Neurology index prediction system and method

Abstract

The invention discloses a neurology index prediction system and method. The method comprises the following steps: firstly, collecting case data of a neurology patient; processing the collected case data to form test data; meanwhile, constructing a neurology index prediction model; inputting test data into the neurology index prediction model for training; and finally, inputting the case data which does not participate in training into the trained neurology index prediction model, and generating a patient neurology disease prediction result from the case data which does not participate in training through the neurology index prediction model. The prediction comprehensiveness is improved through multi-modal data fusion, the detection sensitivity of early functional abnormalities is remarkably improved, the prediction model is more accurate and stable through high-dimensional nonlinear correlation mining and model generalization optimization, non-invasive low cost and clinical interpretability can be considered, invasive detection or high-cost image equipment is not needed, and the method is suitable for large-scale popularization and application. The method is suitable for large-scale screening of basic medical institutions.

Description

Technical Field

[0001] The present invention relates to the field of neurological index prediction, and particularly to a neurological index prediction system and method. Background Art

[0002] Early diagnosis and intervention of neurological diseases (such as Alzheimer's disease, Parkinson's disease, stroke, etc.) are the key to improving the prognosis of patients. However, the existing diagnostic techniques have significant limitations:

[0003] 1. Strong imaging dependence: Clinically, MRI, CT and other imaging examinations are mainly relied on to identify structural lesions. However, such techniques have low sensitivity to early functional abnormalities, and the equipment cost is high and the penetration rate at the grass-roots level is insufficient, resulting in diagnostic delays.

[0004] 2. Low acceptance of invasive detection: Although cerebrospinal fluid detection can provide specific biomarkers (such as β-amyloid protein), lumbar puncture is invasive and the patient compliance is poor, making it difficult to be used as a routine screening method.

[0005] 3. Insufficient prediction ability of single index: Although potential biomarkers such as tau protein and inflammatory factors (such as IL-6) in blood have been found in recent studies, the specificity and sensitivity of single index are limited, and the interaction of multiple systems (such as the association between metabolic syndrome and cerebrovascular lesions) is ignored.

[0006] 4. Poor generalization of traditional models: Prediction models constructed based on statistical methods such as logistic regression are difficult to process high-dimensional and non-linear data, resulting in low applicability across populations. For example, the association between CNTN4 gene mutation and Parkinson's disease varies significantly among different ethnic groups, and such complex interaction features are easily missed by traditional analysis methods.

[0007] Although machine learning techniques have been applied to neurological disease prediction, the existing solutions have the following defects:

[0008] 1. Single data dimension: Most models only integrate imaging or genomic data, and do not systematically incorporate multi-source medical indexes such as metabolic indexes (such as HbA1c) and cardiovascular parameters (such as ambulatory blood pressure). However, studies have shown that abnormal glucose and lipid metabolism can accelerate the process of neurodegenerative diseases.

[0009] 2. Lack of dynamic monitoring: There is a lack of long-term tracking and time-series analysis of physiological indexes, and the progressive changes in the pre-disease stage cannot be captured. For example, the increase in blood pressure variability often occurs several years earlier than the occurrence of stroke.

[0010] 3. Insufficient interpretability: Although black-box models (such as deep neural networks) can improve the accuracy, it is difficult to analyze key prediction factors, which hinders clinical decision support.

[0011] Therefore, there is an urgent need for a neurological indicator prediction system and method that can mine potential correlations between cross-system indicators through advanced algorithms to achieve early warning and accurate risk assessment of neurological diseases. Summary of the invention

[0012] The purpose of the present invention is to provide a neurological index prediction system and method to solve the problems existing in the above-mentioned prior art.

[0013] To achieve the above object, the present invention provides the following solutions:

[0014] The present invention provides a method for predicting neurological indicators, comprising the following steps:

[0015] S1. Data collection: collect case data of patients in the Department of Neurology;

[0016] S2. Data processing: processing the collected case data to form test data;

[0017] S3. Construct a prediction model for neurology indicators;

[0018] S4. Input the test data into the neurology index prediction model for training;

[0019] S5. Input the case data that have not participated in the training into the trained neurology index prediction model, and use the neurology index prediction model to generate the patient's neurology disease prediction results for the case data that have not participated in the training.

