A method and system for predicting the risk of hemiplegia for a patient with brain trauma

Abstract

The method provided by the embodiment of the application is a hemiplegia risk prediction method and system suitable for brain trauma patients, the method obtains patient information and medical data of brain trauma patients uploaded by different medical nodes, screens high-risk hemiplegia patients through a comprehensive risk assessment score, constructs and pre-trains a hemiplegia prediction model to obtain a brain trauma patient hemiplegia risk score, automatically evaluates and records the hemiplegia risk score through a first smart contract, triggers an early warning according to a preset risk threshold and sends early warning information to a medical node and a patient, and automatically triggers hemiplegia prediction model updating and recording through a second smart contract. The method provided by the application calculates a risk score by comprehensively utilizing hemiplegia symptom keywords and dynamic physiological indexes, realizes early screening of high-risk hemiplegia patients, realizes accurate risk prediction of image data based on a prediction model, automatically triggers an early warning, and improves the timeliness of hemiplegia risk intervention for brain trauma patients.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data processing, and particularly relates to a hemiplegia risk prediction method and system suitable for brain trauma patients.

BACKGROUND

[0002] Brain trauma is one of the important reasons for causing nervous system dysfunction, and hemiplegia is a common and serious manifestation in its sequelae. There is a complex pathological relationship between brain trauma and hemiplegia, including direct damage leading to loss of nerve function and secondary pathological processes causing secondary damage.

[0003] At present, in the field of hemiplegia risk assessment after brain trauma, traditional methods mainly rely on doctors to make comprehensive judgments on patient symptoms, image data and physiological indicators by hand, which not only has strong subjectivity and low efficiency, but also is difficult to respond to changes in the condition in a timely manner, resulting in that part of the high-risk patients cannot obtain early intervention. With the development of medical imaging technology, deep learning and data analysis, using automatic means for risk assessment and prediction has become a research hotspot.

SUMMARY

[0004] Therefore, the embodiments of the present application provide a hemiplegia risk prediction method and system suitable for brain trauma patients.

[0005] In a first aspect, the embodiments of the present application provide a hemiplegia risk prediction method suitable for brain trauma patients, which comprises:

[0006] S1, obtaining patient information and medical data of brain trauma patients uploaded by different medical nodes, and performing preprocessing and encryption processing and then distributing and storing in a multi-center cloud platform;

[0007] S2, calculating a comprehensive risk assessment score according to the medical data, and screening high-risk patients with hemiplegia through the comprehensive risk assessment score;

[0008] S3, performing image feature extraction on the high-risk patients with hemiplegia, constructing and pre-training a hemiplegia prediction model, inputting the extracted image features into the hemiplegia prediction model, and obtaining a hemiplegia risk score of the brain trauma patient;

[0009] S4, automatically evaluating and recording the hemiplegia risk score through a first smart contract, triggering an early warning according to a preset risk threshold and sending the early warning information to the medical nodes and the patient, and automatically triggering the hemiplegia prediction model update and recording through a second smart contract.

[0010] As described above, the aspect and any possible implementation manner further provide an implementation manner, and the S1 comprises:

[0011] S11, extract medical image data from the acquired medical data of the brain trauma patient uploaded by different medical nodes, remove image noise using an adaptive filtering method, enhance the contrast of the image using a histogram equalization method, perform image registration to complete alignment on multi-time or multi-modal medical image data, and segment the brain region using an image segmentation algorithm to remove non-background regions;

[0012] S12, standardize the medical data, including normalization and missing value completion;

[0013] S13, generate a hash value by hashing the unique identifier of the cloud storage node, each cloud storage node is distributed according to the position of the hash value in the hash ring to form a hash ring, and the hash value is calculated according to each target patient information as a globally unique identifier of the patient, the nearest cloud storage node is found on the hash ring according to the globally unique identifier, and the medical data of the target patient is distributed to the cloud storage node, and the copy of the target patient medical data is copied clockwise to the subsequent N redundant nodes on the hash ring, when the medical data is updated, the incremental information is pushed to the redundant nodes;

[0014] S14, if a new cloud storage node is added, calculate the hash value of the new node and add it to the hash ring, only migrate the original data interval covered by the new node in the hash ring, use the double-write strategy of old node and new node simultaneous writing during migration, after migration, the new node is included in the subsequent N redundant targets, if a node failure is detected, switch the read-write request to the nearest redundant copy node, if the redundant copy data is complete, directly promote it to a new master node and supplement new redundant copies, if the redundant copy is not complete, recover the data from other copies, after the node is recovered, synchronize the incremental data from the redundant copy to the latest state.

[0015] According to any possible implementation of the above-mentioned aspect, further provided is an implementation, and the S2 specifically comprises:

[0016] S21, construct a hemiplegia symptom keyword library, extract hemiplegia symptom keywords from the medical data of the brain trauma patient according to the hemiplegia symptom keyword library and count the occurrence frequency to generate a keyword list;

[0017] S22, select a physiological index related to hemiplegia and changing over time as a dynamic change feature, and calculate a dynamic change feature value from the medical data of the brain trauma patient;

[0018] S23, standardize the keyword list and the dynamic change feature value, and calculate a comprehensive risk assessment score according to the keyword list and the dynamic change feature value, wherein the calculation formula of the comprehensive risk assessment score is as follows:

[0019]

[0020] Wherein, C is the comprehensive risk assessment score, γ1 and γ2 are weight coefficients, F is the keyword frequency of hemiplegia symptoms, k i is the frequency of the i th hemiplegia symptom, w i is the weight of the i th symptom, n is the number of selected hemiplegia symptoms, Δf j (t) is the dynamic change characteristic value of patient physiological data, λ j is the weight of the j th dynamic change characteristic, m is the number of selected dynamic change characteristics;

[0021] S24, draw the ROC curve based on historical patient data, select the T value when the Youden index is maximum as the initial risk threshold, take T / K as the final risk threshold, K is a set constant, compare the size of the comprehensive risk assessment score C and the final risk threshold T / K, when C≥T / K, it is screened as a high-risk patient with hemiplegia.

