Gastric cancer patient management system and method based on postoperative rehabilitation data analysis

Abstract

The present invention discloses a gastric cancer patient management system and method based on postoperative rehabilitation data analysis, which specifically relates to the field of postoperative rehabilitation management, and is used to solve the problem of delayed division of rehabilitation stages. The collected rehabilitation data is subjected to discrete wavelet transform, and noise filtering and missing value repair are implemented to ensure the integrity and continuity of the data, providing a high-quality data basis for subsequent analysis. The multi-order difference and local influence matrix are used to comprehensively evaluate the impact of interpolation points on the details of the original data, accurately capture the trend changes before and after the interpolation points, and effectively improve the reliability and accuracy of data repair. In addition, by correcting the similarity range between the patient's rehabilitation data sequence and the preset standard sequence, the ambiguity of the rehabilitation stage switching timing is eliminated, and the hidden Markov model is combined to analyze the state transition probability, accurately predict the switching timing of the rehabilitation stage, avoid the lag effect of stage division, and enable the rehabilitation plan to dynamically adapt to the actual needs of the patient.

Description

Technical Field

[0001] The present invention relates to the field of postoperative rehabilitation management, and more specifically, to a gastric cancer patient management system and method based on postoperative rehabilitation data analysis. Background Art

[0002] The postoperative rehabilitation stage of gastric cancer can be roughly divided into multiple stages, including the initial postoperative rehabilitation, functional recovery, nutritional supplementation, and exercise recovery. The rehabilitation goals and intervention measures of each stage are completely different. For example, in the initial rehabilitation stage, the focus is on maintaining the stability of vital signs, while in the nutritional supplementation period, more attention is paid to the patient's weight recovery and nutrient absorption. However, due to individual differences in patients (such as age, basal metabolism, postoperative complications, etc.) and the nonlinear characteristics of their rehabilitation progress, the transition of rehabilitation stages does not follow a unified time node. Simply dividing the stages based on a single or static rehabilitation data trend is prone to a lag effect, that is, the system fails to capture small changes in rehabilitation progress in a timely manner, thereby delaying stage transitions. This lag will cause the rehabilitation plan to not meet the actual needs of the patient, which may prolong the patient's rehabilitation period and even increase the risk of complications.

[0003] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a gastric cancer patient management system and method based on postoperative rehabilitation data analysis. By performing discrete wavelet transform on the collected rehabilitation data, implementing noise filtering and missing value repair, the integrity and continuity of the data are ensured, and a high-quality data basis is provided for subsequent analysis. The multi-order difference and local influence matrix are used to comprehensively evaluate the impact of the interpolation points on the details of the original data, accurately capture the trend changes before and after the interpolation points, and effectively improve the reliability and accuracy of data repair. In addition, by correcting the similarity range between the patient's rehabilitation data sequence and the preset standard sequence, the ambiguity of the timing of switching the rehabilitation stage is eliminated, and the hidden Markov model is combined to analyze the state transition probability, accurately predict the switching timing of the rehabilitation stage, avoid the lag effect of stage division, and enable the rehabilitation plan to dynamically adapt to the actual needs of the patient, thereby improving the rehabilitation efficiency and reducing the risk of postoperative complications, so as to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] S1: By collecting patients' physiological data, behavioral data, and subjective data, feature selection algorithms are used to identify key variables.

[0007] S2: Filter noise and repair missing values ​​of the collected data.

[0008] S3: Use multi-order differences and local influence matrices to comprehensively evaluate the restoration impact of the interpolation points, and issue a signal prompt based on the evaluation results.

[0009] S4: Correct the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminate the ambiguity of the rehabilitation stage switching timing, combine the hidden Markov model to analyze the state transition probability, and predict the rehabilitation stage switching timing.

[0010] In a preferred embodiment, step S1 includes the following contents:

[0011] S1.1, collect patients’ physiological, behavioral, and subjective data in real time through wearable devices and mobile applications.

[0012] S1.2, use the random forest algorithm to perform feature selection and importance assessment on the collected data.

[0013] S1.3, perform data integration on identified key variables to unify physiological, behavioral, and subjective data into a standardized data framework.

[0014] S1.4, output the processed key variables and their importance scores as a structured data file.