[0020] Preferably, in step S1, the case data includes:

[0021] Patient demographic information: age, sex, weight, height, BMI, family medical history;

[0022] Clinical indicator data: blood pressure, heart rate, blood sugar, blood lipids, homocysteine;

[0023] Imaging data: MRI, CT, carotid ultrasound;

[0024] Laboratory examination data: blood routine, coagulation function, C-reactive protein;

[0025] Scale assessment information: NIHSS, MMSE, MoCA, UPDRS;

[0026] Medication record data: history of antiplatelet drugs, anticoagulants, antihypertensive drugs, and thrombolytic therapy;

[0027] Time series data: long-term blood pressure fluctuation trends, blood sugar dynamic monitoring data, and epileptic seizure frequency records.

[0028] Preferably, step S2 comprises:

[0029] S21. Data cleaning, performing missing value processing and outlier detection on the collected data;

[0030] S22. Data standardization and encoding, performing normalization processing and categorical feature processing on the cleaned data;

[0031] S23. Feature engineering, performing time series processing and dimensionality reduction and feature selection on the data.

[0032] Preferably, in step S21, for the missing value processing of continuous variables, the mean or median is used for filling, and for the missing value processing of categorical variables, the mode is used for filling or a new "unknown" category is added, and features with a high missing rate are directly removed.

[0033] Preferably, in step S21, the Z-score method is used for outlier detection.

[0034] Preferably, in step S3, the neurological index prediction model uses the logistic regression algorithm, and its formula is:

[0035]

[0036] where z = β0 + β1x1 + … + βnxn, β0 is the risk baseline, β1…βn are the feature weights, and x1…xn are the feature quantities.

[0037] Preferably, step S4 includes:

[0038] S41. Dividing the test data into a training set, a validation set, and a test set;

[0039] S42. Inputting the training set, the validation set, and the test set into the model for training;

[0040] S43. Optimizing the model by introducing a loss function.

[0041] Preferably, in step S41, the proportion of the training set is 70%, the proportion of the validation set is 15%, and the proportion of the test set is 15%.

[0042] Preferably, in step S43, the cross-entropy loss function is used as the loss function, and its formula is:

[0043] L = -y i logp i ;

[0044] where is the predicted probability distribution pi, and yi is the true label.

[0045] The present invention also provides a neurological index prediction system, including:

[0046] A data acquisition module, used for acquiring the case data of neurological patients;

[0047] A data processing module for processing the collected case data to form test data;

[0048] A model construction module for constructing a neurological index prediction model;

[0049] A training module for inputting the test data into the neurological index prediction model for training;

[0050] A prediction and analysis module for inputting the case data that has not participated in training into the trained neurological index prediction model, and generating the prediction results of the patient's neurological diseases for the case data that has not participated in training through the neurological index prediction model.

[0051] The present invention has achieved the following beneficial technical effects compared with the prior art:

[0052] 1. The neurological index prediction system and method provided by the present invention can improve the comprehensiveness of prediction through multi-modal data fusion:

[0053] The data acquisition module integrates multi-source medical indexes of neurological patients, breaks through the limitations of traditional single-dimensional data, and covers the interaction risk factors of multiple systems including metabolism - cardiovascular - immunity.

[0054] Combining the structured extraction of imaging reports and physiological detection data to construct a cross-scale feature system, significantly improving the detection sensitivity of early functional abnormalities.

[0055] 2. The neurological index prediction system and method provided by the present invention can strengthen the early warning ability through dynamic time series modeling:

[0056] The data processing module supports time series processing of long-term tracking data of indexes such as blood pressure variability and blood glucose fluctuations, captures the progressive physiological offset in the prodromal period of the disease through the logistic regression algorithm, and solves the monitoring blind spot of traditional static models.

[0057] Introduce a dynamic risk scoring mechanism, predict the disease occurrence probability within the next 3 - 5 years according to time series data, and provide an advanced time window for clinical intervention.

[0058] 3. The neurological index prediction system and method provided by the present invention can optimize through high-dimensional non-linear association mining and model generalization:

[0059] The model construction module processes static features through gradient boosting trees, and at the same time uses the self-attention mechanism to analyze the non-linear interaction between multiple indexes, significantly improving the prediction stability across populations.