[0022] Aspects and any possible implementation manners described above, further provide an implementation manner, the dynamic change characteristic includes, but is not limited to, intracranial pressure fluctuation index, Glasgow coma score decline rate, limb muscle strength change rate, systolic pressure variability, electroencephalogram bispectral index trend and limb movement frequency any one or more, wherein,

[0023] The formula for calculating the intracranial pressure fluctuation index is σ I represents the standard deviation of the intracranial pressure measurement value in the preset time window, μ I represents the arithmetic mean of the intracranial pressure in the preset time window;

[0024] The formula for calculating the Glasgow coma score decline rate is t i represents the time point of the i th score, G i represents the Glasgow coma score at the corresponding time point, represents the average value of all time points, represents the average value of the Glasgow coma score at all time points;

[0025] The formula for calculating the limb muscle strength change rate is M t represents the t th muscle strength score, N represents the total number of assessments, and when ΔM is negative, it indicates that the muscle strength continues to decline; The formula for calculating the systolic pressure variability is R S =max (S t )-min (S t ), S t represents the systolic pressure measurement value at each time point in the preset sliding window;

[0026] The formula for calculating the electroencephalogram bispectral index trend is E t =α·Bt + (1 - a) · E t-1 , E t denotes the t-th electroencephalogram bispectral index measurement value, a denotes a smoothing factor, E0=B0;

[0027] The limb movement frequency calculation formula is Q denotes the number of peaks of limb activity amplitude exceeding a threshold value detected by an accelerometer, and ΔT denotes a time window.

[0028] Aspects and any possible implementation modes described above, further provide an implementation mode, and the S3 specifically comprises:

[0029] S31, historical medical images of a hemiplegia high-risk patient are acquired, including CT, MRI, CTP images, and a hemiplegia label is labeled according to a follow-up result;

[0030] S32, registration is performed on different images of the same patient, different image sizes / resolutions are unified to a standard, intensity normalization is performed, and a data set is divided into a training set and a verification set;

[0031] S33, a hemiplegia prediction model is constructed and trained in advance by the training set, wherein the hemiplegia prediction model adopts a Stacking structure of three sub-models and a meta-model, the three sub-models process three types of image features of edema, hemorrhage and abnormal cerebral blood flow respectively, each sub-model is trained on an independent data set, and then corresponding features are output in an inference stage for the meta-model to make final fusion, and the meta-model is trained;

[0032] S34, medical images of a new hemiplegia high-risk patient are input, are sent into the three sub-models for inference, and then are fused by the meta-model to obtain a final hemiplegia risk score.

[0033] Aspects and any possible implementation modes described above, further provide an implementation mode, and the constructing and training the hemiplegia prediction model in advance by the training set specifically comprises:

[0034] An edema detection sub-model is constructed: a network structure uses a 3D U-Net for edema region segmentation, an encoder is 5 layers of 3D convolution+ReLU+Downsample, a decoder is 5 layers of deconvolution+skip-connection, an input is a CT low-density area, and an output is a segmentation mask of an edema area, an edema probability and an edema volume are calculated from a segmentation result, a loss function selects a Dice Loss and a BCE / CE for joint use, a Dropout is used to reduce overfitting, an adaptive learning rate is used, a labeled edema area image is used as a training set, a random rotation, translation, cropping and brightness change method is used to enhance data, 100 epochs are trained, and after training, an edema region segmentation accuracy is verified by a verification set.

[0035] The bleeding detection sub-model is constructed: the network structure uses DenseNet 3D, the input is CT plain scan, the output is bleeding probability and bleeding plaque volume, the loss function selects BCE / CE, the CT image with bleeding plaque annotation is used as the training set, the data is enhanced by random rotation, translation, cropping and brightness change, and the bleeding recognition accuracy is verified through the validation set after training;

[0036] The cerebral blood flow detection sub-model is constructed: the network structure uses 3D-CNN, the input is the MRI perfusion sequence, the output is the perfusion abnormality probability, the loss function adopts the BCE / CE loss function, the low perfusion region annotation image is used as the training set, and the time translation and inter-frame noise are used for data enhancement;

[0037] The meta-model is constructed and trained: the random forest is selected as the meta-model, the numerical features are obtained based on the three sub-models, the image feature edema probability, bleeding probability and perfusion abnormality probability are integrated into a feature vector as the input of the meta-model, the sub-model output is extracted from the training set image for each patient, and the "whether hemiplegia" label is paired for training the meta-model, the hyperparameter of the meta-model is optimized, the accuracy is evaluated on the validation set, and the final output of the meta-model is the hemiplegia risk score.

[0038] According to any possible implementation of the aspect as described above, an implementation is further provided, and the first smart contract specifically comprises:

[0039] When a new cerebral trauma patient hemiplegia risk score is generated, an update score function is called to generate or update a patient risk score table to record the hemiplegia risk score data, the hemiplegia risk score data comprising a patient global unique identifier, a score value and an update time;

[0040] The latest hemiplegia risk score of the patient is compared with the risk threshold of the current period, if the threshold is exceeded, the corresponding alarm level is matched, that is, a low-priority alarm is only recorded in an audit log, a medium-priority alarm directly sends a message to a medical node and a patient at the same time, and a high-priority alarm is sent to the medical node for audit first, if the audit is confirmed, the patient is sent again, and the audit log is recorded;

[0041] A follow-up table is generated or updated by the real information uploaded by the medical node after the follow-up period is completed to record whether the patient has hemiplegia after the follow-up period is completed, and the risk threshold is updated when the update period is reached;

[0042] The update score, early warning triggering, audit result, follow-up table and risk threshold updating operations are recorded.

[0043] According to the aspect and any possible implementation manner described above, further provided is an implementation manner, wherein the risk threshold updating formula is as follows:

[0044] H t =H t-1 + θ · (MR t-1 - MR'),

[0045] wherein H t represents the threshold value of the current period, H t-1 represents the risk threshold value of the previous period, the initial risk threshold value is a set value, and θ represents an adjustment coefficient, MR t-1 represents the misjudgment rate statistically obtained in the previous period, and MR' represents the first target misjudgment rate.

[0046] The latest hemiplegia risk score H * t of the patient is compared with the risk threshold H t of the current period, and when H * t ≥ 0.8H t , it is determined as a high-priority alarm, when 0.6H t ≤ H * t < 0.8H t , it is determined as a medium-priority alarm, and when H * t < 0.6H t , it is determined as a low-priority alarm.

[0047] According to the aspect and any possible implementation manner described above, further provided is an implementation manner, wherein the second smart contract specifically comprises:

[0048] According to the follow-up table, if it is judged that the misjudgment rate exceeds the second target misjudgment rate, misjudgment training data set is generated, the hemiplegia prediction model is retrained, model updating record is generated after the updating is completed, and the time, reason, updating content and version information of each model updating are recorded;

[0049] The permission level of data query and model updating is set for the newly added or existing medical node, and then the "modify permission" function is called to change according to the change request;

[0050] The hemiplegia prediction model updating record and the permission change record are recorded.

[0051] In a second aspect, the embodiments of the present application provide a hemiplegia risk prediction system for brain trauma patients using the above method, and the system comprises:

[0052] The data acquisition and storage module is configured to acquire patient information and medical data of brain trauma patients uploaded by different medical nodes, and to pre-process and encrypt the medical data and store the medical data in a multi-center cloud platform in a distributed manner.