[0015] In a preferred embodiment, step S2 includes the following contents:

[0016] S2.1, discrete wavelet transform is applied to the collected time series data, using Daubechies wavelet as basis function to capture the local characteristics of the signal.

[0017] Using the wavelet transform formula, the original signal Converted into multiple sub-signals, including approximation coefficients and detail factor ; The decomposition of each layer is achieved by the following formula: ;in, and represent wavelet and scaling function respectively.

[0018] S2.2, noise identification and processing are performed on the extracted detail coefficients.

[0019] Using soft threshold processing, for each detail coefficient, the following processing is performed: ;in, Represents the processed detail coefficient.

[0020] S2.3, reconstruct the processed signal and restore the denoised time series.

[0021] Using inverse wavelet transform, the retained approximation coefficients and processed detail coefficients are synthesized back into the original signal: ;in, represents the denoised signal.

[0022] S2.4, for missing values ​​in the data set, the mean interpolation method is used to process them; specifically, the mean of adjacent time points is used for calculation, and the formula is as follows: ;in is a missing value, Indicates a point in time, and are adjacent known values.

[0023] In a preferred embodiment, step S3 includes the following contents:

[0024] S3.1, for each interpolation point, the local trend change is captured by multi-order difference formula.

[0025] The second-order difference formula is used to calculate the trend differences of multiple points before and after the repair point.

[0026] The inflection points and extreme points are captured by third-order differences, quantifying high-order local changes.

[0027] S3.2, combining the differences of different orders before and after the repair point, calculate the cumulative function of the influencing factor and integrate the differences of each order.

[0028] S3.3, for each interpolation point, consider its overall impact on the surrounding adjacent data points; construct a local impact matrix .

[0029] S3.3, evaluate the repair impact of the entire data set and calculate the global repair impact score , summarizing the local influence matrix of each interpolation point: ;in, is the determinant of the local influence matrix.

[0030] Represents the first time point of interpolation and repair data, that is, the time point when missing data first appears in the time series and is interpolated.

[0031] Represents the last time point of interpolation and repair data, that is, the last time missing data appeared in the time series and was interpolated.

[0032] S3.4, standardize the global repair impact score to obtain the final repair impact factor , the specific formula is: .

[0033] in, is the original estimated data before interpolation, Represents the difference between the imputed value and the original estimated value and is used to normalize the impact factor.

[0034] S3.5, comparing the repair impact factor with the first repair impact threshold and the second repair impact threshold respectively.

[0035] If the repair impact factor is less than the first repair impact threshold, then there is a certain deviation between the repaired data and the original data. It is necessary to send a correction signal and adjust the similarity, and then present a repair signal prompt.

[0036] In a preferred embodiment, step S4 includes the following contents:

[0037] S4.1, if a repair signal is obtained, a preliminary calculation is performed on the similarity between the current rehabilitation data sequence and the preset standard sequence, and the fuzzy interval is identified, that is, the stage switching situation that cannot be accurately judged in the standard similarity calculation; the similarity threshold interval is defined .

[0038] Use historical data and patient-specific rehabilitation pathways to dynamically set upper and lower thresholds and , the specific setting formula is: ;in is the reference similarity, and is the adjustment factor.

[0039] S4.2, when the similarity enters the fuzzy interval, the similarity range interval is corrected; the correction process is based on the repair impact factor and correction factor , the similarity interval is adjusted in the following way:

[0040] The fuzzy interval is dynamically corrected using the calculation results of the above-mentioned repair influence factors; the specific correction formula is: ; ;in, and Indicates the upper and lower thresholds of the corrected similarity range.

[0041] S4.3, judge the modified similarity range interval and establish a switching judgment function through fuzzy logic; similarity value Compare with the upper and lower threshold intervals of the corrected similar range interval to define the fuzzy membership function Used to judge the switching status of the rehabilitation stage; the membership function formula is: ; The membership function is used to evaluate whether the patient's current recovery status meets the switching conditions; if the membership If the difference between the positive and negative values ​​and 1 is less than the corresponding threshold, it means that the recovery status is clear and the stage switching should be triggered; otherwise, no switching is required.

