Artificial intelligence-based trauma care plan generation method and system

Abstract

The embodiment of the present application provides a method and system for generating a trauma care plan based on artificial intelligence, which obtains the first basic medical data of the target object to be treated for trauma care, wherein the first basic medical data includes individual characteristic data, genetic data, trauma text data, and medical history text data. The first basic medical data is subjected to data preprocessing and reconstruction operations to generate the second basic medical data of the target object. The second basic medical data is input into a pre-built nursing analysis model to obtain at least one subsequent care stage label for the target object. Based on the subsequent care stage label, the nursing knowledge base, and the historical care data of multiple reference objects, the recommended trauma care plan for the target object is determined, thereby improving the accuracy of the generation of the trauma care plan.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a method and system for generating a trauma care plan based on artificial intelligence. Background Art

[0002] Trauma care is a crucial part of the medical field, and its quality directly affects the patient's recovery and quality of life. In the process of formulating existing trauma care programs, there are many problems and shortcomings.

[0003] Currently, most trauma care plans rely primarily on physicians' clinical experience and limited routine examination data. Doctors typically focus solely on routine information such as the patient's basic vital signs and the superficial appearance of the injury, paying insufficient attention to the patient's individual characteristics, genetic profile, and detailed medical history. This one-sided approach to information acquisition makes it difficult for care plans to fully account for each patient's uniqueness and provide precision care. For example, genetic differences between patients may lead to different responses to the same treatment, but existing care plans often overlook this critical factor.

[0004] Existing methods for categorizing and determining trauma care stages rely primarily on subjective judgment by physicians. Differences in experience and judgment criteria among physicians result in inaccurate and inconsistent categorization of trauma care stages. This inaccurate categorization can lead to mismatches between care plans and patients' actual conditions, compromising care effectiveness.

[0005] Furthermore, existing trauma care plans often lack the ability to fully utilize historical nursing data during their development. The updating and improvement of nursing knowledge bases is also relatively lagging, unable to promptly reflect the latest research findings and clinical experience. This makes care plans lack scientific and forward-looking development, making it difficult to meet the increasingly diverse care needs of patients.

[0006] In summary, existing technologies have problems in the formulation of trauma care plans, such as insufficient data utilization, lack of personalization of care plans, and inaccurate judgment of care stages. Summary of the Invention

[0007] In view of this, the purpose of this application is to provide a trauma care plan generation method and system based on artificial intelligence to improve the accuracy of trauma care plan generation.

[0008] In a first aspect, the present application provides a method for generating a trauma care plan based on artificial intelligence, comprising:

[0009] Acquiring first basic medical data of a target subject to be treated for trauma care, wherein the first basic medical data includes individual characteristic data, gene data, trauma text data, and medical history text data;

[0010] performing data preprocessing and reconstruction operations on the first basic medical data to generate second basic medical data of the target object;

[0011] inputting the second basic medical data into a pre-built nursing analysis model to obtain at least one subsequent nursing stage label of the target object;

[0012] A recommended wound care plan for the target subject is determined based on the subsequent care stage label, a care knowledge base, and historical care data of a plurality of reference subjects.

[0013] In a second aspect, the present application provides an artificial intelligence-based wound care plan generation system, which includes a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the artificial intelligence-based wound care plan generation system implements the aforementioned artificial intelligence-based wound care plan generation method.

[0014] In a third aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer program product, which, when executed by a processor, is used to implement the method as described in any one of the first aspects.

[0016] The artificial intelligence-based trauma care plan generation method and system provided in this application overcomes the limitations of existing technologies that rely solely on limited routine examination data by comprehensively acquiring primary basic medical data, including individual characteristic data, genetic data, trauma text data, and medical history text data. By collecting patient information from multiple dimensions, this method provides a rich and comprehensive data foundation for developing personalized trauma care plans. Data preprocessing and reconstruction operations are performed on the primary basic medical data to generate secondary basic medical data, effectively addressing existing data issues such as data fragmentation, noise, and erroneous information, improving data quality and usability, and ensuring the accuracy and reliability of subsequent analysis. The secondary basic medical data is input into a pre-built nursing analysis model to obtain at least one subsequent care stage label for the target patient, avoiding the drawbacks of existing methods that rely on subjective judgment and lead to inaccurate care stage classification, and can more objectively and accurately determine the patient's care stage. Based on the subsequent care stage label, a nursing knowledge base, and historical nursing data from multiple reference subjects, a recommended trauma care plan for the target patient is determined. This enables personalized and precise generation of trauma care plans, fully utilizing historical nursing data and the latest nursing knowledge, improving the scientific nature and effectiveness of care plans, and better meeting the personalized care needs of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] Figure 1 A flowchart of a method for generating a trauma care plan based on artificial intelligence provided in an embodiment of the present application;

[0019] Figure 2 A schematic diagram of the structure of an artificial intelligence-based trauma care plan generation system provided in an embodiment of the present application.

[0020] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0021] To help those skilled in the art better understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. It is apparent that the described embodiments are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by those skilled in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0022] The terms "first," "second," and so on, in the specification and claims of the present invention and the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or end comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed therein, or may optionally include other steps or elements inherent to such process, method, product, or end.

[0023] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0024] Figure 1 This is a flow chart of a method for generating a wound care plan based on artificial intelligence provided in an embodiment of the present application. It should be understood that in other embodiments, the order of some steps in the method for generating a wound care plan based on artificial intelligence in this embodiment can be shared with each other according to actual needs, or some steps can be omitted or maintained. Figure 1 As shown, the method may include the following steps:

[0025] Step S110: Acquire first basic medical data of a target subject to be treated for trauma care, wherein the first basic medical data includes individual characteristic data, gene data, trauma text data, and medical history text data.

[0026] The process of generating a trauma care plan begins with obtaining the target subject's primary basic medical data. Individual characteristic data encompasses various aspects of the target subject's physical characteristics. From a morphological perspective, this includes the proportional relationships between body parts, such as the length ratios between the torso and limbs. These proportions can be accurately measured using specialized body measurement tools, such as an anthropometric tape. Body fat distribution can also be determined using a body fat meter, which uses the bioelectrical impedance principle to measure fat content in different parts of the body. Physiological functional characteristics include cardiopulmonary function. For example, the heart's pumping function can be comprehensively assessed by measuring cardiac electrical activity using an electrocardiogram (ECG) and cardiac structure and function using echocardiography. Respiratory function can be determined using a spirometer to measure vital capacity, ventilation volume, and other parameters. Furthermore, metabolic function also forms part of the individual characteristic data. For example, basal metabolic rate (BMR) can be calculated using indirect calorimetry, which measures the target subject's oxygen consumption and carbon dioxide production at rest. Alternatively, the BMR can be calculated using relevant formulas. Alternatively, individual characteristic data may include the target subject's age and gender.

[0027] The acquisition of genetic data relies on advanced biotechnology. During the sample collection phase, biological samples from the subject can be obtained using a variety of methods. For blood samples, specialized needles and tubes are used, strictly adhering to aseptic procedures. For saliva samples, the subject rinses their mouth using a specialized saliva collection device. After sample collection, pre-processing is performed, such as centrifugation to extract nucleated cells such as hemocytes. Next, using gene sequencing technologies, such as next-generation sequencing, DNA is fragmented and sequenced using a sequencer, generating a large amount of genetic sequence data. Bioinformatics analysis of this genetic sequence data reveals gene structural and expression characteristics. For example, analyzing the promoter region of a gene can help understand its regulatory mechanisms, while comparing the expression levels of different genes can identify genes associated with trauma recovery.

[0028] Trauma text data primarily comes from doctors' detailed records of target subjects' injuries. For external injuries, these records can include the appearance of the injury, such as its shape (round, oval, or irregular); its edges (uniform or uneven); and its color (including any bruising, redness, or swelling). For internal injuries, these records can be generated based on imaging modalities such as X-rays, CT scans, and magnetic resonance imaging (MRI). X-rays can clearly demonstrate skeletal injuries, and the records can include fracture location and type (e.g., transverse or oblique fractures). CT scans and MRIs can provide a more detailed picture of soft tissue injuries, including injury location (e.g., muscle, tendon, or ligament); and injury severity (e.g., mild, moderate, or severe).

[0029] Medical history text data can be obtained from the target subject's medical records. The medical records record the target subject's past medical history, including the diagnostic characteristics of the disease, the characteristics of the treatment process, and the characteristics of the treatment effect. For chronic diseases that have been suffered, such as hypertension, the medical records will record the blood pressure characteristics at the time of diagnosis, including the numerical characteristics of systolic and diastolic blood pressure; the characteristics of the drugs used during the treatment process, such as the type, dosage, and frequency of medication; the characteristics of the treatment effect, such as the characteristics of blood pressure control, whether the target blood pressure value has been reached. For acute diseases that have been suffered, such as pneumonia, the medical records will record the symptom characteristics at the time of onset, such as the degree of fever, the frequency of coughing, etc.; the characteristics of the treatment measures taken during the treatment process, such as the type of antibiotics used, whether oxygen therapy was performed, etc.; the recovery characteristics after treatment, such as whether the symptoms have disappeared, whether the lung inflammation has been absorbed, etc.

