Gynecological postoperative rehabilitation evaluation method and system based on disease course trajectory modeling

Abstract

The invention relates to the technical field of rehabilitation evaluation, in particular to a gynecological postoperative rehabilitation evaluation method and system based on disease course trajectory modeling. According to the method, by collecting objective physiology, subjective self-evaluation and behavior mode data of a patient, time-frequency domain features are extracted, and rehabilitation state feature vectors are constructed through adaptive weight distribution; searching an optimal clustering center by adopting a particle swarm optimization algorithm, and determining the number and position of rehabilitation attractors by combining the contour coefficient; constructing a disease course trajectory by the feature vectors according to a time sequence, and calculating the convergence probability of the trajectory to each attractor based on a competitive traction model; and establishing a statistical baseline model, identifying physiological fluctuation and pathological deviation through a Mahalanobis distance, and outputting a quantitative evaluation result in combination with a convergence probability. According to the invention, accurate quantitative evaluation and abnormal early warning of the rehabilitation state are realized, and a scientific basis is provided for personalized rehabilitation management.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation assessment technology, specifically to a method and system for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling. Background Technology

[0002] Postoperative rehabilitation assessment in obstetrics and gynecology is an important part of ensuring the quality of patient recovery and reducing the risk of complications, and it is of great significance to promoting the physical and mental health of postpartum women.

[0003] However, existing postoperative rehabilitation assessment methods suffer from the following technical deficiencies: First, the assessment methods rely too heavily on the subjective judgment and experience of medical staff, lacking objective and quantitative assessment standards, which may lead to differences in the judgments of different medical staff regarding the same rehabilitation status; second, traditional methods often use single or a few physiological indicators for assessment, which cannot comprehensively reflect the patient's overall rehabilitation status and are prone to missing important information; third, existing technologies lack modeling and analysis of the dynamic changes in the rehabilitation process, making it difficult to achieve early warning of rehabilitation abnormalities, and often only identifying them when abnormalities have already appeared; fourth, personalized rehabilitation guidance is insufficient, and it is impossible to provide targeted rehabilitation suggestions based on the patient's specific rehabilitation trajectory, which affects the optimization of rehabilitation outcomes. Summary of the Invention

[0004] This invention provides a method and system for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling, which can achieve precise quantitative analysis of rehabilitation status and early warning of abnormalities, providing a scientific basis for personalized rehabilitation management.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A postoperative rehabilitation assessment method for obstetrics and gynecology based on disease course trajectory modeling includes: S100: Collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract time-frequency domain features of each modality data, and construct a rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; S200: The particle swarm optimization algorithm is used to search for the optimal cluster center in the rehabilitation state space, and the number and location of rehabilitation attractors are determined by combining the profile coefficient verification. S300: Construct a disease trajectory from the feature vector of the rehabilitation status according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector; S400: Establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, identify and separate physiological rehabilitation fluctuations from pathological abnormal deviations. Combine the convergence probability of the disease trajectory to each rehabilitation attractor to output a rehabilitation assessment report containing a judgment on the nature of the fluctuations.

[0006] As a preferred embodiment of the present invention, the objective physiological data, subjective self-evaluation data, and behavioral pattern data specifically include: The objective physiological data include body temperature data, heart rate data, wound healing assessment data, white blood cell count data, and C-reactive protein data; The subjective self-assessment data includes pain visual analog scale (VAS) scores, daily activity ability (DDA) scores, sleep quality (SEE) scores, and emotional state (EQ) scores. The behavioral pattern data includes rehabilitation training participation data, medication adherence data, and rehabilitation guidance execution rate data.

[0007] As a preferred embodiment of the present invention, the construction of the rehabilitation state feature vector through the weighted fusion algorithm with adaptive weight allocation specifically includes: The coefficient of variation of each modal data within a preset time window is calculated as a data stability evaluation index. The basic weighting coefficients of each modality data are calculated using the analytic hierarchy process based on the coefficient of variation. The base weighting coefficients are adjusted based on the patient's surgical type and postoperative days; The adjusted weighting coefficients are combined linearly with the corresponding time-frequency domain features to generate a normalized rehabilitation state feature vector.

[0008] As a preferred embodiment of the present invention, the step of using a particle swarm optimization algorithm to search for the optimal cluster center in the rehabilitation state space and combining the silhouette coefficient verification to determine the number and position of rehabilitation attractors specifically includes: Initialize the position vector and velocity vector of each particle in the particle swarm, where the position vector represents the coordinates of the candidate cluster center; Calculate the Euclidean distance between the cluster center and the feature vector of the recovery state based on the current position of the particle, and update the individual optimal position and the global optimal position of the particle based on the principle of minimizing the distance. Calculate the particle's new velocity vector according to the velocity update formula, and calculate the particle's new position vector according to the position update formula; For each clustering number scheme, calculate the corresponding silhouette coefficient value, and select the number of clusters and the cluster center position corresponding to the maximum silhouette coefficient as the number and position of the rehabilitation attractor.