[0042] S4.4, based on the output of the fuzzy membership function, the hidden Markov model is used to perform state transition analysis; the specific steps are as follows:

[0043] The result of fuzzy judgment is used as the observation sequence , input hidden Markov model.

[0044] Apply the Viterbi algorithm to infer the hidden state sequence , and determine the switching timing of the rehabilitation stage based on the state transition probability matrix A;

[0045] The recursive formula of the Viterbi algorithm is: ;

[0046] in Indicates at time Down state The optimal probability of is the state transition probability, is the observed value In Status The probability of the following.

[0047] Indicates the current time The hidden state index of Indicates the previous moment The hidden state index of .

[0048] Based on the optimal state sequence and state transition probability, recommendations are provided for rehabilitation phase switching.

[0049] The gastric cancer patient management system based on postoperative rehabilitation data analysis includes: a data collection module, a cleaning and repair module, a repair evaluation module and a similarity prediction module.

[0050] The data acquisition module collects the patient's physiological data, behavioral data, and subjective data, uses a feature selection algorithm to identify key variables, and sends the collected data sets of key variables to the cleaning and repair module.

[0051] The cleaning and repair module performs noise filtering and missing value repair on the collected data, and sends the repaired data set to the repair evaluation module.

[0052] The repair assessment module uses multi-order differences and local influence matrices to comprehensively evaluate the repair impact of the interpolation points, issues signal prompts based on the evaluation results, and sends the repair impact results and signal prompts to the similarity prediction module.

[0053] The similarity prediction module corrects the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminates the ambiguity of the rehabilitation stage switching timing, combines the hidden Markov model to analyze the state transition probability, and predicts the rehabilitation stage switching timing.

[0054] Technical effects and advantages of the gastric cancer patient management system and method based on postoperative rehabilitation data analysis of the present invention:

[0055] By performing discrete wavelet transform on the collected rehabilitation data, implementing noise filtering and missing value repair, the integrity and continuity of the data are ensured, providing a high-quality data basis for subsequent analysis. The multi-order difference and local influence matrix are used to comprehensively evaluate the impact of the interpolation points on the details of the original data, accurately capturing the trend changes before and after the interpolation points, and effectively improving the reliability and accuracy of data repair. In addition, by correcting the similarity range between the patient's rehabilitation data sequence and the preset standard sequence, the ambiguity of the timing of switching rehabilitation stages is eliminated. The hidden Markov model is combined to analyze the state transition probability, accurately predict the timing of switching rehabilitation stages, avoid the lag effect of stage division, and enable the rehabilitation plan to dynamically adapt to the actual needs of patients, thereby improving rehabilitation efficiency and reducing the risk of postoperative complications. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of the gastric cancer patient management method based on postoperative rehabilitation data analysis of the present invention.

[0057] Figure 2 It is a structural schematic diagram of the gastric cancer patient management system based on postoperative rehabilitation data analysis of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0059] Figure 1 The present invention provides a gastric cancer patient management method based on postoperative rehabilitation data analysis, including:

[0060] S1: By collecting patients' physiological data, behavioral data, and subjective data, feature selection algorithms are used to identify key variables.

[0061] S2: Filter noise and repair missing values ​​of the collected data.

[0062] S3: Use multi-order differences and local influence matrices to comprehensively evaluate the restoration impact of the interpolation points, and issue a signal prompt based on the evaluation results.

[0063] S4: Correct the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminate the ambiguity of the rehabilitation stage switching timing, combine the hidden Markov model to analyze the state transition probability, and predict the rehabilitation stage switching timing.

[0064] In the rehabilitation management process, accurate data collection and effective processing are crucial to the formulation of personalized rehabilitation plans. In order to comprehensively monitor the patient's rehabilitation progress, it is necessary to rely on data input from multiple dimensions, including physiology, behavior, and subjective feedback from patients. These multidimensional data provide the basis for the subsequent division and prediction of rehabilitation stages. However, due to the large amount of data and the large number of variables, how to screen out key variables and effectively integrate them becomes a core issue. Based on this background, step S1 obtains comprehensive patient data through smart devices, combines feature selection with the random forest algorithm, identifies variables that have an important impact on rehabilitation, and provides a solid foundation for subsequent data analysis and rehabilitation plans.

