Chronic obstructive pulmonary disease risk assessment system based on nursing information

Abstract

The invention discloses a chronic obstructive pulmonary disease risk assessment system based on nursing information. The chronic obstructive pulmonary disease risk assessment system comprises a data acquisition module, a feature extraction module, a risk prediction module and a chronic obstructive pulmonary disease risk assessment module. The invention belongs to the field of medical information processing, and particularly relates to a chronic obstructive pulmonary disease risk assessment system based on nursing information, which dynamically adjusts the size of a receptive field based on the change rate of the nursing information, adaptively captures short-term fluctuation and long-term trend, reduces excessive attention to stable nursing data, and improves the risk assessment accuracy. The feature expression ability of a key fluctuation area is improved, and then the accuracy of risk assessment of the chronic obstructive pulmonary disease is improved; penalty parameters are dynamically adjusted according to sample distribution, attention to high-risk patients is enhanced, bias to low-risk patients is relieved, a logarithmic smoothing mechanism is introduced into the low-risk patients to reduce noise interference, and inter-class distances are introduced into a target function to avoid inter-class decision overlapping; and the risk assessment effect of the chronic obstructive pulmonary disease is improved.

Description

Technical Field

[0001] This invention relates to the field of medical information processing, specifically to a chronic obstructive pulmonary disease risk assessment system based on nursing information. Background Technology

[0002] Chronic obstructive pulmonary disease (COPD) risk assessment systems are intelligent, multi-dimensional data analysis tools that can capture dynamic changes in a patient's health status in real time. Combining historical static data with real-time monitoring data, they predict the patient's risk level. Through innovative feature extraction and classification methods, they overcome the limitations of traditional assessment methods, providing patients with more accurate predictions and personalized care services, thereby improving patient prognosis and reducing the burden on healthcare. However, general COPD risk assessment systems suffer from problems such as inappropriate handling of static and dynamic characteristics of patients, inadequate capture of key post-care fluctuations, leading to biased and inaccurate assessment results. Furthermore, they often fail to properly handle imbalanced categories in nursing data, have weak ability to identify high-risk and low-risk patients, and are sensitive to noise, resulting in poor assessment effectiveness. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a chronic obstructive pulmonary disease (COPD) risk assessment system based on nursing information. Addressing the problems of general COPD risk assessment systems that fragment and mishandle patients' static and dynamic characteristics, and fail to capture key post-nursing fluctuations, leading to biased and inaccurate assessment results, this solution dynamically adjusts the receptive field size based on the rate of change of nursing information, adaptively capturing short-term fluctuations and long-term trends, thus improving the ability to extract indicators from key time periods. Furthermore, it utilizes the receptive field attention mechanism to generate attention weights, reducing excessive focus on stable nursing data and improving the ability to express features in key fluctuating regions. To improve the accuracy of chronic obstructive pulmonary disease (COPD) risk assessment, this solution addresses the shortcomings of conventional COPD risk assessment systems, such as inadequate handling of class imbalances in nursing data, weak ability to identify high-risk and low-risk patients, and sensitivity to noise, leading to poor assessment results. This solution dynamically adjusts penalty parameters based on sample distribution, enhancing focus on high-risk patients and reducing bias towards low-risk patients. For low-risk patients, a logarithmic smoothing mechanism is introduced to reduce noise interference. Furthermore, inter-class distance is incorporated into the objective function to avoid overlapping decisions between classes and reduce confusion between high-risk and low-to-medium-risk categories, thereby improving the effectiveness of COPD risk assessment.

[0004] The technical solution adopted by the present invention is as follows: The chronic obstructive pulmonary disease risk assessment system based on nursing information provided by the present invention includes a data acquisition module, a feature extraction module, a risk prediction module and a chronic obstructive pulmonary disease risk assessment module;

[0005] The data acquisition module collects historical nursing data;

[0006] The feature extraction module dynamically adjusts the receptive field size and receptive field attention mechanism to adaptively extract local features of key time periods in the nursing information time series, capturing short-term fluctuations and long-term trends.

[0007] The risk prediction module combines static features and multi-scale dynamic features, utilizes attention mechanisms and global feature splicing to comprehensively integrate local details and global trend information, and establishes a risk prediction model.

[0008] The chronic obstructive pulmonary disease (COPD) risk assessment module uses a completed risk prediction model to assess the risk of COPD based on real-time nursing data.

[0009] Furthermore, in the data acquisition module, the historical nursing data includes patient static characteristic data, patient dynamic nursing data, patient physiological status data, time, and risk assessment level; the risk assessment level includes normal, low risk, medium risk, and high risk; the risk assessment level is used as a data label; and the collected data is cleaned, transformed, and standardized to obtain nursing information time series data.

