Intelligent evaluation method and system for specialized care needs of neurosurgery patients

Abstract

The application provides a neurosurgical patient special care demand intelligent evaluation method and system, relates to the medical care technical field, and comprises the following steps: constructing a patient basic feature vector by collecting physiological and behavioral data of the patient; projecting the patient basic feature vector in a demand space, determining a demand intensity level and a category, and generating a demand state code; aggregating the demand state code, identifying demand conflicts, and correcting the evaluation result based on resource constraints; matching the evaluation result with a scheme library, verifying the optimal care instruction after verification. The application realizes dynamic evaluation of individual and group care demands of patients and optimal allocation of resources.

Description

Technical Field

[0001] This invention relates to the field of medical and nursing technology, and in particular to an intelligent assessment method and system for the specialized care needs of neurosurgical patients. Background Technology

[0002] Postoperative recovery and long-term rehabilitation of neurosurgical patients heavily rely on precise and timely specialist care. Currently, the care needs of neurosurgical patients are typically assessed through the manual observation and experience of healthcare professionals. Nurses regularly record patients' vital signs and, combined with direct observation of their behavior, score them on standardized assessment scales to generate a static needs assessment report. Based on this report, healthcare professionals then develop a general care plan for the patient, referring to established nursing pathways or general guidelines.

[0003] However, the condition of neurosurgical patients is highly dynamic and varies from person to person. Care needs are often reflected in subtle fluctuations in physiological parameters, early signs of abnormal behavioral patterns, and complex states where multiple needs are intertwined. Current technologies for assessing patients are lagging and one-sided, and lack systematic analysis of the correlation and potential conflicts of group needs. This may result in low overall efficiency of nursing plans, making it impossible to achieve optimal allocation of care responses under resource constraints, thereby affecting the overall quality of care and the patient's recovery process. Summary of the Invention

[0004] This invention provides an intelligent assessment method and system for the specialized care needs of neurosurgical patients, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides an intelligent assessment method for the specialist care needs of neurosurgical patients, comprising: Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and mutation components and to construct physiological features. The behavioral data is encoded by action sequence and to construct behavioral pattern features. The physiological features and behavioral pattern features are spliced ​​and nonlinearly mapped to obtain the patient's basic feature vector. Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to the demand representation points is calculated and the demand intensity level is determined. The directional angle corresponding to the demand representation points is calculated and the demand category is determined. The demand intensity level and the demand category are combined and encoded to obtain the demand status code. The demand status codes are aggregated and a group demand distribution matrix is ​​constructed. Demand association patterns are extracted based on the group demand distribution matrix and demand conflict nodes are determined. Based on the demand conflict nodes and preset resource capacity constraints, resource allocation weights are calculated and the demand status codes are corrected to obtain the demand evaluation results. Calculate the matching score between the demand assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

[0006] In one alternative implementation, Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and abrupt change components, and physiological features are constructed. The behavioral data is encoded into action sequences to construct behavioral pattern features. The physiological features and behavioral pattern features are concatenated and nonlinearly mapped to obtain the patient's basic feature vector, including: Physiological data of neurosurgical patients are collected and segmented to obtain time-series data segments. Frequency components are extracted and periodic fluctuation components are identified by frequency domain transformation of the time-series data segments. The gradient of the time-series data segments is calculated and abrupt components are identified. The amplitude envelope of the periodic fluctuation components is calculated and a periodic feature vector is constructed. The peak position of the abrupt components is calculated and an abrupt feature vector is constructed. Physiological characteristics are obtained based on the periodic feature vector and the abrupt feature vector. Behavioral data of neurosurgical patients are collected, and the action identifiers in the behavioral data are arranged to obtain an action sequence. The action sequence is segmented by a sliding window to obtain a set of action segments. The frequency of occurrence of different action segments in the set of action segments is counted and a frequency distribution vector is constructed. The transition probability between adjacent actions in the action sequence is calculated and a state transition matrix is ​​constructed. The state transition matrix is ​​decomposed by feature value to extract the dominant transition pattern and concatenated with the frequency distribution vector to obtain behavioral pattern features. The physiological features and behavioral pattern features are dimensionally aligned and concatenated according to feature type to obtain a hybrid feature vector. The hybrid feature vector is then subjected to nonlinear mapping and dimensionality reduction to obtain the patient's basic feature vector.

[0007] In one alternative implementation, Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to each demand representation point is calculated, and the demand intensity level is determined, including: Covariance analysis is performed on the patient's basic feature vectors to obtain a feature correlation matrix. Eigenvalue decomposition is performed on the feature correlation matrix to extract principal component direction vectors and corresponding eigenvalues. The cumulative contribution rate is calculated based on the eigenvalues ​​to determine the number of effective principal components. An orthogonal basis is constructed based on the principal component direction vectors corresponding to the number of effective principal components, and the coordinate axes of the demand space are determined. The contribution rate of the principal component direction vectors corresponding to the effective principal components is calculated, and the number of dimensions of the demand space is determined. The demand space is constructed based on the orthogonal basis and the number of dimensions. The projection coefficient vector is obtained by performing an inner product operation on the patient's basic feature vector and the orthogonal basis. The coordinate position of the demand representation point in the demand space is determined based on the projection coefficient vector. The coordinate position of the demand representation point is centered and the Euclidean distance between the demand representation point and the origin of the demand space is calculated to obtain the radial distance. The quantile statistics of the radial distance are performed to obtain the distance distribution interval. The intensity threshold boundary is divided based on the distance distribution interval and the radial distance is mapped to the interval corresponding to the intensity threshold boundary to obtain the demand intensity level.

[0008] In one alternative implementation, Calculate the direction angle corresponding to the demand characterization point and determine the demand category. Combine the demand intensity level and the demand category to obtain the demand status code, including: Extract the coordinate components of the demand representation point in the demand space and calculate the direction vector of the demand representation point relative to the origin of the demand space. Calculate the unit direction vector based on the direction vector and determine the angle between the unit direction vector and the basis vectors of each coordinate axis of the demand space to obtain a multi-dimensional direction angle set. Extract the dominant direction angle based on the multi-dimensional direction angle set. Find the angle interval to which the dominant direction angle belongs based on a pre-set mapping table and obtain the corresponding demand category identifier. Calculate the projection length of the unit direction vector in the subspace corresponding to the demand category identifier and determine the demand category based on the projection length. Obtain the level numerical code corresponding to the demand intensity level, perform category index mapping on the demand category to obtain the category numerical code, and determine the coding bit width allocation scheme in combination with the level numerical code. Based on the coding bit width allocation scheme, determine the high-bit coding segment and the low-bit coding segment. Calculate the combined coding value based on the high-bit coding segment and the low-bit coding segment. Calculate the product of the radial distance of the demand characterization point and the dominant direction angle and perform a hash transformation to obtain the check code. Based on the check code and the combined coding value, determine the extended coding structure and perform integrity verification to obtain the demand status code.

[0009] In one alternative implementation, Aggregating the demand status codes and constructing a group demand distribution matrix, extracting demand association patterns based on the group demand distribution matrix and identifying demand conflict nodes includes: Multiple neurosurgical patients' demand status codes are collected and organized into a coding sequence according to patient identification. Frequency statistics are performed on the coding sequence to obtain the demand status frequency distribution and a mapping dictionary is constructed. The unique demand status value in the mapping dictionary is used as the row index and the patient identification is used as the column index to construct a matrix frame. The demand status codes of different patients in the coding sequence are filled into the corresponding row and column positions of the matrix frame and the empty positions are filled with zero values ​​to obtain the group demand distribution matrix. The transpose of the demand distribution matrix is ​​calculated, and a demand state co-occurrence matrix is ​​obtained by combining the demand distribution matrix with the demand state co-occurrence matrix. The diagonal elements of the demand state co-occurrence matrix are zeroed out, and off-diagonal elements are extracted to construct a demand state association strength matrix. The demand state association strength matrix is ​​row-normalized to obtain a demand state transition probability matrix. A directed weighted graph is constructed based on the demand state transition probability matrix, and the demand states are encoded as graph nodes. The directed weighted graph is decomposed into strongly connected components, and demand association patterns are extracted. The difference between demand state pairs corresponding to each element in the demand state association strength matrix is ​​calculated, and high-difference demand state pairs are identified. The set of patients co-occurring with the high-difference demand state pairs in the demand distribution matrix is ​​extracted, and demand conflict nodes are marked.

[0010] In one alternative implementation, Based on the conflicting demand nodes and preset resource capacity constraints, resource allocation weights are calculated, and the demand status codes are corrected to obtain the demand assessment results, including: Extract the set of demand status codes corresponding to the demand conflict nodes and count the number of patients to obtain the conflict scale vector. Extract different total care resource capacities from the preset resource capacity constraints to construct a resource capacity vector. Analyze the resource demand types corresponding to the set of demand status codes and establish a mapping relationship with the resource capacity vector. Calculate the initial resource allocation weight based on the conflict scale vector and the resource capacity vector, and sort the set of demand status codes to obtain a priority sequence. The priority sequence is traversed to extract the corresponding patient identifiers. A patient demand coding group is constructed based on the patient identifiers and the group demand distribution matrix. The coding difference degree between the patient demand coding group and the demand status coding corresponding to the demand conflict node is calculated. A difference degree matrix is ​​constructed based on the coding difference degree, and a patient conflict intensity vector is constructed. A corrected resource allocation weight is calculated based on the patient conflict intensity vector and the initial resource allocation weight, and a resource allocation simulation is performed to determine the amount of allocated resources. The amount of allocated resources and the resource capacity vector are compared in real time. When the amount of allocated resources exceeds the resource capacity vector, the unallocated demand status coding is marked as a coding to be corrected and corrected. The demand assessment result is obtained by combining the amount of allocated resources.