[0009] As a preferred embodiment of the present invention, the step of constructing a disease trajectory from the rehabilitation status feature vector according to a time series specifically includes: The feature vector sequence is arranged in chronological order of acquisition time; For any missing data points, cubic spline interpolation is used, and residual correction based on local polynomial regression is performed after interpolation. The noise reduction of the rehabilitation status feature vector sequence is performed using a sliding window smoothing algorithm; The processed sequence is connected by interpolation points in the recovery state space to form a continuous disease trajectory, and the local curvature is calculated on the disease trajectory by time segment.

[0010] As a preferred embodiment of the present invention, the convergence probability determination step specifically includes: Calculate the Euclidean distance between the current location of the patient's trajectory and each rehabilitation attractor. ; Calculation of traction force based on gravitational field model Satisfy the following formula: ; in For the first The density factor of each recovery attractor is obtained by local point density estimation in cluster analysis; These are normalized weighting coefficients determined based on surgical type and postoperative stage; To avoid singular regularization constants, The time decay constant, The time difference between the current position and the reference time; the composite traction force vector and the rehabilitation state feature vector are obtained by synthesizing the traction force vectors in vector form. And calculate the convergence probability based on the ratio of traction force components. .

[0011] As a preferred embodiment of the present invention, the establishment of a baseline model for physiological fluctuations based on statistical distribution specifically includes: Collect historical rehabilitation data of patients who underwent similar surgeries, and establish data sample sets according to surgical type and postoperative time period; Calculate the mean vector and covariance matrix of the rehabilitation status feature vectors for each data sample set; The maximum likelihood estimation method is used to fit the multivariate normal distribution parameters of the rehabilitation status feature vectors for each time period; A baseline model of physiological fluctuations with pre-set confidence levels is constructed based on the fitted multivariate normal distribution parameters, where the confidence interval boundaries are determined by the critical values ​​of the chi-square distribution.

[0012] As a preferred embodiment of the present invention, the identification and separation of physiological recovery fluctuations and pathological abnormal deviations specifically includes: Calculate the Mahalanobis distance between the current location of the disease trajectory and the mean vector of the baseline model for the corresponding time period; Compare the Mahalanobis distance with the critical distance threshold of the baseline model confidence interval; When the Mahalanobis distance is less than or equal to the critical distance threshold, it is marked as a physiological recovery fluctuation; When the Mahalanobis distance is greater than the critical distance threshold, the projection components of the deviation vector in each rehabilitation attractor direction are calculated, and the directional characteristics of the abnormal deviation are identified based on the distribution pattern of the projection components.

[0013] As a preferred embodiment of the present invention, the output including a rehabilitation assessment report on the nature of fluctuations specifically includes: Construct a rehabilitation trend vector based on the convergence probability distribution of each rehabilitation attractor; By associating and mapping the fluctuation property markers with the rehabilitation trend vector, a rehabilitation state descriptor containing fluctuation type, deviation degree, and convergence direction is generated. A multi-level decision tree structure is used to classify rehabilitation status descriptors and output a rehabilitation assessment report. The rehabilitation assessment report includes assessment information at three levels: rehabilitation progress level, abnormal risk warning, and personalized rehabilitation recommendations.

[0014] This invention also proposes a postoperative rehabilitation assessment system for obstetrics and gynecology based on disease trajectory modeling, comprising: The feature extraction and fusion module is used to collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract the time-frequency domain features of each modality data, and construct the rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; The clustering optimization analysis module is used to search for the optimal cluster center in the rehabilitation state space using the particle swarm optimization algorithm, and to determine the number and location of rehabilitation attractors by combining the profile coefficient verification. The trajectory convergence calculation module is used to construct the disease trajectory from the rehabilitation state feature vector according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector. The fluctuation identification and assessment module is used to establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, it identifies and separates physiological rehabilitation fluctuations from pathological abnormal deviations. Combined with the convergence probability of the disease trajectory to each rehabilitation attractor, it outputs a rehabilitation assessment report containing a judgment on the nature of the fluctuations.

[0015] The beneficial effects of this invention are: 1. This invention innovatively proposes the concept of rehabilitation attractors, identifies typical rehabilitation patterns in the rehabilitation state space using a particle swarm optimization algorithm, and establishes a disease trajectory convergence model based on competitive traction forces. This method abstracts the complex rehabilitation process into a computable dynamic system, predicts rehabilitation trends by calculating the convergence probability of the trajectory to each attractor, and achieves a fundamental shift from traditional qualitative assessment to precise quantitative analysis, improving the accuracy and objectivity of rehabilitation state prediction.