[0010] Furthermore, the feature extraction module specifically includes the following:

[0011] Local feature extraction; extraction of local dependencies in nursing information time series X; extraction of local features using group convolution. Based on the feature distribution of nursing information time series, the size k of the receptive field is dynamically adjusted to achieve adaptive group convolution; and attention weights are generated based on the receptive field attention mechanism. , represented as: ; Finally, the local features are calibrated, as follows: Where Softmax(·) is the Softmax function; ReLU(·) is the ReLU activation function; Norm(·) is the normalization operation; and AvgPool(·) is the average pooling operation. It is a group of convolution operations; and These are the minimum receptive field and the maximum receptive field, respectively. It is the average rate of change in blood oxygen saturation; It is the maximum rate of change in blood oxygen saturation;

[0012] Multi-level feature fusion; for features Multi-scale features are extracted layer by layer, representing: Attention weights are generated based on the SE module. ,express: ; The multi-scale features are adjusted according to their weights, as follows: ;in, It represents multi-scale features extracted layer by layer; S is the number of scales, i is the scale index; Conv(·) is the convolution operation; It is the kernel size; It is an attention score; These are multi-scale features after attention adjustment; It is a static feature The weight matrix; These are bias terms generated by a multilayer perceptron (MLP).

[0013] Global feature fusion; concatenating features from all scales, represented as: ; Static features The dynamic features extracted from the time series are concatenated to form the final feature vector. , represented as: ;in, , and Y represents the features at different scales after weighted adjustment; Concat(·) is the concatenation operation; Attention(·) is the attention mechanism; Y is the concatenated feature.

[0014] Furthermore, the risk prediction module uses the fused feature vector Z generated by the feature extraction module as input data, and performs risk classification based on a multi-class SVM. For each class, a binary SVM is trained to distinguish the current class from other classes, thus obtaining a risk prediction model; specifically, it includes the following:

[0015] Define the objective function as follows: ; The penalty parameters are dynamically adjusted based on the sample distribution, assigning higher weights to high-risk samples, as shown below: ;in, It is the normal vector of the classification hyperplane; It is classification error. This represents the degree of deviation between the o-th sample and the hyperplane classification result; C is the penalty parameter; a is the control parameter; These are the true labels of the samples; It is the sample feature vector; It is the square of the L2 norm; l is the total number of samples; and These are the normal vectors of the classification hyperplanes for categories k1 and y1, respectively;

[0016] Loss handling; imposing a high penalty for misclassification of high-risk patients, expressed as: A small penalty is applied to classification errors of low-risk patients, and a gradient smoothing mechanism is introduced to prevent excessive loss values ​​due to noisy samples; this is expressed as: Where L(·) is the loss function; Smooth(·) is the logarithmic smoothing function;

[0017] Objective function optimization; based on the gradient of the objective function, the weights are updated using the gradient descent algorithm, and the gradient is represented as: This further predicts the patient's risk level for chronic obstructive pulmonary disease, expressed as: ;in, It is the gradient of the objective function; It represents the predicted risk level; T is the matrix transpose; sign(·) is the sign function; Z is the sample set; It represents the sample proportion of each category;

[0018] Classification and determination: The maximum number of iterations and the classification threshold are set in advance. The dataset is divided into a training set and a test set. The risk prediction model is trained based on the training set. When the loss converges or the maximum number of iterations is reached, the risk prediction model training is completed. When the prediction accuracy of the trained risk prediction model on the test set is higher than the classification threshold, the risk prediction model is established. Otherwise, the dataset is re-divided and the initial normal vector parameters are adjusted to train the risk prediction model.

[0019] Furthermore, the chronic obstructive pulmonary disease risk assessment module is based on the established risk prediction model. It collects patient care data in real time, processes it through the feature extraction module, and inputs it into the risk prediction model. The risk assessment level output by the risk prediction model is used as the assessment result. When the assessment result is moderate risk or high risk, the patient is given an early warning.

[0020] The beneficial effects achieved by adopting the above solution are as follows:

[0021] (1) In view of the problems that general chronic obstructive pulmonary disease risk assessment systems have inappropriate handling of patients’ static and dynamic characteristics and inappropriate capture of key fluctuations after nursing care, which leads to one-sided and inaccurate assessment results, this solution dynamically adjusts the size of the receptive field based on the rate of change of nursing information, adaptively captures short-term fluctuations and long-term trends, and improves the ability to extract indicators in key time periods; it uses the receptive field attention mechanism to generate attention weights, reduces excessive attention to stable nursing data, improves the ability to express the characteristics of key fluctuation areas, and thus improves the accuracy of chronic obstructive pulmonary disease risk assessment.