[0065] Step S1 includes the following contents:

[0066] S1.1, through wearable devices and mobile applications, real-time collection of patients' physiological, behavioral and subjective data, including heart rate, blood pressure, exercise volume, dietary intake and pain scores. In the process of data collection, high-frequency sampling technology is used to ensure the timeliness and accuracy of the data. After the collection is completed, data cleaning is performed to remove incomplete or abnormal values, and outliers are identified and eliminated using the mean and standard deviation method to ensure that subsequent analysis is based on high-quality data.

[0067] S1.2, use the random forest algorithm to perform feature selection and importance assessment on the collected data. Random forest constructs multiple decision trees and evaluates the importance of each feature in the classification process to obtain the importance score of each feature. Features with importance scores higher than the set threshold are selected to form a key variable set. This process ensures that the selected features can significantly affect the patient's recovery process and provide an effective basis for subsequent analysis.

[0068] Before using the random forest algorithm, the collected physiological, behavioral, and subjective data are first organized into a structured dataset. Make sure that each column in the dataset corresponds to a feature and each row corresponds to a sample. The following steps are required to process the processed dataset:

[0069] Missing value processing: K nearest neighbor (KNN) interpolation method is used to handle missing values ​​to preserve the integrity of the data.

[0070] Data encoding: One-Hot Encoding is used to convert categorical variables into a numerical form suitable for random forest processing.

[0071] The process of building a random forest model includes the following steps:

[0072] Parameter settings:

[0073] Number of trees: Set to 100, which means building 100 decision trees to enhance the robustness of the model.

[0074] Tree depth: set to 10 to prevent overfitting and ensure that each tree has a moderate depth.

[0075] Minimum number of sample splits: set to 2, indicating that the minimum number of split samples for each tree is 2, allowing for detailed splitting of each node.

[0076] Minimum number of leaf samples: set to 1, indicating that each leaf node of the tree contains at least 1 sample to ensure the availability of the leaf nodes.

[0077] Model training: Use the training set (70% of the data) to train the random forest model by calling the random forest library function and inputting the feature set and target variable for training. This process automatically samples the features and builds a decision tree, and finally integrates the output of all trees to form the final prediction.

[0078] After the random forest model is trained, the feature importance evaluation function in the model is used to determine the influence of each feature. The process includes:

[0079] Importance calculation:

[0080] A tree-based feature importance calculation method is used to evaluate the importance of features using the impact of each tree on sample purity when splitting a node (such as the Gini index or information gain). The importance score of each feature is its total contribution to the split in all trees.

[0081] Result output: The importance scores of all features are summarized and sorted to identify the features that have the greatest impact on patient recovery. A threshold (such as importance score greater than 0.05) is set to screen out key variables, providing a basis for subsequent data analysis.

[0082] S1.3, integrate the data of the identified key variables and unify the physiological, behavioral and subjective data into a standardized data framework. The Z-score standardization method is used to convert the value of each feature into a form with a mean of 0 and a standard deviation of 1 to eliminate the influence of different dimensions on the analysis results. This process ensures that data from different sources are compared and analyzed on the same scale.

[0083] S1.4, output the processed key variables and their importance scores as structured data files for subsequent analysis. Record the processing results and key variable information of each step to form a complete data processing document to ensure data traceability and result transparency. This document provides important support for subsequent analysis and decision-making.

[0084] Step S2 includes the following contents:

[0085] S2.1, apply Discrete Wavelet Transform (DWT) to the collected time series data, using Daubechies wavelet as the basis function to effectively capture the local characteristics of the signal. The specific operations are as follows:

[0086] Decomposition layer number setting: Set the decomposition layer number to 5 to ensure that the low-frequency information and high-frequency noise components in the signal are fully extracted.

[0087] Data decomposition process: Use the wavelet transform formula to transform the original signal Converted into multiple sub-signals, including approximation coefficients (low-frequency part) and detail factor (high frequency part) The decomposition of each layer is achieved by the following formula: ;

[0088] in, and represent wavelet and scaling function respectively.

[0089] S2.2, noise identification and processing of the extracted detail coefficients, the steps are as follows:

[0090] Noise level calculation: Calculate the standard deviation (SD) of the detail coefficients, which is used to evaluate the noise intensity in the signal. Set the threshold The value is 2 times the standard deviation, ensuring that most high frequency noise is removed.