[0016] 2. This invention establishes a precise identification mechanism for physiological rehabilitation fluctuations and pathological abnormal deviations by combining adaptive weighted fusion of multimodal data with a statistical baseline model. The system dynamically adjusts the weights of each modality based on the type of surgery and postoperative stage, and combines Maharanobis distance calculation and deviation vector projection analysis to achieve accurate early warning in the early stages of rehabilitation abnormalities. Compared with the ex-post judgment mode of existing technologies, this invention achieves prospective identification of rehabilitation risks and personalized intervention guidance. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a postoperative rehabilitation assessment method for obstetrics and gynecology based on disease trajectory modeling, according to the present invention. Figure 2 This is a schematic diagram of the structure of a postoperative rehabilitation assessment system for obstetrics and gynecology based on disease trajectory modeling, according to the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] Example 1: As Figure 1 As shown, the present invention provides a postoperative rehabilitation assessment method for obstetrics and gynecology based on disease trajectory modeling, comprising: S100: Collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract time-frequency domain features of each modality data, and construct a rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; Furthermore, the objective physiological data, subjective self-assessment data, and behavioral pattern data specifically include: The objective physiological data include body temperature data, heart rate data, wound healing assessment data, white blood cell count data, and C-reactive protein data; The subjective self-assessment data includes pain visual analog scale (VAS) scores, daily activity ability (DDA) scores, sleep quality (SEE) scores, and emotional state (EQ) scores. The behavioral pattern data includes rehabilitation training participation data, medication adherence data, and rehabilitation guidance execution rate data.

[0020] Furthermore, the construction of the rehabilitation state feature vector through the weighted fusion algorithm with adaptive weight allocation specifically includes: The coefficient of variation of each modal data within a preset time window is calculated as a data stability evaluation index. The basic weighting coefficients of each modality data are calculated using the analytic hierarchy process based on the coefficient of variation. The base weighting coefficients are adjusted based on the patient's surgical type and postoperative days; The adjusted weighting coefficients are combined linearly with the corresponding time-frequency domain features to generate a normalized rehabilitation state feature vector.

[0021] Specifically, the data acquisition system simultaneously acquires three types of data from post-operative obstetric and gynecological patients. After acquisition, the system preprocesses the raw data, using Fourier transform and wavelet transform to extract time-domain statistical features and frequency-domain energy distribution features, respectively. Time-domain features include mean, variance, skewness, and kurtosis, while frequency-domain features include dominant frequency, spectral entropy, and power spectral density. These features comprehensively reflect the distribution patterns of each modality of data in both time and frequency dimensions.

[0022] After feature extraction is complete, the adaptive weight allocation process begins. The system first calculates the coefficient of variation (CV) for each modality within a 24-hour time window as a data stability evaluation index. The calculation formula is as follows: ,in Standard deviation, The value represents the mean. The coefficient of variation reflects the relative volatility of the data; the smaller the value, the more stable the data.

[0023] Based on the calculated coefficient of variation, the system uses the analytic hierarchy process (AHP) to construct a judgment matrix and calculates the basic weight coefficients of each modality using the eigenvector method. The basic weighting coefficients reflect the inherent importance of each modality of data.

[0024] Establish weighted adjustment factors based on the patient's surgical type. Establish a time adjustment factor based on the number of days post-surgery. For example, in this embodiment, for cesarean section, the surgery type adjustment factor is 1.2, the time adjustment factor for 1-3 days post-surgery is 1.3, and the time adjustment factor for 4-7 days post-surgery is 1.0. The final weighting coefficient is determined using the formula... Obtained through calculation.

[0025] After the weight allocation is completed, the system performs a weighted linear combination of the adjusted weight coefficients and the corresponding time-frequency domain features, calculated using the following formula: ,in For the first The feature vectors of each modality are then processed by L2 norm normalization to obtain the final rehabilitation state feature vector.

[0026] This adaptive weighting mechanism can dynamically adjust the importance of each modality's data based on the physiological characteristics of different surgical types and rehabilitation stages, and compared to fixed-weight methods, it can more accurately reflect the patient's true rehabilitation status. Weighting adjustment parameters. and The weighting was determined by analyzing historical case data and statistically analyzing rehabilitation patterns, ensuring the rationality and effectiveness of the weighting allocation.

[0027] S200: The particle swarm optimization algorithm is used to search for the optimal cluster center in the rehabilitation state space, and the number and location of rehabilitation attractors are determined by combining the profile coefficient verification. Furthermore, the step of using particle swarm optimization to search for the optimal cluster center in the rehabilitation state space, and combining this with silhouette coefficient verification to determine the number and location of rehabilitation attractors, specifically includes: Initialize the position vector and velocity vector of each particle in the particle swarm, where the position vector represents the coordinates of the candidate cluster center; Calculate the Euclidean distance between the cluster center and the feature vector of the recovery state based on the current position of the particle, and update the individual optimal position and the global optimal position of the particle based on the principle of minimizing the distance. Calculate the particle's new velocity vector according to the velocity update formula, and calculate the particle's new position vector according to the position update formula; For each clustering number scheme, calculate the corresponding silhouette coefficient value, and select the number of clusters and the cluster center position corresponding to the maximum silhouette coefficient as the number and position of the rehabilitation attractor.