[0022] (2) In view of the problems that general chronic obstructive pulmonary disease risk assessment systems have inappropriate handling of the class imbalance of nursing data, weak ability to identify high-risk and low-risk patients, and sensitivity to noise, which leads to poor assessment results, this scheme dynamically adjusts the penalty parameter according to the sample distribution, enhances the attention to high-risk patients, reduces the bias to low-risk patients, introduces a logarithmic smoothing mechanism for low-risk patients to reduce noise interference, and introduces inter-class distance in the objective function to avoid inter-class decision overlap and reduce confusion between high-risk and medium-low-risk categories; thereby improving the effect of chronic obstructive pulmonary disease risk assessment. Attached Figure Description

[0023] Figure 1 A flowchart illustrating the chronic obstructive pulmonary disease risk assessment system based on nursing information provided by the present invention;

[0024] Figure 2 This is a flowchart of the risk prediction module.

[0025] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] Example 1, see Figure 1 The present invention provides a chronic obstructive pulmonary disease risk assessment system based on nursing information, including a data acquisition module, a feature extraction module, a risk prediction module and a chronic obstructive pulmonary disease risk assessment module;

[0029] The data acquisition module collects historical nursing data and sends the data to the feature extraction module.

[0030] The feature extraction module dynamically adjusts the receptive field size and receptive field attention mechanism to adaptively extract local features of key time periods in the nursing information time series, capturing short-term fluctuations and long-term trends; and sends the data to the risk prediction module.

[0031] The risk prediction module combines static features and multi-scale dynamic features, utilizes attention mechanisms and global feature splicing to comprehensively integrate local details and global trend information to establish a risk prediction model; and sends the data to the chronic obstructive pulmonary disease risk assessment module.

[0032] The chronic obstructive pulmonary disease (COPD) risk assessment module uses a completed risk prediction model to assess the risk of COPD based on real-time nursing data.

[0033] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the data acquisition module, historical nursing data includes patient static characteristic data, patient dynamic nursing data, patient physiological status data, time, and risk assessment level. The risk assessment level includes normal, low risk, moderate risk, and high risk. The risk assessment level is used as a data label. The collected data is cleaned, transformed, and standardized to obtain nursing information time series data. The static characteristic data includes smoking history, occupational exposure, family medical history, disease type of internal medicine and surgical patients, and economic factors. The patient dynamic nursing data includes respiratory system indicators, blood oxygen saturation, cardiovascular indicators, exercise data, environmental risk factors, and sleep characteristics. The patient physiological status data includes blood biochemical indicators and metabolic indicators.

[0034] Example 3, see Figure 1 This embodiment is based on the above embodiment, and the feature extraction module specifically includes the following:

[0035] Local feature extraction; extracting local dependencies in the nursing information time series X, enhancing feature extraction of key periodic fluctuations within 24 hours post-nursing care; using group convolution to extract local features. Based on the feature distribution of nursing information time series, the size k of the receptive field is dynamically adjusted to achieve adaptive group convolution; and attention weights are generated based on the receptive field attention mechanism. , represented as: ; Finally, the local features are calibrated, as follows: Where Softmax(·) is the Softmax function; ReLU(·) is the ReLU activation function; Norm(·) is the normalization operation; and AvgPool(·) is the average pooling operation. It is a group of convolution operations; and These are the minimum receptive field and the maximum receptive field, respectively. It is the average rate of change in blood oxygen saturation; It is the maximum rate of change in blood oxygen saturation;

[0036] Multi-level feature fusion; the risk of chronic obstructive pulmonary disease (COPD) is influenced by multiple levels of features. Multi-scale feature fusion preserves local features while integrating long-term trends, providing comprehensive information for risk assessment. Multi-scale features are extracted layer by layer, representing: Attention weights are generated based on the SE module. ,express: ; The multi-scale features are adjusted according to their weights, as follows: ;in, It represents multi-scale features extracted layer by layer; S is the number of scales, i is the scale index; Conv(·) is the convolution operation; It is the kernel size; It is an attention score; These are multi-scale features after attention adjustment; It is a static feature The weight matrix; These are bias terms generated by a multilayer perceptron (MLP).

[0037] Global feature fusion; concatenating features from all scales, represented as: ; Static features The dynamic features extracted from the time series are concatenated to form the final feature vector. , represented as: ;in, , and Y represents the features at different scales after weighted adjustment; Concat(·) is the concatenation operation; Attention(·) is the attention mechanism; Y is the concatenated feature.