[0091] Threshold processing: Soft threshold processing (SoftThresholding) is used. For each detail coefficient, the following processing is performed: ;

[0092] Represents the processed detail coefficient. The purpose of threshold processing is to reduce or eliminate small amplitude noise, thereby retaining more important signal parts.

[0093] This process effectively removes high-frequency noise below the threshold while retaining stronger signal components and reducing signal distortion.

[0094] S2.3, reconstruct the processed signal to restore the denoised time series. The specific steps include:

[0095] Inverse wavelet transform: Use inverse wavelet transform (IDWT) to synthesize the retained approximation coefficients and processed detail coefficients back into the original signal: ;in, represents the denoised signal;

[0096] Through this step, the denoised signal is reconstructed, the noise is effectively filtered out, and the signal quality is improved.

[0097] S2.4, for the missing values ​​in the data set, the mean interpolation method is used to process them. The specific process is as follows:

[0098] Missing value identification: Identify the location of missing values ​​in the data set by setting a threshold and create an index list of missing values.

[0099] Mean interpolation calculation: For each missing value, the mean of adjacent time points is used for imputation, and the formula is as follows: ;

[0100] in is a missing value, Indicates a point in time, and This method ensures data continuity while reducing information loss caused by interpolation.

[0101] Step S3 includes the following contents:

[0102] S3.1, for each interpolation point, use multi-order difference formulas to capture more complex local trend changes. The specific steps are as follows:

[0103] The second-order difference formula is used to calculate the trend difference of multiple points before and after the repair point. The formula is: ;

[0104] The second-order difference reveals the curvature or inflection point characteristics of the data, capturing the difference in the trend before and after the repair point. If the trend is gentle, the difference value is small; if the trend changes dramatically, the difference value is large.

[0105] Further extended to the third-order difference, the formula is: ;

[0106] The third-order difference captures more complex inflection points and extreme points, quantifying higher-order local changes.

[0107] S3.2, combined with the differences in different orders before and after the repair point, calculate the cumulative function of the impact factor , combining all order differences. The specific formula is: ;

[0108] in, and To adaptively adjust the coefficients, the two coefficients are dynamically adjusted according to the data characteristics to balance the contribution of the second-order and third-order differences to the local trend changes.

[0109] S3.3, for each interpolation point, not only its individual local difference is calculated, but also its overall impact on the surrounding adjacent data points needs to be considered. Construct the local influence matrix , the specific formula is: ;

[0110] The local impact matrix captures the impact of the interpolation point on multiple time points before and after, and accumulates these local impacts in matrix form to form a comprehensive impact assessment of the interpolation point.

[0111] S3.3, evaluate the repair impact of the entire data set and calculate the global repair impact score , summarizing the local influence matrix of each interpolation point: ;

[0112] in, It is the determinant value of the local influence matrix, reflecting the overall degree of local trend change and its influence at this point.

[0113] Represents the first time point of interpolation and repair data, that is, the time point when missing data first appears in the time series and is interpolated.

[0114] Represents the last time point of interpolation and repair data, that is, the last time missing data appeared in the time series and was interpolated.

[0115] S3.4, in order to make the repair impact factor more interpretable and universal, the global repair impact score is standardized to obtain the final repair impact factor , the specific formula is: ;

[0116] in, is the original estimated data before interpolation, It represents the difference between the imputed value and the original estimated value and is used to normalize the impact factor to make it more comparable.

[0117] S3.5, comparing the restoration impact factor with the first restoration impact threshold and the second restoration impact threshold respectively;

[0118] If the repair impact factor is less than the first repair impact threshold, it indicates that the data repair has little impact on the details of the original data, and the repaired data maintains a high degree of continuity and consistency with the original data. Therefore, there is no need to correct the similarity between the current patient rehabilitation data sequence and the preset standard sequence, and the rehabilitation stage can continue to be judged according to the original similarity calculation;

[0119] If the repair impact factor is less than the first repair impact threshold, it means that the data repair has a moderate impact on the details of the original data. At this time, there is a certain deviation between the repaired data and the original data. It is necessary to send a correction signal and adjust the similarity to ensure that the similarity evaluation can reflect the impact of the repair on the data. This correction is to ensure the accuracy of the judgment of the switching timing of the rehabilitation stage, avoid errors caused by the repair impact, and show a repair signal prompt.