[0028] Specifically, after obtaining the recovery state feature vector, the process of determining the recovery attractor begins. First, the particle swarm size is set to an integer multiple of the square root of the number of data samples. In the recovery state space, the position vector and velocity vector of each particle are randomly initialized, where the position vector represents the coordinates of the candidate cluster center in multidimensional space. The dimension of the position vector is consistent with the dimension of the recovery state feature vector.

[0029] In each iteration, the Euclidean distance between the cluster center and all recovery state feature vectors is calculated based on the particle's current position. Based on the principle of distance minimization, each recovery state feature vector is assigned to the nearest cluster center, forming the clustering result. The sum of squared distances within each cluster is calculated as the fitness function; a smaller fitness value indicates better clustering performance.

[0030] Based on the fitness evaluation results, the individual optimal position and global optimal position of the particle are updated. The individual optimal position records the best cluster center position in the particle's history, and the global optimal position records the best cluster center position among all particles.

[0031] According to the speed update formula: ; Calculate the new velocity vector of the particle, where As inertia weights, a linear decreasing strategy can be used to calculate them, decreasing linearly with the number of iterations to balance global search and local search capabilities; and With all learning factors set to 2.0, this configuration belongs to the full-type particle swarm optimization algorithm, which can maintain a balance between convergence speed and search performance, and can also be adjusted according to actual needs; and It is a random number; The optimal position for the individual; This is the globally optimal position. Then, the position is updated according to the formula. Calculate the new position vector of the particle. When the change in the global optimal fitness value during continuous iterations is less than the preset convergence threshold, convergence is determined and the iteration process is terminated.

[0032] After particle swarm optimization is completed, the optimal number and location of healing attractors are determined. The particle swarm optimization process described above is performed for each scheme with different numbers of clusters to obtain the corresponding clustering results.

[0033] For each clustering result, a silhouette coefficient is calculated to evaluate the clustering quality. The silhouette coefficient is calculated using the formula: ; in For the sample Average distance from similar samples For the sample The average distance to the nearest outlier sample. The mean of the silhouette coefficients of all samples is calculated as the overall silhouette coefficient value for this clustering scheme.

[0034] The number of clusters and the location of the cluster centers corresponding to the maximum silhouette coefficients are selected as the final number and location of rehabilitation attractors. These rehabilitation attractors represent typical rehabilitation patterns in the rehabilitation state space, with each attractor corresponding to a specific rehabilitation state type, such as rapid rehabilitation, normal rehabilitation, slow rehabilitation, and abnormal rehabilitation state.

[0035] By leveraging the global search capabilities of the particle swarm optimization algorithm and the objective evaluation of the contour coefficient, the attractor distribution that best reflects the recovery pattern can be automatically identified, providing a reliable reference benchmark for subsequent disease trajectory analysis.

[0036] S300: Construct a disease trajectory from the feature vector of the rehabilitation status according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector; Furthermore, the specific steps of constructing a disease trajectory from the recovery status feature vectors according to a time series include: The feature vector sequence is arranged in chronological order of acquisition time; For any missing data points, cubic spline interpolation is used, and residual correction based on local polynomial regression is performed after interpolation. The noise reduction of the rehabilitation status feature vector sequence is performed using a sliding window smoothing algorithm; The processed sequence is connected by interpolation points in the recovery state space to form a continuous disease trajectory, and the local curvature is calculated on the disease trajectory by time segment.

[0037] Specifically, the rehabilitation status feature vectors collected at different time points are first arranged in chronological order to form time series data. Since data gaps may occur during the actual data collection process, cubic spline interpolation is used to supplement the missing data points. Cubic spline interpolation fits the known data points by constructing piecewise cubic polynomials, ensuring that the interpolation results have continuous first and second derivatives at connection points.

[0038] After interpolation is completed, residual correction based on local polynomial regression is performed. This method improves interpolation accuracy by fitting a low-order polynomial in the neighborhood of each interpolation point, calculating the residual between the interpolation result and the fitted result, and making corresponding corrections to the interpolation.

[0039] Subsequently, a sliding window smoothing algorithm was used to denoise the feature vector sequence of the rehabilitation state. A window length of 5 time points was selected, and a weighted moving average method was used to assign different weights to the data points within the window, with the center point having the largest weight and decreasing towards both ends, effectively filtering out high-frequency noise.

[0040] Finally, the processed sequence is connected by interpolation points in the recovery state space to form a continuous disease trajectory, and the local curvature is calculated on the trajectory segmented by time. The local curvature is obtained by calculating the second derivative of the trajectory at that point, reflecting the degree of aggression or slowness of the changes in the recovery state.

[0041] Furthermore, after the disease trajectory is constructed, the competitive traction force exerted by each rehabilitation attractor on each location point on the trajectory is calculated. First, the Euclidean distance between the current location point on the disease trajectory and each rehabilitation attractor is calculated. .