[0011] In one alternative implementation, Calculating the matching score between the demand assessment result and the care plan templates in the preset care plan library, sorting the care plan templates based on the matching score and determining the candidate care plan sequence, and verifying the feasibility of the candidate care plan sequence to determine the optimal care instruction and issue it for execution includes: Care plan templates are obtained from a pre-set care plan library, and the care action type and resource configuration information corresponding to each care plan template are extracted. The needs assessment results are semantically mapped with the care action type and resource configuration information corresponding to the care plan templates, and the needs coverage is calculated. The structural similarity between the needs assessment results and the care plan templates is calculated. Based on the needs coverage and the structural similarity, dual constraints are constructed and a comprehensive matching score is calculated. Based on the comprehensive matching score, the care plan templates in the care plan library are sorted and a candidate care plan sequence is constructed. The process involves traversing the candidate care plan sequence and extracting the resource requirement list of the care plan template. It also involves querying the available resource status in the resource capacity constraints and marking resource reachability. The execution dependencies of the care plan template are extracted, and dependency integrity is determined based on the requirement assessment results. The executability of the care plan template is determined based on the resource reachability and dependency integrity. Finally, the optimal care instruction is generated and issued for execution based on the care plan template with the best executability.

[0012] A second aspect of this invention provides an intelligent assessment system for the specialized care needs of neurosurgical patients, comprising: The feature construction unit is used to collect physiological and behavioral data of neurosurgical patients, perform time-domain decomposition on the physiological data to obtain periodic fluctuation components and mutation components and construct physiological features, encode the behavioral data into action sequences and construct behavioral pattern features, and concatenate and nonlinearly map the physiological features and behavioral pattern features to obtain the patient's basic feature vector. The demand encoding unit is used to construct a demand space based on the patient's basic feature vector and perform projection mapping to obtain demand representation points, calculate the radial distance corresponding to the demand representation points and determine the demand intensity level, calculate the direction angle corresponding to the demand representation points and determine the demand category, and combine the demand intensity level and the demand category to obtain the demand status code. The demand assessment unit is used to aggregate the demand status codes and construct a group demand distribution matrix, extract demand association patterns and determine demand conflict nodes based on the group demand distribution matrix, calculate resource allocation weights based on the demand conflict nodes and preset resource capacity constraints, and correct the demand status codes to obtain the demand assessment results. The care matching unit is used to calculate the matching score between the needs assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

[0013] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] In this invention, by performing time-domain decomposition on physiological data to separate periodic fluctuation components and abrupt change components, the stability and sudden abnormalities of the patient's physiological state can be more precisely characterized. By fusing and nonlinearly mapping physiological characteristics with behavioral pattern characteristics, the generated patient basic feature vector integrates multimodal information, providing a comprehensive and high-dimensional data foundation for subsequent needs analysis. By calculating the radial distance of representation points, the urgency or intensity level of the patient's needs can be objectively and quantitatively determined. Aggregating individual needs state codes to construct a group needs distribution matrix allows for macro-level analysis of the overall needs of the patient group. Combined with preset resource capacity constraints, resource allocation weights are calculated and individual needs state codes are corrected, ensuring that the final needs assessment results not only reflect individual situations but also consider group coordination and resource feasibility, making the assessment results more reasonable. By calculating the matching score between the needs assessment results and preset care plan templates, potential solutions highly compatible with the current needs can be quickly screened. Feasibility verification of candidate sequences is performed, and by comprehensively considering execution conditions, resource consumption, and expected effects, the optimal care instruction is finally determined. This achieves a closed loop from intelligent needs assessment to automatic solution generation and optimization, improving the timeliness and rationality of care response. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent assessment method for the specialized care needs of neurosurgical patients according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the demand state coding analysis of the intelligent assessment method for the specialized care needs of neurosurgical patients, as described in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0019] Figure 1 This is a flowchart illustrating the intelligent assessment method for the specialized care needs of neurosurgical patients according to an embodiment of the present invention. Figure 1 As shown, the method includes: Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and mutation components and to construct physiological features. The behavioral data is encoded by action sequence and to construct behavioral pattern features. The physiological features and behavioral pattern features are spliced ​​and nonlinearly mapped to obtain the patient's basic feature vector. Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to the demand representation points is calculated and the demand intensity level is determined. The directional angle corresponding to the demand representation points is calculated and the demand category is determined. The demand intensity level and the demand category are combined and encoded to obtain the demand status code. The demand status codes are aggregated and a group demand distribution matrix is ​​constructed. Demand association patterns are extracted based on the group demand distribution matrix and demand conflict nodes are determined. Based on the demand conflict nodes and preset resource capacity constraints, resource allocation weights are calculated and the demand status codes are corrected to obtain the demand evaluation results. Calculate the matching score between the demand assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

[0020] In one alternative implementation, Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and abrupt change components, and physiological features are constructed. The behavioral data is encoded into action sequences to construct behavioral pattern features. The physiological features and behavioral pattern features are concatenated and nonlinearly mapped to obtain the patient's basic feature vector, including: Physiological data of neurosurgical patients are collected and segmented to obtain time-series data segments. Frequency components are extracted and periodic fluctuation components are identified by frequency domain transformation of the time-series data segments. The gradient of the time-series data segments is calculated and abrupt components are identified. The amplitude envelope of the periodic fluctuation components is calculated and a periodic feature vector is constructed. The peak position of the abrupt components is calculated and an abrupt feature vector is constructed. Physiological characteristics are obtained based on the periodic feature vector and the abrupt feature vector. Behavioral data of neurosurgical patients are collected, and the action identifiers in the behavioral data are arranged to obtain an action sequence. The action sequence is segmented by a sliding window to obtain a set of action segments. The frequency of occurrence of different action segments in the set of action segments is counted and a frequency distribution vector is constructed. The transition probability between adjacent actions in the action sequence is calculated and a state transition matrix is ​​constructed. The state transition matrix is ​​decomposed by feature value to extract the dominant transition pattern and concatenated with the frequency distribution vector to obtain behavioral pattern features. The physiological features and behavioral pattern features are dimensionally aligned and concatenated according to feature type to obtain a hybrid feature vector. The hybrid feature vector is then subjected to nonlinear mapping and dimensionality reduction to obtain the patient's basic feature vector.

[0021] Physiological data of neurosurgical patients were collected, including heart rate, blood pressure, respiratory rate, body temperature, and blood oxygen saturation. Segmented sampling technology was used, dividing the physiological data into time-series segments with each 60-second time window. For example, the heart rate data of a neurosurgical patient monitored postoperatively showed sampling values ​​of 72, 75, 78, 76, 75, 77, 79, 80, 78, and 76 beats per minute within one minute.

[0022] Frequency components are extracted by performing a frequency domain transformation on the acquired time-series data segment. The Fast Fourier Transform (FFT) method is used to transform the time-series data segment from the time domain to the frequency domain, obtaining the intensity distribution of components at different frequencies. In the frequency domain, based on the energy distribution of the frequency components, fluctuation components with obvious periodicity are identified. In the previous example, frequency domain analysis identified a fluctuation component with a period of approximately 20 seconds and an amplitude of approximately 2 Hz. The gradient of the time-series data segment is calculated to identify abrupt changes. Gradient calculation uses the difference method between adjacent data points; when the absolute value of the gradient exceeds a preset threshold, it is determined to be an abrupt change point. In the previous example, the gradient threshold was set to 3 Hz, and an abrupt change point was identified between the 7th and 8th seconds, with the heart rate abruptly changing from 76 to 80 beats per minute.

[0023] Calculate the amplitude envelope of the periodic fluctuation component and construct a periodic eigenvector. The amplitude envelope is obtained by extracting the peak points of the waveform and smoothing them together. The peak, trough, average value, and fluctuation frequency of the envelope are used to construct the periodic eigenvector. In the example above, the periodic eigenvector contains a peak frequency of 80 times / minute, a trough frequency of 72 times / minute, an average value of 76.6 times / minute, and a main fluctuation period of 20 seconds.

[0024] The peak positions of the mutation components are calculated, and mutation feature vectors are constructed. The temporal location, mutation amplitude, duration, and recovery time of each mutation point are recorded to form the mutation feature vector. In the previous example, the mutation feature vector includes a mutation location at the 7th second, a mutation amplitude of 4 times / minute, a duration of 2 seconds, and a recovery time of 5 seconds. Physiological characteristics are obtained based on the periodicity feature vector and the mutation feature vector. The two feature vectors are weighted and fused according to their importance to obtain comprehensive physiological characteristics containing both periodicity and mutation information. In the previous example, the physiological characteristics include heart rate fluctuation characteristics, mutation intensity, and recovery ability.

[0025] Behavioral data of neurosurgical patients is collected, including limb movements, changes in posture, eating, toileting, medication administration, and other daily activities. The action markers in the behavioral data are sorted chronologically to obtain action sequences. Taking the behavioral record of a neurosurgical patient within 24 hours post-surgery as an example, the action sequence can be represented as: bed rest - temperature measurement - medication administration - eating - sitting up - getting out of bed - toileting - bed rest - blood pressure measurement - medication administration - bed rest. The action sequence is segmented using a sliding window to obtain a set of action fragments. The window size is set to 3 consecutive actions, and the sliding step size is 1 action. In the aforementioned example, the following action fragments can be obtained: bed rest - temperature measurement - medication administration, temperature measurement - medication administration - eating, medication administration - eating - sitting up.

[0026] The frequency of different action segments in the action segment set is statistically analyzed, and a frequency distribution vector is constructed. The number of times each action segment occurs throughout the observation period is calculated and normalized to obtain the frequency distribution. In the example above, bed rest occurs 3 times, medication occurs 2 times, and other actions each occur once, and a frequency distribution vector is constructed.

