Hemiplegia risk prediction method and system suitable for brain trauma patient

Abstract

The embodiment of the invention provides a hemiplegia risk prediction method and system suitable for a brain trauma patient. The method comprises the following steps: acquiring patient information and medical data of the brain trauma patient uploaded by different medical nodes; screening high-risk patients with hemiplegia through the comprehensive risk assessment scores; constructing and pre-training a hemiplegia prediction model to obtain a hemiplegia risk score of the brain trauma patient; the hemiplegia risk score is automatically evaluated and recorded through the first intelligent contract, early warning is triggered according to a preset risk threshold value, early warning information is sent to the medical node and the patient, and updating and recording of the hemiplegia prediction model are automatically triggered through the second intelligent contract. According to the method, the risk score is calculated by comprehensively utilizing the hemiplegia symptom keyword and the dynamic physiological index, early screening of the hemiplegia high-risk patient is achieved, accurate risk prediction and automatic early warning of the image data are achieved based on the prediction model, and the timeliness of hemiplegia risk intervention of the brain trauma patient is improved.

Description

Technical Field

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

Background Art

[0002] Traumatic brain injury is one of the important causes of neurological dysfunction. Hemiplegia is a relatively common and serious manifestation clinically among its sequelae. There is a complex pathological connection between traumatic brain injury and hemiplegia, including both the loss of nerve function caused by direct injury and the secondary injury caused by secondary pathological processes.

[0003] Currently, in the field of hemiplegia risk assessment after traumatic brain injury, traditional methods mainly rely on doctors' manual comprehensive judgment of patients' symptoms, imaging data, and physiological indicators. There are not only problems such as strong subjectivity and low efficiency, but also it is difficult to respond to changes in the condition in a timely manner, resulting in some high-risk patients failing to receive early intervention. With the development of medical imaging technology, deep learning, and data analysis, using automated means for risk assessment and prediction has become a research hotspot.

Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a hemiplegia risk prediction method and system suitable for patients with traumatic brain injury.

[0005] In a first aspect, embodiments of the present invention provide a hemiplegia risk prediction method suitable for patients with traumatic brain injury, the method comprising:

[0006] S1. Obtain the patient information and medical data of patients with traumatic brain injury uploaded by different medical nodes, perform preprocessing and encryption processing, and then store them distributively in a multi-center cloud platform;

[0007] S2. Calculate a comprehensive risk assessment score based on the medical data, and screen out high-risk hemiplegia patients through the comprehensive risk assessment score;

[0008] S3. Extract image features for high-risk hemiplegia patients, construct and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain the hemiplegia risk score of patients with traumatic brain injury;

[0009] S4. Automatically evaluate and record the hemiplegia risk score through a first smart contract, trigger an alarm according to a preset risk threshold and send the alarm information to medical nodes and patients, and automatically trigger the update and record of the hemiplegia prediction model through a second smart contract.

[0010] As described above in the aspect and any possible implementation manner, a further implementation manner is provided, where the S1 includes:

[0011] S11. Extract medical image data from the medical data of patients with traumatic brain injury uploaded from different medical nodes, remove image noise using an adaptive filtering method, enhance the contrast of the images using a histogram equalization method, perform image registration on multi-temporal or multi-modal medical image data to complete alignment, use an image segmentation algorithm to segment the brain region, and eliminate non-background regions;

[0012] S12. Standardize the medical data, including normalization and missing value imputation;

[0013] S13. Generate a hash value after hashing 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. Calculate the hash value based on each target patient's information as the globally unique identifier of the patient. Find the nearest cloud storage node on the hash ring according to the globally unique identifier, and allocate the medical data of the target patient to the cloud storage node. And the medical data copy of the target patient is copied clockwise to the subsequent N redundant nodes on the hash ring. When the medical data is updated, push incremental information 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 on the hash ring. When migrating, adopt a double-write strategy of writing by the old node and the new node at the same time. After migration, include the new node in the subsequent N redundant targets. If a node failure is detected, switch the read and write requests to the nearest redundant replica node. If the redundant replica data is complete, directly promote it to the new master node and supplement new redundant replicas. If the redundant replica is incomplete, recover the data from other replicas. After the node recovers, synchronize the incremental data from the redundant replica to the latest state.

[0015] For the aspects and any possible implementation manners described above, a further implementation manner is provided. The S2 specifically includes:

[0016] S21. Construct a keyword library for hemiplegia symptoms. Extract hemiplegia symptom keywords from the medical data of patients with traumatic brain injury according to the keyword library for hemiplegia symptoms and count the occurrence frequencies to generate a keyword list;

[0017] S22. Select physiological indicators related to hemiplegia and changing over time as dynamic change features, and calculate dynamic change feature values from the medical data of patients with traumatic brain injury;

[0018] S23. Standardize the keyword list and the dynamic change feature values, and calculate a comprehensive risk assessment score according to the keyword list and the dynamic change feature values. Among them, the calculation formula of the comprehensive risk assessment score is as follows:

[0019]

[0020] 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 the i-th symptom of hemiplegia, w i is the weight of the i-th symptom, n is the number of selected hemiplegia symptoms, Δf j (t) is the eigenvalue of the dynamic change 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;

[0021] S24. Draw an ROC curve based on historical patient data, select the T value when the Youden index is the largest as the initial risk threshold, use T / K as the final risk threshold, where 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, screen it as a high-risk hemiplegia patient.