[0042] The traction force of each rehabilitation attractor on the current position point is calculated based on the improved gravitational field model. The formula for calculating traction force is: ; in: For the first The density factor of each rehabilitation attractor is obtained by local point density estimation in cluster analysis. The calculation method is to count the number of sample points in a spherical region centered on the attractor and with a radius equal to the average intra-class distance. These are normalized weighting coefficients determined based on the type of surgery and postoperative stage; for example, for cesarean section, the weighting coefficients are determined 1-3 days postoperatively. A value of 1.2 was used to reinforce the influence of objective physiological data, 4-7 days post-surgery. The value is 1.0, and it is taken more than 8 days after the operation. The value is 0.9. For other surgical types, the weighting coefficient can be adjusted accordingly. To avoid regularization parameters with a denominator of zero, a value of one-hundredth of the minimum inter-class distance is used. λ is a time decay constant, determined based on the time characteristics of the rehabilitation process, used to simulate the natural decay of traction force over time. This represents the time interval between the current time and the reference time, in days.

[0043] After calculating each traction force, the vectors of each traction force are vectored together to obtain the total traction force vector. The direction of the traction force is from the current position point to the corresponding rehabilitation attractor.

[0044] The convergence probability is calculated based on the proportion of each traction force in the total intensity of all traction forces: ,in This represents the magnitude of the traction force vector. The convergence probability reflects the strength of the tendency for the disease trajectory to converge to a specific rehabilitation attractor; a larger probability value indicates a higher likelihood of convergence to that attractor.

[0045] This mathematical description based on a physical gravitational field model can quantify the dynamic evolution of the disease trajectory in the rehabilitation state space, providing a precise mathematical basis for subsequent rehabilitation assessment.

[0046] S400: Establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, identify and separate physiological rehabilitation fluctuations from pathological abnormal deviations. Combine the convergence probability of the disease trajectory to each rehabilitation attractor to output a rehabilitation assessment report containing a judgment on the nature of the fluctuations.

[0047] Furthermore, the establishment of a baseline model for physiological fluctuations based on statistical distribution specifically includes: Collect historical rehabilitation data of patients who underwent similar surgeries, and establish data sample sets according to surgical type and postoperative time period; Calculate the mean vector and covariance matrix of the rehabilitation status feature vectors for each data sample set; The maximum likelihood estimation method is used to fit the multivariate normal distribution parameters of the rehabilitation status feature vectors for each time period; A baseline model of physiological fluctuations with pre-set confidence levels is constructed based on the fitted multivariate normal distribution parameters, where the confidence interval boundaries are determined by the critical values ​​of the chi-square distribution.

[0048] Specifically, historical rehabilitation data of patients undergoing similar surgeries are first collected, and data sample sets are established according to the type of surgery and the postoperative time period. For example, in this embodiment, for cesarean section surgery, the postoperative rehabilitation process is divided into three time periods: acute phase (1-3 days), recovery phase (4-7 days), and stable phase (8 days or more), and a separate sample set is established for each time period.

[0049] For each set of data samples, calculate the mean vector μ and covariance matrix of the rehabilitation status feature vectors. The mean vector is calculated using the arithmetic mean: The covariance matrix is ​​obtained through the formula Calculation, where For the sample size, For the first The feature vector of each sample.

[0050] The maximum likelihood estimation method was used to fit the multivariate normal distribution parameters of the rehabilitation status feature vectors for each time period. After confirming that the data conformed to the multivariate normal distribution assumption through normality tests (such as the Shapiro-Wilke test), the calculated mean vector μ and covariance matrix were... As a multivariate normal distribution The parameters.

[0051] A baseline model for physiological fluctuations was constructed by selecting a 95% confidence level based on commonly used standards in medical statistics.

[0052] for The confidence ellipsoid is determined by a critical value of the Mahalanobis distance for a 3D multivariate normal distribution. The critical distance threshold is calculated as follows: ,in For the number of dimensions The chi-square distribution critical value with a significance level of 0.05. The feature vector of the recovery state.

[0053] The physiological fluctuation baseline model is defined as: in the rehabilitation state space, with the mean vector Centered on, the distance to Maharanobis is less than or equal to The ellipsoidal region covers 95% of the normal rehabilitation state, and its shape and orientation are determined by the covariance matrix. The eigenvalues ​​and eigenvectors are determined.

[0054] Furthermore, the identification and separation of physiological recovery fluctuations from pathological abnormal deviations specifically includes: Calculate the Mahalanobis distance between the current location of the disease trajectory and the mean vector of the baseline model for the corresponding time period; Compare the Mahalanobis distance with the critical distance threshold of the baseline model confidence interval; When the Mahalanobis distance is less than or equal to the critical distance threshold, it is marked as a physiological recovery fluctuation; When the Mahalanobis distance is greater than the critical distance threshold, the projection components of the deviation vector in each rehabilitation attractor direction are calculated, and the directional characteristics of the abnormal deviation are identified based on the distribution pattern of the projection components.

[0055] Specifically, after the baseline model is established, the automatic identification of fluctuation characteristics begins. The Mahalanobis distance between the current location of the disease trajectory and the mean vector of the baseline model for the corresponding time period is calculated using the following formula: ,in This is the current location. The baseline mean. Let be the covariance matrix.

