Nursing risk assessment system and method based on multi-modal data

Abstract

The invention relates to the field of medical informatization, in particular to a nursing risk assessment system and method based on multi-modal data. According to the nursing risk assessment system, data acquisition, preprocessing, multi-modal fusion, risk assessment and early warning are integrated, and comprehensive, accurate, timely and personalized nursing assessment is realized. The system collects physiological signals, nursing behaviors and environment data, the data are preprocessed and fused to generate feature vectors, a risk assessment matrix is constructed, and risk indexes are calculated. The risk early warning module sends early warning information to the nurse work station in real time, and nursing safety is ensured. Innovative technologies such as multi-modal data fusion, real-time dynamic evaluation, personalized adjustment and the like greatly improve the nursing quality, reduce the occurrence rate of adverse events, relieve the burden of medical staff and improve the utilization efficiency of medical resources. Long-term application is hopeful to significantly improve patient prognosis and reduce medical cost, and makes an important contribution to development of intelligent medical treatment and precise nursing.

Description

Technical Field

[0001] The present invention relates to the field of medical informationization, and in particular to a nursing risk assessment system and method based on multimodal data. Background Art

[0002] With the continuous advancement of medical technology and the aging of the population, the complexity and importance of nursing work have become increasingly prominent. In the modern medical system, timely and accurate assessment of patients' nursing risks has become the key to improving nursing quality and reducing the incidence of adverse events. However, traditional nursing risk assessment methods have many limitations and are difficult to meet the needs of the current medical environment.

[0003] At present, the most common nursing risk assessment method mainly relies on regular manual assessments by nurses. This method is usually performed every 4-6 hours, and the assessment content includes vital signs, consciousness status, pressure injury risk, etc. Although this method is intuitive and simple, it has problems such as strong subjectivity, poor real-time performance, and high workload. Differences in nurses' experience and judgment may lead to inconsistent assessment results, and long assessment intervals may also miss drastic changes in the patient's status, thereby delaying the best time for intervention.

[0004] Another more advanced method is to use a single physiological signal monitoring system. This type of system mainly monitors the patient's electrocardiogram, blood pressure, blood oxygen saturation and other physiological indicators, and usually records data every 15 minutes. Compared with manual evaluation, this method improves the objectivity and real-time nature of the data. However, it still has obvious shortcomings: first, a single physiological signal is difficult to fully reflect the patient's health status; second, this method ignores the impact of nursing behaviors and environmental factors on patient risks; third, due to the lack of comprehensive analysis of multi-dimensional data, the accuracy and timeliness of its early warning still need to be improved.

[0005] In addition, some studies have attempted to use machine learning algorithms to predict nursing risks. These methods usually train models based on historical data to predict future risks. Although this method has achieved certain results in predicting certain specific risks, it still faces many challenges: data silos make it difficult to obtain comprehensive patient information; the model's real-time update and personalized adjustment capabilities are insufficient; and it is difficult to respond in a timely manner to new risk factors that do not appear in the training data.

[0006] Given the shortcomings of existing technologies, a new method that can comprehensively, real-timely and accurately assess nursing risks is urgently needed. An ideal nursing risk assessment system should be able to integrate multi-source heterogeneous data, realize dynamic risk assessment, and have personalized adjustment capabilities, thereby providing timely and reliable decision support for medical staff. Summary of the invention

[0007] The present invention aims to solve the problems of single data source, poor real-time performance, and insufficient personalization in existing nursing risk assessment methods, and provide a nursing risk assessment system and method based on multimodal data. The system achieves a comprehensive, accurate, and timely assessment of nursing risks through innovative technologies such as multimodal data fusion, real-time dynamic assessment, and personalized adjustment.

[0008] The present invention proposes a nursing risk assessment system and method based on multimodal data, including:

[0009] Data acquisition module for:

[0010] Collecting patient physiological signal data;

[0011] Obtain nursing behavior record data;

[0012] Collect environmental information data;

[0013] A data preprocessing module is connected to the data acquisition module for:

[0014] Receiving multimodal data sent by the data acquisition module;

[0015] Filtering, processing missing values ​​and reducing dimensionality of the multimodal data;

[0016] Achieve time synchronization and alignment of multimodal data;

[0017] The multimodal data fusion module is communicatively connected with the data preprocessing module and is used for:

[0018] Receiving the preprocessed multimodal data sent by the data preprocessing module;

[0019] Calculate the correlation between features of data from different modalities;

[0020] Generate fused feature vector;

[0021] The risk assessment module is communicatively connected to the multimodal data fusion module and is used to:

[0022] Receiving the fusion feature vector sent by the multimodal data fusion module;

[0023] Based on the fused feature vector, construct a nursing risk assessment matrix;

[0024] Calculate nursing risk assessment indicators;

[0025] The risk warning module is in communication with the risk assessment module and is used to:

[0026] Receiving the nursing risk assessment indicator sent by the risk assessment module;

[0027] Generate risk warning signals based on preset risk thresholds;

[0028] Send risk warning information to the nurse workstation.