[0022] For the aspects and any possible implementation manners as described above, a further implementation manner is provided. The dynamic change features include but are not limited to any one or more of the intracranial pressure fluctuation index, the Glasgow Coma Scale decline rate, the limb muscle strength change rate, the systolic blood pressure variability, the bispectral index trend, and the limb movement frequency. Among them,

[0023] The calculation formula for the intracranial pressure fluctuation index is σ I represents the standard deviation of the intracranial pressure measurement values within a preset time window, and μ I represents the arithmetic mean of the intracranial pressure within a preset time window;

[0024] The calculation formula for the Glasgow Coma Scale decline rate is t i represents the time point of the i-th score, and G i represents the Glasgow Coma Scale score at the corresponding time point, represents the average value of all time points, represents the average value of the Glasgow Coma Scale scores at all time points;

[0025] The calculation formula for the limb muscle strength change rate is M t represents the muscle strength score at the t-th time, N represents the total number of evaluations. When ΔM is negative, it indicates a continuous decline in muscle strength; the calculation formula for the systolic blood pressure variability is R S =max(S t )-min(S t ), S t represents the systolic blood pressure measurement values at each time point within a preset sliding window;

[0026] The calculation formula for the bispectral index trend is E t =α·Bt +(1 - α)·E t-1 ,E t represents the bispectral index measurement value at the t-th time, α represents the smoothing factor, and E0 = B0;

[0027] The formula for calculating the limb movement frequency is Q represents the peak number of times the limb activity amplitude detected by the accelerometer exceeds the threshold, and ΔT represents the time window.

[0028] For the aspects and any possible implementation manners as described above, a further implementation manner is provided. Specifically, S3 includes:

[0029] S31. Obtain the historical medical images of patients at high risk of hemiplegia, including CT, MRI, and CTP images, and perform hemiplegia label marking according to the follow-up results;

[0030] S32. Register different images of the same patient, unify different image sizes / resolutions to a standard, perform intensity normalization, and divide the dataset into a training set and a validation set;

[0031] S33. Construct and pre-train a hemiplegia prediction model through the training set. Among them, 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 image features: edema, hemorrhage, and abnormal cerebral blood flow. Each sub-model is trained on an independent dataset, and then outputs the corresponding features for the meta-model to perform final fusion during the inference stage, and the meta-model is trained;

[0032] S34. Input the medical images of new patients at high risk of hemiplegia, send them into the three sub-models for inference respectively, and then perform fusion through the meta-model to obtain the final hemiplegia risk score.

[0033] For the aspects and any possible implementation manners as described above, a further implementation manner is provided. Specifically, constructing and pre-training the hemiplegia prediction model through the training set includes:

[0034] Construct an edema detection sub-model: The network structure uses 3D U-Net for edema region segmentation. The encoder is 5 layers of 3D convolution + ReLU + Downsample, and the decoder is 5 layers of transposed convolution + skip-connection. The input is the CT low-density area, and the output is the segmentation mask of the edema region. Calculate the edema probability and edema volume from the segmentation result. The loss function selects the combined use of Dice Loss and BCE / CE. Use Dropout to reduce overfitting, use an adaptive learning rate, use the annotated edema region images as the training set, use methods such as random rotation, translation, cropping, and brightness change to enhance the data, train for 100 epochs, and verify the accuracy of edema region segmentation through the validation set after training;

[0035] Construct a blood hemorrhage detection sub - model: The network structure uses DenseNet 3D. The input is the plain CT scan, and the output is the hemorrhage probability and the volume of hemorrhage plaques. The loss function selects BCE / CE. Use CT images with hemorrhage plaque annotations as the training set, and use methods such as random rotation, translation, cropping, and brightness change to augment the data. After training, verify the hemorrhage recognition accuracy through the validation set;

[0036] Construct a cerebral blood flow detection sub - model: The network structure uses 3D - CNN. The input is the MRI perfusion sequence, and the output is the probability of perfusion abnormality. The loss function adopts the BCE / CE loss function. Use images with low - perfusion area annotations as the training set, and use time translation and inter - frame noise for data augmentation;

[0037] Construct and train a meta - model: Select random forest as the meta - model. Based on the three sub - models, obtain numerical features. Integrate the imaging features of edema probability, hemorrhage probability, and perfusion abnormality probability into a feature vector as the input of the meta - model. Extract the sub - model outputs from the training set images for each patient and pair them with the "whether hemiplegia" label for training the meta - model. Tune the hyperparameters of the meta - model and evaluate the accuracy on the validation set. The final output of the meta - model is the hemiplegia risk score.

[0038] For the aspects and any possible implementation manners as described above, a further implementation manner is provided. The first smart contract specifically includes:

[0039] When generating a new hemiplegia risk score for a patient with traumatic brain injury, call the update score function to generate or update the patient risk score table to register the hemiplegia risk score data, where the hemiplegia risk score data includes the patient's globally unique identifier, the score value, and the update time;

[0040] Compare the patient's latest hemiplegia risk score with the risk threshold of the current cycle. If it exceeds the threshold, match the corresponding alarm level, that is: low - priority alarms are only recorded in the audit log, medium - priority alarms directly send messages to both the medical node and the patient, high - priority alarms are first sent to the medical node for review. If confirmed after review, then sent to the patient and recorded in the audit log;

[0041] Generate or update the follow - up form through the real information uploaded by the medical node after the follow - up period expires to record the real result of whether the patient has hemiplegia after the follow - up period expires, and update the risk threshold when the update cycle arrives;

[0042] Record the update score, warning trigger, audit result, follow - up form, and risk threshold update operation.

[0043] For the aspects and any possible implementation described above, a further implementation is provided. The risk threshold update formula is as follows:

[0044] H t =H t-1 +θ·(MR t-1 -MR′),

[0045] where H t represents the threshold of the current cycle, H t-1 represents the risk threshold of the previous cycle, the initial risk threshold is a 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;

[0046] Compare the latest hemiplegia risk score H * t of the patient with the risk threshold H t of the current cycle. 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. When H * t <0.6H t , it is determined as a low-priority alarm.

[0047] For the aspects and any possible implementation described above, a further implementation is provided. The second smart contract specifically includes:

[0048] Compare the previous prediction result according to the follow-up form. If it is determined that the misjudgment rate exceeds the second target misjudgment rate, generate a misjudgment training data set, retrain the hemiplegia prediction model, and generate a model update record after the update is completed, and record the time, reason, updated content and version information of each model update;

[0049] Set the permission levels for data query and model update for new or existing medical nodes, and then call the "modify permission" function for changes according to the change request;

[0050] Record the hemiplegia prediction model update record and the permission change record.