[0030] Step S120: performing data preprocessing and reconstruction operations on the first basic medical data to generate second basic medical data of the target object.

[0031] After obtaining the first basic medical data, since the data may have various problems, such as inconsistent data format, missing data, data noise, etc., it is necessary to perform data preprocessing and reconstruction operations on it to generate second basic medical data that is more suitable for subsequent analysis.

[0032] The first step in data preprocessing is data cleaning. For individual characteristic data, the cleaning process first addresses inconsistent data formats. For example, different measuring devices may record data in different units. For example, some devices may record height in centimeters, while others may record height in meters. In this case, all height data must be converted to the same unit. This can be accomplished by programming a program to perform the conversion based on the unit conversion relationship. For genetic data, there may be data noise caused by sequencing errors. Base quality filtering can be used when cleaning genetic data. The sequencer assigns a quality score to each base. By setting a quality threshold, bases with a quality score below the threshold are considered unreliable and filtered out. Trauma text data may contain typos and inappropriate expressions. Natural language processing techniques, such as lexical analysis and syntactic analysis, can be used to process the text. First, a lexical analysis tool is used to break the text into words. Words are then spelled correctly and any spelling errors are corrected. Next, syntactic analysis is used to check the structure of sentences and adjust any inappropriate sentence structures. Medical history text data may contain duplicate data. By establishing a data index, each record can be uniquely identified, and then the key information of different records, such as disease name, diagnosis time, etc., can be compared to merge or delete duplicate records.

[0033] Handling missing values ​​is also a crucial step in data cleaning. For missing values ​​in individual characteristic data, if the data is continuous, such as missing blood pressure values, methods such as mean imputation and interpolation can be used. For example, mean imputation can calculate the mean of all non-missing values ​​for the characteristic and then use this mean to impute missing values. For missing values ​​in genetic data, due to the unique nature of genetic data, methods based on genetic databases can be used. By comparing with known genetic databases, gene sequences similar to the target gene can be found, and then the missing values ​​can be filled with corresponding data from these similar gene sequences. For missing values ​​in trauma text data, if the description of certain symptoms is missing, reasonable inferences can be made based on the doctor's clinical experience and other relevant information. For example, if a medical record lacks a description of the degree of pain after trauma, but records the severity of the injury and the use of analgesics, the pain level can be inferred based on analgesic usage. For missing values ​​in medical history text data, if the records of treatment efficacy are missing, subsequent follow-up visits or communication with the patient can be used to complete the missing values.

[0034] Data normalization is another important step in data preprocessing. Different features in individual feature data may have different value ranges and dimensions, which can affect subsequent analysis. For example, height values ​​may range from tens of centimeters to over two meters, while weight values ​​may range from a few kilograms to hundreds of kilograms. To eliminate this discrepancy, the Z-score normalization method can be used. For each feature, the mean and standard deviation are calculated. Then, for each data point, the mean is subtracted and divided by the standard deviation to obtain the normalized data. Similar normalization methods can be used for genetic data, as gene expression levels can vary significantly. The expression values ​​of each gene are normalized to make expression levels comparable across genes. For trauma text data and medical history text data, text vectorization can be used for normalization. Common methods for converting text data into vector representations include the bag-of-words model and the TF-IDF model. The bag-of-words model treats each word in the text as a feature and counts the frequency of each word in the text to form a vector. The TF-IDF model is based on the bag-of-words model and takes into account the importance of words in the entire text collection. By calculating the word frequency and inverse document frequency, it obtains a vector that better reflects the text characteristics.

[0035] Data reconstruction involves integrating and reorganizing different types of data based on data preprocessing to generate secondary basic medical data. For individual feature data, genetic data, trauma text data, and medical history text data, feature concatenation can be used for reconstruction. First, the standardized individual feature data vector, genetic data vector, trauma text data vector, and medical history text data vector are concatenated in a specific order to form a longer vector. During the concatenation process, care must be taken to ensure that the dimensions of each vector are properly aligned. Furthermore, to highlight the importance of different data types, different weights can be assigned to each vector. For example, trauma text data may be more important in generating trauma care plans and therefore warrant a higher weight. During concatenation, the trauma text data vector is multiplied by a larger weight coefficient before being concatenated with the other vectors. Alternatively, feature fusion can be used for data reconstruction. For example, a feature fusion model can be constructed, taking different types of data as input. The model learns the relationships between the different data types and outputs a fused feature vector. This fused feature vector integrates information on individual characteristics, genetic characteristics, trauma characteristics, medical history characteristics, etc., and is more suitable for subsequent nursing analysis models.

[0036] Step S130: inputting the second basic medical data into a pre-built nursing analysis model to obtain at least one subsequent nursing stage label of the target object.

[0037] After generating the target subject's secondary basic medical data, it is fed into a pre-built nursing analysis model to obtain a label for the target subject's subsequent care stage. The nursing analysis model outputs at least one label for the target subject's subsequent care stage, representing the target subject's likely future care stage and providing a crucial basis for developing a recommended trauma care plan.

[0038] For example, the nursing analysis model can output three subsequent nursing stage labels for the target subject, for example, label 1 for the emergency stage, label 2 for the recovery stage, label 3 for the convalescent stage, and so on.

[0039] Step S140: Determine a recommended wound care plan for the target subject based on the subsequent care stage label, the care knowledge base, and the historical care data of multiple reference subjects.

[0040] Step S141: Based on the subsequent care stage label and the historical care data of the multiple reference objects, obtain the candidate care measure text of the target object from the care knowledge base, wherein the candidate care measures included in the candidate care measure text include at least one of fluid replacement measures, analgesic measures, anti-infection measures, and rehabilitation training measures.

[0041] After obtaining the target subject's subsequent care stage label, the system then combines historical care data from multiple reference subjects to identify candidate nursing action texts related to the target subject from the nursing knowledge base. The nursing knowledge base is a database that stores a wealth of nursing knowledge and experience, including various nursing actions for different care stages and trauma situations.

[0042] The historical nursing data of multiple reference subjects records the nursing measures adopted by other subjects at similar nursing stages and the corresponding nursing results. By analyzing this historical nursing data, we can understand which nursing measures are more effective at specific nursing stages.

[0043] Based on the target patient's subsequent care phase label, matching nursing measures are searched in the nursing knowledge base. At the same time, successful nursing measures in the same or similar care phases in historical nursing data are referenced to screen candidate nursing measures suitable for the target patient. These candidate nursing measures may include fluid replacement measures, such as different types of fluid replacement solutions, the speed and volume of fluid replacement; analgesic measures, such as the type, dosage, and method of administration of analgesics; anti-infection measures, such as the type and duration of antibiotics used; and rehabilitation training measures, such as the training items, intensity, and frequency.

[0044] Step S142: extracting key medical features of the target object from the first basic medical data of the target object.

[0045] Extract key medical features from the target subject's primary basic medical data. This primary basic medical data includes individual characteristic data, genetic data, trauma text data, and medical history text data. From this data, filter out key information relevant to trauma care.

[0046] For individual feature data, key medical features may include, for example, basic physiological indicators of the body, such as heart rate, blood pressure, respiratory rate, etc. These indicators can reflect the physical condition of the target subject. Key medical features in genetic data may include, for example, gene expression characteristics related to trauma recovery, which can affect the target subject's response to nursing measures. Key medical features in trauma text data may include, for example, the location, type, and severity of the trauma. This information is crucial for selecting appropriate nursing measures. Key medical features in medical history text data may include, for example, the target subject's past illnesses and treatments. This information can help determine whether the target subject has potential health risks and influence the selection of nursing measures.

[0047] Step S143: Match the key medical features and the candidate nursing measure texts using a preset matching rule, and select target candidate nursing measures from the candidate nursing measure texts whose matching degree with the key medical features is greater than or equal to a preset matching degree threshold.

[0048] Step S1431: Obtain a set of multidimensional feature vectors corresponding to each candidate nursing measure in the candidate nursing measure text, wherein the multidimensional feature vector set includes a physiological impact vector, a risk level vector, an expected efficacy vector, and a resource consumption vector, wherein the physiological impact vector is generated by normalizing the quantitative impact parameters of the candidate nursing measure on the preset physiological indicator system and then splicing them, and the risk level vector is obtained by logarithmically transforming the complication probability distribution corresponding to the candidate nursing measure and reducing its dimension.