[0038] By performing the above operations, this solution addresses the problems of general chronic obstructive pulmonary disease (COPD) risk assessment systems, such as the fragmentation and improper handling of patients' static and dynamic characteristics, and the inadequate capture of key post-nursing fluctuations, leading to biased and inaccurate assessment results. It dynamically adjusts the receptive field size based on the rate of change in nursing information, adaptively capturing short-term fluctuations and long-term trends, thus improving the ability to extract indicators from key time periods. Furthermore, it utilizes the receptive field attention mechanism to generate attention weights, reducing excessive focus on stable nursing data and improving the ability to express characteristics of key fluctuating regions, thereby enhancing the accuracy of COPD risk assessment.

[0039] Example 4, see Figure 1 and Figure 2This embodiment is based on the above embodiment. The risk prediction module uses the fused feature vector Z generated by the feature extraction module as input data and performs risk classification based on a multi-class SVM. For each category, a binary SVM is trained to distinguish the current category from other categories, thus obtaining the risk prediction model. Specifically, it includes the following:

[0040] Define the objective function as follows: ; The penalty parameters are dynamically adjusted based on the sample distribution, assigning higher weights to high-risk samples, as shown below: ;in, It is the normal vector of the classification hyperplane; It is classification error. This represents the degree of deviation between the o-th sample and the hyperplane classification result; C is the penalty parameter; a is the control parameter; These are the true labels of the samples; It is the sample feature vector; It is the square of the L2 norm; l is the total number of samples; and These are the normal vectors of the classification hyperplanes for categories k1 and y1, respectively;

[0041] Loss handling; imposing a high penalty for misclassification of high-risk patients, expressed as: A small penalty is applied to classification errors of low-risk patients, and a gradient smoothing mechanism is introduced to prevent excessive loss values ​​due to noisy samples; this is expressed as: Where L(·) is the loss function; Smooth(·) is the logarithmic smoothing function;

[0042] Objective function optimization; based on the gradient of the objective function, the weights are updated using the gradient descent algorithm, and the gradient is represented as: This further predicts the patient's risk level for chronic obstructive pulmonary disease, expressed as: ;in, It is the gradient of the objective function; It represents the predicted risk level; T is the matrix transpose; sign(·) is the sign function; Z is the sample set; It represents the sample proportion of each category;

[0043] Classification and determination: The maximum number of iterations and the classification threshold are set in advance. The dataset is divided into a training set and a test set. The risk prediction model is trained based on the training set. When the loss converges or the maximum number of iterations is reached, the risk prediction model training is completed. When the prediction accuracy of the trained risk prediction model on the test set is higher than the classification threshold, the risk prediction model is established. Otherwise, the dataset is re-divided and the initial normal vector parameters are adjusted to train the risk prediction model.

[0044] By performing the above operations, this solution addresses the problems of poor assessment results in general chronic obstructive pulmonary disease (COPD) risk assessment systems, such as inadequate handling of class imbalance in nursing data, weak ability to identify high-risk and low-risk patients, and sensitivity to noise. Instead, it dynamically adjusts the penalty parameter based on sample distribution, enhancing attention to high-risk patients and reducing bias towards low-risk patients. For low-risk patients, a logarithmic smoothing mechanism is introduced to reduce noise interference. Furthermore, inter-class distance is incorporated into the objective function to avoid overlapping decisions between classes and reduce confusion between high-risk and low-to-medium-risk categories, thereby improving the effectiveness of COPD risk assessment.

[0045] Example 5, see Figure 1 This embodiment is based on the above embodiment. The chronic obstructive pulmonary disease risk assessment module is based on the established risk prediction model. It collects patient care data in real time, processes it through the feature extraction module, and inputs it into the risk prediction model. The risk assessment level output by the risk prediction model is used as the assessment result. When the assessment result is moderate risk or high risk, the patient is given an early warning.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0048] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A chronic obstructive pulmonary disease risk assessment system based on nursing information, characterized by: The system includes a data acquisition module, a feature extraction module, a risk prediction module and a COPD risk assessment module; The data collection module collects historical nursing data; The feature extraction module adaptively extracts local features of key periods in the nursing information time series by dynamically adjusting the receptive field size and the receptive field attention mechanism, capturing short-term fluctuations and long-term trends; The risk prediction module combines static features and multi-scale dynamic features, uses attention mechanism and global feature splicing, comprehensively integrates local details and global trend information, and establishes a risk prediction model; The chronic obstructive pulmonary disease risk assessment module implements chronic obstructive pulmonary disease risk assessment on real-time nursing data based on the established risk prediction model; The feature extraction module includes the following contents: local feature extraction; extracting local dependencies in the nursing information time series X; extracting local features using group convolution According to the characteristic distribution of the nursing information time series, the size of the receptive field k is dynamically adjusted to achieve adaptive group convolution; and the attention weight is generated based on the receptive field attention mechanism , expressed as: ; ; Finally, calibrate the local features, expressed as: ; Among them, Softmax(·) is the Softmax function; ReLU(·) is the ReLU activation function; Norm(·) is the normalization operation; AvgPool(·) is the average pooling operation; is a group convolution operation; and are the minimum receptive field and the maximum receptive field respectively; is the average rate of change of blood oxygen saturation; is the maximum rate of change of blood oxygen saturation.

