CN-122025189-A - Method, device, equipment and medium for predicting emergency prognosis of cerebral apoplexy

Abstract

The invention relates to the technical field of intelligent medical assistance, in particular to a method, a device, equipment and a medium for predicting emergency prognosis of cerebral apoplexy, wherein the method comprises the steps of collecting historical emergency treatment data, historical basic health data and historical gene data of a historical patient; the method comprises the steps of determining historical data source cross characteristics based on historical emergency treatment data, historical basic health data and historical gene data, constructing a prediction model based on the historical data source cross characteristics by adopting a deep learning algorithm and a random forest algorithm to predict three aspects of patients to obtain historical prediction results, constructing a loss function based on the three aspects of patients based on the historical prediction results, optimizing the prediction model based on the loss function to obtain an optimized prediction model, obtaining multi-source target data of target patients, obtaining target prediction results through the optimized prediction model, breaking through the limitation of the existing single target prediction, and improving the accuracy and reliability of prognosis prediction.

Inventors

• LIU XUEMEI
• LUO BIN
• YUE XIAOBO

Assignees

• 四川互慧软件有限公司

Dates

Publication Date

20260512

Application Date

20260413

Claims (10)

1. 1. A method for predicting the prognosis of an emergency treatment for stroke, comprising: Collecting historical emergency treatment data, historical basic health data and historical gene data of a historical patient, wherein the historical gene data is gene locus information related to cerebral apoplexy prognosis; determining historical data source intersection characteristics based on the historical emergency treatment data, the historical base health data, and the historical gene data; Based on the cross characteristics of the historical data sources, a deep learning algorithm and a random forest algorithm are adopted to construct a prediction model for respectively predicting the acute phase complication risk, the distant nerve function recovery condition and the recovery potential of a patient to obtain a historical prediction result; Constructing a loss function based on acute phase complication risk, distant nerve function recovery condition and rehabilitation potential based on the historical prediction result; optimizing the prediction model based on the loss function to obtain an optimized prediction model; acquiring target emergency treatment data, target basic health data and target gene data of a target patient; and obtaining a target prediction result of the emergency prognosis of the target patient based on the target emergency treatment data, the target basic health data, the target gene data and the optimization prediction model.
2. 2. The method of claim 1, further comprising, after deriving a target prognosis for the emergency prognosis for the target patient based on the target emergency treatment data, target underlying health data, target gene data, and an optimization prediction model: Generating a rehabilitation plan for the target patient through a rehabilitation training model based on the target prediction result, wherein the rehabilitation training model is obtained based on training; Collecting rehabilitation training data of a target patient through wearable equipment according to the rehabilitation plan; Based on the rehabilitation training data, updating parameters of the rehabilitation training model in real time for updating a rehabilitation plan.
3. 3. The method of claim 2, wherein updating parameters of the rehabilitation training model in real time for updating a rehabilitation program based on the rehabilitation training data comprises: Determining a total loss gradient at the current moment based on the rehabilitation training data; Based on the total loss gradient at the current moment and the parameters of the rehabilitation training model at the moment before the current moment, the parameters of the rehabilitation training model are updated in real time according to the following calculation formula: ; Wherein, the As the parameters of the rehabilitation training model at the moment before the current moment, For the parameters of the rehabilitation training model at the current moment, As a forgetting factor, In order to adapt the rate of learning to the user, The total loss gradient at the current moment is obtained by deriving the loss function, Regularizing the coefficients for L2.
4. 4. The method of claim 1, wherein the historical data source intersection characteristic is determined based on the historical emergency treatment data, the historical base health data, and the historical gene data, and wherein the historical data source intersection characteristic is derived in accordance with the following calculation: ; Wherein, the For any one of the historical emergency treatment data, the historical basic health data and the historical gene data, 3 represents the number of three types of data, For the standardized and normalized processing results of any of the historical emergency treatment data, the historical base health data, and the historical gene data, In order for the attention to be weighted, In order to cross the features across the data sources, For the operation of the convolution of the ID, For the weight of the interaction feature, Cross-characteristics the historical data sources.
5. 5. The method of claim 1, wherein a prediction model is constructed based on the historical data source cross feature by adopting a deep learning algorithm and a random forest algorithm, so as to respectively predict the acute phase complication risk, the distant nerve function recovery condition and the rehabilitation potential of the patient to obtain a historical prediction result, and the historical prediction result is obtained according to the following calculation formula: ; Wherein, the In order to pass through the output of the deep learning algorithm, In order to pass through the output of the random forest algorithm, In order to dynamically modify the term(s), In order to correct the coefficient of the coefficient, For the weights corresponding to the deep learning algorithm, Is the weight corresponding to the random forest algorithm and , Is a historical prediction result.
6. 6. The method of claim 1, wherein constructing a loss function based on acute phase complication risk, distant nerve function recovery, and rehabilitation potential based on the historical prediction results, specifically constructing a loss function based on acute phase complication risk, distant nerve function recovery, and rehabilitation potential according to the following calculation formula: ; Wherein, the A two-class cross entropy for risk of acute phase complications, Ordered cross entropy for the recovery level of distant nerve function, For the mean square error of the rehabilitation potential, For the task to be associated with a regular term, A predictive task corresponding to the risk of acute phase complications, A predictive task corresponding to a level of recovery of distant nerve function, In response to the predicted task of rehabilitation potential, As the weight corresponding to the cross entropy of the two categories of the acute phase complication risk, Weights corresponding to the ordered cross entropy of the distant nerve function recovery level, For the weight corresponding to the mean square error of the rehabilitation potential, For the regularization coefficient(s), Is the loss function.
7. 7. The method of claim 6, wherein the task association regularization term is computed in particular according to the following calculation formula: ; Wherein, the As a result of the covariance, And The method is a prediction result of any two of a prediction task of acute stage complication risk, a prediction task of distant nerve function recovery level and a prediction task of rehabilitation potential.
8. 8. An apparatus for emergency prognosis prediction of stroke, comprising: The acquisition module is used for acquiring historical emergency treatment data, historical basic health data and historical gene data of a historical patient, wherein the historical gene data is gene locus information related to cerebral apoplexy prognosis; the determining module is used for determining the cross characteristics of the historical data sources based on the historical emergency treatment data, the historical basic health data and the historical gene data; The first construction module is used for constructing a prediction model by adopting a deep learning algorithm and a random forest algorithm based on the cross characteristics of the historical data sources so as to respectively predict the acute phase complication risk, the distant nerve function recovery condition and the rehabilitation potential of the patient and obtain a historical prediction result; The second construction module is used for constructing a loss function based on acute phase complication risk, distant nerve function recovery condition and rehabilitation potential based on the historical prediction result; the first obtaining module is used for optimizing the prediction model based on the loss function to obtain an optimized prediction model; the acquisition module is used for acquiring target emergency treatment data, target basic health data and target gene data of a target patient; and the second obtaining module is used for obtaining a target prediction result of the emergency prognosis of the target patient based on the target emergency treatment data, the target basic health data, the target gene data and the optimization prediction model.
9. 9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1-7 when the program is executed by the processor.
10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-7.