[0053] The risk assessment module is configured to calculate a comprehensive risk assessment score according to the medical data, and to screen high-risk patients with hemiplegia through the comprehensive risk assessment score.

[0054] The image feature processing module is configured to extract image features of the high-risk patients with hemiplegia, construct and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain a hemiplegia risk score of the brain trauma patient.

[0055] The first smart contract module is configured to automatically evaluate and record the obtained hemiplegia risk score, and trigger an early warning based on a preset risk threshold, and send the early warning information to the medical node and the patient.

[0056] The second smart contract module is configured to update the hemiplegia prediction model when the triggering condition is met and record the update information.

[0057] The above technical solution has the following beneficial effects:

[0058] The method and system for predicting the risk of hemiplegia in brain trauma patients are provided in the method, the distributed cloud platform and the hash ring structure are used to realize the encryption preprocessing, distributed storage and redundant backup of medical data, the risk score is calculated by comprehensively utilizing the key words of hemiplegia symptoms and dynamic physiological indexes, the early screening of high-risk patients with hemiplegia is realized, the precise risk prediction of image data is realized based on the prediction model, and the automatic recording of the risk score, the triggering of the early warning and the adaptive updating of the model are realized through the smart contract. The integrated technology realizes the precise prediction and automatic early warning of the risk of hemiplegia in brain trauma patients, and improves the timeliness of the intervention of the risk of hemiplegia in brain trauma patients. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0060] Figure 1 A flowchart of a method for predicting the risk of hemiplegia in brain trauma patients is provided.

[0061] Figure 2A schematic block diagram of a hemiplegia risk prediction system for brain trauma patients provided by an embodiment of the present application;

[0062] Figure 3 A hardware structure schematic diagram of a hemiplegia risk prediction system for brain trauma patients provided by an embodiment of the present application.

CONCRETE IMPLEMENTATION

[0063] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described below in detail with specific embodiments and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0064] Please refer to Figure 1 , which is a flowchart of a hemiplegia risk prediction method for brain trauma patients provided by an embodiment of the present application, as shown in Figure 1 , the method comprises the following steps:

[0065] S1, obtaining patient information and medical data of brain trauma patients uploaded by different medical nodes, and performing preprocessing and encryption processing, and then distributing and storing in a multi-center cloud platform;

[0066] S2, calculating a comprehensive risk assessment score according to the medical data, and screening high-risk patients of hemiplegia through the comprehensive risk assessment score;

[0067] S3, performing image feature extraction on the high-risk patients of hemiplegia, constructing and pre-training a hemiplegia prediction model, inputting the extracted image features into the hemiplegia prediction model, and obtaining a hemiplegia risk score of the brain trauma patients;

[0068] S4, automatically evaluating and recording the hemiplegia risk score through a first smart contract, triggering an early warning according to a preset risk threshold, and sending the early warning information to the medical nodes and the patients, and automatically triggering the hemiplegia prediction model update and recording through a second smart contract.

[0069] The embodiment of the application realizes distributed storage, processing and redundant backup of medical data through a distributed cloud platform, realizes early screening of high-risk patients with hemiplegia by using a risk assessment score, realizes accurate hemiplegia risk prediction of image data based on a prediction model, and realizes automatic recording of the risk score, early warning triggering and adaptive updating of the model through a smart contract. Through accurate management, screening, prediction and automatic early warning of the hemiplegia risk of brain trauma patients, the timeliness and pertinence of clinical intervention are improved. Through end-to-end automatic processing from patient data collection to risk prediction, early warning triggering and model updating, the influence of manual intervention and subjective judgment is reduced, and intelligent processing is realized.

[0070] In the preferred embodiment of the application, a hemiplegia risk prediction method suitable for brain trauma patients is provided, which comprises the following steps:

[0071] S1, patient information and medical data of brain trauma patients uploaded by different medical nodes are acquired, and after preprocessing and encryption processing, the medical data are distributed stored in a multi-center cloud platform;

[0072] S2, a comprehensive risk assessment score is calculated according to the medical data, and high-risk patients with hemiplegia are screened through the comprehensive risk assessment score;

[0073] S3, image feature extraction is performed on the high-risk patients with hemiplegia, a hemiplegia prediction model is constructed and pre-trained, the extracted image features are input into the hemiplegia prediction model, and a hemiplegia risk score of the brain trauma patient is obtained.

[0074] In the preferred embodiment of the application, the S1 comprises:

[0075] S11, medical image data are extracted from the acquired medical data of brain trauma patients uploaded by different medical nodes, adaptive filtering method is used to remove image noise, histogram equalization method is used to enhance the contrast of the image, image registration is performed on the multi-phase or multi-modal medical image data to complete alignment, and the brain region is segmented out by using an image segmentation algorithm to eliminate non-background regions;

[0076] S12, the medical data are standardized, including normalization and missing value completion;

[0077] S13, a hash value is generated by hash operation using the unique identifier of the cloud storage node, each cloud storage node is distributed in a hash ring according to the position of the hash value, a hash value is obtained by hash calculation according to each target patient information as a globally unique identifier of the patient, the nearest cloud storage node is found on the hash ring according to the globally unique identifier, and the medical data of the target patient are distributed to the cloud storage node, and a copy of the medical data of the target patient is copied clockwise to the subsequent N redundant nodes on the hash ring, and when the medical data are updated, the incremental information is pushed to the redundant nodes;

[0078] S14, if a new cloud storage node is added, the hash value of the new node is calculated and added to the hash ring, only the original data interval covered by the new node on the hash ring is migrated, a double-write strategy of simultaneous writing of the old node and the new node is adopted during migration, after migration is completed, the new node is included in the subsequent N redundant targets, if node failure is detected, the read and write requests are switched to the nearest redundant copy node, if the redundant copy data is complete, it is directly promoted to a new master node and new redundant copies are supplemented, if the redundant copy is not complete, data is recovered from other copies, after the node is recovered, the incremental data is synchronized from the redundant copy to the latest state.

[0079] The preprocessing technologies of adaptive filtering, histogram equalization, image registration and segmentation are used in the embodiment of the application, so that the noise is effectively reduced, the contrast is improved, and the key region of the brain is highlighted, thereby providing high-quality data for subsequent feature extraction and model training; the medical data is normalized and the missing values are completed, so that the consistency and integrity of the multi-source data are ensured, thereby improving the robustness of the subsequent risk assessment and prediction model; the hash operation is used to generate a globally unique identifier, and a hash ring storage structure is constructed, so that the distributed storage and redundant backup of the data on the multi-center cloud platform are realized, the data security is ensured, and the efficiency of data access and update is improved; the double-write strategy and independent synchronization channel are used to effectively support the data migration of the new node and the real-time update of the hot data, so that the system has good expansibility and high availability.

