Dynamic subspace-based symptom adaptive test item selection method and system

Abstract

The application relates to a dynamic subspace-based symptom adaptive evaluation item selection method and system, which comprises the following steps: constructing a somatic symptom evaluation item library; collecting initial features of a cancer patient, outputting a multidimensional somatic symptom vector based on the initial features; generating a candidate evaluation item set; obtaining an updated feature matrix and a multidimensional somatic symptom vector based on an evaluation result, and generating an initial candidate evaluation item set based on the feature matrix and the multidimensional somatic symptom vector of the current round; sequentially performing dynamic subspace selection, evaluation subspace selection and dynamic orthogonalization selection on the initial candidate evaluation item set to generate a candidate evaluation item set, until iteration convergence is achieved, an optimal multidimensional somatic symptom vector is obtained, and an evaluation report is output. The application can reduce the number of evaluation items, reduce the evaluation burden of cancer patients, improve the efficiency of multidimensional somatic symptom evaluation, reduce the redundancy of evaluation items, improve the data processing efficiency, and realize individualized evaluation paths.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a method and system for selecting symptom adaptive assessment items based on dynamic subspace. Background Technology

[0002] Cancer patients often experience a variety of physical symptoms during disease progression, surgery, radiotherapy, chemotherapy, targeted therapy, immunotherapy, and palliative care, including fatigue, pain, sleep disturbances, nausea and vomiting, decreased appetite, constipation, diarrhea, cough, shortness of breath, numbness, and skin reactions. These physical symptoms are not isolated but rather multidimensional, coexisting across different cancer types, treatment methods, treatment stages, and time points. Clinically, assessing the physical symptoms of cancer patients typically employs various methods, such as paper-based questionnaires, electronic questionnaires, fixed-item scales, and interviews with healthcare professionals. Existing methods for assessing physical symptoms can be categorized into two main types:

[0003] The first type is the fixed-item assessment method. This method requires pre-setting fixed items, and all patients answer the same assessment items. The total score or scores for each dimension of physical symptoms are calculated based on the patient's assessment results. This method is simple to operate and easy to promote in clinical practice, but its number of items is fixed and cannot be dynamically adjusted according to the patient's cancer type, treatment method, symptom presentation, and previous assessment results.

[0004] The second category is electronic item-based assessment. This method transforms paper questionnaires into electronic questionnaires accessible via mobile phones, tablets, web pages, or hospital information systems. Patients can answer the assessment items independently through these devices, and the system automatically stores the results. While this method improves data collection convenience, most items still rely on fixed items or simple skipping logic, and it cannot dynamically select the most valuable assessment items from a large-scale assessment item database based on the patient's real-time responses.

[0005] Therefore, when facing the multi-dimensional physical characteristic assessment of cancer patients, existing technologies still have problems such as redundant assessment items, heavy assessment burden, low accuracy of physical characteristic estimation, low assessment efficiency, and difficulty in dynamically updating assessment items. Summary of the Invention

[0006] The purpose of this invention is to address the problems in existing technologies when assessing the multidimensional physical characteristics of cancer patients, such as redundant assessment items, heavy assessment burden, low accuracy of physical characteristic estimation, low efficiency of the assessment process, and difficulty in dynamically updating assessment items. This invention provides a symptom-adaptive assessment item selection method and system based on dynamic subspace.

[0007] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:

[0008] The symptom adaptive assessment item selection method based on dynamic subspace includes the following steps:

[0009] Step 1: Construct a database of physical symptom assessment items;

[0010] Step 2: Collect initial features P from cancer patients, and output a multi-dimensional somatic symptom vector Y based on the initial features P. (1) Based on the initial feature P and the multi-dimensional somatic symptom vector Y (1) Generate a set S of candidate evaluation items (1) ;

[0011] Step 3: Cancer patients answer the candidate assessment item set S (1) The first round of evaluation results R were subsequently obtained. (1) Based on the results of the first round of evaluation R (1) Obtain updated feature matrices and multi-dimensional somatic symptom vectors;