Description

Method, device, equipment and medium for predicting emergency prognosis of cerebral apoplexy Technical Field The invention relates to the technical field of intelligent medical assistance, in particular to a method, a device, equipment and a medium for predicting emergency prognosis of cerebral apoplexy. Background Cerebral apoplexy is an acute cerebrovascular disease, has the characteristics of high morbidity, high disability rate, high death rate and high recurrence rate, and is one of the main causes of disability and death in the global scope. With the aging of population structures and the evolution of life style, the incidence of cerebrovascular diseases is in a continuously rising situation, and the disease becomes the first fatal and disabling disease. Although the mortality rate of acute ischemic stroke patients has been significantly reduced in recent years, the disability burden of stroke survivors has increased and rehabilitation has limited effectiveness. The single risk or rehabilitation potential of the acute cerebrovascular disease can be predicted in the prior art, but the single risk or rehabilitation potential has limitations, and the accuracy and reliability of the prediction cannot be ensured. Therefore, how to improve the accuracy and reliability of the emergency prognosis prediction of cerebral apoplexy is a technical problem to be solved urgently. Disclosure of Invention In view of the above, the present invention provides a method, apparatus, device and medium for emergency prognosis prediction of stroke that overcomes or at least partially solves the above-mentioned problems. In a first aspect, the invention provides a method for prognosis prediction of cerebral apoplexy in emergency, comprising: Collecting historical emergency treatment data, historical basic health data and historical gene data of a historical patient, wherein the historical gene data is gene locus information related to cerebral apoplexy prognosis; determining historical data source intersection characteristics based on the historical emergency treatment data, the historical base health data, and the historical gene data; Based on the cross characteristics of the historical data sources, a deep learning algorithm and a random forest algorithm are adopted to construct a prediction model for respectively predicting the acute phase complication risk, the distant nerve function recovery condition and the recovery potential of a patient to obtain a historical prediction result; Constructing a loss function based on acute phase complication risk, distant nerve function recovery condition and rehabilitation potential based on the historical prediction result; optimizing the prediction model based on the loss function to obtain an optimized prediction model; acquiring target emergency treatment data, target basic health data and target gene data of a target patient; and obtaining a target prediction result of the emergency prognosis of the target patient based on the target emergency treatment data, the target basic health data, the target gene data and the optimization prediction model. Preferably, after obtaining the target prediction result of the emergency prognosis based on the target emergency treatment data, the target basic health data, the target gene data and the optimization prediction model, the method further comprises: Generating a rehabilitation plan for the target patient through a rehabilitation training model based on the target prediction result, wherein the rehabilitation training model is obtained based on training; Collecting rehabilitation training data of a target patient through wearable equipment according to the rehabilitation plan; Based on the rehabilitation training data, updating parameters of the rehabilitation training model in real time for updating a rehabilitation plan. Preferably, updating parameters of the rehabilitation training model in real time based on the rehabilitation training data for updating a rehabilitation plan includes: Determining a total loss gradient at the current moment based on the rehabilitation training data; Based on the total loss gradient at the current moment and the parameters of the rehabilitation training model at the moment before the current moment, the parameters of the rehabilitation training model are updated in real time according to the following calculation formula: Wherein, the As the parameters of the rehabilitation training model at the moment before the current moment,For the parameters of the rehabilitation training model at the current moment,As a forgetting factor,In order to adapt the rate of learning to the user,The total loss gradient at the current moment is obtained by deriving the loss function,Regularizing the coefficients for L2. Preferably, the historical data source intersection characteristic is determined based on the historical emergency treatment data, the historical basic health data and the historical gene data, and the hi