[0080] In the preferred embodiment of the application, the method further comprises:

[0081] S15, when the number of updates of the medical data of the patient within a preset time is greater than a preset update threshold, the patient is marked as a hot data patient, and the following operations are performed: a dedicated master node is allocated for the hot data patient, the incremental data is batch pushed to the redundant copy node on the hash ring through the asynchronous message queue, and the incremental data is directly written to the master node in real time; a global version number is added to the medical data of the hot data patient, and the version number is incremented each time the data is updated; if the redundant copy node receives conflicting version data, the old data is automatically discarded and an alarm is triggered; if the version numbers are the same but the contents are different, a consistency checking protocol is started, the latest data is selected based on the timestamp and the majority copy consensus; the read and write requests of the hot data are preferentially routed to the dedicated master node, so that the synchronization delay influence of the redundant copy is reduced.

[0082] When the patient data update frequency exceeds the preset threshold, the embodiment of the application marks the data as hotspot data, and the exclusive master node directly performs real-time write update, effectively reducing the redundant copy synchronization delay caused by frequent updates; the asynchronous message queue batch pushes the incremental data to the redundant node, not only smoothing the data transmission load, but also ensuring the fast and consistent update of data in the distributed storage system; by adding a global version number to the hotspot data, the version is incremented each time the data is updated, providing a clear identification for the data state; when the redundant copy node receives the conflict version data, the old data is automatically discarded and an alarm is triggered; for the case where the version numbers are the same but the contents are different, the latest data is selected based on the timestamp and the majority copy consensus through the consistency checking protocol, thereby effectively solving the data conflict problem and ensuring the data consistency and integrity; the read and write requests of the hotspot data are preferentially routed to the exclusive master node, which maximally reduces the response delay caused by multi-node synchronization and improves the processing efficiency and overall performance of the system in the high-frequency data update scenario.

[0083] In the preferred embodiment of the application, S2 specifically comprises:

[0084] S21, constructing a hemiplegia symptom keyword library, extracting hemiplegia symptom keywords from the medical data of the brain trauma patient according to the hemiplegia symptom keyword library and counting the occurrence frequency to generate a keyword list;

[0085] S22, selecting a physiological index related to hemiplegia and changing over time as a dynamic change feature, and calculating a dynamic change feature value from the medical data of the brain trauma patient;

[0086] S23, standardizing the keyword list and the dynamic change feature value, and calculating a comprehensive risk assessment score according to the keyword list and the dynamic change feature value, wherein the calculation formula of the comprehensive risk assessment score is as follows:

[0087]

[0088] Wherein C is the comprehensive risk assessment score, γ1 and γ2 are weight coefficients, F is the hemiplegia symptom keyword frequency, k i is the frequency of the i-th hemiplegia symptom, w i is the weight of the i-th symptom, n is the number of selected hemiplegia symptoms, Δf j (t) is the dynamic change feature value of the patient's physiological data, λ j is the weight of the j-th dynamic change feature, and m is the number of selected dynamic change features.

[0089] S24, draw the ROC curve based on the historical patient data, select the T value when the Youden index is maximum as the initial risk threshold, take T / K as the final risk threshold, K is a set constant, compare the size of the comprehensive risk assessment score C and the final risk threshold T / K, when C≥T / K, it is screened as a high risk patient with hemiplegia.

[0090] The embodiment of the application realizes quantitative evaluation of patient risk from static and dynamic aspects by combining text mining and physiological dynamic feature calculation, improves the comprehensiveness of risk judgment; in addition, the comprehensive risk assessment score is calculated through standardization processing and mathematical formula calculation based on a weight coefficient, manual subjective intervention is reduced, and the risk assessment is more scientific and objective; the ROC curve is drawn by using historical data, the T value corresponding to the maximum Youden index is determined, the initial risk threshold is set, and the actual data is dynamically adjusted to provide accurate basis for screening of high risk patients.

[0091] The keyword library of hemiplegia symptoms can be selected as follows:

[0092]

[0093] The physiological indexes related to hemiplegia and changing over time can include, but are not limited to, any one or more of intracranial pressure fluctuation index, Glasgow coma score decline rate, limb muscle strength change rate, systolic blood pressure variability, electroencephalogram bispectral index trend and limb movement frequency. It should be noted that the selection of physiological indexes can also be adjusted according to actual conditions.

[0094] Specifically, the dynamic change characteristic value is calculated as follows. The calculation formula of the intracranial pressure fluctuation index is σ I , which represents the standard deviation of the intracranial pressure measurement value in the preset time window, μ I , which represents the arithmetic mean of the intracranial pressure in the preset time window, V I The larger the value is, the worse the intracranial pressure stability is; the calculation formula of the Glasgow coma score decline rate is t i , which represents the time point of the i-th score, G i , which represents the Glasgow coma score at the corresponding time point, , which represents the average value of all time points, , which represents the average value of the Glasgow coma score at all time points, when S G <0 indicates that the consciousness state is deteriorating; the calculation formula of the limb muscle strength change rate is M t , which represents the t-th muscle strength score, N represents the total number of assessments, and when ΔM is negative, it indicates that the muscle strength continues to decrease; the calculation formula of the systolic blood pressure variability is R S =max(S t) - min(S t ), S t represents the systolic blood pressure measurement value at each time point in the preset sliding window, R S The larger the value of R t represents that the blood pressure fluctuation is severe; the trend calculation formula of the electroencephalogram bispectrum index is E t = α·B t-1 +(1-α)·E t , E m represents the t-th electroencephalogram bispectrum index measurement value, α represents a smoothing factor, E0=B0, and when ΔM continues to decrease, it indicates that the brain function inhibition is aggravated; the calculation formula of the limb movement frequency is Q represents the number of peaks of limb activity amplitude exceeding a threshold value detected by an accelerometer, ΔT represents a time window, F m decreases, which indicates that the motor function is significantly impaired.

[0095] The embodiment of the present application specifies the intracranial pressure fluctuation index, the Glasgow coma score decrease rate, the limb muscle strength change rate, the systolic blood pressure variability, the electroencephalogram bispectrum index trend and the limb movement frequency, and gives the corresponding calculation formula. Through multiple dynamic indexes, the changes of the physiological state of the patient in the preset time window can be comprehensively reflected, the complex pathological changes caused by brain trauma can be effectively captured, and a quantitative basis for risk scoring can be provided. Through standardization and mathematical modeling of each dynamic index, the reasonable allocation of the influence weight of different indexes is realized, and it is ensured that the risk assessment can take into account each physiological data.