[0049] For each candidate nursing measure in the candidate nursing measure text, its corresponding multidimensional feature vector set needs to be obtained. The multidimensional feature vector set contains feature information of multiple aspects and is used to comprehensively describe the characteristics of the candidate nursing measures.

[0050] The physiological impact vector reflects the impact of the candidate nursing measure on the preset physiological indicator system. The preset physiological indicator system can include common physiological indicators such as heart rate, blood pressure, and blood oxygen saturation. For each candidate nursing measure, there will be a series of quantitative impact parameters, which represent the degree of impact of the nursing measure on each physiological indicator. In order to make the impact parameters of different physiological indicators comparable, these parameters need to be normalized. Normalization can map the value of the parameter to a specific range, such as between 0 and 1. Then, the normalized parameters are spliced ​​to obtain the physiological impact vector.

[0051] The risk rating vector measures the potential risk associated with a candidate care action. Each candidate care action has a corresponding complication probability distribution, representing the probability of each complication occurring. To process these probability distributions, a logarithmic transformation is first performed to convert the probability values ​​into logarithmic form. This transformation compresses the range of probability values ​​while highlighting differences between them. Next, dimensionality reduction techniques, such as principal component analysis (PCA), are used to reduce the high-dimensional probability distribution data to a lower dimension, yielding the risk rating vector.

[0052] The expected efficacy vector represents the expected therapeutic effect of a candidate nursing intervention. Based on historical nursing data and clinical experience, the efficacy of each candidate nursing intervention can be evaluated to obtain a series of evaluation indicators. These evaluation indicators are combined into the expected efficacy vector.

[0053] The resource consumption vector reflects the resources required by the candidate nursing measures, such as human, material, and financial resources. The resources required for each candidate nursing measure can be quantitatively evaluated, and the evaluation results can be combined into a resource consumption vector.

[0054] Step S1432: Extract the dynamic response vector corresponding to the key medical feature of the target object. The dynamic response vector is generated by projecting the product matrix of the gene expression regulation coefficient corresponding to the gene data in the first basic medical data and the tissue damage degree score corresponding to the trauma text data into a preset vector space after singular value decomposition.

[0055] Dynamic response vectors corresponding to key medical features are extracted from the target subject's primary basic medical data. The dynamic response vectors combine information from genetic data and trauma text data to reflect the target subject's potential response to nursing interventions.

[0056] First, we obtain the gene expression regulation coefficient from the gene data. This coefficient indicates the degree to which gene expression levels are regulated by various factors. Then, we obtain the tissue damage score from the trauma text data. The tissue damage score provides a comprehensive assessment based on information such as the location, type, and severity of the trauma.

[0057] Multiplying the gene expression regulation coefficient with the tissue injury score yields a product matrix that encompasses the combined effects of gene and trauma information.

[0058] Perform singular value decomposition on the product matrix. Singular value decomposition is a matrix decomposition technique that decomposes a matrix into the product of three matrices: the left singular matrix, the singular value matrix, and the right singular matrix. Through singular value decomposition, important information can be extracted from the matrix, more clearly demonstrating the key relationship between genes and trauma.

[0059] Finally, the decomposed matrix is ​​projected onto a preset vector space. This is a predefined vector space into which the projection operation maps the information in the matrix, resulting in a dynamic response vector. This dynamic response vector integrates information from both genetic and traumatic text data, reflecting the target subject's potential response to nursing interventions under the combined influence of genetics and trauma. Analyzing the dynamic response vector can provide a better understanding of the target subject's physiological state and potential response to different nursing interventions, thereby providing a basis for selecting appropriate nursing interventions. For example, if the dynamic response vector has a large value in a certain dimension, it indicates that the target subject may have a more pronounced response to the nursing intervention in that dimension, and relevant nursing interventions can be prioritized when formulating a nursing plan.

[0060] Step S1433: Calculate the bidirectional weighted similarity between the dynamic response vector and each vector in the multidimensional feature vector set, where the positive weight is determined based on the medical priority strategy predefined in the nursing knowledge base, and the negative weight is dynamically adjusted based on the actual effect deviation rate of similar nursing measures in the historical nursing data.

[0061] Calculate the bidirectional weighted similarity between the dynamic response vector and each vector in the multidimensional feature vector set. Bidirectional weighted similarity takes into account both forward and reverse weights to more accurately measure the similarity between the dynamic response vector and the multidimensional feature vector.

[0062] Positive weights are determined based on a predefined medical prioritization strategy within the nursing knowledge base. This strategy assigns different weights to different feature dimensions based on different medical conditions and care goals. For example, in some cases, physiological impact may be more important, so the positive weight of the physiological impact vector will be relatively high.

[0063] The inverse weight is dynamically adjusted based on the actual effect deviation rate of similar nursing measures in historical nursing data. The actual effect deviation rate indicates the difference between the actual effect of similar nursing measures and the expected effect. If the actual effect deviation rate of a nursing measure is large, it indicates that the effect of the nursing measure is unstable, and the corresponding inverse weight will be relatively low.

[0064] For each vector in the multidimensional feature vector set, its similarity with the dynamic response vector is calculated. This similarity can be calculated using common similarity metrics such as cosine similarity and Euclidean distance. The calculated similarity is then multiplied by the forward and reverse weights to obtain bidirectional weighted similarity.

[0065] Step S1434: construct a dynamic matching surface based on the joint distribution characteristics of the bidirectional weighted similarity, perform topological clustering on the candidate nursing measure texts on the dynamic matching surface, and screen out candidate nursing measures that are in a preset high-density area and meet the curvature continuity constraint as the target candidate nursing measures.

[0066] A dynamic matching surface is constructed based on the joint distribution characteristics of bidirectional weighted similarities. This surface is a high-dimensional surface whose shape and characteristics are determined by the distribution of bidirectional weighted similarities. Each point on this surface represents a candidate care measure, and the height of the point indicates the degree of match between the candidate care measure and the target patient.

[0067] Topological clustering is performed on the candidate nursing action texts on the dynamic matching surface. Topological clustering is a clustering method based on spatial topology. It can group points on the surface according to their spatial distribution. Through topological clustering, candidate nursing actions with high matching scores can be grouped together.

[0068] Candidate care measures that fall within a pre-defined high-density region and satisfy the curvature continuity constraint are selected as target candidate care measures. The pre-defined high-density region represents a high-matching region on the dynamic matching surface; the candidate care measures within this region have a relatively high match with the target object. The curvature continuity constraint requires a certain degree of continuity in the distribution of candidate care measures on the surface, avoiding the selection of isolated, unstable candidate care measures.

[0069] Step S144: determining the risk level of the target candidate nursing measure based on the target candidate nursing measure and the key medical characteristics, adjusting the target candidate nursing measure based on the risk level of the target candidate nursing measure, and generating a trauma nursing measure for the target object.

[0070] After selecting candidate nursing measures, the risk level of the candidate measures is determined based on the target patient's key medical characteristics. Key medical characteristics reflect the target patient's physical condition and health risks, and are important for assessing the risk of nursing measures.

[0071] For each candidate care measure, the potential risk is assessed by considering its own risk factors and the target patient's key medical characteristics. For example, if the target patient already has certain underlying diseases, certain care measures may aggravate the symptoms of these diseases, thereby increasing the risk.

[0072] Based on the assessed risk level, target candidate nursing measures are adjusted. If the risk level is high, the intensity of the nursing measures can be appropriately reduced or the method of nursing measures can be adjusted; if the risk level is low, the original nursing measures can be maintained or the intensity can be appropriately increased.

[0073] By adjusting target candidate nursing measures, trauma nursing measures suitable for target subjects are generated. These trauma nursing measures should take into account both nursing effects and the physical tolerance and risk factors of the target subjects.

[0074] Step S145: Integrate the wound care measures of the target object to generate the wound care recommendation plan for the target object.

[0075] The generated trauma care measures are integrated to generate a recommended trauma care plan for the target patient. The integration process needs to consider the coordination and rationality of the care measures to ensure that the various care measures can cooperate with each other to achieve the best care effect.

[0076] First, trauma care measures should be classified and prioritized. They can be categorized by type, such as fluid replacement, analgesia, anti-infection, and rehabilitation training, and then prioritized based on the chronological order and importance of care.

[0077] Next, a nursing plan is developed. This plan includes detailed information such as the timing, frequency, and methods of implementing nursing measures. For example, for fluid rehydration measures, the interval, rate, and volume of rehydration should be clearly defined; for rehabilitation training measures, a training schedule and a plan for increasing the intensity of the training should be established.

[0078] Finally, the classified trauma care measures and nursing plans are integrated to form a complete trauma care recommendation plan, which provides specific guidance and suggestions for the target patient's trauma care.