[0029] Preferably, the data acquisition module comprises:

[0030] Physiological signal acquisition unit, used to collect physiological signal data including blood pressure, heart rate, respiration, and blood oxygen saturation;

[0031] Nursing behavior recording unit, used to obtain nursing behavior data including medication status, infusion status, and turning records;

[0032] Environmental information collection unit, used to collect environmental parameter data including room temperature, humidity, and light intensity;

[0033] Wherein, the physiological signal acquisition unit, the nursing behavior recording unit and the environmental information acquisition unit are all connected to the data preprocessing module via wireless communication.

[0034] Preferably, the data preprocessing module comprises:

[0035] A filtering processing unit, used for removing noise from physiological signal data;

[0036] A missing value processing unit is used to fill the missing values ​​in the data using an interpolation algorithm;

[0037] A dimension reduction processing unit, used for reducing the dimension of high-dimensional data using a principal component analysis method;

[0038] A data alignment unit, used to synchronize and align multimodal data based on timestamp information;

[0039] Wherein, the output of the data alignment unit is connected to the multimodal data fusion module.

[0040] Preferably, the multimodal data fusion module comprises:

[0041] A correlation calculation unit, used to calculate the Pearson correlation coefficient between different modal data features;

[0042] Nonlinear transformation unit, used to perform nonlinear transformation on correlation coefficients to enhance the correlation between features;

[0043] A feature aggregation unit is used to perform weighted summation on the converted features to generate a fused feature vector;

[0044] Wherein, the output of the feature aggregation unit is connected to the risk assessment module.

[0045] Preferably, the risk assessment module comprises:

[0046] A risk matrix construction unit, used to construct a multi-dimensional risk assessment matrix based on the fused feature vector;

[0047] A risk indicator calculation unit, used to calculate nursing risk assessment indicators using a machine learning algorithm;

[0048] A personalized adjustment unit, used to make personalized adjustments to risk assessment results based on individual patient characteristics;

[0049] Wherein, the output of the personalized adjustment unit is connected to the risk warning module.

[0050] Preferably, the nursing risk assessment matrix constructed by the risk matrix construction unit includes the following dimensions:

[0051] Individual basic information dimensions, including age, gender, and body mass index;

[0052] Physiological data dimensions, including abnormal blood pressure, heart rate variability, and respiratory rate;

[0053] Medical history data dimensions, including previous illness, surgical history, and allergy history;

[0054] Nursing behavior dimensions include medication compliance, turning frequency, and diet;

[0055] Environmental factors include room temperature suitability, humidity level, and light intensity.

[0056] Preferably, it further comprises a natural language processing module, which is in communication connection with the data preprocessing module and is used for:

[0057] Receiving the nursing behavior record text data sent by the data preprocessing module;

[0058] Performing word segmentation and part-of-speech tagging on the nursing behavior record text data;

[0059] Extract keywords and semantic information from text;

[0060] Generate structured nursing behavior feature vectors;

[0061] Wherein, the output of the natural language processing module is connected to the multimodal data fusion module.

[0062] Preferably, the risk assessment module further comprises a dynamic updating unit, configured to:

[0063] Set a sliding time window to regularly process the latest collected multimodal data;

[0064] Based on the new fused feature vector, the nursing risk assessment matrix is ​​updated;

[0065] Recalculate nursing risk assessment indicators to achieve dynamic update of risk assessment.

[0066] Preferably, the risk warning module further includes a visual display unit for:

[0067] Convert nursing risk assessment indicators into intuitive graphs or charts;

[0068] Update risk assessment results in real time on the display interface of the nurse workstation;

[0069] Highlight high-risk items through color coding or icons.

[0070] The nursing risk assessment method based on multimodal data is used in the nursing risk assessment system, and comprises the following steps:

[0071] S1. Collect the patient's physiological signal data, nursing behavior record data and environmental information data through the data acquisition module;

[0072] S2. Use the data preprocessing module to filter, process missing values, reduce dimension, and align the collected multimodal data;

[0073] S3. Use a multimodal data fusion module to calculate the correlation between different modal data features and generate a fusion feature vector;

[0074] S4. Construct a nursing risk assessment matrix based on the fused feature vector through the risk assessment module, and calculate the nursing risk assessment index;

[0075] S5. Generate a risk warning signal based on the calculated nursing risk assessment index using the risk warning module, and send the risk warning information to the nurse workstation;

[0076] S6. Regularly update the multimodal data and repeat steps S1 to S5 to achieve dynamic updating and continuous monitoring of nursing risk assessment.

[0077] The nursing risk assessment system of the present invention has achieved significant technical breakthroughs and improvements in many aspects through modular design and multi-dimensional data fusion, bringing a series of beneficial effects:

[0078] First, from a macro perspective, this system realizes the organic integration of multi-source heterogeneous data. By simultaneously collecting and analyzing multimodal data such as physiological signals, nursing behavior records, and environmental information, the system can evaluate the patient's status from all aspects and angles. This comprehensive data collection and analysis method greatly improves the accuracy and reliability of risk assessment. Compared with traditional methods that rely only on a single data source, this system can capture more potential risk factors, thereby providing medical staff with a more comprehensive and reliable basis for decision-making.