[0120] If the repair impact factor is greater than the second repair impact threshold, it means that the data repair has too much impact on the details of the original data, and the difference between the repaired data and the original data has exceeded the acceptable range, resulting in the repaired data no longer being able to reflect the true recovery status. The repaired data will be judged as invalid data and need to be discarded.

[0121] Step S4 includes the following contents:

[0122] S4.1, if the repair signal prompt is obtained, the similarity between the current rehabilitation data sequence and the preset standard sequence is preliminarily calculated to identify the fuzzy interval, that is, the stage switching situation that cannot be accurately judged in the standard similarity calculation. In order to deal with this ambiguity, a similarity threshold interval is defined ,This interval represents the area where the rehabilitation data may be at different stages but cannot be clearly judged.

[0123] Use historical data and patient-specific rehabilitation pathways to dynamically set upper and lower thresholds and , avoiding the limitations of static thresholds. The specific setting formula is: ;

[0124] in is the reference similarity, and It is an adjustment coefficient, which is adaptively adjusted according to individual differences of patients.

[0125] The similarity between the current rehabilitation data sequence and the preset standard sequence refers to the degree of similarity between the actual rehabilitation data of the current patient and an ideal or preset rehabilitation standard data. This preset standard sequence is usually established based on historical data or an ideal rehabilitation path to represent the normal rehabilitation process. Analyzing and calculating the similarity between the two can help evaluate whether the patient's rehabilitation progress is in line with expectations and whether it is on the ideal rehabilitation track. By comparing the similarity, it can be determined whether the patient's rehabilitation plan needs to be adjusted or whether the time to switch to a certain rehabilitation stage has been reached. Similarity analysis can quantify the patient's rehabilitation progress, provide data support, avoid subjective judgment, and make the rehabilitation plan more accurate and personalized.

[0126] S4.2, when the similarity enters the fuzzy interval, the similarity range interval is corrected. The correction process is based on the repair impact factor and correction factor , the similarity interval is adjusted in the following way:

[0127] The fuzzy interval is dynamically corrected using the calculation results of the above-mentioned repair influence factors. The specific correction formula is:

[0128] ;

[0129] ;

[0130] in, and Indicates the upper and lower thresholds of the corrected similarity range.

[0131] Through correction, the upper and lower thresholds of the fuzzy interval are adjusted accordingly according to the impact of the repair data to eliminate the uncertainty caused by the repair data.

[0132] S4.3, judge the corrected similarity range interval and establish a switching judgment function through fuzzy logic. Compare with the upper and lower threshold intervals of the corrected similar range interval to define the fuzzy membership function Used to determine the switching status of the rehabilitation stage. The membership function formula is: ;

[0133] The membership function is used to evaluate whether the patient's current recovery status meets the switching conditions. Close to 1, it means that the recovery status is clear and the stage switch should be triggered; Close to 0, indicating no switching is required.

[0134] S4.4, based on the output of the fuzzy membership function, the hidden Markov model (HMM) is used to perform state transition analysis. The state transition matrix A and observation probability matrix B of the hidden Markov model have been trained with historical data before. This step infers the most likely state sequence based on the currently observed similarity membership. The specific steps are as follows:

[0135] The result of fuzzy judgment is used as the observation sequence , input hidden Markov model.

[0136] Apply the Viterbi Algorithm to infer the hidden state sequence , and determine the switching timing of the rehabilitation stage based on the state transition probability matrix A.

[0137] The recursive formula of the Viterbi algorithm is: ;

[0138] in Indicates at time Down state The optimal probability of is the state transition probability, is the observed value In Status The probability of the following.

[0139] Indicates the current time The hidden state index of . That is, in the recursive step, at the current moment The possible hidden states .

[0140] Indicates the previous moment The hidden state index of . That is, at time The possible hidden states in the hidden Markov model are .

[0141] Based on the optimal state sequence and state transition probability, the hidden Markov model analysis results can provide suggestions for rehabilitation stage switching.