[0051] In a second aspect, an embodiment of the present invention provides a hemiplegia risk prediction system for brain trauma patients using the above method. The system includes:

[0052] A data acquisition and storage module, configured to obtain patient information and medical data of patients with traumatic brain injury uploaded by different medical nodes, perform preprocessing and encryption processing, and then store them distributively in a multi-center cloud platform;

[0053] A risk assessment module, configured to calculate a comprehensive risk assessment score based on medical data, and screen out patients with high risk of hemiplegia through the comprehensive risk assessment score;

[0054] An image feature processing module, which extracts image features for patients with high risk of hemiplegia, constructs and pre-trains a hemiplegia prediction model, inputs the extracted image features into the hemiplegia prediction model, and obtains the hemiplegia risk score of patients with traumatic brain injury;

[0055] A first smart contract module, configured to automatically evaluate and record the obtained hemiplegia risk score, trigger an alarm based on a preset risk threshold, and send the alarm information to medical nodes and patients;

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

[0057] One of the above technical solutions has the following beneficial effects:

[0058] In the method of the embodiment of the present invention, a hemiplegia risk prediction method and system suitable for patients with traumatic brain injury are proposed. Through a distributed cloud platform and a hash ring structure, the method realizes the encrypted preprocessing, distributed storage and redundant backup of medical data, comprehensively uses hemiplegia symptom keywords and dynamic physiological indicators to calculate risk scores, realizes the early screening of patients with high risk of hemiplegia, realizes the accurate risk prediction of image data based on a prediction model, and realizes the automatic recording of risk scores, alarm triggering and model adaptive update through smart contracts. The integrated technology of the present invention realizes the accurate prediction and automatic alarm of the hemiplegia risk of patients with traumatic brain injury, and improves the timeliness of hemiplegia risk intervention for patients with traumatic brain injury.

Description of the Drawings

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0060] Figure 1 It is a schematic flow chart of a hemiplegia risk prediction method suitable for patients with traumatic brain injury provided by an embodiment of the present invention;

[0061] Figure 2Schematic block diagram of a hemiplegia risk prediction system for brain injury patients provided by an embodiment of the present invention;

[0062] Figure 3 Schematic hardware structure diagram of a hemiplegia risk prediction system for brain injury patients provided by an embodiment of the present invention.

Detailed implementation manners

[0063] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0064] Please refer to Figure 1 , which is a schematic flowchart of a hemiplegia risk prediction method for brain injury patients provided by an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0065] S1. Obtain the patient information and medical data of brain injury patients uploaded by different medical nodes, perform preprocessing and encryption processing, and then store them distributedly in a multi-center cloud platform;

[0066] S2. Calculate a comprehensive risk assessment score based on the medical data, and screen out high-risk hemiplegia patients through the comprehensive risk assessment score;

[0067] S3. Extract image features for high-risk hemiplegia patients, construct and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain the hemiplegia risk score of the brain injury patient;

[0068] S4. Automatically evaluate and record the hemiplegia risk score through the first smart contract, trigger an alarm according to a preset risk threshold and send the alarm information to the medical node and the patient, and automatically trigger the update and record of the hemiplegia prediction model through the second smart contract.

[0069] In the embodiments of the present invention, distributed storage, processing, and redundant backup of medical data are achieved through a distributed cloud platform. Early screening of patients at high risk of hemiplegia is realized by using risk assessment scores. Accurate prediction of hemiplegia risk for imaging data is achieved based on a prediction model. Automatic recording of risk scores, triggering of early warnings, and adaptive update of the model are realized through smart contracts. Through accurate management, screening, prediction, and automated early warning of the hemiplegia risk of patients with traumatic brain injury, the timeliness and pertinence of clinical intervention are improved. Through end-to-end automated processing from patient data collection to risk prediction, early warning triggering, and model update, the influence of manual intervention and subjective judgment is reduced, and intelligent processing is realized.

[0070] In a preferred embodiment of the present invention, a method for predicting the risk of hemiplegia applicable to patients with traumatic brain injury is provided. The method includes the following steps:

[0071] S1. Obtain the patient information and medical data of patients with traumatic brain injury uploaded by different medical nodes, perform preprocessing and encryption processing, and then store them distributively in a multi-center cloud platform;

[0072] S2. Calculate a comprehensive risk assessment score based on the medical data, and screen out patients at high risk of hemiplegia through the comprehensive risk assessment score;

[0073] S3. Extract image features for patients at high risk of hemiplegia, construct and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain the hemiplegia risk score of patients with traumatic brain injury.

[0074] In a preferred embodiment of the present invention, S1 includes:

[0075] S11. Extract medical image data from the medical data of patients with traumatic brain injury uploaded by different medical nodes obtained, use an adaptive filtering method to remove image noise, use a histogram equalization method to enhance the contrast of the image, perform image registration on multi-temporal or multi-modal medical image data to complete alignment, use an image segmentation algorithm to segment the brain region, and eliminate non-background regions;

[0076] S12. Perform standardization processing on the medical data, including normalization and filling of missing values;

[0077] S13. Generate a hash value through 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. Calculate the hash value as the globally unique identifier of the patient according to each target patient information. Find the nearest cloud storage node on the hash ring according to the globally unique identifier, and assign the medical data of the target patient to the cloud storage node. And the medical data copy of the target patient is copied clockwise to the subsequent N redundant nodes on the hash ring. When the medical data is updated, push incremental information to the redundant nodes;

[0078] 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 range covered by the new node on the hash ring. When migrating, adopt the dual-write strategy of writing simultaneously from the old node and the new node. After migration is completed, include the new node in the subsequent N redundant target nodes. If a node failure is detected, switch the read / write request to the nearest redundant replica node. If the redundant replica data is complete, directly promote it to the new master node and supplement a new redundant replica. If the redundant replica is incomplete, recover the data from other replicas. After the node is recovered, synchronize the incremental data from the redundant replica to the latest state.