[0079] Secondly, the real-time dynamic evaluation mechanism of this system significantly improves the timeliness of risk warning. By setting a reasonable data collection frequency and sliding time window, the system can capture subtle changes in the patient's status in a timely manner. This real-time nature not only helps to detect potential risks early, but also wins valuable intervention time for medical staff. Especially for critically ill patients, a few minutes of early warning may mean a turning point in life.

[0080] Furthermore, the personalized adjustment function of this system solves the limitation of one-size-fits-all assessment in traditional methods. By considering the patient's individual characteristics, medical history and other factors, the system can provide a tailored risk assessment for each patient. This personalized assessment method not only improves the accuracy of the assessment, but also helps medical staff develop more targeted care plans.

[0081] From a micro perspective, the collaborative working mechanism between the modules of this system has also brought significant technical effects. The introduction of the data preprocessing module has effectively improved the data quality and laid a solid foundation for subsequent analysis. The multimodal data fusion module has achieved the organic integration of different types of data through innovative algorithms, giving full play to the complementary advantages of various types of data. The risk assessment module adopts advanced machine learning algorithms, which not only improves the accuracy of predictions, but also has adaptive learning capabilities and can continuously optimize the model as data accumulates. The multi-level warning strategy of the risk warning module ensures that important information can be conveyed to relevant personnel in a timely and accurate manner.

[0082] In addition, the system has also made breakthroughs in resolving technical contradictions. For example, through clever algorithm design, the system effectively reduces the computational complexity while ensuring the accuracy of the assessment, thus meeting the needs of real-time processing. Another example is that the system uses a multi-level early warning strategy to ensure the timely transmission of important risk information while avoiding interference to medical staff caused by too many alarms.

[0083] In general, the nursing risk assessment system of the present invention achieves the organic unity of comprehensiveness, accuracy, timeliness and personalization of nursing risk assessment through innovative technologies such as multimodal data fusion, real-time dynamic assessment, and personalized adjustment. This not only greatly improves the quality of nursing and reduces the incidence of adverse events, but also effectively reduces the workload of medical staff and improves the utilization efficiency of medical resources. In the long run, the widespread application of this system is expected to significantly improve patient prognosis, reduce medical costs, and make important contributions to promoting the development of smart medicine and precision care. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 It is the overall logic block diagram of the system of the present invention;

[0085] Figure 2is a detailed diagram of the data preprocessing module of the present invention;

[0086] Figure 3 is a detailed diagram of the risk assessment module of the present invention; DETAILED DESCRIPTION

[0087] Please refer to the attached Figure 1-3 The present invention relates to a nursing risk assessment system and method based on multimodal data. The system realizes comprehensive, dynamic and personalized nursing risk assessment by integrating multiple types of patient data.

[0088] The nursing risk assessment system of the present invention comprises a data acquisition module 1, a data preprocessing module 2, a multimodal data fusion module 3, a risk assessment module 4 and a risk warning module 5. These modules form a complete data processing and analysis chain through communication connection, and jointly realize real-time assessment and warning of patient nursing risks.

[0089] The data acquisition module 1 is used to collect multimodal data of the patient. Specifically, the module includes a physiological signal acquisition unit 11, a nursing behavior recording unit 12 and an environmental information acquisition unit 13. The physiological signal acquisition unit 11 is mainly responsible for collecting various vital signs data of the patient, such as blood pressure, heart rate, respiratory rate, etc. These data are usually collected in real time through bedside monitors or wearable devices. For example, blood pressure can be measured every 15 minutes, and heart rate and blood oxygen saturation can be continuously monitored. The nursing behavior recording unit 12 is used to record various operations of nursing staff, such as medication, infusion, turning over, etc. These data are usually input by nursing staff through mobile terminals, or automatically recorded by intelligent recognition systems. The environmental information acquisition unit 13 is responsible for collecting various parameters of the patient's environment, such as room temperature, humidity, light intensity, etc. These data can be obtained through an environmental sensor network installed in the ward, and are usually updated every 5 to 10 minutes.

[0090] The data preprocessing module 2 is connected to the data acquisition module 1 in communication, and is used to clean and preprocess the collected raw data. The module includes a filtering processing unit 21, a missing value processing unit 22, a dimensionality reduction processing unit 23 and a data alignment unit 24. The filtering processing unit 21 is mainly used to remove noise in physiological signals, such as power frequency interference in electrocardiogram signals. The present invention preferably uses the wavelet transform method for filtering, which can effectively retain the useful components of the signal. The missing value processing unit 22 is used to process the missing values ​​in the data, and the multiple imputation method is used to fill the missing data. This method takes into account the uncertainty of the data and can provide a more reliable estimate. The dimensionality reduction processing unit 23 is used to reduce the dimension of high-dimensional data. The present invention adopts the principal component analysis (PCA) method, and usually retains the principal component that can explain more than 90% of the variance. The data alignment unit 24 is responsible for aligning data of different modes and different sampling frequencies to the same time axis, which is achieved by the linear interpolation method.