[0079] Optionally, after determining the recommended wound care plan for the target subject based on the subsequent care stage label, the care knowledge base, and the historical care data of multiple reference subjects, the embodiment of the present application may further include the following subsequent steps:

[0080] Step S150: monitoring the multi-source heterogeneous data of the target object during the process of the target object being cared for by the trauma care recommendation program, wherein the multi-source heterogeneous data includes physiological parameter data, rehabilitation function data, medical imaging data, and nursing execution data.

[0081] After determining the recommended trauma care plan for the target patient, it is necessary to monitor the multi-source heterogeneous data of the target patient during the care process. Multi-source heterogeneous data comes from different data sources and has different data types and formats.

[0082] Physiological parameter data can be collected in real time through various physiological monitoring devices, such as heart rate monitors, blood pressure monitors, and blood oxygen saturation monitors. These devices can continuously record the target subject's heart rate, blood pressure, blood oxygen saturation and other physiological parameters, reflecting the target subject's physical condition.

[0083] Rehabilitation function data is obtained by assessing the target patient's rehabilitation function. For example, for patients with fractures, their recovery can be assessed through methods such as joint range of motion measurement and muscle strength testing. Rehabilitation function data can help determine the effectiveness of nursing care plans and adjust nursing measures in a timely manner.

[0084] Medical imaging data is acquired through imaging techniques such as X-rays, CT scans, and magnetic resonance imaging (MRI). These data can clearly show the internal structure of the target body and the recovery of injuries, making them crucial for determining the extent of injury healing and the presence of complications.

[0085] Nursing execution data is the data recorded by nursing staff during the implementation of nursing measures, including the implementation time, frequency, and method of the nursing measures. Nursing execution data can reflect the implementation of nursing measures and ensure that the nursing plan is implemented correctly.

[0086] Step S160: determining the target object's disease condition change characteristics and rehabilitation effect characteristics based on the multi-source heterogeneous data of the target object.

[0087] Analyze the collected multi-source heterogeneous data to determine the target subject's disease progression and rehabilitation outcome characteristics. For example, AI models can be used to analyze multi-source heterogeneous data to output the target subject's disease progression and rehabilitation outcome characteristics.

[0088] For physiological parameter data, time series analysis can be used to determine whether the target subject's physical condition is stable. For example, abnormal fluctuations in physiological parameters such as heart rate and blood pressure may indicate a change in the target subject's condition. Statistical indicators such as the mean and standard deviation of physiological parameters can be calculated and compared with the normal range to assess the target subject's physical condition.

[0089] Rehabilitation function data can directly reflect the rehabilitation outcomes of the target subjects. By comparing rehabilitation function assessment results at different time points, such as increases in joint range of motion and muscle strength, the effectiveness of rehabilitation training measures can be assessed. Indicators such as the improvement rate of rehabilitation function can be calculated to quantify the rehabilitation effect.

[0090] Medical imaging data can provide intuitive information about disease progression. By comparing medical images taken at different times, such as those showing fracture healing and tissue repair, we can assess disease trends. By analyzing and extracting features from medical images, we can identify key features of disease progression.

[0091] Nursing execution data can help analyze the relationship between nursing interventions, disease progression, and recovery outcomes. Inadequate implementation of nursing interventions can negatively impact recovery outcomes; conversely, effective implementation can promote improvement. By analyzing nursing execution data, we can identify factors influencing recovery outcomes and provide a basis for adjusting nursing plans.

[0092] Specifically, for physiological parameter data, assume that the collected heart rate data is a time series set, denoted as HR = {hr_1, hr_2, ..., hr_n}, where hr_i represents the heart rate value at the i-th time point. To analyze heart rate fluctuations, the mean (HR) and standard deviation (Std(HR)) of this time series can be calculated. The mean reflects the average level of heart rate, while the standard deviation reflects the degree of heart rate fluctuation. The calculated mean and standard deviation are compared with the normal heart rate range. If the mean deviates significantly from the normal range or the standard deviation is too large, it indicates abnormal heart rate fluctuations, which may indicate unstable medical condition of the target patient.

[0093] For rehabilitation function data, take joint range of motion as an example. Assume that the target subject's joint range of motion is measured at different time points t_1, t_2, ..., t_m, resulting in a joint range of motion data set ROM = {rom_1, rom_2, ..., rom_m}. The difference in joint range of motion between adjacent time points, Δrom_i = rom_{i+1} - rom_i, can be calculated. These differences reflect changes in joint range of motion. If most Δrom_i values ​​are positive, it indicates that joint range of motion is gradually increasing and rehabilitation training is achieving some results. Conversely, if a large number of negative values ​​are present, adjustments to the rehabilitation training program may be necessary.

[0094] Medical imaging data may require different focus for different types of trauma. For example, for fractures, the focus can be on observing changes in the fracture line and the formation of callus. Medical images taken at different times can be compared to identify features such as the location and width of the fracture line and the size and density of the callus. By analyzing changes in these features, the progress of fracture healing can be assessed.

[0095] For nursing execution data, assuming nursing measures include taking medication on time and regular rehabilitation training, information such as the execution time, frequency, and quality of these measures can be recorded. For example, for the nursing measure of taking medication on time, the deviation between the actual time of each medication and the prescribed time can be recorded, as well as whether the medication was taken on time and in the correct dosage. If significant deviations in the execution of a nursing measure are found, it may be necessary to strengthen nursing staff training or adjust nursing arrangements.

[0096] Step S170: Determine an adjustment target for the recommended trauma care plan based on the characteristics of the disease change and the characteristics of the rehabilitation effect.

[0097] After determining the characteristics of the target subject's condition changes and rehabilitation effects, the adjustment goals of the trauma care recommendation plan can be determined based on these characteristics.

[0098] If the characteristics of the patient's condition change indicate a worsening of their condition, such as increased abnormal fluctuations in physiological parameters or poor recovery from trauma as shown by medical imaging, adjustments may include strengthening treatment, changing treatment methods, or adding additional nursing care. For example, if the heart rate remains elevated and blood pressure is unstable, increased cardiovascular monitoring and treatment may be necessary; if the fracture site is healing slowly, adjustments to the intensity of rehabilitation training or the use of adjunctive treatments may be necessary.

[0099] If the rehabilitation outcome characteristics indicate that rehabilitation is progressing smoothly, the goal of adjustment may be to maintain the current care plan and optimize it appropriately. For example, if joint range of motion has significantly increased and muscle strength has gradually recovered, the difficulty and intensity of rehabilitation training can be appropriately increased to further promote recovery.

[0100] If certain nursing interventions are found to be ineffective, adjustments may be made to improve their implementation or to replace them. For example, if a medication is found to have significant side effects and ineffectiveness, a medication change could be considered. If rehabilitation training methods are not suitable for the target patient, the training program could be adjusted.

[0101] Specifically, if physiological parameter data indicates that the target subject's blood pressure remains consistently above the normal range, an adjustment target can be set to bring blood pressure back within the normal range within a certain period of time. To achieve this goal, consideration can be given to increasing the dose of antihypertensive medication or switching to a more effective antihypertensive medication, while also increasing the frequency of blood pressure monitoring.

[0102] When functional rehabilitation data show that the target subject's muscle strength is recovering slowly, the goal may be to improve muscle strength over the next period of time. This can be achieved by increasing the intensity and frequency of rehabilitation training or introducing new rehabilitation training programs.

[0103] When nursing execution data show that a certain nursing measure often has execution deviations, the adjustment goal can be to improve the execution accuracy of the nursing measure.

[0104] Step S180: updating the wound care recommendation plan based on the adjustment target.

[0105] Based on the identified adjustment goals, the recommended trauma care plan is updated. The update process requires comprehensive consideration of various factors to ensure that the new care plan can achieve the adjustment goals without placing an excessive burden on the target patients.

[0106] If the goal is to strengthen treatment, the updated care plan can include increasing the dose of medication, extending the duration of treatment, or introducing new treatments. For example, if anti-infection treatment is needed, the dose of antibiotics can be increased or a more effective antibiotic can be used. If fracture healing is needed, the frequency of physical therapy can be increased or new physical therapy techniques can be introduced.

[0107] If the goal is to optimize rehabilitation training, the content, intensity, and frequency of rehabilitation training can be adjusted when updating the care plan. For example, if the goal is to improve joint mobility, the number of range of motion training items and duration can be increased; if the goal is to increase muscle strength, the intensity and number of strength training sets can be increased.

[0108] If the goal is to improve the implementation of nursing measures, when updating the nursing plan, you can improve the nursing process, strengthen the training of nursing staff, or increase the supervision mechanism. For example, you can develop more detailed nursing operation standards, conduct regular training and assessment of nursing staff, and establish a feedback mechanism for nursing implementation.

