-relapsed samples, call the response values ​​of the two types of samples under the channel, calculate the degree of response difference between the channels, and generate the channel response difference matrix. S502: Based on the channel response difference matrix, filter the channel indexes with response differences, extract the number order in the original channel list, call the original signal matrix, adjust the channel dimension order, and generate a recombined channel response matrix; S503: Based on the recombined channel response matrix, extract all channel response values ​​according to the sample index, construct an input structure with channels as features, integrate them to form a unified two-dimensional data format, and generate a joint input vector set of relapse risk features.

10. The metabolomics-radiomics method for predicting the risk of recurrence of chronic subdural hematoma according to claim 9, characterized in that, The relapsed sample refers to the sample that was marked as having a relapse event in the study, such as a patient or lesion that has relapsed in medical image analysis. The response signal of the channel usually refers to the signal intensity, characteristic value or statistical response recorded in a channel or region in metabolic imaging (such as PET, fMRI); The original signal matrix refers to the set of signal response data initially collected before sample metabolic imaging analysis, without reordering or filtering.

Images