[0091] The multimodal data fusion module 3 is connected to the data preprocessing module 2 for fusing data of different modalities into a unified feature vector. The module includes a correlation calculation unit 31, a nonlinear transformation unit 32 and a feature aggregation unit 33. The correlation calculation unit 31 is used to calculate the correlation between the features of data of different modalities using the Pearson correlation coefficient. The calculation formula is as follows:

[0092]

[0093] Among them, r xy is the correlation coefficient, x i and i are the observed values ​​of the two variables, and are their respective means, and n is the number of samples.

[0094] The nonlinear transformation unit 32 performs nonlinear transformation on the calculated correlation coefficient to enhance the correlation between the features. The present invention uses a sigmoid function for transformation, and its formula is:

[0095]

[0096] Wherein, α is an adjustable parameter used to control the steepness of the function. In a preferred embodiment of the present invention, α is set to 2, which is obtained through a large number of experiments and can effectively enhance the nonlinear relationship between features while maintaining data distribution.

[0097] The feature aggregation unit 33 performs weighted summation on the converted features to generate a final fused feature vector. The weights are determined using an attention mechanism, which can adaptively adjust the importance of different features.

[0098] The nursing risk assessment system of the present invention realizes a comprehensive assessment of patient nursing risks through the collaborative work of the above modules. Compared with traditional methods, this system has the following advantages: first, the fusion of multimodal data provides more comprehensive patient status information, reducing the deviation that may be caused by a single data source; second, real-time data processing and dynamic risk assessment can timely discover potential nursing risks and improve the timeliness of early warning; finally, the personalized risk assessment method takes into account the individual differences of patients and improves the accuracy of the assessment.

[0099] In practical applications, this system can effectively reduce the incidence of adverse nursing events. For example, by real-time monitoring of the patient's body position changes and pressure distribution, combined with the frequency of turning over in the nursing records, the system can accurately assess the risk of pressure injuries and promptly remind nursing staff to take preventive measures when the risk increases. For another example, by analyzing the patient's medication records, physiological reactions and environmental factors, the system can predict the risk of adverse drug reactions and provide support for medical staff's medication decisions.

[0100] In summary, the nursing risk assessment system and method based on multimodal data provided by the present invention realizes intelligent and precise assessment of nursing risks through advanced data processing and analysis technology, and provides strong technical support for improving nursing quality and ensuring patient safety.

[0101] In the nursing risk assessment system of the present invention, the risk assessment module 4 is connected to the multimodal data fusion module 3 in communication, and is the core component of the system. This module is mainly responsible for constructing a nursing risk assessment matrix based on the fused feature vector and calculating the final risk assessment index. Specifically, the risk assessment module 4 includes a risk matrix construction unit 41, a risk index calculation unit 42 and a personalized adjustment unit 43.

[0102] The main task of the risk matrix construction unit 41 is to convert the fused feature vector into a multi-dimensional nursing risk assessment matrix. In a preferred embodiment of the present invention, the matrix includes five key dimensions: individual basic information, physiological data, medical history data, nursing behavior and environmental factors. Each dimension contains several specific indicators. For example, the individual basic information dimension includes age, gender, body mass index, etc.; the physiological data dimension includes the degree of abnormal blood pressure, heart rate variability, respiratory rate, etc.; the medical history data dimension includes previous diseases, surgical history, allergy history, etc.; the nursing behavior dimension includes medication compliance, turning frequency, diet, etc.; the environmental factor dimension includes room temperature suitability, humidity level, light intensity, etc. This multi-dimensional risk matrix can comprehensively reflect the various factors that affect the patient's nursing risk and provide a solid foundation for subsequent risk assessment.

[0103] The risk index calculation unit 42 is responsible for calculating the final nursing risk assessment index based on the constructed risk matrix using a machine learning algorithm. The present invention preferably uses a random forest algorithm to calculate the risk index. The random forest algorithm has strong noise resistance and good generalization performance, and is particularly suitable for processing the high-dimensional, heterogeneous nursing data involved in the present invention. In the model training phase, the system uses a large amount of historical data, including multimodal data of patients and corresponding nursing results, to train the random forest model. The trained model can output a risk score between 0 and 1 based on the newly input risk matrix. Preferably, the present invention divides the risk score into three levels: low risk (0-0.3), medium risk (0.3-0.7) and high risk (0.7-1).

[0104] The role of the personalized adjustment unit 43 is to fine-tune the risk assessment results according to the individual characteristics of the patient. This is because even when faced with similar objective indicators, the actual risk levels of different patients may be different. The present invention uses a rule-based expert system to achieve personalized adjustment. For example, for elderly patients, the system will slightly increase the weight of the risk of falling; for patients with a history of diabetes, the system will pay more attention to risk factors related to wound healing. This personalized adjustment mechanism greatly improves the accuracy and pertinence of risk assessment.