[0079] In the embodiments of the present invention, through preprocessing technologies such as adaptive filtering, histogram equalization, image registration, and segmentation, the noise is effectively reduced, the contrast is improved, and the key brain regions are highlighted, providing high-quality data for subsequent feature extraction and model training; normalizing and filling in missing values for medical data to ensure the consistency and integrity of multi-source data, thereby improving the robustness of subsequent risk assessment and prediction models; using hash operations to generate globally unique identifiers and constructing a hash ring storage structure to achieve distributed storage and redundant backup of data on a multi-center cloud platform, which not only guarantees data security but also improves the efficiency of data access and update; using the dual-write strategy and independent synchronization channels to effectively support data migration of newly added nodes and real-time update of hot data, ensuring that the system has good scalability and high availability.

[0080] In the preferred embodiments of the present invention, the method of the present invention further includes:

[0081] S15. When the number of updates of a patient's medical data within a preset time is greater than a preset update threshold, mark the patient as a hot data patient and perform the following operations: allocate an exclusive master node for the hot data patient, directly write the real-time update to this master node, and batch push the incremental data to the redundant replica nodes on the hash ring through an asynchronous message queue; add a global version number to the medical data of the hot data patient, and increment the version number each time an update occurs; if a redundant replica node receives conflicting version data, automatically discard the old data and trigger an alarm; if the version numbers are the same but the content is inconsistent, start a consistency check protocol and select the latest data based on the time stamp and majority replica consensus; preferentially route the read / write requests for hot data to the exclusive master node to reduce the impact of synchronization delay of redundant replicas.

[0082] In the embodiment of the present invention, when the update frequency of patient data exceeds a preset threshold, it is marked as hot data, and the exclusive master node directly performs real-time writing and updating, effectively reducing the redundancy copy synchronization delay caused by frequent updates; while the asynchronous message queue batches and pushes incremental data to the redundant nodes, which not only smooths the data transmission load, but also ensures fast and consistent updates of data in the distributed storage system; by adding a global version number to the hot data, the version is incremented each time an update is made, providing a clear identifier for the data state; when the redundant copy node receives data with a conflicting version, 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 inconsistent, the latest data is selected through a consistency check protocol based on the time stamp and the consensus of the majority of replicas, thus effectively solving the data conflict problem and ensuring data consistency and integrity; the read and write requests for hot data are preferentially routed to the exclusive master node, minimizing the response delay caused by multi-node synchronization and improving the processing efficiency and overall performance of the system in the scenario of high-frequency data updates.

[0083] In a preferred embodiment of the present invention, step S2 specifically includes:

[0084] S21. Construct a keyword library for hemiplegia symptoms, extract hemiplegia symptom keywords from the medical data of patients with brain trauma according to the keyword library for hemiplegia symptoms, and count the occurrence frequencies to generate a keyword list;

[0085] S22. Select physiological indicators related to hemiplegia and changing over time as dynamic change features, and calculate dynamic change feature values from the medical data of patients with brain trauma;

[0086] S23. Standardize the keyword list and the dynamic change feature values, and calculate a comprehensive risk assessment score according to the keyword list and the dynamic change feature values. The calculation formula of the comprehensive risk assessment score is as follows:

[0087]

[0088] where 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 the i-th symptom of hemiplegia, 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 an ROC curve based on historical patient data, select the T value when the Youden index is the largest as the initial risk threshold, and use T / K as the final risk threshold, where 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, screen them as patients at high risk of hemiplegia.

[0090] In the embodiment of the present invention, by combining text mining and the calculation of physiological dynamic characteristics, the risk of patients is quantitatively evaluated from both static and dynamic aspects, improving the comprehensiveness of risk judgment. In addition, by standardizing the processing and calculating the comprehensive risk assessment score using a mathematical formula based on weight coefficients, the manual subjective intervention is reduced, making the risk assessment more scientific and objective. Use historical data to draw an ROC curve and determine the T value corresponding to the largest Youden index, set the initial risk threshold, and dynamically adjust according to actual data, providing an accurate basis for screening high-risk patients.

[0091] Among them, a reference example of the keyword library for hemiplegia symptoms is as follows:

[0092]

[0093] Among them, selecting physiological indicators related to hemiplegia and changing over time as dynamic change characteristics may include but are not limited to any one or more of the intracranial pressure fluctuation index, the decline rate of Glasgow Coma Scale, the change rate of limb muscle strength, the systolic blood pressure variability, the trend of bispectral index, and the limb movement frequency. It should be noted that the selection of physiological indicators can also be adjusted according to the actual situation.

[0094] Specifically, the calculation of the dynamic change characteristic value refers to the following. The calculation formula for the intracranial pressure fluctuation index is σ I represents the standard deviation of the intracranial pressure measurement values within a preset time window, and μ I represents the arithmetic mean of the intracranial pressure within a preset time window. The larger V I is, the worse the intracranial pressure stability. The calculation formula for the decline rate of Glasgow Coma Scale is t[[ID=2,6]] i represents the time point of the i-th score, and G i represents the Glasgow Coma Scale corresponding to the time point, represents the average value of all time points, represents the average value of the Glasgow Coma Scale scores at all time points. When S G <0, it indicates the deterioration of the consciousness state. The calculation formula for the change rate of limb muscle strength is M t represents the muscle strength score at the t-th time, and N represents the total number of evaluations. When ΔM is negative, it indicates a continuous decrease in muscle strength. The calculation formula for the systolic blood pressure variability is R S = max(S t)-min(S t ),S t represents the systolic blood pressure measurement values at each time point within a preset sliding window, and R S The larger the value, the more severe the blood pressure fluctuation; the formula for calculating the bispectral index trend is E t = α·B t +(1 - α)·E t-1 where E t represents the bispectral index measurement value at the t-th time, α represents the smoothing factor, E0 = B0, and when ΔM continuously decreases, it indicates the aggravation of brain function inhibition; the formula for calculating the limb movement frequency is Q represents the peak number of times the limb activity amplitude detected by the accelerometer exceeds the threshold, ΔT represents the time window, and F m A decrease indicates significant impairment of motor function.