[0060] The model training module integrates adversarial verification and transfer learning strategies, and reduces the impact of data distribution shift on model performance through adversarial sample generation and domain adaptation training.

[0061] 4. The neurology index prediction system and method provided by the present invention can achieve both non-invasive, low-cost and clinical interpretability:

[0062] The prediction module only relies on routine physical examination indicators and electronic medical record data, without the need for invasive detection or high-cost imaging equipment, and is suitable for large-scale screening in primary medical institutions.

[0063] The model output provides an analysis of feature importance, clarifies key contributing factors, and assists doctors in formulating personalized intervention plans, breaking through the clinical implementation barriers of black-box models. Description of the Drawings

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0065] Figure 1 It is a flowchart of the neurology index prediction method provided by the present invention. Detailed Embodiments

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0067] The purpose of the present invention is to provide a neurology index prediction system and method to solve the problems existing in the prior art.

[0068] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0069] Embodiment 1:

[0070] This embodiment provides a neurology index prediction method, including the following steps:

[0071] S1. Data collection, collecting case data of neurology patients; among them, the case data includes:

[0072] Patient demographic information: age, gender, weight, height, BMI, family medical history;

[0073] Clinical indicator data: blood pressure (systolic / diastolic), heart rate, blood glucose, blood lipids (total cholesterol, LDL, HDL), homocysteine;

[0074] Imaging data: MRI (volume of white matter hyperintensity, degree of brain atrophy), CT (area of infarction), carotid ultrasound (plaque thickness);

[0075] Laboratory test data: blood routine (white blood cells, platelets), coagulation function (INR, D-dimer), C-reactive protein (CRP);

[0076] Scale assessment information: NIHSS (stroke score), MMSE (cognitive function), MoCA (mild cognitive impairment), UPDRS (Parkinson's);

[0077] Medication record data: antiplatelet drugs (aspirin / clopidogrel), anticoagulant drugs (warfarin), antihypertensive drugs, history of thrombolytic therapy;

[0078] Time series data: long-term blood pressure fluctuation trend, dynamic blood glucose monitoring data, epilepsy seizure frequency record;

[0079] Attention should be paid to the relevance and availability of indicators during data collection to avoid missing important data points;

[0080] S2. Data processing, the collected case data is processed to form test data; specifically:

[0081] S21. Data cleaning, perform missing value processing and outlier detection on the collected data; among them, for continuous variables, missing values are filled with the mean or median, for categorical variables, missing values are filled with the mode or a new "unknown" category, and features with a high missing rate are directly removed; outlier detection uses the Z-score method;

[0082] S22. Data standardization and encoding, perform normalization processing and categorical feature processing on the cleaned data;

[0083] S23. Feature engineering, perform time series processing and dimensionality reduction and feature selection on the data; among them, time series processing can use a sliding window to statistically calculate the continuous mean and fluctuation variance, or use Fourier transform to extract periodic features; dimensionality reduction and feature selection can use PCA to combine highly correlated indicators, or use L1 regularization to screen factors that have a significant impact on the target variable (such as recurrence risk), or use the random forest feature evaluation index (such as the volume of white matter hyperintensity) to weight;

[0084] S3. Construct a neurological index prediction model; in this embodiment, the neurological index prediction model uses a logistic regression algorithm, and its formula is:

[0085]

[0086] Among them, z = β0 + β1x1 + … + βnxn, where β0 is the risk baseline, β1…βn are feature weights, and x1…xn are the number of features;

[0087] Of course, when facing the above-mentioned index data, those skilled in the art can also use algorithms such as random forest, XGBoost, and multi-classification, regression, and time series prediction algorithms to design models specifically;

[0088] S4. Input the test data into the neurology index prediction model for training; including:

[0089] S41. Divide the test data into a training set, a validation set, and a test set; among them, the proportion of the training set is 70%, the proportion of the validation set is 15%, and the proportion of the test set is 15%;

[0090] S42. Input the training set, the validation set, and the test set into the model for training;

[0091] S43. Optimize the model by introducing a loss function; the loss function uses the cross-entropy loss function, and its formula is:

[0092] L = -y i logp i ;

[0093] Among them, is the predicted probability distribution of pi, and yi is the true label;

[0094] S5. Input the case data that has not participated in the training into the trained neurology index prediction model, and generate the prediction result of the patient's neurology disease for the case data that has not participated in the training through the neurology index prediction model.