[0056] The calculated Mahalanobis distance is compared with the critical distance threshold of the baseline model confidence interval. When the Mahalanobis distance is less than or equal to the critical distance threshold, the current fluctuation is determined to be a physiological recovery fluctuation, indicating that the patient's recovery status is within the normal range.

[0057] When the Mahalanobis distance exceeds a critical distance threshold, it indicates an abnormal deviation, requiring further analysis of the deviation's nature. The deviation vector is then calculated. The projection components in each direction of the rehabilitation attractor. Let the first... The direction vectors of the rehabilitation attractors are: Then the projection of the deviation vector in that direction is: .

[0058] The directional characteristics of abnormal deviations are identified based on the distribution pattern of the projected components. When the deviation vector is mainly projected in the direction of the rapid recovery attractor, it indicates that the patient's recovery speed is exceeding expectations; when it is mainly projected in the direction of the abnormal recovery attractor, it indicates that there may be a recovery obstacle.

[0059] Furthermore, the output, which includes a rehabilitation assessment report that assesses the nature of fluctuations, specifically includes: Construct a rehabilitation trend vector based on the convergence probability distribution of each rehabilitation attractor; By associating and mapping the fluctuation property markers with the rehabilitation trend vector, a rehabilitation state descriptor containing fluctuation type, deviation degree, and convergence direction is generated. A multi-level decision tree structure is used to classify rehabilitation status descriptors and output a rehabilitation assessment report. The rehabilitation assessment report includes assessment information at three levels: rehabilitation progress level, abnormal risk warning, and personalized rehabilitation recommendations.

[0060] Specifically, after identifying the fluctuation characteristics, the identification results are comprehensively analyzed along with the convergence probabilities of the disease trajectory towards each rehabilitation attractor. A rehabilitation trend vector is then constructed based on the convergence probability distribution of each rehabilitation attractor. ,in To the first The convergence probability of a recovery attractor.

[0061] By associating fluctuation characteristics (physiological fluctuations or pathological abnormal deviations) with rehabilitation trend vectors, a rehabilitation state descriptor is generated, containing the fluctuation type, degree of deviation, and direction of convergence. The descriptor structure is as follows: ; A multi-layer decision tree structure is used to classify rehabilitation status descriptors. The first layer performs preliminary classification based on fluctuation type, the second layer assesses the degree of deviation based on Mahalanobis distance value, and the third layer determines the rehabilitation trend by combining convergence probability distribution.

[0062] The final output is a rehabilitation assessment report, which includes assessment information at three levels: rehabilitation progress level, abnormal risk warning, and personalized rehabilitation recommendations. The rehabilitation progress level is determined based on the dominant direction of the convergence probability, the abnormal risk warning is given based on the threshold exceeding the Mahalanobis distance, and the personalized rehabilitation recommendations provide targeted guidance based on the projection pattern of the deviation vector.

[0063] By combining statistical baseline models with dynamic convergence analysis, this method can accurately distinguish between normal rehabilitation fluctuations and abnormal pathological deviations, and provide a quantitative assessment of rehabilitation status, thus providing a scientific basis for clinical decision-making.

[0064] Example 2: In traditional postoperative rehabilitation management, the obstetrics and gynecology department of a certain hospital mainly relies on the subjective judgment of medical staff and routine indicator monitoring. This leads to problems such as inaccurate rehabilitation assessment, delayed identification of abnormalities, and a lack of personalized guidance. Especially for patients after cesarean section, traditional methods often focus only on single indicators such as body temperature and wound healing, making it difficult to comprehensively grasp the patient's overall recovery process, resulting in some abnormalities being overlooked or detected too late. Therefore, the hospital introduced this invention: a postoperative rehabilitation assessment system for obstetrics and gynecology based on disease trajectory modeling. Figure 2 As shown, it includes: The feature extraction and fusion module is used to collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract the time-frequency domain features of each modality data, and construct the rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; The clustering optimization analysis module is used to search for the optimal cluster center in the rehabilitation state space using the particle swarm optimization algorithm, and to determine the number and location of rehabilitation attractors by combining the profile coefficient verification. The trajectory convergence calculation module is used to construct the disease trajectory from the rehabilitation state feature vector according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector. The fluctuation identification and assessment module is used to establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, it identifies and separates physiological rehabilitation fluctuations from pathological abnormal deviations. Combined with the convergence probability of the disease trajectory to each rehabilitation attractor, it outputs a rehabilitation assessment report containing a judgment on the properties of the feature vector of the fluctuation rehabilitation state.

[0065] In a typical implementation, a 32-year-old post-cesarean section patient underwent a 7-day rehabilitation status monitoring and assessment: Patient basic information: A 32-year-old primiparous woman underwent a successful cesarean section and was in good physical condition before surgery. During a certain period on postoperative days 1-2: the system collected multimodal data including the patient's temperature (37.2°C), heart rate (82 bpm), good wound healing, and a VAS pain score of 6. Through an adaptive weight allocation algorithm, the system automatically increased the weight of objective physiological data (adjustment coefficient 1.3), and the constructed recovery status feature vector showed that the patient was in a normal postoperative acute reaction phase. The particle swarm optimization algorithm identified four recovery attractors: rapid recovery, normal recovery, slow recovery, and abnormal recovery. The patient's disease trajectory converged to the normal recovery attractor with a 78% probability, and the Mahalanobis distance was 1.8 (less than the critical threshold of 2.1), which the system determined to be physiological recovery fluctuation.