[0027] The transition probabilities between adjacent actions in the action sequence are calculated, and a state transition matrix is ​​constructed. The frequency of transitions from one action to another is counted, and conditional probabilities are calculated. In the previous example, the transition probability from bed rest to measuring body temperature is 0.33, the transition probability from bed rest to measuring blood pressure is 0.33, and the transition probability from bed rest to taking medication is 0. Eigenvalue decomposition is performed on the state transition matrix to extract the dominant transition pattern. The eigenvalues ​​and eigenvectors of the transition matrix are calculated, and the eigenvector corresponding to the largest eigenvalue is selected as the dominant transition pattern. In the previous example, the dominant transition pattern shows that the patient's behavior pattern mainly revolves around the cycle of bed rest - measuring vital signs - taking medication - bed rest.

[0028] The dominant transfer pattern is concatenated with the frequency distribution vector to obtain the behavioral pattern features. Feature normalization is performed during the concatenation process to ensure comparability of features with different dimensions. In the previous example, the behavioral pattern features reflect that the patient is less active, mainly resting in bed and receiving regular vital sign monitoring and medication. The physiological features and behavioral pattern features are dimensionally aligned and concatenated according to feature type to obtain a mixed feature vector. Dimensional alignment is achieved through feature selection or feature expansion to match the dimensions of the two types of features. The adjusted features are then concatenated in the order of physiological features first, followed by behavioral features, to obtain the mixed feature vector.

[0029] Nonlinear mapping and dimensionality reduction are performed on the mixed feature vectors to obtain the patient's basic feature vectors. The nonlinear mapping uses radial basis function kernels to transform the features into a high-dimensional space, and then principal component analysis is used to reduce the dimensionality, retaining principal components that explain 90% of the variance, to obtain the patient's basic feature vectors with a dimension of approximately 10 to 15.

[0030] In this embodiment, by performing time-domain decomposition on the physiological data of neurosurgical patients and identifying periodic fluctuation components and mutation components, and simultaneously performing action sequence encoding and pattern extraction on behavioral data, collaborative modeling of physiological state information and behavioral pattern information is achieved. This effectively improves the completeness and information expression ability of patient state description, enhances the sensitivity and recognition ability of changes in patient state, and by extracting periodic fluctuation features and mutation features in physiological data processing, both long-term stable change trends and short-term abnormal fluctuations in physiological signals can be effectively captured, thereby improving the ability to identify potential physiological abnormal changes. By constructing the frequency distribution of action segments and state transition relationships on behavioral data and extracting dominant transition patterns, the structural regularity and behavioral evolution characteristics in behavioral sequences can be effectively characterized, which can more deeply reflect the dynamic change law of behavioral patterns and improve the ability to identify behavioral pattern differences and abnormal behaviors.

[0031] In one alternative implementation, Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to each demand representation point is calculated, and the demand intensity level is determined, including: Covariance analysis is performed on the patient's basic feature vectors to obtain a feature correlation matrix. Eigenvalue decomposition is performed on the feature correlation matrix to extract principal component direction vectors and corresponding eigenvalues. The cumulative contribution rate is calculated based on the eigenvalues ​​to determine the number of effective principal components. An orthogonal basis is constructed based on the principal component direction vectors corresponding to the number of effective principal components, and the coordinate axes of the demand space are determined. The contribution rate of the principal component direction vectors corresponding to the effective principal components is calculated, and the number of dimensions of the demand space is determined. The demand space is constructed based on the orthogonal basis and the number of dimensions. The projection coefficient vector is obtained by performing an inner product operation on the patient's basic feature vector and the orthogonal basis. The coordinate position of the demand representation point in the demand space is determined based on the projection coefficient vector. The coordinate position of the demand representation point is centered and the Euclidean distance between the demand representation point and the origin of the demand space is calculated to obtain the radial distance. The quantile statistics of the radial distance are performed to obtain the distance distribution interval. The intensity threshold boundary is divided based on the distance distribution interval and the radial distance is mapped to the interval corresponding to the intensity threshold boundary to obtain the demand intensity level.

[0032] Covariance analysis was performed on the aforementioned baseline feature vectors of neurosurgical patients to obtain a feature correlation matrix. Assume that the baseline feature vector of a neurosurgical patient contains 12 dimensions, corresponding to key indicators such as physiological stability, spontaneous breathing ability, cardiovascular function, neurological reflexes, level of consciousness, activity level, and self-care ability. The covariance among these 12 feature dimensions was calculated, forming a 12×12 feature correlation matrix. Standardized feature values ​​were used in the covariance calculation to ensure comparability of features with different dimensions.

[0033] Eigenvalue decomposition is performed on the feature correlation matrix to extract the principal component direction vectors and their corresponding eigenvalues. The eigenvalue decomposition uses the Jacobi iterative method, with a convergence threshold of 0.0001 and a maximum of 100 iterations. After decomposition, 12 eigenvalues ​​and their corresponding eigenvectors are obtained. In the previous example, the 12 eigenvalues, arranged from largest to smallest, are: 5.21, 3.76, 1.94, 0.87, 0.56, 0.31, 0.15, 0.09, 0.05, 0.03, 0.02, and 0.01, with a total eigenvalue sum of 13.

[0034] The cumulative contribution rate is calculated based on eigenvalues ​​to determine the number of effective principal components. The cumulative contribution rate is determined by sorting the eigenvalues ​​from largest to smallest and calculating the proportion of the sum of the top k eigenvalues ​​to the total sum of eigenvalues. A cumulative contribution rate threshold of 85% is set to determine the number of effective principal components. In the previous example, the sum of the first four eigenvalues ​​is 11.78, accounting for 90.6% of the total sum of eigenvalues, exceeding the preset threshold; therefore, the number of effective principal components is determined to be four.

[0035] An orthogonal basis is constructed based on the principal component direction vectors corresponding to the number of effective principal components, and the coordinate axes of the demand space are determined. The first four principal component direction vectors are used as the orthogonal basis to form the coordinate axes of the demand space. Each coordinate axis represents a major influencing factor of a type of care demand. In the aforementioned example, the first principal component direction vector is mainly related to physiological stability and neural reflexes; the second principal component direction vector is mainly related to activity level and self-care ability; the third principal component direction vector is mainly related to respiratory function and cardiovascular function; and the fourth principal component direction vector is mainly related to level of consciousness.

[0036] The contribution rate of the principal component direction vectors corresponding to the effective principal components is calculated to determine the number of dimensions in the demand space. The contribution rate of each principal component is calculated by dividing the eigenvalue by the sum of the eigenvalues. In the previous example, the contribution rates of the four effective principal components are 40.1%, 28.9%, 14.9%, and 6.7%, respectively. Based on the contribution rate distribution, the number of dimensions in the demand space is determined. A contribution rate threshold of 10% is set for each dimension, and dimensions with contribution rates greater than the threshold are selected as dimensions of the demand space. Therefore, the number of dimensions in the demand space is determined to be three, corresponding to physiological care needs, daily activity care needs, and respiratory and circulatory care needs, respectively. The total contribution rate of these three dimensions is 83.9%.

[0037] A demand space is constructed based on an orthogonal basis and the number of dimensions. The direction vectors of the first three principal components are used as coordinate axes to construct a three-dimensional space. Each coordinate axis represents a major influencing factor for a type of care demand, and the coordinate value reflects the intensity of that type of demand. The origin of the demand space represents the average demand level, and a coordinate value of zero indicates that the demand in that dimension is at an average level.

[0038] The projection coefficient vector is obtained by performing an inner product operation on the patient's basic feature vector and the orthogonal basis. The 12-dimensional patient basic feature vector is then projected onto the three principal component direction vectors to obtain a 3-dimensional projection coefficient vector. The projection calculation uses the vector inner product method, i.e., the dot product operation between the feature vector and the principal component direction vector. In the aforementioned example, the basic feature vector of a neurosurgical patient on the third day after surgery, after being projected onto the required space, yields a projection coefficient vector of (2.46, 1.85, 0.93).

[0039] The coordinates of demand representation points in the demand space are determined based on the projection coefficient vector. The projection coefficient vector serves as the coordinates of the demand representation points in the demand space, representing the patient's specific location across the three dimensions of care demand. In the aforementioned example, the patient's coordinates are 2.46 in the physiological care demand dimension, 1.85 in the daily activities care demand dimension, and 0.93 in the respiratory and circulatory care demand dimension.

[0040] The coordinates of the demand representation point are centered, and the radial distance is obtained by calculating the Euclidean distance between the demand representation point and the origin of the demand space. Centering is achieved by subtracting the mean of historical patient samples across all dimensions. The Euclidean distance is calculated using the square root of the sum of squares of the coordinates in each dimension. In the example above, after centering, the patient's coordinates are adjusted to (1.73, 1.21, 0.45), and the calculated Euclidean distance is 2.16.

[0041] The radial distance was statistically analyzed using quantiles to obtain the distance distribution intervals. Based on radial distance data from 1000 neurosurgical patients in a historical patient database, the 10%, 25%, 50%, 75%, and 90% quantiles were calculated to obtain the distance distribution intervals. In the aforementioned example, the quantile results were: 10% quantile 0.62, 25% quantile 0.95, 50% quantile 1.43, 75% quantile 2.08, and 90% quantile 2.74.

[0042] The intensity threshold boundaries are defined based on distance distribution intervals, and the radial distance is mapped to the intervals corresponding to these threshold boundaries to obtain the intensity level of the need. Using the distance distribution intervals as intensity threshold boundaries, five levels are defined: mild, moderate, severe, and serious. Radial distances less than or equal to the 10th percentile are classified as mild; between the 10th and 25th percentiles as mild; between the 25th and 75th percentiles as moderate; between the 75th and 90th percentiles as severe; and greater than the 90th percentile as serious. In the aforementioned example, the patient's radial distance of 2.16 falls between the 75th and 90th percentiles, corresponding to a severe intensity level of need, indicating that the patient has a high need for specialist care and requires closer monitoring and a personalized care plan.