[0142] The recovery stage switching recommendation can be explained by specific numerical analysis. For example, suppose the hidden Markov model analysis results show that the current patient is in the recovery stage , and the state transition probability matrix A , indicating that from the recovery stage Transfer to recovery phase The probability is 85%. At the same time, the observed patient recovery data sequence During the rehabilitation phase The corresponding observation probability , indicating the observed value and rehabilitation stage The match is 90%.

[0143] Based on the above analysis, the state transition analysis of HMM can suggest entering the rehabilitation stage Because the state transition probability of 85% and the observation matching degree of 90% both show that the patient's recovery progress is close to the recovery stage Therefore, it is recommended to adjust the rehabilitation program to adapt to the patient's new rehabilitation stage needs, such as switching from the functional recovery stage to the consolidation stage. This specific numerical analysis can provide reliable data support for the switching of rehabilitation stages, not only based on the probability of state transition, but also to ensure the accuracy of decision-making through the matching degree of observed data. Example 2

[0144] Figure 2 The present invention provides a gastric cancer patient management system based on postoperative rehabilitation data analysis, comprising: a data acquisition module, a cleaning and repair module, a repair evaluation module and a similarity prediction module.

[0145] The data acquisition module collects the patient's physiological data, behavioral data, and subjective data, uses a feature selection algorithm to identify key variables, and sends the collected data sets of key variables to the cleaning and repair module.

[0146] The cleaning and repair module performs noise filtering and missing value repair on the collected data, and sends the repaired data set to the repair evaluation module.

[0147] The repair assessment module uses multi-order differences and local influence matrices to comprehensively evaluate the repair impact of the interpolation points, issues signal prompts based on the evaluation results, and sends the repair impact results and signal prompts to the similarity prediction module.

[0148] The similarity prediction module corrects the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminates the ambiguity of the rehabilitation stage switching timing, combines the hidden Markov model to analyze the state transition probability, and predicts the rehabilitation stage switching timing.