[0096] In the preferred embodiment of the present application, the S3 specifically comprises:

[0097] S31, acquiring historical medical images of a hemiplegia high-risk patient, including CT, MRI and CTP images, and performing hemiplegia label marking according to the follow-up results;

[0098] S32, registering different images of the same patient, unifying different image sizes / resolutions to a standard, performing intensity normalization, and dividing the data set into a training set and a validation set;

[0099] S33, constructing and pre-training a hemiplegia prediction model through the training set, wherein the hemiplegia prediction model adopts a Stacking structure of three sub-models and a meta-model, the three sub-models process three types of image features of edema, hemorrhage and abnormal cerebral blood flow respectively, each sub-model is trained on an independent data set, and then outputs corresponding features for the meta-model to make final fusion in the inference stage, and the meta-model is trained;

[0100] S34, inputting medical images of a new hemiplegia high-risk patient, respectively sending them into the three sub-models for inference, and then performing fusion through the meta-model to obtain a final hemiplegia risk score.

[0101] The embodiment of the application constructs a set of representative and trainable image data set through the acquisition, labeling and standardization processing of patient historical medical images, processes different image features by using three sub-models of edema, hemorrhage and abnormal cerebral blood flow, fuses the results of each sub-model through a meta-model, realizes multi-angle risk assessment, and significantly improves the overall prediction accuracy. After inputting new patient image data, the risk score of hemiplegia is quickly obtained, which supports early intervention and personalized treatment plan for high-risk patients in the clinic.

[0102] In the preferred embodiment of the application, the hemiplegia prediction model is constructed and trained by the training set in advance, specifically comprising:

[0103] The edema detection sub-model is constructed: the network structure uses 3D U-Net for edema region segmentation, the encoder is 5 layers of 3D convolution + ReLU + Downsample, the decoder is 5 layers of deconvolution + skip-connection, the input is MRI or CT low-density area, and the output is the segmentation mask of the edema area. The edema probability and edema volume are calculated from the segmentation result, the loss function selects DiceLoss and BCE / CE for joint use, Dropout is used to reduce overfitting, adaptive learning rate is used, the labeled edema region image is used as the training set, random rotation, translation, cropping, brightness change method is used to enhance the data, and the edema region segmentation accuracy is verified through the validation set after 100 epochs of training.

[0104] The hemorrhage detection sub-model is constructed: the network structure uses DenseNet 3D, the input is CT plain scan, the output is the probability of hemorrhage and the volume of hemorrhagic plaque, the loss function selects BCE / CE, the CT image with hemorrhagic plaque annotation is used as the training set, random rotation, translation, cropping, brightness change method is used to enhance the data, and the hemorrhage recognition accuracy is verified through the validation set after training.

[0105] The cerebral blood flow detection sub-model is constructed: the network structure uses 3D-CNN, the input is CTP or MRI perfusion sequence, the output is the probability of perfusion abnormality, the loss function adopts BCE / CE loss function, the low perfusion region annotation image is used as the training set, and time translation and inter-frame noise are used for data enhancement.

[0106] The meta-model is constructed and trained: the random forest is selected as the meta-model, the numerical features are obtained based on the three sub-models, the image features of edema probability, hemorrhage probability and perfusion abnormality probability are integrated into a feature vector as the input of the meta-model, the sub-model output is extracted from the training set image of each patient, and the "hemiplegia or not" label is paired for training the meta-model. The meta-model is optimized for hyperparameters, the accuracy is evaluated on the validation set, and the final output of the meta-model is the risk score of hemiplegia.

[0107] The embodiments of the present application respectively adopt 3D U-Net, DenseNet 3D and 3D-CNN to perform fine-grained detection on different pathological features, ensure that each sub-model can accurately quantify the image features concerned by itself, that is, edema probability, hemorrhage probability and perfusion abnormalities, and the application of data enhancement, joint loss function and adaptive learning rate and other strategies effectively reduces the risk of overfitting and improves the adaptability of the model to different patients and different image data; the random forest meta-model is used to fuse the features output by each sub-model to realize comprehensive judgment on multi-dimensional risk information, thereby improving the reliability and accuracy of the final hemiplegia risk score.

[0108] In the preferred embodiments of the present application, the first smart contract specifically comprises:

[0109] When a new brain trauma patient hemiplegia risk score is generated, an update score function is called to generate or update a patient risk score table to record the hemiplegia risk score data, which includes a patient globally unique identifier, a score value and an update time;

[0110] The latest hemiplegia risk score of the patient is compared with the risk threshold of the current period, and if it exceeds the threshold, it is matched to the corresponding alarm level, that is, a low-priority alarm is only recorded in the audit log, a medium-priority alarm directly sends a message to the medical node and the patient at the same time, and a high-priority alarm is first sent to the medical node for audit, and if confirmed after the audit, it is then sent to the patient and recorded in the audit log;

[0111] A follow-up table is generated or updated by the real information uploaded by the medical node after the follow-up period is completed to record whether the patient has hemiplegia after the follow-up period is completed, and the risk threshold is updated when the update period is reached;

[0112] The update score, early warning triggering, audit result, follow-up table and risk threshold update operation are recorded.

[0113] The embodiments of the present application automatically record the patient risk score, update operation log and early warning triggering process by using the smart contract, realize real-time monitoring and automatic response to the risk state, and reduce manual intervention; by comparing the risk score with the preset risk threshold, different early warning levels are distinguished, and corresponding response measures are taken under different risk conditions to improve the pertinence of patient intervention; all operations are recorded by the smart contract to provide reliable basis for subsequent data auditing, model adjustment and quality control.

[0114] In addition, the risk threshold update formula is as follows:

[0115] H t = H t-1 + theta * (MR t-1 - MR'), wherein H represents the hemiplegia risk score, MR represents the risk score of the patient in the current period, MR' represents the risk score of the patient in the last period, and theta represents the learning rate.

[0116] Among them, H t Indicates the threshold of the current cycle, H t-1 represents the risk threshold of the previous cycle. The initial risk threshold is the set value. θ represents the adjustment coefficient. MR t-1 represents the misjudgment rate counted in the previous cycle, and MR′ represents the first target misjudgment rate;

[0117] Comparison of patients' latest hemiplegia risk score H * t The risk threshold H of the current cycle t , when H * t ≥0.8h t When 0.6H t ≤H * t <0.8H t When H * t <0.6H t , it is considered a low priority alarm.

[0118] The risk threshold update formula of the embodiment of the present invention automatically adjusts the risk threshold based on historical data and actual misjudgment rate, so that the system can dynamically adapt to changes in clinical data and maintain the sensitivity and specificity of risk assessment; and the risk threshold update mechanism prompts the entire system to be continuously optimized in long-term operation, reducing the misjudgment rate and improving the accuracy of early warning.