[0043] In this embodiment, principal component direction vectors are extracted through eigenvalue decomposition. Effective principal components are selected based on cumulative contribution rates to construct an orthogonal basis and form a demand space. This allows the original multi-dimensional and correlated feature data to be re-expressed under a low-correlation structure, improving the effectiveness and structural clarity of feature expression. This makes the representation of demand states more stable and interpretable. By projecting the patient's basic feature vector onto the constructed orthogonal basis to obtain projection coefficients, and determining the position of demand representation points in the demand space, the patient's multi-dimensional state can be expressed in a unified spatial coordinate form. This allows for a more objective demand representation result under the combined effect of multi-dimensional features, thereby improving the accuracy and consistency of demand state characterization. By centering the demand representation points and calculating their radial distance from the origin of the demand space, and then combining quantile statistics to construct distance distribution intervals and delineate intensity threshold boundaries, the determination of demand intensity is adaptively divided based on the overall sample distribution characteristics. This better adapts to the differences in feature distribution among different patient groups, improving the objectivity and discriminative ability of demand intensity grading.

[0044] In one alternative implementation, Calculate the direction angle corresponding to the demand characterization point and determine the demand category. Combine the demand intensity level and the demand category to obtain the demand status code, including: Extract the coordinate components of the demand representation point in the demand space and calculate the direction vector of the demand representation point relative to the origin of the demand space. Calculate the unit direction vector based on the direction vector and determine the angle between the unit direction vector and the basis vectors of each coordinate axis of the demand space to obtain a multi-dimensional direction angle set. Extract the dominant direction angle based on the multi-dimensional direction angle set. Find the angle interval to which the dominant direction angle belongs based on a pre-set mapping table and obtain the corresponding demand category identifier. Calculate the projection length of the unit direction vector in the subspace corresponding to the demand category identifier and determine the demand category based on the projection length. Obtain the level numerical code corresponding to the demand intensity level, perform category index mapping on the demand category to obtain the category numerical code, and determine the coding bit width allocation scheme in combination with the level numerical code. Based on the coding bit width allocation scheme, determine the high-bit coding segment and the low-bit coding segment. Calculate the combined coding value based on the high-bit coding segment and the low-bit coding segment. Calculate the product of the radial distance of the demand characterization point and the dominant direction angle and perform a hash transformation to obtain the check code. Based on the check code and the combined coding value, determine the extended coding structure and perform integrity verification to obtain the demand status code.

[0045] Extract the coordinate components of the demand representation point in the demand space, and calculate the direction vector of the demand representation point relative to the origin of the demand space. The coordinate components of the demand representation point in the demand space include projection coefficient values ​​in each dimension. In the previous example, the coordinate components of the neurosurgical patient in the three-dimensional demand space are 2.46, 1.85, and 0.93. The direction vector of the demand representation point relative to the origin of the demand space is constructed by directly using the aforementioned coordinate component values. For the current patient, the direction vector is 2.46, 1.85, and 0.93.

[0046] The unit direction vector is calculated based on the direction vector, and the angles between the unit direction vector and the basis vectors of each coordinate axis in the demand space are determined to obtain a multidimensional set of direction angles. The unit direction vector is calculated by dividing the direction vector by its magnitude. The magnitude of the direction vector is equal to the square root of the sum of the squares of the coordinate components. In the previous example, the magnitude of the direction vector is 3.17, resulting in unit direction vectors of 0.78, 0.58, and 0.29. The angles between the unit direction vector and the basis vectors of each coordinate axis in the demand space are calculated to obtain a multidimensional set of direction angles. The angles are determined by the vector dot product operation, i.e., the inverse cosine of the dot product of the unit direction vector and each coordinate axis basis vector. In the previous example, the angles between the unit direction vector and the three coordinate axis basis vectors are 38.7 degrees, 54.6 degrees, and 73.1 degrees, respectively.

[0047] The dominant direction angle is extracted based on a multi-dimensional set of direction angles. The dominant direction angle is the smallest angle value in the multi-dimensional set of direction angles, representing the coordinate axis direction that the demand representation point is closest to in the demand space. In the example above, the dominant direction angle is 38.7 degrees, corresponding to the first coordinate axis, namely the physiological care demand dimension.

[0048] The system uses a pre-defined mapping table to find the angle range to which the dominant direction angle belongs and obtains the corresponding demand category identifier. The pre-defined mapping table defines the correspondence between angle ranges and demand category identifiers. The mapping table contains three angle ranges: 0 degrees to 42 degrees corresponds to physiological care needs, 42 degrees to 62 degrees corresponds to daily activity care needs, and 62 degrees to 90 degrees corresponds to respiratory and circulatory care needs. In the example above, the dominant direction angle of 38.7 degrees falls within the 0-42 degree range, and the corresponding demand category identifier is physiological care need.

[0049] The projection length of the unit direction vector within the subspace corresponding to the demand category identifier is calculated, and the demand category is determined based on this projection length. The subspace projection length is calculated by the inner product of the unit direction vector and the coordinate axis basis vectors corresponding to the category identifier. In the previous example, the projection length of the unit direction vector on the physiological care demand dimension is 0.78. A projection length threshold of 0.7 is set; if the projection length is greater than the threshold, this dimension is confirmed as the primary demand category. In the previous example, the projection length of 0.78 is greater than the threshold of 0.7, and the patient's demand category is determined to be physiological care demand.

[0050] Obtain the numerical code corresponding to the intensity level of the need. Map the need categories using a category index to obtain category numerical codes, and combine this with the intensity level numerical codes to determine the bit width allocation scheme. The intensity level of the need is mapped to a numerical code: 1 for mild, 2 for moderate, 3 for moderate, 4 for severe, and 5 for serious. In the previous example, the patient's intensity level is moderate, corresponding to a numerical code of 4. The need category is converted to a numerical code through index mapping: 1 for physiological care needs, 2 for daily activity care needs, and 3 for respiratory and circulatory care needs. In the previous example, the patient's need category is physiological care needs, corresponding to a numerical code of 1. The bit width allocation scheme specifies that the category numerical code occupies 3 bits, and the intensity level numerical code occupies 3 bits.

[0051] The high-order and low-order coding segments are determined based on the coding bit-width allocation scheme, and the combined coding value is calculated based on these segments. The high-order coding segment consists of category numerical codes, occupying 3 bits of coding space; the low-order coding segment consists of grade numerical codes, also occupying 3 bits of coding space. The coding bit-width allocation scheme ensures that the category numerical code is in the high-order bits and the grade numerical code is in the low-order bits in the combined coding value. The combined coding value is obtained by left-shifting the high-order coding segment by the bit width of the low-order coding segment and then performing a bitwise OR operation with the low-order coding segment. In the example above, category numerical code 1 is used as the high-order coding segment, and grade numerical code 4 is used as the low-order coding segment. Left-shifting the high-order coding segment by 3 bits yields 8, which is then bitwise ORed with the low-order coding segment 4 to obtain the combined coding value 12. This coding structure design ensures that combinations of different categories and intensity levels have unique combined coding values, facilitating subsequent querying and identification.

[0052] The checksum is obtained by calculating the product of the radial distance and the dominant direction angle of the demand representation point and performing a hash transformation. The product of the radial distance and the dominant direction angle is used as the input to the hash function. The hash transformation uses modulo division, taking the product value modulo the prime number 97 to obtain the checksum. In the previous example, the product of the radial distance 2.16 and the dominant direction angle 38.7 degrees is 83.59, and taking the modulo of 97 yields the checksum 83.

[0053] The extended coding structure is determined based on the checksum and combined coding value, and integrity verification is performed to obtain the requirement status code. The extended coding structure includes a checksum field and a combined coding value field. The checksum field occupies 7 bits, and the combined coding value field occupies 6 bits. The extended coding structure is obtained by left-shifting the checksum by the bit width of the combined coding value field and then performing a bitwise OR operation with the combined coding value. In the previous example, the checksum 83 left-shifted by 6 bits yields 5312, which is then bitwise ORed with the combined coding value 12 to obtain the extended coding value 5324. Integrity verification is achieved by calculating the checksum of the extended coding value and comparing it with a preset value. The checksum is calculated by summing each digit. In the previous example, the checksum of the extended coding value 5324 is 14, which equals the preset value; therefore, the verification passes, and the requirement status code is determined to be 5324.

[0054] In this embodiment, by extracting the coordinate components of the demand representation points in the demand space and calculating the direction vector relative to the origin, further obtaining the unit direction vector and analyzing its angle with the basis vectors of each coordinate axis, the structural differences between different demand types can be reflected more accurately, improving the accuracy and distinguishability of demand category identification. The demand category is determined by calculating the projection length of the unit direction vector in the corresponding category subspace. This ensures that demand category determination not only relies on angle interval division but also incorporates the contribution of the direction vector in the target subspace, reducing the probability of misclassification of samples in boundary regions and improving the reliability of demand category identification results. This is achieved by analyzing the demand intensity... The system uses numerical encoding of the grade and allocates bit width based on the demand category to construct a combined encoding value from high-bit and low-bit encoding segments. This allows demand category information and demand intensity information to be expressed in a unified encoding structure, improving the structuring and compact representation of demand status information and facilitating subsequent rapid retrieval, matching, and data exchange. By combining the radial distance and dominant direction angle of the demand representation point and performing hash transformation to generate a check code, and then combining the combined encoding value to construct an extended encoding structure for integrity verification, the system can effectively improve the reliability and security of data transmission and storage, reducing the risks of encoding errors or data tampering.