[0095] In the embodiments of the present invention, by specifying indicators such as the intracranial pressure fluctuation index, the rate of decline in the Glasgow Coma Scale, the change rate of limb muscle strength, the systolic blood pressure variability, the bispectral index trend, and the limb movement frequency, and giving the corresponding calculation formulas, multiple dynamic indicators can comprehensively reflect the changes in the physiological state of the patient within a preset time window, effectively capture the complex pathological changes caused by brain trauma, and provide a quantitative basis for risk scoring; through the standardization and mathematical modeling of each dynamic indicator, a reasonable distribution of the influence weights of different indicators is achieved, ensuring that the risk assessment can take into account various physiological data.

[0096] In the preferred embodiment of the present invention, the S3 specifically includes:

[0097] S31. Obtain the historical medical images of patients at high risk of hemiplegia, including CT, MRI, and CTP images, and perform hemiplegia label marking according to the follow-up results;

[0098] S32. Register different images of the same patient, unify different image sizes / resolutions to a standard, and perform intensity normalization, and divide the data set into a training set and a validation set;

[0099] S33. Construct and pre-train a hemiplegia prediction model through the training set. Among them, 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 image features: edema, hemorrhage, and abnormal cerebral blood flow. Each sub-model is trained on an independent data set, and then outputs the corresponding features for the meta-model to perform final fusion during the inference stage, and train the meta-model;

[0100] S34. Input the medical images of new patients at high risk of hemiplegia, send them into the three sub-models for inference respectively, and then perform fusion through the meta-model to obtain the final hemiplegia risk score.

[0101] In the embodiments of the present invention, by acquiring, annotating, and standardizing the historical medical images of patients, a set of representative and trainable image datasets are constructed. Three sub-models for edema, hemorrhage, and abnormal cerebral blood flow are used to process different image features respectively, and then the results of each sub-model are fused by a meta-model to achieve multi-angle risk assessment, significantly improving the overall prediction accuracy. After inputting the image data of a new patient, the hemiplegia risk score can be quickly obtained, which supports the early intervention for high-risk patients and the formulation of personalized treatment plans in clinical practice.

[0102] In a preferred embodiment of the present invention, the construction and pre-training of the hemiplegia prediction model through a training set specifically include:

[0103] Construct an edema detection sub-model: The network structure uses 3D U-Net for edema region segmentation. The encoder is 5 layers of 3D convolution + ReLU + Downsample, and the decoder is 5 layers of transposed convolution + skip-connection. The input is the low-density area of MRI or CT, and the output is the segmentation mask of the edema region. The edema probability and edema volume are calculated from the segmentation results. The loss function selects the combined use of DiceLoss and BCE / CE. Dropout is used to reduce overfitting, and an adaptive learning rate is used. The annotated edema region images are used as the training set, and data is enhanced using methods such as random rotation, translation, cropping, and brightness change. Train for 100 epochs, and after training, verify the accuracy of edema region segmentation through the validation set;

[0104] Construct a hemorrhage detection sub-model: The network structure uses DenseNet 3D. The input is the plain CT scan, and the output is the hemorrhage probability and the volume of hemorrhage plaques. The loss function selects BCE / CE. The CT images with hemorrhage plaque annotations are used as the training set, and data is enhanced using methods such as random rotation, translation, cropping, and brightness change. After training, verify the accuracy of hemorrhage recognition through the validation set;

[0105] Construct a cerebral blood flow detection sub-model: The network structure uses 3D-CNN. The input is the CTP or MRI perfusion sequence, and the output is the probability of perfusion abnormality. The loss function uses the BCE / CE loss function. The low-perfusion region annotated images are used as the training set, and data is enhanced using time translation and inter-frame noise;

[0106] Construct and train a meta-model: Select random forest as the meta-model. Based on the three sub-models, numerical features are obtained. 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. Based on the sub-model outputs extracted from the training set images for each patient, they are paired with the "whether hemiplegic" label for training the meta-model. Hyperparameters of the meta-model are tuned, and the accuracy is evaluated on the validation set. The final output of the meta-model is the hemiplegia risk score.

[0107] In the embodiments of the present invention, 3D U-Net, DenseNet 3D, and 3D-CNN are respectively used for fine-grained detection of different pathological features to ensure that each sub-model can accurately quantify the imaging features it focuses on, namely the probability of edema, the probability of hemorrhage, and perfusion abnormalities. The application of strategies such as data augmentation, combined loss function, and adaptive learning rate effectively reduces the risk of overfitting and improves the adaptability of the model to different patients and different imaging data. A random forest meta-model is used to fuse the features output by each sub-model to achieve a comprehensive judgment of multi-dimensional risk information, thereby improving the reliability and accuracy of the final hemiplegia risk score.

[0108] In a preferred embodiment of the present invention, the first smart contract specifically includes:

[0109] When generating a new hemiplegia risk score for a brain trauma patient, the update score function is called to generate or update the patient risk score table to register the hemiplegia risk score data, where the hemiplegia risk score data includes the patient's globally unique identifier, the score value, and the update time;

[0110] Compare the patient's latest hemiplegia risk score with the risk threshold of the current cycle. If it exceeds the threshold, the corresponding alarm level is matched, that is: low-priority alarms are only recorded in the audit log, medium-priority alarms directly send messages to both the medical node and the patient, high-priority alarms are first sent to the medical node for review, and if confirmed after review, they are then sent to the patient and recorded in the audit log;

[0111] Generate or update the follow-up form through the real information uploaded by the medical node after the follow-up period expires to record the real result of whether the patient has hemiplegia after the follow-up period expires, and update the risk threshold when the update cycle arrives;

[0112] Record the update score, warning trigger, audit result, follow-up form, and risk threshold update operation.

[0113] The embodiments of the present invention use a smart contract to automatically record the patient risk score, update operation logs, and warning trigger process, realizing real-time monitoring and automatic response to the risk status, reducing manual intervention; by comparing the risk score with the preset risk threshold, different warning levels of low, medium, and high are distinguished to ensure that corresponding response measures are taken in different risk situations, improving the pertinence of patient intervention; all operations are recorded through the smart contract, providing a 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 +θ·(MR t-1 -MR′),

[0116] Among them, H t represents the threshold value of the current cycle, H t-1 represents the risk threshold value of the previous cycle, the initial risk threshold value is a 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] Compare the latest hemiplegia risk score H of the patient * t with the risk threshold value H of the current cycle t , 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. When H * t <0.6H t , it is determined as a low-priority alarm.