[0109] When updating care plans, it's also important to consider the individual differences and tolerance of the target patient. Different targets may respond differently to treatment and care measures, so adjustments to the plan need to be tailored to the specific circumstances of the target patient. For example, for older or more frail targets, increasing treatment intensity requires caution to avoid placing excessive strain on the body.

[0110] Specifically, let's assume the goal is to control the target subject's blood sugar to a normal range within a certain period of time. When updating the care plan, you can first analyze the current diet and medication status. If dietary control is insufficient, you can adjust the diet to reduce the intake of high-sugar foods and increase dietary fiber. If medication is ineffective, you can consider increasing the dosage of the hypoglycemic medication or changing the type of medication. At the same time, increase the frequency of blood sugar monitoring to adjust the treatment plan in a timely manner.

[0111] Assume the goal is to improve the effectiveness of rehabilitation training. When updating the care plan, adjustments can be made to the rehabilitation training plan based on the patient's physical condition and recovery progress. For example, targeted rehabilitation training programs, such as balance and coordination training, can be added. The intensity and frequency of training can be adjusted, and the difficulty can be gradually increased based on the patient's response. Furthermore, guidance and supervision of rehabilitation training should be strengthened to ensure that the patient correctly performs the training movements.

[0112] The following is a detailed introduction to the construction process of the aforementioned nursing analysis model. In this application, the construction process of the nursing analysis model may include the following steps:

[0113] Step S210: constructing an initial nursing analysis model and obtaining sample nursing data for training the initial nursing analysis model.

[0114] Among them, the sample nursing data includes static sample data and time series sample data. The static sample data includes sample individual characteristic data, sample gene data, sample trauma text data, sample medical history text data, and sample medical image data. The time series sample data includes rehabilitation progress data, physiological indicator data, and historical nursing logs. The initial nursing analysis model includes a neural network branch, an LSTM branch, and a classifier branch.

[0115] When building the initial care analysis model, the model structure must be determined first. The initial care analysis model consists of a neural network branch, an LSTM branch, and a classifier branch. The neural network branch primarily processes static sample data, the LSTM branch processes time series sample data, and the classifier branch performs classification based on the features output by the first two branches to obtain labels for subsequent care stages and rehabilitation outcomes.

[0116] When acquiring sample nursing data, static sample data is acquired in a similar manner to primary basic medical data. Sample individual characteristic data is also acquired through physical measurements and physiological function testing, including physical morphological characteristics and physiological function characteristics. Sample genetic data is acquired from biological samples of the sample subject using gene sequencing technology. Sample trauma text data is acquired by doctors through detailed records of the sample subject's trauma, including characteristics of the trauma's appearance, size, and degree of injury. Sample medical history text data is acquired from the sample subject's medical records and records the sample subject's past medical history, including diagnostic characteristics of the disease, treatment process characteristics, and treatment effect characteristics. Sample medical image data is acquired through various imaging examination methods, such as X-rays, CT scans, and magnetic resonance imaging. The images contain structural information within the sample subject's body, which is important for determining the severity of the trauma and recovery status.

[0117] The rehabilitation progress data in the time series sample data is obtained through regular examinations and assessments during the recovery process of the sample subjects. For example, patients with fractures undergo regular X-ray examinations to observe the healing of the fracture and record the blurring of the fracture line and the formation of callus. Physiological indicator data, such as heart rate, blood pressure, and blood oxygen saturation, are collected in real time through various physiological monitoring devices. Historical nursing logs in the time series sample data record nursing measures, nursing time, patient responses, and other information recorded by nursing staff during the care process.

[0118] Step S220: Input the static sample data into the neural network branch, and obtain a static sample feature vector corresponding to the static sample data through the multi-layer convolution layer, maximum pooling layer, and fully connected layer in the neural network branch.

[0119] Step S221: input the static sample data into the neural network branch, and obtain a sample feature map corresponding to the static sample data through the convolution operation of the multi-layer convolution layer in the neural network branch.

[0120] After static sample data is input into a neural network branch, it first passes through multiple convolutional layers for convolution operations. A convolutional layer consists of multiple convolution kernels, each of which is a small matrix. During the convolution operation, the convolution kernel slides over the input data, multiplies it element-by-element with a local region of the input data, and then sums the results to produce a convolution result. For the sample's individual feature data, assuming it is a multidimensional vector, the convolution kernel slides over this vector, extracting feature relationships between different dimensions. For sample medical image data, the convolution kernel slides over the image's pixel matrix, extracting local features such as edge features and texture features. Through convolution operations with multiple convolution kernels, multiple different convolution results can be obtained, which form the sample feature map. Each convolution kernel can be considered a feature extractor, and different convolution kernels can extract different types of features. For example, a convolution kernel can extract local features associated with different types of trauma and the physiological condition of the target subject.

[0121] Step S222: downsampling the sample feature map through the maximum pooling layer to generate a downsampled sample feature map.

[0122] After obtaining the sample feature map, it is fed into the max pooling layer for downsampling. The max pooling layer divides the sample feature map into multiple non-overlapping regions and selects the maximum value in each region as the output for that region. This aims to reduce the dimensionality of the feature map while preserving important feature information. For example, a large sample feature map can be reduced in size through max pooling, reducing computational effort. Furthermore, by selecting the maximum value, the most significant features in the feature map are retained, enhancing the robustness of the model.

[0123] Step S223: Mapping the downsampled sample feature map to a preset dimension through the fully connected layer to obtain the static sample feature vector corresponding to the static sample data.

[0124] After processing by the max pooling layer, a downsampled sample feature map is obtained. This downsampled sample feature map is then fed into a fully connected layer. Each neuron in the fully connected layer is connected to all elements in the downsampled sample feature map. The fully connected layer maps the downsampled sample feature map into a vector space of preset dimensions, yielding a static sample feature vector. The choice of preset dimensions needs to be adjusted based on the specific task and model requirements. Through the mapping of the fully connected layer, the feature information in the sample feature map can be integrated into a single vector, facilitating subsequent processing and analysis.

[0125] Optionally, a residual connection mechanism can be introduced into the neural network branches to address the vanishing gradient problem in deep networks and improve feature transfer efficiency. Alternatively, an attention mechanism can be introduced to dynamically adjust the weights corresponding to the channel features in the neural network branches to highlight key features (i.e., key nursing indicators such as lactate levels and blood oxygen saturation).

[0126] Step S230: Input the time series sample data into the LSTM branch, and obtain a time series sample feature sequence corresponding to the time series sample data through the bidirectional LSTM layer in the LSTM branch.

[0127] Step S231: input the time series sample data into the LSTM branch, and obtain multiple time series sample subsequences corresponding to the time series sample data according to a preset sliding window.

[0128] After the time series sample data is input into the LSTM branch, it is partitioned according to the preset sliding window. The preset sliding window size and step size are two important parameters. Assume the sliding window size is a fixed time period, and the step size represents the distance of each slide. For example, for physiological indicator data, with one hour as the time unit, the sliding window size is set to 6 hours, and the step size is set to 1 hour. Then, starting from the starting point of the time series sample data, 6 hours of data are taken as a time series sample subsequence. The window is then slid backward by 1 hour, and another 6 hours of data are taken as the next time series sample subsequence. This process continues until the entire time series sample data is traversed, resulting in multiple time series sample subsequences.

[0129] Step S232: recursively process the time series sample subsequence through the forward LSTM in the bidirectional LSTM layer to generate a previous hidden state sequence.

[0130] The acquired time series sample subsequence is input into the forward LSTM in the bidirectional LSTM layer. The forward LSTM is a recurrent neural network that processes sequential data and remembers historical information within the sequence. In the forward LSTM, the input for each time step is the time series sample subsequence element of the current time step, along with the hidden state from the previous time step. Through recursive computation, the forward LSTM updates the hidden state of the current time step. As time steps progress, a series of hidden states are generated, forming the previous hidden state sequence. This previous hidden state sequence contains information about the time series sample subsequence from previous to next, reflecting the historical development trend of the sequence.

[0131] Step S233: recursively process the time series sample subsequence through the backward LSTM in the bidirectional LSTM layer to generate a subsequent hidden state sequence.

[0132] In addition to the forward LSTM, the bidirectional LSTM layer also includes a backward LSTM. The backward LSTM processes data in a similar manner to the forward LSTM, but from back to front. The time series sample subsequences are fed into the backward LSTM in reverse order. Each time step receives the reversed elements of the current time step and the hidden state of the next time step. Through recursive computation, the backward LSTM updates the hidden state of the current time step, generating a subsequent hidden state sequence. This subsequent hidden state sequence contains information about the time series sample subsequence from back to front, reflecting the future development trend of the sequence.