[0055] In one alternative implementation, Aggregating the demand status codes and constructing a group demand distribution matrix, extracting demand association patterns based on the group demand distribution matrix and identifying demand conflict nodes includes: Multiple neurosurgical patients' demand status codes are collected and organized into a coding sequence according to patient identification. Frequency statistics are performed on the coding sequence to obtain the demand status frequency distribution and a mapping dictionary is constructed. The unique demand status value in the mapping dictionary is used as the row index and the patient identification is used as the column index to construct a matrix frame. The demand status codes of different patients in the coding sequence are filled into the corresponding row and column positions of the matrix frame and the empty positions are filled with zero values ​​to obtain the group demand distribution matrix. The transpose of the demand distribution matrix is ​​calculated, and a demand state co-occurrence matrix is ​​obtained by combining the demand distribution matrix with the demand state co-occurrence matrix. The diagonal elements of the demand state co-occurrence matrix are zeroed out, and off-diagonal elements are extracted to construct a demand state association strength matrix. The demand state association strength matrix is ​​row-normalized to obtain a demand state transition probability matrix. A directed weighted graph is constructed based on the demand state transition probability matrix, and the demand states are encoded as graph nodes. The directed weighted graph is decomposed into strongly connected components, and demand association patterns are extracted. The difference between demand state pairs corresponding to each element in the demand state association strength matrix is ​​calculated, and high-difference demand state pairs are identified. The set of patients co-occurring with the high-difference demand state pairs in the demand distribution matrix is ​​extracted, and demand conflict nodes are marked.

[0056] Demand status codes for multiple neurosurgical patients were collected and organized into a coding sequence according to patient identifiers. Frequency statistics were performed on the coding sequences to obtain the demand status frequency distribution, and a mapping dictionary was constructed. Demand status codes were collected from 50 patients in the neurosurgical intensive care unit. Each patient had 7 consecutive days of records, totaling 350 coding data points. Patient identifiers were represented by 4-digit numeric codes, ranging from 1001 to 1050. The coding sequence was organized according to the patient identifier order. For example, the 7 coding data points for patient 1001 were 5324, 5324, 5318, 5318, 5312, 5306, and 5306, while the 7 coding data points for patient 1002 were 6428, 6428, 6428, 6422, 6422, 6416, and 6416. Frequency statistics were performed on the aforementioned 350 coding data points to obtain the demand status frequency distribution. Statistical results show that the five most frequent demand status codes are 5324, 6428, 4226, 5318, and 6422, appearing 42, 38, 35, 33, and 30 times respectively. A mapping dictionary is constructed based on these frequency statistics, mapping different demand status codes to consecutive integer index values. In the example above, there are 15 different demand status codes. The mapping dictionary maps 5324 to 1, 6428 to 2, and so on until the last code is mapped to 15.

[0057] A matrix framework is constructed using the unique demand state values ​​in the mapping dictionary as row indices and patient identifiers as column indices. The demand state codes for different patients in the coded sequence are then filled into the corresponding row and column positions of the matrix framework, with empty positions filled with zero values ​​to obtain the group demand distribution matrix. A 15x50 matrix framework is constructed, with row indices corresponding to 15 different demand state codes and column indices corresponding to 50 patients. 350 coded data entries are filled into the corresponding positions in the matrix. If a patient exhibits a certain demand state code during the observation period, the corresponding position in the matrix is ​​filled with the number of times that demand state code appeared during that patient's observation period; if a patient does not exhibit a certain demand state code, the corresponding position is filled with 0. In the previous example, patient 1001's coded sequence 5324 appears twice, so the position with row index 1 and column index 1 in the matrix is ​​filled with the value 2; patient 1001 does not exhibit the code 6428, so the position with row index 2 and column index 1 in the matrix is ​​filled with the value 0. After filling, the group demand distribution matrix is ​​obtained.

[0058] The demand state co-occurrence matrix is ​​calculated by combining the transpose of the group demand distribution matrix with the group demand distribution matrix. The transpose of the group demand distribution matrix is ​​a 50x15 matrix, where the row indices correspond to 50 patients and the column indices correspond to 15 demand state codes. The demand state co-occurrence matrix is ​​calculated by matrix multiplication of the group demand distribution matrix and its transpose, resulting in a 15x15 square matrix. Each element in the co-occurrence matrix represents the weighted sum of the co-occurrence frequencies of the two demand state codes across all patients. In the previous example, the value of the element in the first row and second column of the co-occurrence matrix is ​​78, indicating that the weighted sum of the co-occurrence frequencies of demand state codes 5324 and 6428 across all patients is 78.

[0059] The demand state co-occurrence matrix is ​​zeroed out along its diagonal, and the off-diagonal elements are extracted to construct a demand state association strength matrix. Zeroing out the diagonal elements eliminates the influence of co-occurrence frequencies between different demand states by setting the values ​​of the main diagonal elements of the co-occurrence matrix to 0. The off-diagonal elements constitute the demand state association strength matrix, which shows the degree of association between different demand states. In the previous example, the element in the first row and second column of the demand state association strength matrix has a value of 78, and the element in the second row and first column has a value of 78, indicating a strong association between demand state codes 5324 and 6428.

[0060] The demand state association strength matrix is ​​row-normalized to obtain the demand state transition probability matrix. Row normalization is achieved by dividing each element of the matrix by the sum of the elements in that row, ensuring that the sum of the elements in each row is 1. In the previous example, the sum of the elements in the first row of the demand state association strength matrix is ​​267. After row normalization, the value of the element in the second column of the first row becomes 0.292, indicating that the probability of transitioning from demand state code 5324 to 6428 is 0.292. The same operation is performed on all rows to obtain the complete demand state transition probability matrix.

[0061] A directed weighted graph is constructed based on the demand state transition probability matrix, with demand state codes used as graph nodes. The directed weighted graph contains 15 nodes, corresponding to 15 different demand state codes. The edges connecting nodes point from the source state to the target state, and the weight of each edge is the corresponding transition probability value in the transition probability matrix. To improve visualization, a transition probability threshold of 0.1 is set, and only edges with transition probabilities greater than the threshold are retained. In the example above, the edge weight from node 5324 to node 6428 is 0.292, which is greater than the threshold, so this edge is retained.

[0062] Strongly connected component decomposition (SCT) was performed on the directed weighted graph to extract demand association patterns. The SCT employed a depth-first search algorithm to identify strongly connected subgraphs within the graph. In the aforementioned example, the SCT result contained three subgraphs: the first subgraph contained nodes 5324, 6428, and 5318, indicating a close transition relationship between these three demand states; the second subgraph contained nodes 4226 and 6422, indicating a close transition relationship between these two demand states; and the third subgraph contained the remaining independent nodes. Each strongly connected component represents a typical demand association pattern, reflecting common demand state evolution patterns observed in clinical practice.

[0063] The degree of difference between demand state pairs corresponding to each element in the demand state association strength matrix is ​​calculated, and high-difference demand state pairs are identified. The degree of difference between demand state pairs is calculated by analyzing the category code and intensity code corresponding to the two demand state codes. The demand state code is parsed into category code and intensity code, the category difference value and intensity difference value are calculated, and the combined value is used to obtain the comprehensive degree of difference. In the previous example, demand state code 5324 is parsed into category code 1 and intensity code 4, and demand state code 4226 is parsed into category code 2 and intensity code 6. The category difference value is 1, the intensity difference value is 2, and the comprehensive degree of difference is 1.5. A degree of difference threshold of 1.2 is set, and demand state pairs with a degree of difference greater than the threshold are identified as high-difference demand state pairs. In the previous example, the degree of difference between 5324 and 4226 is 1.5, which is greater than the threshold, and they are identified as high-difference demand state pairs.

[0064] Extract the set of patients with co-occurring high-difference demand state pairs in the group demand distribution matrix and label demand conflict nodes. For each high-difference demand state pair, find the set of patients in the group demand distribution matrix that simultaneously contain both state codes. In the previous example, demand state pairs 5324 and 4226 co-occurred in 8 patients, with corresponding patient identifiers of 1003, 1007, 1015, 1022, 1028, 1034, 1041, and 1045. Analysis of the clinical records corresponding to these patients revealed that 5 of them experienced demand conflicts during treatment, such as a priority conflict between physiological care needs and daily activity care needs. These patients were labeled as demand conflict nodes.

[0065] In this embodiment, by collecting demand state codes from multiple neurosurgical patients and constructing a coding sequence according to patient identifiers, and further constructing a group demand distribution matrix through frequency statistics and mapping dictionaries, the demand states of different patients are organized and expressed in a unified data structure. This enables centralized modeling and overall analysis of multi-patient demand information, thereby improving the systematic nature and data utilization efficiency of group demand feature identification. By performing matrix operations on the group demand distribution matrix to obtain a demand state co-occurrence matrix, and extracting off-diagonal elements to construct a demand state association strength matrix, the co-occurrence relationship between different demand states is quantitatively expressed, enabling a more objective characterization of the degree of association between different demand states and improving the accuracy and quantifiability of demand relationship analysis. By performing strongly connected component decomposition on the directed weighted graph, stable association patterns between demand states can be identified, thereby revealing the potential organizational relationships in the group demand structure. This can more effectively mine the structured association features between demand states, improving the depth and reliability of demand pattern recognition. By calculating the degree of difference between demand state pairs and identifying highly different demand state pairs, and then combining the group demand distribution matrix to extract the corresponding co-occurring patient set and mark demand conflict nodes, the conflict structure between demands can be effectively discovered, improving the ability to identify differences and potential contradictions in group demands.

[0066] Figure 2 This is a flowchart illustrating the demand state coding analysis of the intelligent assessment method for the specialized care needs of neurosurgical patients, as described in an embodiment of the present invention.