[0066] On postoperative days 3-4, traditional assessment methods showed that the patient's indicators were generally normal, but our system detected subtle changes in the rehabilitation status feature vector. The local curvature of the disease trajectory increased, the convergence probability to the normal rehabilitation attractor decreased to 52%, and the convergence probability to the slow rehabilitation attractor increased to 35%. The Mahalanobis distance increased to 2.3 (exceeding the critical threshold), and the system identified the abnormal deviation trend in advance. The system analyzed the projection components of the deviation vector in each attractor direction and found that the main projection was in the slow rehabilitation direction, warning of potential rehabilitation obstacles and recommending strengthened rehabilitation guidance and close monitoring.

[0067] On days 5-7 post-surgery, based on system alerts, medical staff intensified their monitoring of the patient and discovered a mild tendency towards intestinal adhesions, which affected normal activities. The rehabilitation plan was promptly adjusted, and targeted rehabilitation training was added. After intervention, the patient's disease trajectory re-converged towards normal rehabilitation, with the convergence probability recovering to 71% by day 7, and the Maharanobis distance decreasing to 1.9, indicating a significant improvement in rehabilitation status.

[0068] This typical case demonstrates that the system successfully identified abnormal recovery trends as early as the 3rd day after surgery, providing an early warning 2 days earlier than traditional methods. This early intervention saved valuable time and enabled personalized rehabilitation guidance.

[0069] During the three-month application period, the hospital conducted rehabilitation assessments on 120 post-operative obstetric and gynecological patients. The system operated stably and achieved good application results: it successfully provided personalized rehabilitation guidance for 115 patients with normal recovery; it provided early warning for 16 patients with abnormal recovery trends; and the distribution of rehabilitation status identified by the system was reasonable: 68% normal recovery, 17% rapid recovery, 12% slow recovery, and 3% abnormal recovery.

[0070] This invention successfully addresses the key problems of traditional rehabilitation assessment, such as strong subjectivity, delayed early warning, and lack of personalized guidance. Through rehabilitation attractor modeling and disease trajectory analysis, the system can quantify the complex rehabilitation process into a computable mathematical model, realizing a shift from qualitative judgment to quantitative analysis. This invention introduces rehabilitation attractors, abstracting typical rehabilitation patterns in the rehabilitation state space into attractors with competing traction forces. By calculating the convergence probability of the disease trajectory to each attractor, the rehabilitation trend can be accurately predicted. Combined with the fluctuation identification mechanism of the statistical baseline model, it effectively distinguishes between physiological rehabilitation fluctuations and pathological abnormal deviations, providing a new technical approach for postoperative rehabilitation management.

[0071] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease course trajectory modeling, characterized in that, include: S100: Collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract time-frequency domain features of each modality data, and construct a rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; S200: The particle swarm optimization algorithm is used to search for the optimal cluster center in the rehabilitation state space, and the number and location of rehabilitation attractors are determined by combining the profile coefficient verification. S300: Construct a disease trajectory from the feature vector of the rehabilitation status according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector; S400: Establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, identify and separate physiological rehabilitation fluctuations from pathological abnormal deviations. Combine the convergence probability of the disease trajectory to each rehabilitation attractor to output a rehabilitation assessment report containing a judgment on the nature of the fluctuations.

2. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The objective physiological data, subjective self-assessment data, and behavioral pattern data specifically include: The objective physiological data include body temperature data, heart rate data, wound healing assessment data, white blood cell count data, and C-reactive protein data; The subjective self-assessment data includes pain visual analog scale (VAS) scores, daily activity ability (DDA) scores, sleep quality (SEE) scores, and emotional state (EQ) scores. The behavioral pattern data includes rehabilitation training participation data, medication adherence data, and rehabilitation guidance execution rate data.

3. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The construction of the rehabilitation status feature vector through the weighted fusion algorithm with adaptive weight allocation specifically includes: The coefficient of variation of each modal data within a preset time window is calculated as a data stability evaluation index. The basic weighting coefficients of each modality data are calculated using the analytic hierarchy process based on the coefficient of variation. The base weighting coefficients are adjusted based on the patient's surgical type and postoperative days; The adjusted weighting coefficients are combined linearly with the corresponding time-frequency domain features to generate a normalized rehabilitation state feature vector.

4. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The step of using particle swarm optimization to search for the optimal cluster center in the rehabilitation state space, and combining silhouette coefficient verification to determine the number and location of rehabilitation attractors, specifically includes: Initialize the position vector and velocity vector of each particle in the particle swarm, where the position vector represents the coordinates of the candidate cluster center; Calculate the Euclidean distance between the cluster center and the feature vector of the recovery state based on the current position of the particle, and update the individual optimal position and the global optimal position of the particle based on the principle of minimizing the distance. Calculate the particle's new velocity vector according to the velocity update formula, and calculate the particle's new position vector according to the position update formula; For each clustering number scheme, calculate the corresponding silhouette coefficient value, and select the number of clusters and the cluster center position corresponding to the maximum silhouette coefficient as the number and position of the rehabilitation attractor.

5. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The specific steps of constructing a disease trajectory from the recovery status feature vector according to a time series include: The feature vector sequence is arranged in chronological order of acquisition time; For any missing data points, cubic spline interpolation is used, and residual correction based on local polynomial regression is performed after interpolation. The noise reduction of the rehabilitation status feature vector sequence is performed using a sliding window smoothing algorithm; The processed sequence is connected by interpolation points in the recovery state space to form a continuous disease trajectory, and the local curvature is calculated on the disease trajectory by time segment.

6. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The convergence probability determination step specifically includes: Calculate the Euclidean distance between the current location of the patient's trajectory and each rehabilitation attractor. ; Calculation of traction force based on gravitational field model Satisfy the following formula: ； in For the first The density factor of each recovery attractor is obtained by local point density estimation in cluster analysis; These are normalized weighting coefficients determined based on surgical type and postoperative stage; To avoid singular regularization constants, The time decay constant, The time difference between the current position and the reference time; the composite traction force vector and the rehabilitation state feature vector are obtained by synthesizing the traction force vectors in vector form. And calculate the convergence probability based on the ratio of traction force components. .

7. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The establishment of a baseline model for physiological fluctuations based on statistical distribution specifically includes: Collect historical rehabilitation data of patients who underwent similar surgeries, and establish data sample sets according to surgical type and postoperative time period; Calculate the mean vector and covariance matrix of the rehabilitation status feature vectors for each data sample set; The maximum likelihood estimation method is used to fit the multivariate normal distribution parameters of the rehabilitation status feature vectors for each time period; A baseline model of physiological fluctuations with pre-set confidence levels is constructed based on the fitted multivariate normal distribution parameters, where the confidence interval boundaries are determined by the critical values ​​of the chi-square distribution.

8. The method for postoperative rehabilitation assessment in obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The identification and separation of physiological recovery fluctuations from pathological abnormal deviations specifically includes: Calculate the Mahalanobis distance between the current location of the disease trajectory and the mean vector of the baseline model for the corresponding time period; Compare the Mahalanobis distance with the critical distance threshold of the baseline model confidence interval; When the Mahalanobis distance is less than or equal to the critical distance threshold, it is marked as a physiological recovery fluctuation; When the Mahalanobis distance is greater than the critical distance threshold, the projection components of the deviation vector in each rehabilitation attractor direction are calculated, and the directional characteristics of the abnormal deviation are identified based on the distribution pattern of the projection components.

9. The postoperative rehabilitation assessment method for obstetrics and gynecology based on disease trajectory modeling according to claim 1, characterized in that, The output, which includes a rehabilitation assessment report that determines the nature of fluctuations, specifically includes: Construct a rehabilitation trend vector based on the convergence probability distribution of each rehabilitation attractor; By associating and mapping the fluctuation property markers with the rehabilitation trend vector, a rehabilitation state descriptor containing fluctuation type, deviation degree, and convergence direction is generated. A multi-level decision tree structure is used to classify rehabilitation status descriptors and output a rehabilitation assessment report. The rehabilitation assessment report includes assessment information at three levels: rehabilitation progress level, abnormal risk warning, and personalized rehabilitation recommendations.

10. A postoperative rehabilitation assessment system for obstetrics and gynecology based on disease course trajectory modeling, characterized in that, include: The feature extraction and fusion module is used to collect objective physiological data, subjective self-evaluation data and behavioral pattern data of obstetric and gynecological postoperative patients, extract the time-frequency domain features of each modality data, and construct the rehabilitation status feature vector through a weighted fusion algorithm with adaptive weight allocation; The clustering optimization analysis module is used to search for the optimal cluster center in the rehabilitation state space using the particle swarm optimization algorithm, and to determine the number and location of rehabilitation attractors by combining the profile coefficient verification. The trajectory convergence calculation module is used to construct the disease trajectory from the rehabilitation state feature vector according to the time series, calculate the competitive traction force of each rehabilitation attractor on the disease trajectory, and determine the convergence probability of the trajectory to each attractor based on the synthesis of the traction force vector. The fluctuation identification and assessment module is used to establish a baseline model of physiological fluctuations based on statistical distribution. By comparing the actual fluctuation amplitude of the disease trajectory with the degree of deviation from the baseline model, it identifies and separates physiological rehabilitation fluctuations from pathological abnormal deviations. Combined with the convergence probability of the disease trajectory to each rehabilitation attractor, it outputs a rehabilitation assessment report containing a judgment on the nature of the fluctuations.

Images