[0119] In a preferred embodiment of the present invention, the second smart contract specifically includes:

[0120] Compare the previous prediction results with the follow-up table. If the misjudgment rate exceeds the second target misjudgment rate, generate a misjudgment training dataset and retrain the hemiplegia prediction model. After the update is completed, generate a model update record and record the time, reason, update content and version information of each model update.

[0121] To set the permission level for data query and model update for new or existing medical nodes, call the "Modify Permissions" function to make changes based on the change request;

[0122] Record the update records and permission change records of the hemiplegia prediction model.

[0123] The second smart contract of the embodiment of the present application realizes dynamic updating and self-adaptive improvement of the model by automatically monitoring the prediction misjudgment rate and triggering the model retraining mechanism, so as to ensure that the prediction system always meets the clinical actual demand; for the newly added or existing medical nodes, the intelligent contract automatically changes the permission and records, ensures that the data query and model updating operation are carried out under strict permission control; the time, reason, content and version information of each model updating are recorded, ensuring that the whole improvement process is open and transparent, which is helpful for subsequent technical optimization and problem tracking.

[0124] Specifically, the misjudgment training data set is generated, and the hemiplegia prediction model is retrained, which can be realized by the following steps: collecting the true follow-up results of patients from the follow-up data uploaded by each medical node, synchronously obtaining the historical prediction risk scores of the system for these patients, comparing the prediction results with the actual follow-up results, and identifying the cases of prediction errors; according to the preset misjudgment rate threshold, it is judged whether the current system misjudgment rate exceeds the second target misjudgment rate, for the misjudgment patients, mark them as "misjudgment cases", and record their corresponding original data, image data, dynamic physiological data and prediction score and other key information; all the data of the misjudgment cases are summarized to generate a special misjudgment training data set, which ensures that the data set contains patient global unique identifier, preprocessed medical image features, dynamic physiological indicators, keyword statistical data and original prediction score and other information, and the data is preprocessed to ensure data consistency and integrity; the image data in the misjudgment training data set is subjected to image registration, segmentation and standardization processing, and the key features such as edema, hemorrhage and abnormal cerebral blood flow are extracted, for text data, keyword extraction and frequency statistics are performed according to the keyword library of hemiplegia symptoms, and dynamic physiological indicators are standardized, data augmentation is used to expand data samples and reduce overfitting risk, data augmentation such as rotation, translation, cropping, brightness change and the like; adjust the parameters and weights of the comprehensive risk assessment score, so that the calculation score of the real hemiplegia patients is higher; combine the generated misjudgment training data set with the original training data as new training samples, input, fine-tune the training of each sub-model, and use enhanced data to further improve the robustness of the model, output the updated sub-model, use random forest or other ensemble learning methods to train the meta-model, form a new Stacking structure, and generate a comprehensive hemiplegia risk score; use an independent validation set to verify the updated prediction model, evaluate its performance on ROC curve, accuracy, sensitivity, specificity and other indicators, when the verification result meets the preset requirements, the updated model is deployed to the system to replace the original hemiplegia prediction model; the second intelligent contract automatically generates model update records, records the time, reason, update content and version information of the update, the system continues to monitor the subsequent prediction results and follow-up data in real time, calculates the misjudgment rate regularly, if the misjudgment rate exceeds the target threshold again, a new misjudgment training data generation and model retraining process is automatically triggered, forming a closed loop feedback mechanism, continuously improving the prediction performance of the system.

[0125] The first smart contract module and the second smart contract module are implemented based on a multi-center cloud platform, and specifically can include: an event listening layer that captures medical data update events in real time through a message queue; a contract execution layer that encapsulates early warning logic and model update logic through microservices, and provides external calls through a REST API; an audit storage layer that records all contract operations using a WORM mode log system, with a timestamp and digital signature attached; and an authority control layer that limits data access and operation authority of nodes based on an OAuth2 protocol and an RBAC model. It should be noted that the smart contract can also be executed using a blockchain underlying architecture, for example, using a consortium chain, with only medical institutions authorized to join as nodes, and then executing the first smart contract and the second smart contract.

[0126] The embodiment of the application further provides a device for implementing each step and method in the method embodiment.

[0127] Please refer to Figure 2 , which is Figure 2 A schematic block diagram of a hemiplegia risk prediction system for brain trauma patients is provided in the embodiment of the application. The system includes:

[0128] A data acquisition and storage module 210 is configured to acquire patient information and medical data of brain trauma patients uploaded by different medical nodes, and to perform preprocessing and encryption processing and then distribute the data for storage in a multi-center cloud platform;

[0129] A risk assessment module 220 is configured to calculate a comprehensive risk assessment score according to the medical data, and to screen high-risk patients for hemiplegia through the comprehensive risk assessment score;

[0130] An image feature processing module 230 is configured to perform image feature extraction on the high-risk patients for hemiplegia, to construct and pre-train a hemiplegia prediction model, to input the extracted image features into the hemiplegia prediction model, and to obtain a hemiplegia risk score of the brain trauma patient;

[0131] A first smart contract module 240 is configured to automatically evaluate and record the obtained hemiplegia risk score, and to trigger an early warning based on a preset risk threshold, and to send early warning information to medical nodes and patients;

[0132] A second smart contract module 250 is configured to update the hemiplegia prediction model when a triggering condition is met and to record the update information.

[0133] The application provides a brain trauma patient hemiplegia risk prediction system that integrates data acquisition, risk assessment, medical image processing, smart contract early warning, and model updating.

[0134] Since each unit module in the embodiment can perform Figure 1The method shown, the part of the embodiment not described in detail, can refer to the related description of Figure 1

[0135] Please refer to Figure 3 , which is a hardware structure diagram of a hemiplegia risk prediction system for brain trauma patients provided by an embodiment of the application. The data prediction device includes at least one processor and a memory, the at least one processor is coupled with the memory, and is used to read and execute instructions in the memory to execute the data prediction method provided by the embodiment of the application.

[0136] In a third aspect, an embodiment of the application provides a computer readable medium. The computer readable medium stores program code, when the computer program code runs on the computer, so that the computer executes the hemiplegia risk prediction method for brain trauma patients provided by the embodiment of the application.

[0137] At the hardware level, the device can include a processor, and optionally further include an internal bus, a network interface, a memory. The memory can include a memory, such as a random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory. Of course, the device can also include other hardware required by the business.

[0138] The processor, network interface and memory can be connected to each other through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc.

[0139] The memory is used to store programs. Specifically, the program can include program code, and the program code includes computer operation instructions. The memory can include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0140] ​The steps of the method disclosed by the embodiments of the present application can be directly embodied as hardware code processing executed by a hardware code processor, or executed by a combination of hardware and software modules in the code processor. The software modules can be located in storage media in the art such as random access memories (RAMs), flash memories, read only memories (ROMs), programmable read only memories (PROMs), electrically programmable read only memories (EPROMs), electrically erasable and programmable read only memories (EEPROMs), registers, or the like. The storage media is located in the storage, and the processor reads information in the storage media to execute the steps of the above-mentioned method in combination with hardware.