[0067] In one alternative implementation, Based on the conflicting demand nodes and preset resource capacity constraints, resource allocation weights are calculated, and the demand status codes are corrected to obtain the demand assessment results, including: Extract the set of demand status codes corresponding to the demand conflict nodes and count the number of patients to obtain the conflict scale vector. Extract different total care resource capacities from the preset resource capacity constraints to construct a resource capacity vector. Analyze the resource demand types corresponding to the set of demand status codes and establish a mapping relationship with the resource capacity vector. Calculate the initial resource allocation weight based on the conflict scale vector and the resource capacity vector, and sort the set of demand status codes to obtain a priority sequence. The priority sequence is traversed to extract the corresponding patient identifiers. A patient demand coding group is constructed based on the patient identifiers and the group demand distribution matrix. The coding difference degree between the patient demand coding group and the demand status coding corresponding to the demand conflict node is calculated. A difference degree matrix is ​​constructed based on the coding difference degree, and a patient conflict intensity vector is constructed. A corrected resource allocation weight is calculated based on the patient conflict intensity vector and the initial resource allocation weight, and a resource allocation simulation is performed to determine the amount of allocated resources. The amount of allocated resources and the resource capacity vector are compared in real time. When the amount of allocated resources exceeds the resource capacity vector, the unallocated demand status coding is marked as a coding to be corrected and corrected. The demand assessment result is obtained by combining the amount of allocated resources.

[0068] The conflict scale vector is obtained by extracting the set of demand state codes corresponding to the demand conflict nodes and counting the number of patients. A demand conflict node refers to a patient with multiple different demand states within the same observation period. The corresponding demand state codes are extracted from the marked demand conflict nodes in the aforementioned embodiments to obtain a set of demand state codes. This code set contains five codes: 5324, 4226, 6428, 5318, and 6422, corresponding to five demand states: severe physiological care needs, mild daily activity care needs, severe respiratory and circulatory care needs, moderate physiological care needs, and moderate respiratory and circulatory care needs, respectively. The number of patients involved in each demand state code is counted to construct the conflict scale vector. In the aforementioned example, code 5324 involves 24 patients, code 4226 involves 18 patients, code 6428 involves 22 patients, code 5318 involves 15 patients, and code 6422 involves 19 patients. The conflict scale vector is (24, 18, 22, 15, 19).

[0069] A resource capacity vector is constructed by extracting the total capacity of different care resources from the preset resource capacity constraints. The preset resource capacity constraints include the total capacity of various care resources that the hospital's neurosurgery department can provide. In the example above, the hospital's neurosurgery department can provide 16 specialist nurses, 22 monitoring devices, 12 auxiliary care personnel, 8 rehabilitation therapy devices, and a daily usage limit of 60 units for special drugs and consumables. The resource capacity vector constructed by extracting the aforementioned resource capacity values ​​is (16, 22, 12, 8, 60).

[0070] The resource demand types corresponding to the demand status codes are analyzed, and a mapping relationship between them and the resource capacity vector is established. Demand status codes are analyzed into category codes and intensity codes, which are further mapped to specific resource demand types. In the aforementioned example, code 5324 is analyzed as a relatively heavy physiological care demand, corresponding to resource demands of 2 specialist nurses, 1 monitoring device, and 4 units of special drugs and consumables per day; code 4226 is analyzed as a relatively light daily activity care demand, corresponding to resource demands of 1 specialist nurse, 1 assistive care worker, and 1 rehabilitation therapy device; code 6428 is analyzed as a relatively heavy respiratory and circulatory care demand, corresponding to resource demands of 2 specialist nurses, 2 monitoring devices, and 5 units of special drugs and consumables per day. The mapping relationship between demand status codes and resource demands is established, forming a resource demand mapping table.

[0071] The initial resource allocation weights are calculated based on the conflict scale vector and resource capacity vector, and the priority sequence is obtained by sorting the demand status codes. The initial resource allocation weights are calculated by dividing the conflict scale vector by the corresponding resource capacity vector. The utilization ratio of each type of resource for each demand status code is calculated, and the overall resource utilization rate is obtained by summing them. In the example above, the overall resource utilization rate of code 5324 is 1.13, that of code 4226 is 0.82, that of code 6428 is 1.28, that of code 5318 is 0.71, and that of code 6422 is 0.95. The demand status codes are sorted in descending order according to the overall resource utilization rate, resulting in the priority sequence 6428, 5324, 6422, 4226, 5318.

[0072] The priority sequence is traversed to extract the corresponding patient identifiers, and a patient demand coding group is constructed based on the patient identifiers and the group demand distribution matrix. Demand status codes are extracted one by one from the priority sequence, and patient identifiers with those codes are found. In the previous example, the patient identifiers corresponding to code 6428 include 22 patients such as 1002, 1005, and 1009; the patient identifiers corresponding to code 5324 include 24 patients such as 1001, 1003, and 1007. For each patient identifier, all demand status codes for that patient during the observation period are extracted from the group demand distribution matrix to construct a patient demand coding group. In the previous example, the demand coding group for patient 1002 is 6428, 6428, 6428, 6422, 6422, 6416, and 6416; the demand coding group for patient 1001 is 5324, 5324, 5318, 5318, 5312, 5306, and 5306.

[0073] The coding dissimilarity between the patient's demand coding group and the demand state coding corresponding to the demand conflict node is calculated. A dissimilarity matrix and a patient conflict intensity vector are constructed based on this dissimilarity. The coding dissimilarity is calculated by comparing the category coding and intensity coding of two demand state codes. In the previous example, the dissimilarity between coding 5324 and coding 6428 is 1.41, and the dissimilarity between coding 5324 and coding 4226 is 1.5. For each patient, the dissimilarity between each coding in the corresponding demand coding group and all demand state codings corresponding to the conflict node is calculated, and a dissimilarity matrix is ​​constructed. The patient conflict intensity value is determined by the number of elements in the dissimilarity matrix whose dissimilarity is below a threshold. A dissimilarity threshold of 0.8 is set, and the conflict intensity value for each patient is calculated. In the previous example, the conflict intensity value for patient 1002 is 3, and the conflict intensity value for patient 1001 is 2. The patient conflict intensity vector is obtained by summing the conflict intensity values ​​of all patients.

[0074] Based on the patient conflict intensity vector and the initial resource allocation weights, a corrected resource allocation weight is calculated, and a resource allocation simulation is performed to determine the amount of resources already allocated. The corrected resource allocation weight is calculated by multiplying the initial resource allocation weight by a conflict intensity weighting factor. The conflict intensity weighting factor is set according to the patient conflict intensity vector; the higher the conflict intensity, the larger the weighting factor. In the aforementioned example, the initial resource allocation weight for code 6428 is 1.28, the conflict intensity weighting factor is 1.15, and the corrected resource allocation weight is 1.47; the initial resource allocation weight for code 5324 is 1.13, the conflict intensity weighting factor is 1.1, and the corrected resource allocation weight is 1.24. Resources are allocated preferentially according to the ranking result of the corrected resource allocation weights, the allocation process is simulated, and the amount of resources already allocated is recorded. In the aforementioned example, priority was given to allocating the resource needs corresponding to code 6428, involving 22 patients. The allocated resources are 44 specialist nurses, 44 monitoring devices, and 110 units of special drugs and consumables per day. Next, the resource needs corresponding to code 5324 were allocated, involving 24 patients. The total allocated resources are 92 specialist nurses, 68 monitoring devices, and 206 units of special drugs and consumables per day.

[0075] The allocated resource quantity and resource capacity vector are compared in real time. When the allocated resource quantity exceeds the resource capacity vector, the unallocated demand status code is marked as a code to be corrected and then corrected. The demand assessment result is obtained by combining the allocated resource quantity. During the resource allocation process, it is checked in real time whether the allocated quantity of each type of resource exceeds the resource capacity. In the example above, the allocated quantity of 92 specialist nursing staff exceeded the resource capacity by 16, the allocated quantity of 68 monitoring devices exceeded the resource capacity by 22, and the allocated quantity of 206 units of special drugs and consumables exceeded the resource capacity by 60 units. Codes 6422, 4226, and 5318 that have not been allocated are marked as codes to be corrected. The codes to be corrected are adjusted according to the resource excess, which is achieved by reducing the demand intensity or changing the demand category. In the example above, code 6422 is corrected to 6416, indicating a reduction in demand intensity; code 4226 is corrected to 4220, indicating a change in demand category. After correction, the resource allocation is recalculated to ensure that the total resource allocation does not exceed the resource capacity, and the demand assessment results are obtained, including the demand status code of each patient, the resource allocation status, and the satisfaction score.

[0076] In this embodiment, the initial resource allocation weight is calculated based on the conflict scale vector and the resource capacity vector, and the demand state code set is prioritized. This enables the resource allocation process to be scheduled in an orderly manner according to the demand scale and resource constraints, and can more reasonably reflect the urgency and resource occupation relationship between different demand states, thereby improving the rationality and efficiency of resource allocation. By extracting the patient demand code group and calculating the coding difference degree between it and the demand conflict node, a difference degree matrix and a patient conflict intensity vector are constructed, which can quantify the degree of conflict between different patient demands and effectively improve the balance and coordination of resource allocation. By comparing the allocated resource amount with the resource capacity vector in real time during the resource allocation process, and marking and correcting the demand state codes that exceed the resource capacity limit, the resource allocation process can be dynamically adjusted under the capacity constraint condition, which can avoid the problem of over-occupancy or resource imbalance during the resource allocation process, and improve the controllability and stability of resource allocation.

[0077] In one alternative implementation, Calculating the matching score between the demand assessment result and the care plan templates in the preset care plan library, sorting the care plan templates based on the matching score and determining the candidate care plan sequence, and verifying the feasibility of the candidate care plan sequence to determine the optimal care instruction and issue it for execution includes: Care plan templates are obtained from a pre-set care plan library, and the care action type and resource configuration information corresponding to each care plan template are extracted. The needs assessment results are semantically mapped with the care action type and resource configuration information corresponding to the care plan templates, and the needs coverage is calculated. The structural similarity between the needs assessment results and the care plan templates is calculated. Based on the needs coverage and the structural similarity, dual constraints are constructed and a comprehensive matching score is calculated. Based on the comprehensive matching score, the care plan templates in the care plan library are sorted and a candidate care plan sequence is constructed. The process involves traversing the candidate care plan sequence and extracting the resource requirement list of the care plan template. It also involves querying the available resource status in the resource capacity constraints and marking resource reachability. The execution dependencies of the care plan template are extracted, and dependency integrity is determined based on the requirement assessment results. The executability of the care plan template is determined based on the resource reachability and dependency integrity. Finally, the optimal care instruction is generated and issued for execution based on the care plan template with the best executability.