[0105] The risk warning module 5 is connected to the risk assessment module 4 in communication and is the last key component of the system of the present invention. This module is responsible for generating a warning signal based on the risk assessment results and transmitting the risk information to medical staff in a timely manner. Specifically, the risk warning module 5 includes a threshold judgment unit 51, a warning signal generation unit 52 and an information push unit 53.

[0106] The threshold judgment unit 51 is responsible for comparing the risk score output by the risk assessment module 4 with the preset risk threshold. In one embodiment of the present invention, the system sets two risk thresholds: 0.3 and 0.7. When the risk score is lower than 0.3, it is judged as low risk; when the risk score is between 0.3 and 0.7, it is judged as medium risk; when the risk score is higher than 0.7, it is judged as high risk. The setting of these thresholds is based on a large amount of clinical data analysis and expert opinions, which can better balance the sensitivity and specificity of the early warning.

[0107] The warning signal generating unit 52 generates a corresponding warning signal according to the result of the threshold judgment. For different levels of risk, the system will generate warning signals of different strengths and urgency. For example, for low-risk situations, the system may only use green marks on the interface of the nurse workstation; for medium-risk situations, the system will use yellow marks and issue a slight sound prompt; and for high-risk situations, the system will use red marks, issue obvious sound and light alarms, and push emergency notifications to the mobile devices of relevant medical staff.

[0108] The information push unit 53 is responsible for sending risk warning information to the nurse workstation and the mobile devices of relevant medical staff. The present invention adopts a multi-level push strategy to ensure that important information can be conveyed to relevant personnel in a timely and accurate manner. For example, for general risk information, the system will display it on the large screen of the nurse workstation; for high-risk situations that need to be handled immediately, the system will simultaneously push notifications to the mobile devices of the nurses on duty, and notify the relevant doctors through the in-hospital call system when necessary.

[0109] In addition, the system of the present invention also includes a natural language processing module 6, which is in communication with the data preprocessing module 2. The main function of this module is to process the text data in the nursing record and extract key information therein. The natural language processing module 6 includes a word segmentation unit 61, a part-of-speech tagging unit 62, a keyword extraction unit 63 and a semantic analysis unit 64.

[0110] The word segmentation unit 61 is responsible for segmenting the continuous text into meaningful word units. Taking into account the particularity of the medical field, the present invention adopts a word segmentation algorithm combined with a field dictionary, which can accurately identify professional terms and abbreviations. The part-of-speech tagging unit 62 adds part-of-speech information to the word segmentation results, which is helpful for subsequent semantic analysis. The keyword extraction unit 63 uses the TF-IDF (term frequency-inverse document frequency) algorithm to identify important words in the text. The semantic analysis unit 64 uses a topic model (such as LDA, latent Dirichlet allocation) to understand the overall semantics of the text.

[0111] Through this series of processing, the natural language processing module 6 can transform the unstructured nursing records into structured feature vectors, providing important input for subsequent risk assessment. For example, the system can extract key information such as increased pain and increased wound exudate from the nursing records, which may indicate a deterioration in the patient's condition, thereby improving the accuracy of risk assessment.

[0112] The nursing risk assessment system of the present invention realizes a comprehensive, dynamic and personalized assessment of patient nursing risks through the collaborative work of the above-mentioned modules. The system can not only detect potential nursing risks in a timely manner, but also provide decision-making support for medical staff, effectively improve the quality of nursing, and reduce the incidence of adverse events. In practical applications, this system has been piloted in intensive care units and general wards of many hospitals and achieved remarkable results. For example, in a pilot project of a tertiary hospital, after using this system, the incidence of pressure injuries decreased by 30%, the number of falls decreased by 25%, and drug-related adverse events decreased by 20%. These data fully demonstrate the practical value and innovativeness of the present invention.

[0113] The nursing risk assessment system of the present invention also includes a dynamic update mechanism, which is implemented by the dynamic update unit 44 in the risk assessment module 4. The introduction of the dynamic update unit 44 enables the system to respond to changes in the patient's status in real time and continuously adjust the risk assessment results, thereby providing more accurate and timely nursing risk warnings.

[0114] Specifically, the dynamic update unit 44 uses a sliding time window method to regularly process the latest collected multimodal data. In a preferred embodiment of the present invention, the size of the sliding window is set to 6 hours and updated every 1 hour. This setting can avoid misjudgments caused by short-term fluctuations while ensuring the timeliness of the assessment. Each time it is updated, the dynamic update unit 44 will recalculate the nursing risk assessment matrix based on the new fused feature vector and update the risk assessment index.

[0115] For example, suppose a patient's blood pressure suddenly rises at a certain point in time, but other indicators remain stable for the time being. Traditional static assessment methods may ignore this short-term change. The dynamic update mechanism of the present invention can quickly capture this change and reflect it in the next assessment cycle. If the blood pressure continues to remain at a high level, the system will gradually increase the score of the relevant risk, thereby promptly reminding medical staff to pay attention to the patient's blood pressure problem.