[0133] Step S234: concatenate the preceding hidden state sequence and the succeeding hidden state sequence to obtain a time series sample feature sequence corresponding to the time series sample data.

[0134] The preceding hidden state sequence and the subsequent hidden state sequence are concatenated to obtain a time series sample feature sequence. This concatenation can be done by connecting the two sequences along a dimension, so that the time series sample feature sequence contains both forward and backward information from the time series sample subsequences. This allows for more comprehensive capture of feature information in the time series sample data, improving the model's ability to process time series data.

[0135] Step S240: According to each time step of the time series sample feature sequence, the static sample feature vector and the time series sample feature sequence are concatenated to generate a target joint feature vector.

[0136] After obtaining the static sample feature vector and the time series sample feature sequence, they are concatenated for each time step of the time series sample feature sequence. Specifically, the static sample feature vector is concatenated with the vector for each time step in the time series sample feature sequence. This way, the vector for each time step contains both the feature information of the static sample and the time series feature information for the current time step. This concatenation method generates a target joint feature vector. This target joint feature vector combines the feature information of both the static sample data and the time series sample data, providing a richer feature representation for subsequent classification tasks.

[0137] Step S250: inputting the target joint feature vector into the classifier branch to obtain a subsequent care stage label set and a rehabilitation effect label set corresponding to the sample care data.

[0138] The target joint feature vector is input into the classifier branch. The classifier branch can use a common classification algorithm, such as the softmax classifier. The softmax classifier processes the target joint feature vector and calculates the probability of each category (follow-up care stage and rehabilitation outcome category). Based on the probability values, the category with the highest probability is selected as the prediction result. For example, for follow-up care stages, there may be categories such as initial care stage, mid-term care stage, and late care stage; for rehabilitation outcomes, there may be categories such as good, fair, and poor. Through processing in the classifier branch, a follow-up care stage label set and a rehabilitation outcome label set corresponding to the sample nursing data are obtained. For example, for multiple fractures, the follow-up care stage label set may include labels for the emergency period, recovery period, and rehabilitation period. The rehabilitation outcome label set may include labels for physiological indicator recovery (such as stable vital signs, fluctuating vital signs, abnormal vital signs, or resolved inflammation, persistent inflammation), functional recovery ability labels (such as bed dependency, bed mobility, ambulatory, or unable to care for oneself, partial independence, or basic independence), etc.

[0139] Step S260: The initial nursing analysis model is trained based on the subsequent nursing stage label set, the rehabilitation effect label set, the preset verification set, the target loss function, and the target optimizer to obtain the nursing analysis model.

[0140] After obtaining the subsequent care phase label set and the rehabilitation outcome label set corresponding to the sample nursing data, we can begin training the initial nursing analysis model. The training process requires combining a preset validation set, a target loss function, and a target optimizer to continuously adjust the model parameters to improve model performance.

[0141] A validation set is a subset of sample data used to validate the model's performance on unseen data. The data distribution of the validation set should be similar to that of the training set to accurately assess the model's generalization ability. During training, the initial nursing analysis model's predictions on the validation set data are compared with the true labels in the validation set to evaluate the model's performance.

[0142] The target loss function is used to measure the difference between the model's prediction and the true label. For classification tasks, a commonly used loss function is the cross-entropy loss function. The cross-entropy loss function can measure the difference between two probability distributions. The loss value is obtained by calculating the cross-entropy between the category probability distribution predicted by the model and the true category probability distribution. The smaller the loss value, the closer the model's prediction result is to the true label. During the training process, the goal is to minimize the value of the loss function, that is, to make the model's prediction result as accurate as possible. For regression tasks, the mean squared error loss function can be used. For example, for label classification prediction in the follow-up care stage, the cross-entropy loss function can be used, and for rehabilitation effect prediction, the mean squared error loss function can be used. That is, the target loss function includes the cross-entropy loss function and the mean squared error loss function. For example, the target loss function can be obtained by fusion of preset adjustment parameters to comprehensively optimize the performance of the model, that is, the target loss function = α*cross entropy loss function + (1-α)*mean squared error loss function, where α can be set to 0.7, for example.

[0143] The objective optimizer adjusts the model's parameters based on the loss function. Common optimizers include stochastic gradient descent (SGD) and its variants, such as Adagrad, Adadelta, and Adam. The Adam optimizer, for example, combines the advantages of Adagrad and RMSProp to adaptively adjust the learning rate for each parameter. In each training iteration, the optimizer calculates the update amount for each parameter based on the gradient of the loss function and then updates the model's parameters.

[0144] The specific training steps are as follows:

[0145] First, the sample nursing data is divided into a training set and a validation set. The division ratio can be adjusted according to the specific situation, but it must be ensured that the validation set can fully represent the distribution of the sample data.

[0146] Next, initialize the parameters of the initial nursing analysis model. There are various ways to initialize these parameters, such as random initialization, Xavier initialization, and He initialization. Different initialization methods are suitable for different model structures and activation functions. Choosing the right initialization method can accelerate model convergence.

[0147] Next, training iterations begin. In each iteration, a batch of sample data from the training set is input into the initial nursing analysis model. After processing by the neural network branch, LSTM branch, and classifier branch, a predicted set of labels for subsequent nursing stages and rehabilitation outcomes is obtained.

[0148] Calculate the loss between the predicted label set and the true label set using the target loss function. For example, the cross entropy loss function takes the predicted class probability distribution and the true class probability distribution as input to calculate the loss value.

[0149] The target optimizer is used to calculate the parameter update amount based on the loss value. The optimizer uses a specific algorithm to calculate the amount of update required for each parameter based on the gradient information of the loss function.

[0150] Based on the calculated update amount, update the parameters of the initial nursing analysis model. Add the corresponding update amount to each parameter to complete the parameter update.

[0151] After completing a certain number of iterations, the model is validated using the validation set. The validation set data is fed into the updated model to obtain predictions. Evaluation metrics such as accuracy, recall, and F1 value are then calculated between the predictions and the true labels in the validation set.

[0152] If the evaluation indicators meet the preset conditions, such as the accuracy reaches a certain threshold, the model training is considered completed and the nursing analysis model is obtained; otherwise, the training iterations continue until the conditions are met.

[0153] Optionally, during training, you can employ techniques to improve model performance. For example, you can use a learning rate decay strategy to gradually reduce the learning rate as training progresses, allowing the model to converge more stably. You can also employ an early stopping strategy to prevent overfitting by stopping training early if the evaluation metric on the validation set does not improve significantly within a certain number of iterations.

[0154] In one possible implementation, during the model training process, the parameters of the model training can be updated by the group trend factor to further improve the accuracy of the model output. The method of the present application can also include the following steps:

[0155] Step S310: Acquire target historical nursing data of a plurality of sample subjects that meet a preset similarity condition within a preset time period.

[0156] The preset similarity condition includes that the trauma similarity is less than or equal to a first threshold, and / or that the individual feature similarity is less than or equal to a second threshold, and the historical nursing data includes trauma nursing measures and rehabilitation progress.

[0157] To further optimize the training of the initial nursing analysis model, target historical nursing data for multiple sample subjects that meet preset similarity criteria within a preset time period can be obtained. The preset time period can be set based on actual needs, for example, selecting nursing data from a certain period in the past to ensure the timeliness and relevance of the data.

[0158] The preset similarity conditions include trauma similarity and individual feature similarity. Trauma similarity can be determined by comparing features such as trauma location, trauma type, and trauma severity across sample subjects. For example, trauma location can be categorized and coded, trauma type features extracted, and trauma severity quantitatively assessed. The distance between trauma features between sample subjects is then calculated. When the distance is less than or equal to a first threshold, the trauma similarity is considered to meet the conditions.

[0159] Individual feature similarity can be achieved by comparing individual feature data of sample objects, such as age, gender, and basic physiological indicators. Similarly, these individual features can be quantified to calculate the individual feature distance between sample objects. When the distance is less than or equal to a second threshold, the individual feature similarity is considered to meet the conditions.

[0160] The target historical nursing data includes trauma care measures and rehabilitation progress. Trauma care measures record the various nursing methods used by the sample subjects during the nursing process, such as medication, physical therapy, and rehabilitation training. Rehabilitation progress records the sample subjects' recovery during the nursing process, such as changes in physiological parameters and improvements in rehabilitation function.

[0161] Step S320: extracting the historical nursing feature vector of each of the sample objects in the target historical nursing data, and obtaining the nursing stage label corresponding to each of the sample objects.

[0162] Extract each sample subject's historical nursing feature vector from the target historical nursing data. For trauma nursing measures, different nursing measures can be coded to form a nursing measure vector. For example, if the nursing measures include medication A, medication B, and physical therapy C, then the use of medication A can be coded as 1, and the non-use of medication as 0, and so on, forming a binary nursing measure vector.