[0078] The system retrieves care plan templates from a pre-defined care plan library and extracts the care action types and resource configuration information corresponding to each template. The care plan library contains standard care plan templates for different neurosurgical needs. In the aforementioned example, the library contains 85 care plan templates, covering common neurosurgical disease types and care need scenarios. Each care plan template contains multiple care action types and corresponding resource configuration information. For example, care plan template A105 corresponds to the physiological care needs of patients with traumatic brain injury, including care action types such as vital sign monitoring, pupil observation, head positioning adjustment, and drainage tube care, with corresponding resource configuration information of 2 specialist nurses, 1 monitoring device, and 3 units of special drugs and consumables per day; care plan template B078 corresponds to the respiratory and circulatory care needs of patients with cerebrovascular disease, including care action types such as airway management, blood oxygen monitoring, postural drainage, and airway humidification, with corresponding resource configuration information of 2 specialist nurses, 2 monitoring devices, and 4 units of special drugs and consumables per day. Extract the care action types and resource configuration information from all care plan templates to form a care plan feature library.

[0079] The needs assessment results are semantically mapped to the corresponding care action types and resource allocation information of the care plan template, and the needs coverage is calculated. The needs assessment results include the patient's needs status code and resource allocation. In the example above, the needs assessment result for patient 1001 is code 5324, indicating a relatively heavy physiological care need, with allocated resources of 2 specialist nurses, 1 monitoring device, and 4 units of special medications and consumables per day. Semantic mapping is performed by matching the semantic parsing of the needs status code with the care action types of the care plan template. Semantic parsing uses a pre-trained medical domain word vector model to calculate the similarity between the semantics of the needs status and the semantics of the care actions. In the example above, code 5324 is parsed as a relatively heavy physiological care need, with a semantic similarity of 0.87 with the care action type of care plan template A105 and 0.52 with the care action type of template B078. Needs coverage is calculated based on the degree of matching between the resource needs in the needs assessment results and the resource allocation information of the care plan template. In the aforementioned example, the degree of matching between the needs assessment result of patient 1001 and the resource configuration information of template A105 is 0.92, and the degree of matching between the resource configuration information of template B078 is 0.68.

[0080] The structural similarity between the needs assessment results and the care plan template is calculated. Structural similarity is determined by comparing the similarity between the needs assessment results and the care plan template in structural features such as care action sequences, timing, and execution frequency. A graph structure matching algorithm is used to represent the care action sequences as a directed graph, and the similarity of the graph structure is calculated. In the aforementioned example, the structural similarity between patient 1001's needs assessment results and template A105 is 0.85, and the structural similarity with template B078 is 0.63.

[0081] A dual constraint condition based on demand coverage and structural similarity is constructed, and a comprehensive matching score is calculated. The dual constraint condition sets weight coefficients for demand coverage and structural similarity to comprehensively consider their impact on the matching result. The weight for demand coverage is 0.6, and the weight for structural similarity is 0.4. The comprehensive matching score is calculated by weighted averaging. In the aforementioned example, the comprehensive matching score between patient 1001's demand assessment result and template A105 is 0.89, and the comprehensive matching score with template B078 is 0.66.

[0082] Candidate care plan sequences were constructed by sorting care plan templates in the care plan library based on their comprehensive matching scores. The 85 care plan templates were then sorted in descending order according to their comprehensive matching scores, and the top 10 were selected as candidate care plan sequences. In the aforementioned example, the candidate care plan sequences for patient 1001 were A105, A108, A112, C034, A119, B052, C029, B078, A125, and C041, with corresponding comprehensive matching scores of 0.89, 0.85, 0.82, 0.79, 0.76, 0.73, 0.70, 0.66, 0.63, and 0.60, respectively.

[0083] Iterate through the candidate care plan sequence and extract the resource requirements list for each care plan template. Examine each care plan template in the candidate care plan sequence one by one and extract its resource requirements list. In the previous example, the resource requirements list for care plan template A105 is 2 specialist nurses, 1 monitoring device, and 3 units of daily usage of special medications and consumables; the resource requirements list for care plan template A108 is 2 specialist nurses, 1 monitoring device, 1 auxiliary caregiver, and 4 units of daily usage of special medications and consumables.

[0084] Query the available resource status in the resource capacity constraints and mark resource reachability. Resource capacity constraints include the total capacity and current availability of various resources. In the example above, the total capacity of specialist nursing staff is 16, with 4 currently available; the total capacity of monitoring equipment is 22 units, with 6 currently available; the total capacity of auxiliary care staff is 12, with 5 currently available; and the daily usage limit for special drugs and consumables is 60 units, with 15 currently available. Resource reachability is determined by comparing the resource requirements of the care plan template with the current available resource quantity. In the example above, all resource requirements of care plan template A105 can be met, and resource reachability is marked as 1; care plan template A108 requires 1 auxiliary care staff member, which is currently met, and resource reachability is marked as 1.

[0085] The execution dependencies of the care plan templates are extracted, and dependency integrity is determined based on the needs assessment results. Execution dependencies refer to the preconditions that must be met before the care plan can be executed. In the aforementioned example, the execution dependencies of care plan template A105 include stable patient vital signs and a neurological score greater than 8; the execution dependencies of care plan template A108 include stable patient vital signs, a neurological score greater than 6, and a patent drainage tube. Dependency integrity is determined by comparing the patient status information in the needs assessment results with the execution dependencies. In the aforementioned example, patient 1001 has stable vital signs, a neurological score of 12, and a patent drainage tube; the dependency integrity of both care plan templates A105 and A108 is marked as 1.

[0086] The executability of care plan templates is determined based on resource reachability and dependency integrity. Optimal care instructions are generated and issued based on the most executable template. Executability is determined by the product of resource reachability and dependency integrity. In the previous example, the executability of care plan templates A105 and A108 is 1. Considering both executability and overall matching score, the optimal care plan template is selected. In the previous example, the overall matching score of care plan template A105 (0.89) is higher than that of A108 (0.85), and both have the same executability; therefore, A105 is selected as the optimal care plan template. Specific care instructions are generated based on the optimal care plan template, including care goals, care actions, execution frequency, resource allocation, and precautions. In the aforementioned example, the generated care instructions include: the care goal is to maintain stable vital signs and prevent increased intracranial pressure; care actions include monitoring vital signs hourly, observing pupils every 2 hours, adjusting head position every 4 hours, and providing drainage tube care three times per shift; resource allocation includes 2 specialist nurses, 1 monitoring device, and 3 units of special medications and consumables per day; precautions include observing changes in pupil size and being alert for signs of brain herniation. The care instructions are distributed to relevant medical staff for execution via the hospital information system.

[0087] In this embodiment, the structural similarity between the needs assessment results and the care plan template is calculated, and a dual constraint condition is constructed based on the needs coverage to form a comprehensive matching score. This ensures that the care plan selection not only considers the coverage of the needs content but also the matching relationship between the needs structure and the plan structure. This allows for a more comprehensive assessment of the fit between the care plan and the patient's needs, thereby improving the rationality and stability of the plan selection results. By analyzing the resource needs list of each candidate care plan sequence and judging resource accessibility based on resource capacity constraints, the care plan can meet the current resource supply conditions while matching the needs. Resource limitations can be identified in advance during the plan selection stage, thereby improving the feasibility of plan execution and reducing resource conflicts or execution interruptions. The executability of the care plan is determined based on resource accessibility and dependency integrity, and the plan with the best executability is selected to generate care instructions and issue them for execution. This ensures that the final output care plan not only highly matches the needs but also has the actual execution conditions. This not only improves the accuracy of care plan selection but also enhances the feasibility and stability of plan execution, thereby effectively improving the scientific nature and implementation efficiency of neurosurgical patient care decisions.

[0088] A second aspect of this invention provides an intelligent assessment system for the specialized care needs of neurosurgical patients, comprising: The feature construction unit is used to collect physiological and behavioral data of neurosurgical patients, perform time-domain decomposition on the physiological data to obtain periodic fluctuation components and mutation components and construct physiological features, encode the behavioral data into action sequences and construct behavioral pattern features, and concatenate and nonlinearly map the physiological features and behavioral pattern features to obtain the patient's basic feature vector. The demand encoding unit is used to construct a demand space based on the patient's basic feature vector and perform projection mapping to obtain demand representation points, calculate the radial distance corresponding to the demand representation points and determine the demand intensity level, calculate the direction angle corresponding to the demand representation points and determine the demand category, and combine the demand intensity level and the demand category to obtain the demand status code. The demand assessment unit is used to aggregate the demand status codes and construct a group demand distribution matrix, extract demand association patterns and determine demand conflict nodes based on the group demand distribution matrix, calculate resource allocation weights based on the demand conflict nodes and preset resource capacity constraints, and correct the demand status codes to obtain the demand assessment results. The care matching unit is used to calculate the matching score between the needs assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