[0118] The risk threshold update formula of the embodiment of the present invention automatically adjusts the risk threshold according to historical data and the actual misjudgment rate, so that the system can dynamically adapt to the changes of clinical data and maintain the sensitivity and specificity of risk assessment; and the risk threshold update mechanism promotes the continuous optimization of the entire system during long-term operation, reduces the misjudgment rate, and improves the warning accuracy.

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

[0120] Compare the previous prediction result according to the follow-up form. If it is judged that the misjudgment rate exceeds the second target misjudgment rate, generate a misjudgment training data set, retrain the hemiplegia prediction model, generate a model update record after the update is completed, and record the time, reason, updated content and version information of each model update;

[0121] Set the permission levels for data query and model update for new or existing medical nodes, and then call the "modify permission" function for modification according to the change request;

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

[0123] The second smart contract of the embodiment of the present invention automatically monitors the prediction misjudgment rate, triggers the model retraining mechanism, realizes the dynamic update and adaptive improvement of the model, so as to ensure that the prediction system always conforms to the actual clinical needs; for new or existing medical nodes, the smart contract automatically changes the permissions and records them, ensuring that data query and model update operations are carried out under strict permission control; records the time, reason, content and version information of each model update, ensuring that the entire improvement process is open and transparent, which helps subsequent technical optimization and problem tracking.

[0124] Specifically, generating a misjudgment training data set and retraining the hemiplegia prediction model can be achieved through the following steps: collecting the actual follow-up results of patients from the follow-up data uploaded by each medical node, synchronously obtaining the system's historical predicted risk scores for these patients, comparing the predicted results with the actual follow-up results, and identifying cases with incorrect predictions; judging whether the current system misjudgment rate exceeds the second target misjudgment rate based on the preset misjudgment rate threshold, marking the misjudged patients as "misjudgment cases", and recording their corresponding original data, imaging data, dynamic physiological data, and prediction scores and other key information; aggregating the data of all misjudgment cases to generate a dedicated misjudgment training data set, ensuring that the data set contains information such as the patient's global unique identifier, preprocessed medical imaging features, dynamic physiological indicators, keyword statistics, and original prediction scores, and preprocessing the data to ensure data consistency and integrity; performing image registration, segmentation, and standardization on the imaging data in the misjudgment training data set, extracting key features such as edema, hemorrhage, and abnormal cerebral blood flow, and for text data, performing keyword extraction and frequency statistics based on the hemiplegia symptom keyword library, standardizing the dynamic physiological indicators, and using data enhancement to expand the data. Samples, reduce the risk of overfitting, data enhancement methods such as rotation, translation, cropping, brightness change, etc.; adjust the parameters and weights of the comprehensive risk assessment score to make the calculated score of real hemiplegic patients higher; combine the generated misjudgment training data set with the original training data as new training sample input, perform fine-tuning training on each sub-model separately, and use enhanced data to further improve the robustness of the model. For the updated sub-model output, use random forest or other integrated learning methods to train the meta-model to form a new Stacking structure to generate a comprehensive hemiplegia risk score; use an independent validation set to verify the updated predictions The model is verified and its performance in terms of ROC curve, accuracy, sensitivity, specificity and other indicators is evaluated. When the verification results meet the preset requirements, the updated model is deployed to the system to replace the original hemiplegia prediction model. The model update record is automatically generated through the second smart contract to record 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, and regularly calculates the misjudgment rate. If it is found that the misjudgment rate exceeds the target threshold again, it automatically triggers the generation of new misjudgment training data and the model retraining process, forming a closed-loop feedback mechanism to continuously improve the system prediction performance.

[0125] The first smart contract module and the second smart contract module of the present invention are implemented based on a multi - center cloud platform, which may specifically include: an event listening layer: capturing medical data update events in real - time through a message queue; a contract execution layer: encapsulating early warning logic and model update logic with microservices and providing external calls through REST APIs; an audit storage layer: recording all contract operations using a WORM - mode logging system, attaching timestamps and digital signatures; a permission control layer: restricting data access and operation permissions of nodes based on the OAuth2 protocol and the RBAC model. It should be noted that, if necessary, the underlying architecture of the blockchain can also be used to execute smart contracts. For example, a consortium blockchain can be used, and only authorized medical institutions are allowed to join as nodes, and then the first smart contract and the second smart contract are executed.

[0126] The embodiments of the present invention further provide device embodiments for implementing each step and method in the above - mentioned method embodiments.

[0127] Please refer to Figure 2 which is Figure 2 a schematic block diagram of a hemiplegia risk prediction system for patients with traumatic brain injury provided by an embodiment of the present invention. The system includes:

[0128] A data acquisition and storage module 210, configured to obtain patient information and medical data of patients with traumatic brain injury uploaded by different medical nodes, perform pre - processing and encryption processing, and then store them distributively in a multi - center cloud platform;

[0129] A risk assessment module 220, configured to calculate a comprehensive risk assessment score based on medical data, and screen out patients with high hemiplegia risk through the comprehensive risk assessment score;

[0130] An image feature processing module 230, extracting image features for patients with high hemiplegia risk, constructing and pre - training a hemiplegia prediction model, inputting the extracted image features into the hemiplegia prediction model, and obtaining a hemiplegia risk score for patients with traumatic brain injury;

[0131] A first smart contract module 240, 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 medical nodes and patients;

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

[0133] The present invention provides a hemiplegia risk prediction system for patients with traumatic brain injury that integrates data acquisition, risk assessment, medical image processing, smart contract early warning, and model update.

[0134] Since each unit module in this embodiment can execute Figure 1For the method shown, for parts not described in detail in this embodiment, reference may be made to the relevant description of Figure 1 .