2. The chronic obstructive pulmonary disease risk assessment system based on nursing information according to claim 1, characterized in that: The feature extraction module specifically includes the following contents: Local feature extraction; Multi-level feature fusion; Features Extract multi-scale features layer by layer, indicating: ; Generate attention weights based on SE module ,express: ; ; Adjust the multi-scale features according to the weights, expressed as: ;in, is the multi-scale feature extracted layer by layer; S is the number of scales, i is the scale index; Conv(·) is the convolution operation; is the convolution kernel size; is the attention score; is the multi-scale feature after attention adjustment; It is a static feature The weight matrix of It is the bias term generated by the multi-layer perceptron MLP; Global feature fusion: concatenate features of all scales, expressed as: ; Static features Combined with the dynamic features extracted from the time series to form the final feature vector , expressed as: ;in, , and are the features of different scales after weighted adjustment; Concat(·) is the concatenation operation; Attention(·) is the attention mechanism; Y is the concatenated feature.

3. The chronic obstructive pulmonary disease risk assessment system based on nursing information according to claim 2, characterized in that: The risk prediction module uses the fused feature vector Z generated by the feature extraction module as input data, performs risk classification based on a multi-classification SVM, and for each category, trains a binary classification SVM to distinguish the current category from other categories to obtain a risk prediction model; specifically, it includes the following contents: Define the objective function, expressed as: ; ; Dynamically adjust the penalty parameter according to the sample distribution to set a higher weight for high-risk samples, expressed as: ;in, is the normal vector of the classification hyperplane; is the classification error, Indicates the degree of deviation of the oth sample from the hyperplane classification result; C is the penalty parameter; a is the control parameter; is the true label of the sample; is the sample feature vector; is the square of the L2 norm; l is the total number of samples; and are the normal vectors of the classification hyperplanes of category k1 and category y1 respectively; Loss handling; impose a high penalty on misclassification of high-risk patients, expressed as: ; A small penalty is imposed on misclassification of low-risk patients, and a gradient smoothing mechanism is introduced to prevent excessive loss values ​​due to noise samples; expressed as: ; Where L(·) is the loss function; Smooth(·) is the logarithmic smoothing function; Objective function optimization; Classification judgment; the maximum number of iterations and classification threshold are set in advance, the data set is divided into a training set and a test set, and the risk prediction model is trained based on the training set. When the loss converges or reaches the maximum number of iterations, the risk prediction model training is completed; when the prediction accuracy of the trained risk prediction model for the test set is higher than the classification threshold, the risk prediction model is established, otherwise the data set is re-divided and the initial normal vector parameters are adjusted to train the risk prediction model.

4. The chronic obstructive pulmonary disease risk assessment system based on nursing information according to claim 3, characterized in that: The objective function optimization is specifically as follows: based on the objective function gradient, the weight is updated using the gradient descent algorithm, and the gradient is expressed as: ; Then predict the patient's COPD risk level, expressed as: ;in, is the objective function gradient; is the predicted risk level; T is the matrix transpose; sign(·) is the sign function; Z is the sample set; is the sample proportion of the class.

5. The chronic obstructive pulmonary disease risk assessment system based on nursing information according to claim 4, characterized in that: In the data acquisition module, the historical nursing data includes patient static characteristic data, patient dynamic nursing data, patient physiological condition data, time and risk assessment level; The risk assessment levels include normal, low risk, medium risk and high risk; the risk assessment levels are used as data labels; and the collected data are cleaned, converted and standardized to obtain nursing information time series data.

6. The chronic obstructive pulmonary disease risk assessment system based on nursing information according to claim 5, characterized in that: The COPD risk assessment module is based on an established risk prediction model, collects patient care data in real time, and inputs the data into the risk prediction model after being processed by a feature extraction module. The risk assessment level output by the risk prediction model is used as the assessment result. When the assessment result is a moderate risk or a high risk, the patient is warned.

Images