[0163] Rehabilitation progress can be quantified by analyzing changes in physiological parameters and improvements in rehabilitation function to form a rehabilitation progress vector. For example, changes in physiological parameters such as heart rate and blood pressure can be standardized, and improvements in rehabilitation functions such as joint range of motion and muscle strength can be quantitatively assessed. These values ​​can then be combined to form a rehabilitation progress vector.

[0164] The nursing measure vector and the rehabilitation progress vector are concatenated to obtain the historical nursing feature vector.

[0165] At the same time, the nursing stage label corresponding to each sample object is obtained. The nursing stage label can be divided according to the recovery status of the sample object and the implementation stage of the nursing measures, such as the initial nursing stage, the mid-term nursing stage, the late nursing stage, etc.

[0166] Step S330: Generate a group trend factor based on the historical nursing feature vector of each sample object, the nursing stage label, and the number of sample objects.

[0167] The group trend factor is used to characterize historical nursing trends.

[0168] A group trend factor is generated based on the historical nursing feature vector, nursing stage label, and number of sample subjects for each sample subject. The group trend factor can reflect the nursing trends at different nursing stages under similar trauma and individual characteristics.

[0169] First, the sample subjects are grouped according to the care stage label. For each sample subject in each care stage, the mean vector of its historical care feature vector is calculated. The mean vector can represent the average care features of the sample subjects in that care stage.

[0170] Next, consider the impact of the number of sample subjects on the mean vector. The greater the number of sample subjects, the more representative the mean vector is of the care trends during that care phase. Each care phase's mean vector can be assigned a weight proportional to the number of sample subjects in that care phase.

[0171] Finally, the weighted mean vectors of each care phase are integrated to form a group trend factor. The group trend factor can be used during the training of the initial care analysis model to guide the model to learn historical care trends and improve the model's prediction accuracy.

[0172] Step S340: During the training process of the initial nursing analysis model, the model parameters in the initial nursing analysis model are updated using the population trend factor.

[0173] During the initial nursing analysis model training process, the group trend factor is used to update the model parameters. The group trend factor contains information about historical nursing trends and can help the model better learn the characteristics and patterns of different nursing stages.

[0174] In each training iteration, the difference between the feature vector output by the current model and the group trend factor is calculated. Similarity metrics such as cosine similarity and Euclidean distance can be used to calculate the degree of difference between the two.

[0175] Adjust the model parameters based on the calculated difference. If the difference is large, it means that the model output deviates significantly from the historical nursing trend, and the model parameters need to be adjusted more significantly. If the difference is small, it means that the model output is close to the historical nursing trend, and the model parameters can be adjusted more slightly.

[0176] The specific parameter adjustment method can adopt optimization algorithms such as gradient descent method. By continuously updating the model parameters, the output of the model gradually approaches the historical nursing trend represented by the group trend factor, thereby improving the performance of the model.

[0177] Step S350: Continue training the initial nursing analysis model based on the updated model parameters.

[0178] After updating the model parameters using the population trend factor, the initial nursing analysis model is trained based on the updated model parameters. The training process is similar to the previous training process and still requires the use of sample nursing data, a preset validation set, a target loss function, and a target optimizer.

[0179] The sample nursing data is fed into the updated model and the loss between the model’s predictions and the true labels is calculated. This is done using the target loss function, which aims to minimize the loss.

[0180] The target optimizer is used to calculate the parameter update amount based on the loss value and update the model parameters. During the parameter update process, the group trend factor can still be incorporated to ensure that the model learns historical nursing trends while learning from the sample data.

[0181] The training iteration process is repeated until the model's performance meets preset conditions, such as the evaluation index on the validation set reaching a certain threshold. Ultimately, a nursing analysis model with better performance is obtained.

[0182] Optionally, for the above-mentioned group trend factors, the weights of the group trend factors used in the initial nursing analysis model training process can also be dynamically adjusted based on the differences in the time series of nursing data to enhance the adaptability of the generated nursing analysis model to the changing trends of nursing data, thereby further improving the accuracy of the nursing analysis model output. The method of the present application may also include the following steps:

[0183] Step S410: Acquire newly added nursing data.

[0184] The time period corresponding to the newly added nursing data is later than the preset time period corresponding to the target historical nursing data.

[0185] As time goes by, new nursing data is continuously generated. Acquiring newly added nursing data that corresponds to a time period later than the preset time period corresponding to the target historical nursing data can provide the latest nursing information and experience, helping to further optimize the nursing analysis model.

[0186] The sources of the newly added nursing data are similar to those of the previous sample nursing data, including static data and time-series data. Static data includes individual characteristic data, genetic data, trauma text data, medical history text data, and medical image data of the newly added sample subjects; time-series data includes rehabilitation progress data, physiological indicator data, and historical nursing logs.

[0187] Step S420: When the difference between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data is greater than or equal to a first difference threshold, a distribution difference measure value between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data is calculated.

[0188] Compare the distribution characteristics of the newly added nursing data with those of the target historical nursing data. Distribution characteristics can be represented by statistical features of the data, such as mean, standard deviation, and variance. Statistical features can be calculated for each feature dimension of the newly added nursing data and the target historical nursing data, and then the differences between these statistical features can be compared.

[0189] When the difference between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data is greater than or equal to the first difference threshold, it indicates that there is a significant difference between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data. In this case, it is necessary to calculate the distribution difference measure between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data.

[0190] Distribution difference measures can be calculated using a variety of methods, such as the Kullback-Leibler divergence and the Wasserstein distance. These methods measure the degree of difference between two distributions. By calculating distribution difference measures, we can quantify the distributional differences between the newly added nursing data and the target historical nursing data.

[0191] Step S430: Obtaining the group trend factor weight corresponding to the distribution characteristics of the newly added nursing data according to the distribution difference measurement value, the preset adjustment parameter, and the group trend factor weight corresponding to the distribution characteristics of the target historical nursing data.

[0192] According to the calculated distribution difference measurement value, the preset adjustment parameter and the group trend factor weight corresponding to the distribution characteristics of the historical data, the group trend factor weight corresponding to the distribution characteristics of the newly added nursing data is obtained.

[0193] The preset adjustment parameter is a pre-set parameter used to control the adjustment range of the group trend factor weight. A larger distribution difference metric value indicates a greater difference in the distribution of the new nursing data and the target historical nursing data, and in this case, a larger adjustment to the group trend factor weight is required.

[0194] An adjustment formula can be used to calculate the weight of the population trend factor corresponding to the distribution characteristics of the newly added nursing data. The adjustment formula can consider the relationship between the distribution difference measure, the preset adjustment parameters, and the weight of the population trend factor corresponding to the distribution characteristics of the historical data. For example, the weight of the population trend factor corresponding to the distribution characteristics of the historical data can be adjusted according to a certain proportion based on the size of the distribution difference measure to obtain the weight of the population trend factor corresponding to the distribution characteristics of the newly added nursing data.

[0195] Step S440: During the training process of the initial nursing analysis model, the initial nursing analysis model is continued to be trained based on the adjusted group trend factor weight and the group trend factor.

[0196] During the initial nursing analysis model training process, the model is further trained based on the adjusted group trend factor weights and group trend factors. The adjusted group trend factor weights reflect the degree of influence of the newly added nursing data on the group trend factor.

[0197] In each training iteration, the group trend factor is weighted according to the adjusted group trend factor weight. The weighted group trend factor can better reflect the trend of the current nursing data.

[0198] Calculate the difference between the eigenvector output by the current model and the weighted population trend factor, and adjust the model parameters based on the difference. Use the target loss function to calculate the loss value, and use the target optimizer to update the model parameters.

[0199] Through continuous training iterations, the model can adapt to the distribution characteristics of newly added nursing data, improving the model's generalization ability and prediction accuracy. Ultimately, a nursing analysis model that can better handle new data is obtained.

[0200] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based wound care plan generation system 100 provided in an embodiment of the present application. Figure 2 As shown, the processor 120 can be used in the artificial intelligence-based wound care plan generation system 100 and used to perform the functions of the present invention.

[0201] The AI-based wound care plan generation system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the AI-based wound care plan generation method of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0202] For example, the AI-based wound care plan generation system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. For example, the AI-based wound care plan generation system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention may be implemented based on these program instructions. The AI-based wound care plan generation system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0203] For ease of explanation, only one processor is described in the artificial intelligence-based wound care plan generation system 100. However, it should be noted that the artificial intelligence-based wound care plan generation system 100 of the present invention may also include multiple processors, and therefore the steps performed by one processor described in the present invention may also be performed jointly or individually by multiple processors. For example, if the processor of the artificial intelligence-based wound care plan generation system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by a single processor. For example, the first processor performs step A and the second processor performs step B, or the first processor and the second processor perform steps A and B together.