[0141] The systems, apparatuses, modules or units disclosed in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0142] For the convenience of description, the above apparatus is described as various units or modules in function for separate description. Of course, functions of each unit or module can be implemented in one or more software and / or hardware in the implementation of the present application.

[0143] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer-usable program code.

[0144] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in a flow or multiple flows and / or blocks.

[0145] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0147] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0148] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) about which the computer stores information about an operating system, application software, and / or the like. Memory is an example of computer readable media.

[0149] Computer readable media includes permanent and non-permanent, moveable and non- moveable media that can be implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that is accessible to a computing device. According to the definition provided herein, computer readable media excludes transitory media, such as modulated data signals and carrier waves.

[0150] It is also to be noted that the terms "comprising", "including", and "having" or variations thereof herein, are intended to be open-ended terms that specify the presence of the stated elements but do not preclude the presence of additional elements. It is also to be noted that the term "consisting of" is intended to be a closed term that specifies the presence of only the stated elements. The term "comprising" is used herein to include the presence of one or more elements that are not specifically named.

[0151] It will be appreciated by those skilled in the art that embodiments of the present application can be provided as a method, system or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-usable program code.

[0152] The present application can be described in the general context of computer- executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices.

[0153] Embodiments of the present application are described with reference to the attached figures, wherein like numerals indicate corresponding elements in the figures. The embodiments of the present application are described more fully with reference to the attached figures.

[0154] The specific embodiments of the present application have been described herein for the purpose of illustration and description, and are not intended to be limiting. Any modifications and variations that come within the scope of the inventive concepts disclosed herein are intended to be included within the scope of the claims. Accordingly, it is expressly intended that all such modifications and variations are to be included within the scope of the present application as provided in the following claims.

Claims

1. A method for predicting the risk of hemiplegia in patients with brain trauma, characterized in that: The method comprises: S1. Obtain patient information and medical data of brain trauma patients uploaded by different medical nodes, pre-process and encrypt them, and then store them in a distributed manner on a multi-center cloud platform; S2. Calculate a comprehensive risk assessment score based on medical data and use the comprehensive risk assessment score to screen patients at high risk of hemiplegia; S3. Extract image features from patients at high risk of hemiplegia, build and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain a hemiplegia risk score for patients with brain trauma; S4. Automatically evaluate and record the hemiplegia risk score through the first smart contract, trigger an early warning based on the preset risk threshold, and send the early warning information to the medical node and the patient. Automatically trigger the update and recording of the hemiplegia prediction model through the second smart contract; The S2 specifically includes: S21. Build a hemiplegia symptom keyword database, extract hemiplegia symptom keywords from medical data of brain trauma patients based on the hemiplegia symptom keyword database, count the occurrence frequencies, and generate a keyword list; S22. Select physiological indicators related to hemiplegia that change over time as dynamic change features, and calculate dynamic change feature values ​​from the medical data of patients with brain trauma; S23. Standardize the keyword list and the dynamic change characteristic value, and calculate a comprehensive risk assessment score based on the keyword list and the dynamic change characteristic value. The calculation formula of the comprehensive risk assessment score is as follows: Among them, C is the comprehensive risk assessment score, γ1 and γ2 are weight coefficients, F is the frequency of hemiplegia symptom keywords, k i is the frequency of occurrence of the ith symptom of hemiplegia, w i is the weight of the ith symptom, n is the number of selected hemiplegic symptoms, Δf j (t) is the dynamic change characteristic value of the patient's physiological data, λ j is the weight of the jth dynamic change feature, and m is the number of selected dynamic change features; S24. Draw the ROC curve based on historical patient data, select the T value when the Youden index is maximum as the initial risk threshold, use T / K as the final risk threshold, and use K as the set constant. Compare the comprehensive risk assessment score C and the final risk threshold T / K. When C ≥ T / K, screen the patient as a high-risk patient for hemiplegia.

2. The method for predicting hemiplegia risk for patients with brain trauma according to claim 1, characterized in that: Said S1 comprises: S11. Extract medical imaging data from the medical data of brain trauma patients uploaded by different medical nodes, remove image noise using adaptive filtering methods, enhance image contrast using histogram equalization methods, perform image registration on multi-temporal or multi-modal medical imaging data to complete alignment, and use image segmentation algorithms to segment brain regions and eliminate non-background areas; S12. Standardize medical data, including normalization and missing value filling; S13. A hash value is generated by performing a hash operation using the unique identifier of the cloud storage node. Each cloud storage node forms a hash ring according to the position distribution of the hash value in the hash ring. A hash value is obtained by performing a hash operation based on the information of each target patient as the patient's globally unique identifier. The nearest cloud storage node is found on the hash ring based on the globally unique identifier, and the medical data of the target patient is assigned to the cloud storage node. A copy of the target patient's medical data is copied clockwise on the hash ring to the subsequent N redundant nodes. When the medical data is updated, the incremental information is pushed to the redundant node. S14. If a new cloud storage node is added, the hash value of the new node is calculated and added to the hash ring. Only the original data interval covered by the new node on the hash ring is migrated. During the migration, a dual-write strategy is adopted in which the old node and the new node are written at the same time. After the migration is completed, the new node is included in the subsequent N redundant targets. If a node failure is detected, the read and write requests are switched to the nearest redundant copy node. If the redundant copy data is complete, it is directly promoted to the new primary node and a new redundant copy is added. If the redundant copy is incomplete, the data is restored from other copies. After the node is restored, the incremental data is synchronized to the latest state from the redundant copy.

3. The method for predicting hemiplegia risk for patients with brain trauma according to claim 1, characterized in that: The dynamic change characteristics include but are not limited to any one or more of intracranial pressure fluctuation index, Glasgow Coma Scale decline rate, limb muscle strength change rate, systolic blood pressure variability, bispectral index trend and limb movement frequency, wherein, The calculation formula of the intracranial pressure fluctuation index is: σ I Represents the standard deviation of intracranial pressure measurements within the preset time window, μ I It represents the arithmetic mean of intracranial pressure within the preset time window; The Glasgow Coma Scale decline rate calculation formula is: t i represents the time point of the i-th scoring, G i represents the Glasgow Coma Scale score at the corresponding time point, represents the average value of all time points, represents the mean of the Glasgow Coma Scale scores at all time points; The formula for calculating the rate of change of limb muscle strength is: M t represents the t-th muscle strength score, N represents the total number of assessments, and when △M is negative, it means that the muscle strength continues to decline; the formula for calculating systolic blood pressure variability is R S =max(S t )-min(S t ), S t Represents the systolic blood pressure measurement value at each time point within the preset sliding window; The bispectral index trend calculation formula is E t =α·B t +(1-α)·E t-1 , E t represents the bispectral index measurement value at time t, α represents the smoothing factor, and E0 = B0; The limb movement frequency calculation formula is: Q represents the number of peaks of limb movement detected by the accelerometer that exceeds the threshold, and △T represents the time window.