[0095] This embodiment also provides a neurology index prediction system, including:

[0096] A data acquisition module, used to acquire the case data of neurology patients;

[0097] A data processing module, used to process the acquired case data to form test data;

[0098] A model construction module, used to construct a neurology index prediction model;

[0099] A training module, used to input the test data into the neurology index prediction model for training;

[0100] A prediction and analysis module, used to input the case data that has not participated in the training into the trained neurology index prediction model, and generate the prediction result of the patient's neurology disease for the case data that has not participated in the training through the neurology index prediction model.

[0101] The present invention uses specific examples to illustrate the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for predicting neurological indicators, characterized in that: The following steps are involved: S1. Data collection: collect case data of patients in the Department of Neurology; S2. Data processing: processing the collected case data to form test data; S3. Construct a prediction model for neurology indicators; S4. Input the test data into the neurology index prediction model for training; S5. Input the case data that have not participated in the training into the trained neurology index prediction model, and use the neurology index prediction model to generate the patient's neurology disease prediction results for the case data that have not participated in the training.

2. The neurological index prediction method according to claim 1, characterized in that: In step S1, the case data includes: Patient demographic information: age, sex, weight, height, BMI, family medical history; Clinical indicator data: blood pressure, heart rate, blood sugar, blood lipids, homocysteine; Imaging data: MRI, CT, carotid ultrasound; Laboratory examination data: blood routine, coagulation function, C-reactive protein; Scale assessment information: NIHSS, MMSE, MoCA, UPDRS; Medication record data: history of antiplatelet drugs, anticoagulants, antihypertensive drugs, and thrombolytic therapy; Time series data: long-term blood pressure fluctuation trends, blood sugar dynamic monitoring data, and epileptic seizure frequency records.

3. The neurological index prediction method according to claim 1, characterized in that: Step S2 includes: S21. Data cleaning: missing value processing and outlier detection of collected data; S22. Data standardization and coding, normalization and classification feature processing of the cleaned data; S23. Feature engineering: time series processing, dimensionality reduction and feature selection of data.

4. The neurological index prediction method according to claim 3, characterized in that: In step S21, the missing values ​​of continuous variables are filled by the mean or median, and the missing values ​​of categorical variables are filled by the mode or a new "unknown" category is added, and features with high missing rate are directly eliminated.

5. The neurological index prediction method according to claim 3, characterized in that: In step S21, the outlier detection adopts the Z-score method.

6. The neurological index prediction method according to claim 1, characterized in that: In step S3, the neurology index prediction model uses a logistic regression algorithm, and its formula is: Among them, z=β0+β1x1+…+βnxn, β0 is the risk baseline, β1…βn are the feature weights, and x1…xn is the number of features.

7. The neurological index prediction method according to claim 1, characterized in that: Step S4 includes: S41. Divide the test data into a training set, a validation set and a test set; S42. Input the training set, validation set and test set into the model for training; S43. Optimize the model by introducing a loss function.

8. The neurological index prediction method according to claim 7, characterized in that: In step S41, the proportion of the training set is 70%, the proportion of the validation set is 15%, and the proportion of the test set is 15%.

9. The neurological index prediction method according to claim 7, characterized in that: In step S43, the loss function adopts the cross entropy loss function, and its formula is: L=-y i logp i ; Among them, is the predicted probability distribution of pi, and yi is the true label.

10. A neurological index prediction system, characterized by: include: Data collection module, used to collect case data of patients in the Department of Neurology; A data processing module is used to process the collected case data to form test data; Model building module, used to build prediction models for neurology indicators; A training module, used to input test data into the neurology index prediction model for training; The prediction analysis module is used to input the case data that have not participated in the training into the trained neurology index prediction model, and generate the patient's neurological disease prediction results from the case data that have not participated in the training through the neurology index prediction model.

Images