[0089] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0090] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0091] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent assessment method for the specialized care needs of neurosurgical patients, characterized in that, include: Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and mutation components and to construct physiological features. The behavioral data is encoded by action sequence and to construct behavioral pattern features. The physiological features and behavioral pattern features are spliced ​​and nonlinearly mapped to obtain the patient's basic feature vector. Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to the demand representation points is calculated and the demand intensity level is determined. The directional angle corresponding to the demand representation points is calculated and the demand category is determined. The demand intensity level and the demand category are combined and encoded to obtain the demand status code. The process involves aggregating the demand status codes and constructing a group demand distribution matrix. Based on this matrix, demand association patterns are extracted and demand conflict nodes are identified. Based on these conflict nodes and preset resource capacity constraints, resource allocation weights are calculated, and the demand status codes are corrected to obtain the demand assessment result, including: Multiple neurosurgical patients' demand status codes are collected and organized into a coding sequence according to patient identification. Frequency statistics are performed on the coding sequence to obtain the demand status frequency distribution and a mapping dictionary is constructed. The unique demand status value in the mapping dictionary is used as the row index and the patient identification is used as the column index to construct a matrix frame. The demand status codes of different patients in the coding sequence are filled into the corresponding row and column positions of the matrix frame and the empty positions are filled with zero values ​​to obtain the group demand distribution matrix. The transpose of the group demand distribution matrix is ​​calculated, and a demand state co-occurrence matrix is ​​obtained by combining the group demand distribution matrix. The diagonal elements of the demand state co-occurrence matrix are zeroed out, and off-diagonal elements are extracted to construct a demand state association strength matrix. The row normalization of the demand state association strength matrix is ​​performed to obtain a demand state transition probability matrix. A directed weighted graph is constructed based on the demand state transition probability matrix, and the demand states are encoded as graph nodes. The directed weighted graph is decomposed into strongly connected components, and demand association patterns are extracted. The difference between demand state pairs corresponding to each element in the demand state association strength matrix is ​​calculated, and high-difference demand state pairs are identified. The set of patients co-occurring high-difference demand state pairs in the group demand distribution matrix is ​​extracted, and demand conflict nodes are marked. Extract the set of demand status codes corresponding to the demand conflict nodes and count the number of patients to obtain the conflict scale vector. Extract different total care resource capacities from the preset resource capacity constraints to construct a resource capacity vector. Analyze the resource demand types corresponding to the set of demand status codes and establish a mapping relationship with the resource capacity vector. Calculate the initial resource allocation weight based on the conflict scale vector and the resource capacity vector, and sort the set of demand status codes to obtain a priority sequence. The priority sequence is traversed to extract the corresponding patient identifiers. Based on the patient identifiers and the group demand distribution matrix, a patient demand coding group is constructed. The coding difference degree between the patient demand coding group and the demand status coding corresponding to the demand conflict node is calculated. Based on the coding difference degree, a difference degree matrix is ​​constructed and a patient conflict intensity vector is constructed. Based on the patient conflict intensity vector and the initial resource allocation weight, a corrected resource allocation weight is calculated and a resource allocation simulation is performed to determine the allocated resource amount. The allocated resource amount and the resource capacity vector are compared in real time. When the allocated resource amount exceeds the resource capacity vector, the unallocated demand status coding is marked as a coding to be corrected and corrected. The demand assessment result is obtained by combining the allocated resource amount. Calculate the matching score between the demand assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

2. The method according to claim 1, characterized in that, Physiological and behavioral data of neurosurgical patients are collected. The physiological data is decomposed in the time domain to obtain periodic fluctuation components and abrupt change components, and physiological features are constructed. The behavioral data is encoded into action sequences to construct behavioral pattern features. The physiological features and behavioral pattern features are concatenated and nonlinearly mapped to obtain the patient's basic feature vector, including: Physiological data of neurosurgical patients are collected and segmented to obtain time-series data segments. Frequency components are extracted and periodic fluctuation components are identified by frequency domain transformation of the time-series data segments. The gradient of the time-series data segments is calculated and abrupt components are identified. The amplitude envelope of the periodic fluctuation components is calculated and a periodic feature vector is constructed. The peak position of the abrupt components is calculated and an abrupt feature vector is constructed. Physiological characteristics are obtained based on the periodic feature vector and the abrupt feature vector. Behavioral data of neurosurgical patients are collected, and the action identifiers in the behavioral data are arranged to obtain an action sequence. The action sequence is segmented by a sliding window to obtain a set of action segments. The frequency of occurrence of different action segments in the set of action segments is counted and a frequency distribution vector is constructed. The transition probability between adjacent actions in the action sequence is calculated and a state transition matrix is ​​constructed. The state transition matrix is ​​decomposed by feature value to extract the dominant transition pattern and concatenated with the frequency distribution vector to obtain behavioral pattern features. The physiological features and behavioral pattern features are dimensionally aligned and concatenated according to feature type to obtain a hybrid feature vector. The hybrid feature vector is then subjected to nonlinear mapping and dimensionality reduction to obtain the patient's basic feature vector.

3. The method according to claim 1, characterized in that, Based on the patient's basic feature vector, a demand space is constructed and projected to obtain demand representation points. The radial distance corresponding to each demand representation point is calculated, and the demand intensity level is determined, including: Covariance analysis is performed on the patient's basic feature vectors to obtain a feature correlation matrix. Eigenvalue decomposition is performed on the feature correlation matrix to extract principal component direction vectors and corresponding eigenvalues. The cumulative contribution rate is calculated based on the eigenvalues ​​to determine the number of effective principal components. An orthogonal basis is constructed based on the principal component direction vectors corresponding to the number of effective principal components, and the coordinate axes of the demand space are determined. The contribution rate of the principal component direction vectors corresponding to the effective principal components is calculated, and the number of dimensions of the demand space is determined. The demand space is constructed based on the orthogonal basis and the number of dimensions. The projection coefficient vector is obtained by performing an inner product operation on the patient's basic feature vector and the orthogonal basis. The coordinate position of the demand representation point in the demand space is determined based on the projection coefficient vector. The coordinate position of the demand representation point is centered and the Euclidean distance between the demand representation point and the origin of the demand space is calculated to obtain the radial distance. The quantile statistics of the radial distance are performed to obtain the distance distribution interval. The intensity threshold boundary is divided based on the distance distribution interval and the radial distance is mapped to the interval corresponding to the intensity threshold boundary to obtain the demand intensity level.

4. The method according to claim 1, characterized in that, Calculate the direction angle corresponding to the demand characterization point and determine the demand category. Combine the demand intensity level and the demand category to obtain the demand status code, including: Extract the coordinate components of the demand representation point in the demand space and calculate the direction vector of the demand representation point relative to the origin of the demand space. Calculate the unit direction vector based on the direction vector and determine the angle between the unit direction vector and the basis vectors of each coordinate axis of the demand space to obtain a multi-dimensional direction angle set. Extract the dominant direction angle based on the multi-dimensional direction angle set. Find the angle interval to which the dominant direction angle belongs based on a pre-set mapping table and obtain the corresponding demand category identifier. Calculate the projection length of the unit direction vector in the subspace corresponding to the demand category identifier and determine the demand category based on the projection length. Obtain the level numerical code corresponding to the demand intensity level, perform category index mapping on the demand category to obtain the category numerical code, and determine the coding bit width allocation scheme in combination with the level numerical code. Based on the coding bit width allocation scheme, determine the high-bit coding segment and the low-bit coding segment. Calculate the combined coding value based on the high-bit coding segment and the low-bit coding segment. Calculate the product of the radial distance of the demand characterization point and the dominant direction angle and perform a hash transformation to obtain the check code. Based on the check code and the combined coding value, determine the extended coding structure and perform integrity verification to obtain the demand status code.

5. The method according to claim 1, characterized in that, Calculating the matching score between the demand assessment result and the care plan templates in the preset care plan library, sorting the care plan templates based on the matching score and determining the candidate care plan sequence, and verifying the feasibility of the candidate care plan sequence to determine the optimal care instruction and issue it for execution includes: Care plan templates are obtained from a pre-set care plan library, and the care action type and resource configuration information corresponding to each care plan template are extracted. The needs assessment results are semantically mapped with the care action type and resource configuration information corresponding to the care plan templates, and the needs coverage is calculated. The structural similarity between the needs assessment results and the care plan templates is calculated. Based on the needs coverage and the structural similarity, dual constraints are constructed and a comprehensive matching score is calculated. Based on the comprehensive matching score, the care plan templates in the care plan library are sorted and a candidate care plan sequence is constructed. The process involves traversing the candidate care plan sequence and extracting the resource requirement list of the care plan template. It also involves querying the available resource status in the resource capacity constraints and marking resource reachability. The execution dependencies of the care plan template are extracted, and dependency integrity is determined based on the requirement assessment results. The executability of the care plan template is determined based on the resource reachability and dependency integrity. Finally, the optimal care instruction is generated and issued for execution based on the care plan template with the best executability.

6. A smart assessment system for the specialized care needs of neurosurgical patients, used to implement the method of any one of claims 1-5, characterized in that, include: The feature construction unit is used to collect physiological and behavioral data of neurosurgical patients, perform time-domain decomposition on the physiological data to obtain periodic fluctuation components and mutation components and construct physiological features, encode the behavioral data into action sequences and construct behavioral pattern features, and concatenate and nonlinearly map the physiological features and behavioral pattern features to obtain the patient's basic feature vector. The demand encoding unit is used to construct a demand space based on the patient's basic feature vector and perform projection mapping to obtain demand representation points, calculate the radial distance corresponding to the demand representation points and determine the demand intensity level, calculate the direction angle corresponding to the demand representation points and determine the demand category, and combine the demand intensity level and the demand category to obtain the demand status code. The demand assessment unit is used to aggregate the demand status codes and construct a group demand distribution matrix, extract demand association patterns and determine demand conflict nodes based on the group demand distribution matrix, calculate resource allocation weights based on the demand conflict nodes and preset resource capacity constraints, and correct the demand status codes to obtain the demand assessment results. The care matching unit is used to calculate the matching score between the needs assessment result and the care plan templates in the preset care plan library, sort the care plan templates based on the matching score and determine the candidate care plan sequence, perform feasibility verification on the candidate care plan sequence to determine the optimal care instruction and issue it for execution.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images