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

[0136] In a third aspect, an embodiment of the present invention provides a computer-readable medium. The computer-readable medium stores program code, and when the computer program code runs on a computer, the computer is caused to execute the hemiplegia risk prediction method for brain trauma patients provided by the embodiment of the present invention.

[0137] At the hardware level, the device may include a processor, and optionally also an internal bus, a network interface, and a memory. Among them, the memory may include internal memory, such as high-speed random access memory (Random-Access Memory, RAM), and may also include non-volatile memory, such as at least one disk memory, etc. Of course, the device may also include other hardware required for other services.

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

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

[0140] The steps of the method disclosed in the embodiments of the present invention can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method.

[0141] The systems, devices, modules or units illustrated in the above embodiments can be specifically implemented by a computer chip or an entity, or implemented by a product 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, when describing the above devices, they are divided into various units or modules according to functions and described separately. Of course, when implementing the present invention, the functions of each unit or module can be implemented in the same or multiple software and / or hardware.

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

[0144] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple 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 operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction means that implements the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 in one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

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

[0148] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0149] Computer-readable media includes both permanent and non-permanent, removable and non-removable media implemented by any method or technology for storing information. The information can be 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 technologies, compact disc read only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0150] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising said element.

[0151] Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system or a computer program product. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may 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-ROM, optical storage, etc.) containing computer-usable program code.

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

[0153] The various embodiments of the present invention are described in a progressive manner. For the same or similar parts among the various embodiments, reference may be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference may be made to the corresponding parts of the method embodiments for the relevant content.

[0154] The above are only the embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various modifications and changes can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims

1. A hemiplegia risk prediction method applicable to patients with traumatic brain injury, characterized in that, The method includes: S1. Obtain the patient information and medical data of patients with traumatic brain injury uploaded by different medical nodes, perform preprocessing and encryption processing, and then store them distributively in a multi-center cloud platform; S2. Calculate the comprehensive risk assessment score based on the medical data, and screen out patients with high risk of hemiplegia through the comprehensive risk assessment score; S3. Extract image features for patients with high risk of hemiplegia, construct and pre-train a hemiplegia prediction model, input the extracted image features into the hemiplegia prediction model, and obtain the hemiplegia risk score of patients with traumatic brain injury; S4. Automatically evaluate and record the hemiplegia risk score through the first smart contract, trigger an alarm according to the preset risk threshold and send the alarm information to the medical node and the patient, and automatically trigger the update and record of the hemiplegia prediction model through the second smart contract.

2. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 1, characterized in that, The S1 includes: S11. Extract medical image data from the medical data of patients with traumatic brain injury uploaded by different medical nodes, use the adaptive filtering method to remove image noise, use the histogram equalization method to enhance the contrast of the image, complete the alignment of multi-temporal or multi-modal medical image data through image registration, use the image segmentation algorithm to segment the brain region, and eliminate non-background regions; S12. Perform standardization processing on the medical data, including normalization and missing value filling; S13. Generate a hash value through the hash operation of 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. Calculate the hash value according to each target patient information as the globally unique identifier of the patient. Find the nearest cloud storage node on the hash ring according to the globally unique identifier, and assign the medical data of the target patient to the cloud storage node. And the medical data copy of the target patient is copied clockwise to the subsequent N redundant nodes on the hash ring. When the medical data is updated, push the incremental information to the redundant nodes; 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 on the hash ring. When migrating, adopt the dual-write strategy of writing by the old node and the new node at the same time. After migration, include the new node in the subsequent N redundant targets. If a node failure is detected, switch the read and write requests to the nearest redundant copy node. If the redundant copy data is complete, directly promote it to the new master node and supplement the new redundant copy. If the redundant copy is incomplete, recover the data from other copies. After the node recovers, synchronize the incremental data from the redundant copy to the latest state.

3. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 1, wherein The S2 specifically includes: S21. Construct a keyword library for hemiplegia symptoms, extract hemiplegia symptom keywords from the medical data of patients with traumatic brain injury according to the keyword library for hemiplegia symptoms and count the occurrence frequency to generate a keyword list; S22. Select physiological indicators related to hemiplegia and changing over time as dynamic change features, and calculate the dynamic change feature values from the medical data of patients with traumatic brain injury; S23. Standardize the keyword list and the dynamic change feature values, and calculate the comprehensive risk assessment score according to the keyword list and the dynamic change feature values. Among them, 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 the i-th symptom of hemiplegia, w i is the weight of the i-th symptom, n is the number of selected hemiplegia symptoms, Δf j (t) is the eigenvalue of the dynamic change 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; S24. Draw an ROC curve based on historical patient data, select the T value when the Youden index is the largest as the initial risk threshold, and use T / K as the final risk threshold, where 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, screen them as patients with high risk of hemiplegia.

4. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 3, wherein The dynamic change characteristics include but are not limited to any one or more of the intracranial pressure fluctuation index, the decreasing rate of Glasgow Coma Scale, the limb muscle strength change rate, the systolic blood pressure variability, the bispectral index trend, and the limb movement frequency. Among them, The calculation formula for the intracranial pressure fluctuation index is σ I represents the standard deviation of the intracranial pressure measurement values within a preset time window, and μ I represents the arithmetic mean of the intracranial pressure within a preset time window; The calculation formula for the decreasing rate of the Glasgow Coma Scale is as follows t i represents the time point of the i-th score, and G i represents the Glasgow Coma Scale score corresponding to the time point, represents the average value of all time points, represents the average value of the Glasgow Coma Scale scores at all time points; The calculation formula for the limb muscle strength change rate is M t represents the muscle strength score at the t-th time, N represents the total number of evaluations. When ΔM is negative, it indicates that the muscle strength continues to decline. The calculation formula for the systolic blood pressure variability is R S = max(S t ) - min(S t ), where S t represents the systolic blood pressure measurement values at each time point within the preset sliding window. The formula for the bispectral index trend is E t = α · B t + (1 - α) · E t-1 , E t represents the bispectral index measurement value at the t-th time, α represents the smoothing factor, and E0 = B0; The calculation formula for the limb movement frequency is Q represents the peak number of times the limb activity amplitude detected by the accelerometer exceeds the threshold, and ΔT represents the time window.

5. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to any one of claims 1 or 4, characterized in that The specific steps of S3 are as follows: S31. Obtain the historical medical images of patients with high risk of hemiplegia, including CT, MRI, and CTP images, and perform hemiplegia label marking according to the follow-up results. S32. Register different images of the same patient, unify different image sizes / resolutions to a standard, and perform intensity normalization. Divide the data set into a training set and a validation set. S33. Construct and pre-train a hemiplegia prediction model through the training set. Among them, 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 image features: edema, hemorrhage, and abnormal cerebral blood flow. Each sub-model is trained on an independent data set, and then outputs the corresponding features for the meta-model to perform final fusion during the inference stage, and train the meta-model. S34. Input the medical images of new patients with high risk of hemiplegia, send them into the three sub-models for inference respectively, and then perform fusion through the meta-model to obtain the final hemiplegia risk score.

6. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 5, wherein The specific steps of constructing and pre-training a hemiplegia prediction model through the training set are as follows: Construct an edema detection sub-model: The network structure uses 3D U-Net for edema region segmentation. The encoder is 5 layers of 3D convolution + ReLU + Downsample, and the decoder is 5 layers of transposed convolution + skip-connection. The input is the CT low-density area, and the output is the segmentation mask of the edema region. Calculate the edema probability and edema volume from the segmentation result. The loss function selects Dice Loss and BCE / CE jointly. Use Dropout to reduce overfitting, use an adaptive learning rate, use the annotated edema region images as the training set, and use methods such as random rotation, translation, cropping, and brightness change to enhance the data. Train for 100 epochs, and verify the accuracy of edema region segmentation through the validation set after training. Construct a hemorrhage detection sub-model: The network structure uses DenseNet 3D. The input is the non-contrast CT scan, and the output is the hemorrhage probability and the volume of hemorrhage plaques. The loss function selects BCE / CE. Use the CT images with hemorrhage plaque annotations as the training set, and use methods such as random rotation, translation, cropping, and brightness change to enhance the data. Verify the accuracy of hemorrhage recognition through the validation set after training. Construct a cerebral blood flow detection sub-model: The network structure uses 3D-CNN. The input is the MRI perfusion sequence, and the output is the probability of perfusion abnormality. The loss function uses the BCE / CE loss function. Use the images with low perfusion region annotations as the training set, and perform data augmentation using time translation and inter-frame noise. Constructing and training the meta-model: Select the random forest as the meta-model, obtain numerical features based on three sub-models, integrate the imaging feature edema probability, hemorrhage probability, and perfusion abnormality probability into a feature vector as the input of the meta-model. Extract the sub-model outputs from the training set images for each patient and pair them with the "hemiplegia or not" label for training the meta-model. Tune the hyperparameters of the meta-model and evaluate the accuracy on the validation set. The final output of the meta-model is the hemiplegia risk score.

7. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 1, wherein The first smart contract specifically includes: When generating a new hemiplegia risk score for a brain trauma patient, call the update score function to generate or update the patient risk score table to register the hemiplegia risk score data, where the hemiplegia risk score data includes the patient's globally unique identifier, score value, and update time; Compare the patient's latest hemiplegia risk score with the risk threshold of the current cycle. If it exceeds the threshold, match the corresponding alarm level, that is: low-priority alarms are only recorded in the audit log, medium-priority alarms directly send messages to both the medical node and the patient, high-priority alarms are first sent to the medical node for review. If confirmed after review, then send to the patient and record in the audit log; Generate or update the follow-up table through the real information uploaded by the medical node after the follow-up period expires to record the real result of whether the patient has hemiplegia after the follow-up period expires, and update the risk threshold when the update cycle arrives; Record the update score, warning trigger, audit result, follow-up table, and risk threshold update operation.

8. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 7, characterized in that, The risk threshold update formula is as follows: H t = H t-1 + θ·(MR t-1 - MR′), Among them, H t represents the threshold value of the current cycle, and H t-1 represents the risk threshold value of the previous cycle. The initial risk threshold value is a set value, θ represents the adjustment coefficient, and MR t-1 represents the misjudgment rate counted in the previous cycle, and MR′ represents the first target misjudgment rate; Compare the patient's latest hemiplegia risk score H * t with the risk threshold H of the current cycle t , when H * t ≥0.8H t , it is considered a high-priority alert. When 0.6H t ≤H * t <0.8H t , it is considered a medium-priority alert. When H * t <0.6H t , it is considered a low-priority alert.

9. The hemiplegia risk prediction method applicable to patients with traumatic brain injury according to claim 8, characterized in that, The second smart contract specifically includes: According to the follow-up table, compare with the previous prediction results. If it is determined that the misjudgment rate exceeds the second target misjudgment rate, generate a misjudgment training data set and retrain the hemiplegia prediction model. After the update is completed, generate a model update record and record the time, reason, updated content, and version information of each model update; Set the permission levels for data query and model update for new or existing medical nodes, and then call the "modify permission" function for change according to the change request; Record the hemiplegia prediction model update record and permission change record.

10. A hemiplegia risk prediction system for brain trauma patients using the method according to any one of claims 1-9, characterized in that, The system includes: A data acquisition and storage module configured to obtain the patient information and medical data of brain trauma patients uploaded by different medical nodes, perform preprocessing and encryption processing, and then distribute and store them in a multi-cloud platform; A risk assessment module configured to calculate a comprehensive risk assessment score based on the medical data and screen out high-risk hemiplegia patients through the comprehensive risk assessment score; An imaging feature processing module, which extracts imaging features for high-risk hemiplegia patients, constructs and pre-trains a hemiplegia prediction model, and inputs the extracted imaging features into the hemiplegia prediction model to obtain the hemiplegia risk score of the brain trauma patient; A first smart contract module configured to automatically evaluate and record the obtained hemiplegia risk score, trigger an alarm based on a preset risk threshold, and send the warning information to the medical node and the patient; A second smart contract module configured to update the hemiplegia prediction model and record the update information when the trigger condition is met.

Images