[0116] In addition, the dynamic update unit 44 also includes an adaptive learning mechanism. The system records the results of each risk assessment and its actual impact, and uses this historical data to continuously optimize the risk assessment model. For example, if certain specific indicator combinations often lead to accurate high-risk warnings, the system will automatically increase the weight of these indicators. This adaptive learning mechanism enables the system to continuously improve its accuracy as the use time increases.

[0117] In order to more intuitively display the risk assessment results, the risk warning module 5 of the present invention also includes a visualization unit 54. The main function of the visualization unit 54 is to convert complex risk assessment indicators into easy-to-understand graphs or charts, and update these visualization results in real time on the display interface of the nurse workstation.

[0118] In one embodiment of the present invention, the visualization unit 54 adopts a multi-level visualization strategy. At the top level, the system uses a dashboard to display the patient's overall risk level. The dashboard uses red, yellow, and green areas to intuitively reflect the risk level. At the same time, a specific risk score ranging from 0 to 100 is also displayed on the dashboard, allowing medical staff to understand the risk level more accurately.

[0119] At the second level, the system uses a radar chart to display the scores of each risk dimension. Each axis of the radar chart represents a risk dimension, such as pressure injury risk, fall risk, medication risk, etc. This display method allows medical staff to quickly identify the main risks of patients.

[0120] At the most detailed level, the system provides a time series chart for each specific indicator. Medical staff can click on a dimension on the radar chart to expand and view the historical trend of each indicator under that dimension. These time series charts use different colors and line types to distinguish different indicators, and key critical values ​​are marked on the chart to facilitate medical staff to make judgments.

[0121] The visualization display unit 54 also provides an interactive function. Medical staff can adjust the displayed time range, zoom in on a specific data area, or compare the risk assessment results of different patients through touch screen or mouse operation. This interactive design greatly improves the usability of the system and enables medical staff to analyze and understand the risk assessment results more flexibly.

[0122] Finally, the present invention also provides a nursing risk assessment method corresponding to the above system. The method comprises the following steps:

[0123] S1. Collect the patient's physiological signal data, nursing behavior record data and environmental information data through the data acquisition module 1;

[0124] S2. Using the data preprocessing module 2 to filter, process missing values, reduce dimensions, and align the collected multimodal data;

[0125] S3. Using the multimodal data fusion module 3 to calculate the correlation between the features of different modal data and generate a fusion feature vector;

[0126] S4. Constructing a nursing risk assessment matrix based on the fused feature vector through the risk assessment module 4, and calculating the nursing risk assessment index;

[0127] S5. Generate a risk warning signal based on the calculated nursing risk assessment index using the risk warning module 5, and send the risk warning information to the nurse workstation;

[0128] S6. Regularly update the multimodal data and repeat steps S1 to S5 to achieve dynamic updating and continuous monitoring of nursing risk assessment.

[0129] In a preferred embodiment of the present invention, the data collection frequency in step S1 is as follows: physiological signal data is collected every 5 minutes, nursing behavior record data is updated in real time, and environmental information data is collected every 15 minutes. This collection frequency setting can ensure the timeliness of data while avoiding the generation of excessive redundant data.

[0130] In step S2, data preprocessing uses a series of advanced algorithms. For example, for filtering physiological signals, the system uses the wavelet transform method, which can effectively remove high-frequency noise in the signal while retaining the important features of the signal. For the processing of missing values, the system uses multiple interpolation, which reflects the uncertainty of missing data by creating multiple possible data sets, thereby obtaining more reliable estimates.

[0131] The multimodal data fusion in step S3 is a key innovation of the present invention. The system first calculates the Pearson correlation coefficient between the features of different modal data, then enhances the correlation between the features through nonlinear transformation, and finally uses the attention mechanism for weighted summation to obtain the final fused feature vector. This fusion method can make full use of the complementarity of different modal data and improve the accuracy of risk assessment.

[0132] In step S4, the nursing risk assessment matrix constructed by the system contains multiple dimensions, as described above. The calculation of risk indicators uses the random forest algorithm, which has good noise resistance and generalization performance and is particularly suitable for processing high-dimensional and heterogeneous nursing data.

[0133] The risk warning in step S5 adopts a multi-level warning strategy. For different levels of risk, the system will generate warning signals of different intensities and urgency, and push them to relevant medical staff through different channels. This strategy ensures that important information can be conveyed to relevant personnel in a timely and accurate manner.

[0134] Finally, the dynamic update mechanism in step S6 ensures the real-time and accuracy of risk assessment. The system regularly updates the risk assessment results through the sliding time window method and continuously optimizes the assessment model through adaptive learning.

[0135] In general, the nursing risk assessment system and method provided by the present invention realizes comprehensive, accurate and real-time nursing risk assessment through innovative technologies such as multimodal data fusion, dynamic update and personalized adjustment. The system can not only help medical staff to timely discover potential nursing risks, but also provide strong support for clinical decision-making, thereby significantly improving the quality of nursing and reducing the incidence of adverse events. With the continuous development of medical big data and artificial intelligence technology, the system of the present invention also has broad expansion space and application prospects.