[0149] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0150] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0151] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0152] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A gastric cancer patient management method based on postoperative rehabilitation data analysis, characterized in that: Includes steps: S1: By collecting patients’ physiological data, behavioral data, and subjective data, feature selection algorithms are used to identify key variables; S2: Filter the collected data for noise and repair missing values; S3: Use multi-order differences and local influence matrices to comprehensively evaluate the restoration impact of the interpolation points, and issue a signal prompt based on the evaluation results; S4: Correct the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminate the ambiguity of the rehabilitation stage switching timing, analyze the state transition probability in combination with the hidden Markov model, and predict the rehabilitation stage switching timing; Step S1 includes the following contents: S1.1, collect patients’ physiological, behavioral, and subjective data in real time through wearable devices and mobile applications; S1.2, use the random forest algorithm to perform feature selection and importance assessment on the collected data; S1.3, perform data integration on identified key variables to unify physiological, behavioral, and subjective data into a standardized data framework; S1.4, output the processed key variables and their importance scores as structured data files; Step S2 includes the following contents: S2.1, discrete wavelet transform is applied to the collected time series data, using Daubechies wavelet as basis function to capture the local characteristics of the signal; Using the wavelet transform formula, the original signal Converted into multiple sub-signals, including approximation coefficients and detail factor ; The decomposition of each layer is achieved by the following formula: ;in, and denote wavelet and scaling function respectively; S2.2, noise identification and processing of the extracted detail coefficients; Using soft threshold processing, for each detail coefficient, the following processing is performed: ;in, represents the processed detail coefficient; S2.3, reconstruct the processed signal and restore the denoised time series; Using inverse wavelet transform, the retained approximation coefficients and processed detail coefficients are synthesized back into the original signal: ;in, represents the denoised signal; S2.4, for missing values ​​in the data set, the mean interpolation method is used to process them; specifically, the mean of adjacent time points is used for calculation, and the formula is as follows: ;in is a missing value, Indicates a point in time, and are adjacent known values; Step S3 includes the following contents: S3.1, for each interpolation point, the local trend change is captured by multi-order difference formula; The trend differences of multiple points before and after the repair point are calculated using the second-order difference formula; Capture inflection points and extreme points through third-order differences and quantify high-order local changes; S3.2, combining the differences of different orders before and after the repair point, calculate the cumulative function of the influencing factor and integrate the differences of each order; S3.3, for each interpolation point, consider its overall impact on the surrounding adjacent data points; construct a local impact matrix ; S3.3, evaluate the repair impact of the entire data set and calculate the global repair impact score , summarizing the local influence matrix of each interpolation point: ;in, is the determinant value of the local influence matrix; Represents the first time point of interpolation and repair data, that is, the time point when missing data first appears in the time series and is interpolated; Represents the last time point of interpolation and repair data, that is, the time point when missing data appeared in the time series for the last time and was interpolated; S3.4, standardize the global repair impact score to obtain the final repair impact factor , the specific formula is: ; in, is the original estimated data before interpolation, It represents the difference between the interpolated value and the original estimated value and is used to normalize the impact factor; S3.5, comparing the restoration impact factor with the first restoration impact threshold and the second restoration impact threshold respectively; If the repair impact factor is less than the first repair impact threshold, then there is a certain deviation between the repaired data and the original data, and a correction signal needs to be issued and the similarity needs to be adjusted, and a repair signal prompt is displayed; Step S4 includes the following contents: S4.1, if a repair signal is obtained, a preliminary calculation is performed on the similarity between the current rehabilitation data sequence and the preset standard sequence, and the fuzzy interval is identified, that is, the stage switching situation that cannot be accurately judged in the standard similarity calculation; the similarity threshold interval is defined ; Use historical data and patient-specific rehabilitation pathways to dynamically set upper and lower thresholds and , the specific setting formula is: ;in is the reference similarity, and is the adjustment factor; S4.2, when the similarity enters the fuzzy interval, the similarity range interval is corrected; the correction process is based on the repair impact factor and correction factor , the similarity interval is adjusted in the following way: The fuzzy interval is dynamically corrected using the calculation results of the above-mentioned repair influence factors; the specific correction formula is: ; ;in, and Indicates the upper and lower thresholds of the similar range after correction; S4.3, judge the modified similarity range interval and establish a switching judgment function through fuzzy logic; similarity value Compare with the upper and lower threshold intervals of the corrected similar range interval to define the fuzzy membership function Used to judge the switching status of the rehabilitation stage; the membership function formula is: ; The membership function is used to evaluate whether the patient's current recovery status meets the switching conditions; if the membership If the difference between the value and 1 is less than the corresponding threshold, it means that the recovery status is clear and the stage switching should be triggered; otherwise, it means that no switching is required; S4.4, based on the output of the fuzzy membership function, the hidden Markov model is used to perform state transition analysis; the specific steps are as follows: The result of fuzzy judgment is used as the observation sequence , input hidden Markov model; Apply the Viterbi algorithm to infer the hidden state sequence , and determine the switching timing of the rehabilitation stage based on the state transition probability matrix A; The recursive formula of the Viterbi algorithm is: ; in Indicates at time Down state The optimal probability of is the state transition probability, is the observed value In Status The probability of the following; Indicates the current time The hidden state index of Indicates the previous moment The hidden state index of Based on the optimal state sequence and state transition probability, recommendations are provided for rehabilitation phase switching.

2. A gastric cancer patient management system based on postoperative rehabilitation data analysis, used to implement the gastric cancer patient management method based on postoperative rehabilitation data analysis as claimed in claim 1, characterized in that: include: Data collection module, cleaning and repair module, repair evaluation module and similarity prediction module; The data acquisition module collects the patient's physiological data, behavioral data, and subjective data, uses a feature selection algorithm to identify key variables, and sends the collected data sets of key variables to the cleaning and repair module; The cleaning and repair module performs noise filtering and missing value repair on the collected data, and sends the repaired data set to the repair evaluation module; The repair assessment module uses multi-order differences and local impact matrices to comprehensively assess the repair impact of the interpolation point, issues a signal prompt based on the assessment result, and sends the repair impact result and signal prompt to the similarity prediction module; The similarity prediction module corrects the similarity range between the current patient rehabilitation data sequence and the preset standard sequence, eliminates the ambiguity of the rehabilitation stage switching timing, combines the hidden Markov model to analyze the state transition probability, and predicts the rehabilitation stage switching timing.

Images