[0204] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the artificial intelligence-based wound care plan generation method described in the aforementioned embodiment.

[0205] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps in the artificial intelligence-based wound care plan generation method described in the aforementioned embodiment.

[0206] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0207] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of storing or storing data.

[0208] Finally, it should be noted that what is disclosed above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for generating a trauma care plan based on artificial intelligence, characterized in that: include: Acquiring first basic medical data of a target subject to be treated for trauma care, wherein the first basic medical data includes individual characteristic data, gene data, trauma text data, and medical history text data; performing data preprocessing and reconstruction operations on the first basic medical data to generate second basic medical data of the target object; Constructing an initial nursing analysis model and obtaining sample nursing data for training the initial nursing analysis model, wherein the sample nursing data includes static sample data and time series sample data, wherein the static sample data includes sample individual feature data, sample gene data, sample trauma text data, sample medical history text data, and sample medical image data, and the time series sample data includes rehabilitation progress data, physiological indicator data, and historical nursing logs, and the initial nursing analysis model includes a neural network branch, an LSTM branch, and a classifier branch; Inputting the static sample data into the neural network branch, and obtaining a static sample feature vector corresponding to the static sample data through multiple convolutional layers, maximum pooling layers, and fully connected layers in the neural network branch; Inputting the time series sample data into the LSTM branch, and obtaining a time series sample feature sequence corresponding to the time series sample data through the bidirectional LSTM layer in the LSTM branch; According to each time step of the time series sample feature sequence, splicing the static sample feature vector and the time series sample feature sequence to generate a target joint feature vector; Inputting the target joint feature vector into the classifier branch to obtain a subsequent care stage label set and a rehabilitation effect label set corresponding to the sample care data; Training the initial nursing analysis model based on the subsequent nursing stage label set, the rehabilitation effect label set, a preset validation set, a target loss function, and a target optimizer to obtain the nursing analysis model; inputting the second basic medical data into a pre-built nursing analysis model to obtain at least one subsequent nursing stage label of the target object; A recommended wound care plan for the target subject is determined based on the subsequent care stage label, a care knowledge base, and historical care data of a plurality of reference subjects.

2. The method according to claim 1, characterized in that The step of inputting the static sample data into the neural network branch and obtaining a static sample feature vector corresponding to the static sample data through multiple convolutional layers, maximum pooling layers, and fully connected layers in the neural network branch includes: Inputting the static sample data into the neural network branch, and obtaining a sample feature map corresponding to the static sample data through the convolution operation of the multi-layer convolution layer in the neural network branch; Downsampling the sample feature map through the maximum pooling layer to generate a downsampled sample feature map; The downsampled sample feature map is mapped to a preset dimension through the fully connected layer to obtain the static sample feature vector corresponding to the static sample data.

3. The method according to claim 1, characterized in that Inputting the time series sample data into the LSTM branch, and obtaining a time series sample feature sequence corresponding to the time series sample data through a bidirectional LSTM layer in the LSTM branch, includes: Inputting the time series sample data into the LSTM branch, and obtaining multiple time series sample subsequences corresponding to the time series sample data according to a preset sliding window; Recursively processing the time series sample subsequence through the forward LSTM in the bidirectional LSTM layer to generate a previous hidden state sequence; Recursively processing the time series sample subsequence through the backward LSTM in the bidirectional LSTM layer to generate a subsequent hidden state sequence; The preceding hidden state sequence and the succeeding hidden state sequence are concatenated to obtain a time series sample feature sequence corresponding to the time series sample data.

4. The method according to claim 1, wherein The method further comprises: Obtaining target historical nursing data for a plurality of sample subjects that meet preset similarity conditions within a preset time period, wherein the preset similarity conditions include trauma similarity being less than or equal to a first threshold and / or individual characteristic similarity being less than or equal to a second threshold, the historical nursing data including trauma nursing measures and rehabilitation progress; Extracting the historical nursing feature vector of each of the sample objects in the target historical nursing data, and obtaining the nursing stage label corresponding to each of the sample objects; generating a group trend factor according to the historical nursing feature vector of each sample object, the nursing stage label, and the number of sample objects, wherein the group trend factor is used to characterize the historical nursing trend; During the training process of the initial nursing analysis model, updating the model parameters in the initial nursing analysis model by using the population trend factor; The initial care analysis model continues to be trained based on the updated model parameters.

5. The method according to claim 4, characterized in that Also includes: Acquire newly added nursing data, where a time period corresponding to the newly added nursing data is later than a preset time period corresponding to the target historical nursing data; When the difference between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data is greater than or equal to a first difference threshold, calculating a distribution difference measure between the distribution characteristics of the newly added nursing data and the distribution characteristics of the target historical nursing data; Obtaining the group trend factor weight corresponding to the distribution feature of the newly added nursing data according to the distribution difference measure, the preset adjustment parameter, and the group trend factor weight corresponding to the distribution feature of the target historical nursing data; During the training process of the initial nursing analysis model, the initial nursing analysis model continues to be trained based on the adjusted group trend factor weight and the group trend factor.

6. The method according to claim 1, characterized in that Determining a recommended wound care plan for the target subject based on the subsequent care stage label, the care knowledge base, and historical care data of a plurality of reference subjects includes: Obtaining, from the nursing knowledge base, candidate nursing measure texts for the target subject based on the subsequent nursing stage labels and the historical nursing data of the multiple reference subjects, wherein the candidate nursing measures included in the candidate nursing measure texts include at least one of fluid replacement measures, analgesic measures, anti-infection measures, and rehabilitation training measures; extracting key medical features of the target object from the first basic medical data of the target object; Matching the key medical features and the candidate nursing measure texts according to a preset matching rule, and selecting target candidate nursing measures from the candidate nursing measure texts whose matching degree with the key medical features is greater than or equal to a preset matching degree threshold; Determining a risk level of the target candidate nursing measure based on the target candidate nursing measure and the key medical characteristics, and adjusting the target candidate nursing measure based on the risk level of the target candidate nursing measure to generate a trauma nursing measure for the target subject; The wound care measures of the target object are integrated to generate the wound care recommendation plan for the target object.

7. The method according to claim 6, characterized in that The step of matching the key medical features and the candidate nursing measure texts using a preset matching rule, and selecting target candidate nursing measures from the candidate nursing measure texts whose matching degree with the key medical features is greater than or equal to a preset matching degree threshold, includes: Obtaining a multidimensional feature vector set corresponding to each candidate nursing measure in the candidate nursing measure text, the multidimensional feature vector set including a physiological impact vector, a risk level vector, an expected efficacy vector, and a resource consumption vector, wherein the physiological impact vector is generated by normalizing and concatenating quantitative impact parameters of the candidate nursing measure on a preset physiological indicator system, and the risk level vector is obtained by logarithmically transforming and then reducing the dimensionality of the complication probability distribution corresponding to the candidate nursing measure; Extracting a dynamic response vector corresponding to the key medical feature of the target subject, where the dynamic response vector is generated by projecting a product matrix of a gene expression regulation coefficient corresponding to the gene data in the first basic medical data and a tissue injury degree score corresponding to the trauma text data onto a preset vector space after performing singular value decomposition; Calculating a bidirectional weighted similarity between the dynamic response vector and each vector in the multidimensional feature vector set, wherein a positive weight is determined based on a predefined medical priority strategy in the nursing knowledge base, and a negative weight is dynamically adjusted based on an actual effect deviation rate of similar nursing measures in the historical nursing data; A dynamic matching surface is constructed according to the joint distribution characteristics of the bidirectional weighted similarity, and the candidate nursing measure texts are topologically clustered on the dynamic matching surface to screen out candidate nursing measures that are in a preset high-density area and meet the curvature continuity constraint condition as the target candidate nursing measures.

8. The method according to claim 1, characterized in that After determining a recommended wound care plan for the target subject based on the subsequent care stage label, the care knowledge base, and historical care data of a plurality of reference subjects, the method further includes: monitoring multi-source heterogeneous data of the target subject during the process of receiving care according to the trauma care recommendation plan, wherein the multi-source heterogeneous data includes physiological parameter data, rehabilitation function data, medical imaging data, and nursing execution data; Determining the target subject's disease condition change characteristics and rehabilitation effect characteristics based on the target subject's multi-source heterogeneous data; Determining the adjustment goals of the recommended trauma care plan based on the characteristics of the disease changes and the characteristics of the rehabilitation effect; The wound care recommendation is updated based on the adjustment goal.

9. An artificial intelligence-based trauma care plan generation system, characterized in that: The invention comprises a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the method for generating a wound care plan based on artificial intelligence according to any one of claims 1 to 8 is implemented.

Images