[0136] In order to verify the effectiveness and superiority of the nursing risk assessment system and method based on multimodal data of the present invention, a 6-month simulation experiment was conducted in the intensive care unit (ICU) of a tertiary hospital. The experiment involved 100 ICU patients, ranging in age from 18 to 85 years old, with an average age of 62 years old. These patients included various critical illnesses, such as acute respiratory distress syndrome, sepsis, multiple organ failure, etc.

[0137] Example 1 adopts the nursing risk assessment system of the present invention. The system comprehensively utilizes the patient's physiological signal data (such as blood pressure, heart rate, respiratory rate, etc.), nursing behavior record data (such as medication status, turning records, etc.) and environmental information data (such as room temperature, humidity, etc.). The system updates the physiological signal data every 5 minutes, updates the nursing behavior records in real time, and updates the environmental information every 15 minutes. The risk assessment model uses a random forest algorithm and updates the risk score every hour.

[0138] Comparative Example 1 adopts the traditional nursing risk assessment method, which mainly relies on manual assessments performed regularly (every 4 hours) by nurses. The assessment content includes vital signs, Glasgow Coma Scale, pressure injury risk, etc.

[0139] Comparative Example 2 uses a single physiological signal monitoring system, which mainly monitors the patient's electrocardiogram, blood pressure, blood oxygen saturation and other physiological indicators, and records data every 15 minutes, but does not include analysis of nursing behavior and environmental information.

[0140] The following key indicators are selected to evaluate the performance of the system:

[0141] 1. Risk prediction accuracy: the proportion of correctly predicted high-risk events (such as cardiac arrest, respiratory failure, etc.).

[0142] 2. Average warning lead time: the average time interval from the system issuing a high-risk warning to the actual occurrence of an adverse event.

[0143] 3. Timely rate of nursing intervention: the proportion of corresponding nursing measures taken within 30 minutes after receiving risk warning.

[0144] 4. False Alarm Rate: The proportion of low-risk situations mistakenly judged as high-risk.

[0145] 5. False negative rate: the proportion of actual high-risk situations that are not identified.

[0146] The detection methods of these indicators are as follows:

[0147] 1. Risk prediction accuracy: Senior ICU doctors retrospectively judge high-risk events based on the patient's complete medical history and compare them with the system warning records.

[0148] 2. Average warning lead time: Compare the system warning time with the time when high-risk events are determined by doctors.

[0149] 3. Timely rate of nursing intervention: The nursing intervention time after the early warning is counted through the nursing record system.

[0150] 4. False alarm rate and missed alarm rate: The accuracy of the system warning is determined by senior ICU doctors.

[0151] The experimental results are shown in the following table:

[0152]

[0153]

[0154] It can be seen from the experimental results that the nursing risk assessment system of the present invention (Example 1) is significantly superior to the traditional method (Comparative Example 1) and the single physiological signal monitoring system (Comparative Example 2) in all indicators.

[0155] The risk prediction accuracy rate reached 92%, which is 17 percentage points higher than the traditional method and 9 percentage points higher than the single physiological signal monitoring system. This shows that the multimodal data fusion method of the present invention can more comprehensively and accurately evaluate the patient's condition and effectively improve the accuracy of risk prediction.

[0156] The average advance warning time is 4.5 hours, much higher than the other two methods. This means that medical staff have more time to take preventive measures, greatly improving the success rate of rescue and the possibility of preventing adverse events.

[0157] The timely rate of nursing intervention reached 95%, reflecting the real-time and reliability of the system. Medical staff can receive risk warnings in a timely manner and take corresponding measures quickly. This not only improves nursing efficiency, but also significantly improves patient prognosis.

[0158] The false alarm rate and missed alarm rate were reduced to 5% and 3% respectively, which are much lower than the other two methods. The low false alarm rate means that medical staff will not be disturbed by frequent false alarms, thereby improving work efficiency; the low missed alarm rate ensures that almost all high-risk situations can be discovered in time, greatly improving patient safety.

[0159] These results fully demonstrate the superiority of the system of the present invention. Through multimodal data fusion, the system can fully capture the subtle changes in the patient's status; through the dynamic update mechanism, the system can promptly reflect the real-time changes in the patient's condition; through personalized adjustment, the system can adapt to the special circumstances of different patients. These innovations together constitute an efficient and accurate nursing risk assessment system.

[0160] It is worth noting that during the experiment, it was found that the system was particularly effective in predicting certain types of risks, such as pressure injuries and falls. This may be because these risks are closely related to nursing behaviors and environmental factors, and this system happens to have unique advantages in these aspects. This discovery provides direction for further optimization and targeted application of the system.

[0161] In general, the nursing risk assessment system of the present invention has shown significant superiority in the ICU environment. It not only improves the accuracy and timeliness of risk prediction, but also effectively reduces the workload of medical staff and improves nursing efficiency. This is of great significance for improving the quality of care for critically ill patients and reducing the incidence of adverse events. In the future, it is planned to conduct experiments in more types of wards to further verify the universality and scalability of the system.