4. The method for predicting hemiplegia risk for patients with brain trauma according to any one of claims 1 or 3, characterized in that: The S3 specifically includes: S31. Obtain historical medical images of patients at high risk of hemiplegia, including CT, MRI, and CTP images, and label hemiplegia based on follow-up results; S32. Register different images of the same patient, unify the different image sizes / resolutions to a standard, perform intensity normalization, and divide the dataset into training and validation sets. S33. Construct and pre-train a hemiplegia prediction model using a training set, wherein the hemiplegia prediction model adopts a stacking structure of three sub-models and a meta-model. The three sub-models respectively process three types of imaging features: edema, hemorrhage, and abnormal cerebral blood flow. Each sub-model is trained on an independent data set, and then outputs corresponding features in the inference stage for final fusion with the meta-model to train the meta-model. S34. Input the medical images of new high-risk patients for hemiplegia, send them to the three sub-models for reasoning, and then fuse them through the meta-model to obtain the final hemiplegia risk score.

5. The method for predicting hemiplegia risk for patients with brain trauma according to claim 4, characterized in that: The construction and pre-training of the hemiplegia prediction model using a training set specifically includes: Constructing an edema detection sub-model: The network structure uses 3D U-Net for edema region segmentation. The encoder is a 5-layer 3D convolution + ReLU + Downsample, and the decoder is a 5-layer deconvolution + skip-connection. The input is the CT low-density area, and the output is the segmentation mask of the edema region. The edema probability and edema volume are inferred from the segmentation results. The loss function uses Dice Loss and BCE / CE in combination. Dropout is used to reduce overfitting. An adaptive learning rate is used. Labeled edema region images are used as the training set. Random rotation, translation, cropping, and brightness change methods are used to enhance the data. Training is completed for 100 epochs. After training, the accuracy of edema region segmentation is verified using the validation set. Construct a hemorrhage detection sub-model: The network structure uses DenseNet 3D, with CT scans as input and bleeding probability and hemorrhage plaque volume as output. The loss function uses BCE / CE. CT images with annotated hemorrhage plaques are used as the training set. Random rotation, translation, cropping, and brightness change methods are used to enhance the data. After training, the accuracy of hemorrhage recognition is verified using a validation set. Construct a cerebral blood flow detection sub-model: The network structure uses 3D-CNN, the input is MRI perfusion sequence, the output is the probability of perfusion abnormality, the loss function adopts the BCE / CE loss function, the low perfusion area annotated images are used as the training set, and time shift and inter-frame noise are used for data enhancement; Meta-model construction and training: A random forest was selected as the meta-model. Numerical features were derived from three sub-models. The imaging features of edema probability, hemorrhage probability, and perfusion abnormality probability were integrated into a feature vector as the meta-model input. Sub-model outputs were extracted from the training set images for each patient and paired with the "hemiplegia" label to train the meta-model. Hyperparameters of the meta-model were optimized, and accuracy was evaluated on the validation set. The final output of the meta-model was the hemiplegia risk score.

6. The method for predicting hemiplegia risk for patients with brain trauma according to claim 1, characterized in that: The first smart contract specifically includes: When a new hemiplegia risk score for a brain injury patient is generated, the update score function is called to generate or update the patient risk score table to register the hemiplegia risk score data, wherein the hemiplegia risk score data includes the patient's globally unique identifier, the score value, and the update time; Compare the patient's latest hemiplegia risk score with the risk threshold of the current cycle. If the threshold is exceeded, the corresponding alarm level is matched. That is, low-priority alarms are only recorded in the audit log, medium-priority alarms are directly sent to the medical node and the patient at the same time, and high-priority alarms are first sent to the medical node for review. If confirmed after review, they are sent to the patient and recorded in the audit log; Generate or update a follow-up table based on the real information uploaded by the medical node after the follow-up period to record the actual results of whether the patient develops hemiplegia after the follow-up period expires, and update the risk threshold when the update cycle is reached; Record updated scores, warning triggers, review results, follow-up tables, and risk threshold update operations.

7. The method for predicting hemiplegia risk for patients with brain trauma according to claim 6, characterized in that: The risk threshold update formula is as follows: A t =H t-1 +θ·(MR t-1 -MR′), Among them, H t Indicates the threshold of the current cycle, H t-1 represents the risk threshold of the previous cycle. The initial risk threshold is the set value. θ represents the adjustment coefficient. MR t-1 represents the misjudgment rate counted in the previous cycle, and MR′ represents the first target misjudgment rate; Comparison of patients' latest hemiplegia risk score H * t The risk threshold H of the current cycle t , when H * t ≥0.8h t When 0.6H t ≤H * t <0.8H t When H * t <0.6H t , it is considered a low priority alarm.

8. The method for predicting hemiplegia risk for patients with brain trauma according to claim 7, characterized in that: The second smart contract specifically includes: Compare the previous prediction results with the follow-up table. If the misjudgment rate exceeds the second target misjudgment rate, generate a misjudgment training dataset and retrain the hemiplegia prediction model. After the update is completed, generate a model update record and record the time, reason, update content and version information of each model update. To set the permission level for data query and model update for new or existing medical nodes, call the "Modify Permissions" function to make changes based on the change request; Record the update records and permission change records of the hemiplegia prediction model.

9. A hemiplegia risk prediction system for patients with brain trauma using the method of claim 1, characterized in that: The system comprises: The data acquisition and storage module is configured to obtain patient information and medical data of brain trauma patients uploaded by different medical nodes, perform pre-processing and encryption, and then store them in a distributed manner in the multi-center cloud platform; a risk assessment module configured to calculate a comprehensive risk assessment score based on the medical data and screen patients with a high risk of hemiplegia using the comprehensive risk assessment score; The image feature processing module extracts image features from patients at high risk of hemiplegia, builds and pre-trains a hemiplegia prediction model, and inputs the extracted image features into the hemiplegia prediction model to obtain a hemiplegia risk score for patients with brain trauma; The first smart contract module is configured to automatically evaluate and record the obtained hemiplegia risk score, trigger an early warning based on a preset risk threshold, and send the early warning information to the medical node and the patient; The second smart contract module is configured to update the hemiplegia prediction model and record the update information when the trigger condition is met.

Images