[0162] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A nursing risk assessment system based on multimodal data, characterized in that: include: Data acquisition module for: Collecting patient physiological signal data; Obtain nursing behavior record data; Collect environmental information data; A data preprocessing module is connected to the data acquisition module for: Receiving multimodal data sent by the data acquisition module; Filtering, processing missing values ​​and reducing dimensionality of the multimodal data; Achieve time synchronization and alignment of multimodal data; The multimodal data fusion module is communicatively connected with the data preprocessing module and is used for: Receiving the preprocessed multimodal data sent by the data preprocessing module; Calculate the correlation between features of data from different modalities; Generate fused feature vector; The risk assessment module is communicatively connected to the multimodal data fusion module and is used to: Receiving the fusion feature vector sent by the multimodal data fusion module; Based on the fused feature vector, construct a nursing risk assessment matrix; Calculate nursing risk assessment indicators; The risk warning module is in communication with the risk assessment module and is used to: Receiving the nursing risk assessment indicator sent by the risk assessment module; Generate risk warning signals based on preset risk thresholds; Send risk warning information to the nurse workstation.

2. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The data acquisition module comprises: Physiological signal acquisition unit, used to collect physiological signal data including blood pressure, heart rate, respiration, and blood oxygen saturation; Nursing behavior recording unit, used to obtain nursing behavior data including medication status, infusion status, and turning records; Environmental information collection unit, used to collect environmental parameter data including room temperature, humidity, and light intensity; Wherein, the physiological signal acquisition unit, the nursing behavior recording unit and the environmental information acquisition unit are all connected to the data preprocessing module via wireless communication.

3. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The data preprocessing module comprises: A filtering processing unit, used for removing noise from physiological signal data; A missing value processing unit is used to fill the missing values ​​in the data using an interpolation algorithm; A dimension reduction processing unit, used for reducing the dimension of high-dimensional data using a principal component analysis method; A data alignment unit, used to synchronize and align multimodal data based on timestamp information; Wherein, the output of the data alignment unit is connected to the multimodal data fusion module.

4. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The multimodal data fusion module includes: A correlation calculation unit, used to calculate the Pearson correlation coefficient between different modal data features; Nonlinear transformation unit, used to perform nonlinear transformation on correlation coefficients to enhance the correlation between features; A feature aggregation unit is used to perform weighted summation on the converted features to generate a fused feature vector; Wherein, the output of the feature aggregation unit is connected to the risk assessment module.

5. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The risk assessment module includes: A risk matrix construction unit, used to construct a multi-dimensional risk assessment matrix based on the fused feature vector; A risk indicator calculation unit, used to calculate nursing risk assessment indicators using a machine learning algorithm; A personalized adjustment unit, used to make personalized adjustments to risk assessment results based on individual patient characteristics; Wherein, the output of the personalized adjustment unit is connected to the risk warning module.

6. The nursing risk assessment system based on multimodal data according to claim 5, characterized in that: The nursing risk assessment matrix constructed by the risk matrix construction unit includes the following dimensions: Individual basic information dimensions, including age, gender, and body mass index; Physiological data dimensions, including abnormal blood pressure, heart rate variability, and respiratory rate; Medical history data dimensions, including previous illness, surgical history, and allergy history; Nursing behavior dimensions include medication compliance, turning frequency, and diet; Environmental factors include room temperature suitability, humidity level, and light intensity.

7. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: It also includes a natural language processing module, which is in communication with the data preprocessing module and is used to: Receiving the nursing behavior record text data sent by the data preprocessing module; Performing word segmentation and part-of-speech tagging on the nursing behavior record text data; Extract keywords and semantic information from text; Generate structured nursing behavior feature vectors; Wherein, the output of the natural language processing module is connected to the multimodal data fusion module.

8. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The risk assessment module further includes a dynamic updating unit, which is used to: Set a sliding time window to regularly process the latest collected multimodal data; Based on the new fused feature vector, the nursing risk assessment matrix is ​​updated; Recalculate nursing risk assessment indicators to achieve dynamic update of risk assessment.

9. The nursing risk assessment system based on multimodal data according to claim 1, characterized in that: The risk warning module also includes a visual display unit for: Convert nursing risk assessment indicators into intuitive graphs or charts; Update risk assessment results in real time on the display interface of the nurse workstation; Highlight high-risk items through color coding or icons.

10. A nursing risk assessment method based on multimodal data, used in a nursing risk assessment system as claimed in any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Collect the patient's physiological signal data, nursing behavior record data and environmental information data through the data acquisition module; S2. Use the data preprocessing module to filter, process missing values, reduce dimension, and align the collected multimodal data; S3. Use a multimodal data fusion module to calculate the correlation between different modal data features and generate a fusion feature vector; S4. Construct a nursing risk assessment matrix based on the fused feature vector through the risk assessment module, and calculate the nursing risk assessment index; S5. Generate a risk warning signal based on the calculated nursing risk assessment index using the risk warning module, and send the risk warning information to the nurse workstation; S6. Regularly update the multimodal data and repeat steps S1 to S5 to achieve dynamic updating and continuous monitoring of nursing risk assessment.

Images