[0012] Step 4, starting from the second round of evaluation, based on the results of the previous round of evaluation R (k-1) Update the feature matrix X for this round. (k) and multidimensional somatic symptom vector Y (k) And based on the feature matrix X of this round (k) and multidimensional somatic symptom vector Y (k) Generate an initial set of candidate evaluation items C (k) For the initial candidate evaluation item set C (k) The dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection are performed sequentially to generate a candidate evaluation item set S. (k) ;

[0013] Step 5: Cancer patients answer the candidate assessment item set S (k) The evaluation result R was subsequently obtained. (k) Return to step 4 and continue iteratively until convergence is achieved, yielding the optimal multidimensional somatic symptom vector Y. (K) Output the evaluation report.

[0014] Furthermore, the somatic symptom assessment item library contains several assessment items, which are divided into two categories: symptom assessment items and influencing factor items.

[0015] Each symptom assessment item includes the item number, item name, item text, the physical symptom dimension to which it belongs, applicable cancer type, applicable treatment method, applicable treatment stage, scoring range, item discrimination level, item difficulty level, item information content level, item cost parameter, item redundancy coefficient, item clinical weight, whether it is a core assessment item, whether repeated assessment is allowed, and update time.

[0016] Each influencing factor entry includes entry number, entry name, variable type, value range, coding rules, applicable cancer type, applicable treatment method, applicable treatment stage, handling method for missing values, entry clinical weight, and update time.

[0017] Furthermore, in step 2, the initial characteristics include age, gender, cancer type, time since diagnosis, disease stage, treatment method, treatment stage, current treatment cycle days, self-reported physical symptoms, past physical symptoms, and past assessment results.

[0018] Furthermore, in step 2, based on the initial feature P and the multi-dimensional somatic symptom vector Y... (1) Generate a set S of candidate evaluation items (1) The steps include:

[0019] Based on the initial characteristics P and multidimensional somatic symptom vector Y of cancer patients (1) An initial set of candidate assessment items C is selected from the somatic symptom assessment item database. (1) The filtering rules include:

[0020] (1) The applicable cancer types for the assessment items match the cancer types in the initial characteristics;

[0021] (2) The applicable treatment stage of the assessment item matches the treatment stage in the initial feature P;

[0022] (3) The physical symptoms dimension to which the assessment item belongs is related to the self-reported physical symptoms in the initial feature P;

[0023] (4) The information content level of the evaluation item is higher than the preset threshold;

[0024] (5) The cost parameter of the evaluation item is lower than the preset threshold;

[0025] (6) Each self-reported somatic symptom in the initial feature P retains at least one assessment item belonging to the same somatic symptom dimension;

[0026] Based on the initial candidate evaluation item set C (1) For each assessment item, the item information level, item cost parameter, item redundancy coefficient, item clinical weight, and whether it is a core assessment item are comprehensively scored and ranked. The top L1 assessment items are selected to generate a candidate assessment item set S. (1) L1 is less than or equal to C (1) The number of evaluation items.

[0027] Furthermore, the updated feature matrix X in step 4 (k) For: X (k) =[P,R (1) ,...,R (k-1)], where P is the initial feature, R (1) For the first round of assessment results, R (k-1) This is the result of the (k-1)th round of assessment.

[0028] Furthermore, step 4, the step of performing dynamic subspace selection, includes:

[0029] Calculate the initial candidate evaluation item set C (k) The dynamic utility value of each evaluation item:

[0030]

[0031] Where j is the initial set of candidate evaluation items C (k) Index of the j-th assessment item; U j Let I be the dynamic utility value of the j-th evaluation item; j A represents the information gain of the j-th evaluation item; j For the j-th assessment item, the multidimensional somatic symptom vector Y (k) Coverage; M j D represents the degree of matching between the j-th assessment item and the current round feature matrix of cancer patients; j The redundancy between the j-th assessment item and the assessment items already answered by the cancer patient; B j Let be the item cost parameter for the j-th evaluation item; α1, α2, α3, α4, and α5 are the preset weights;

[0032] Based on the dynamic utility value, the initial candidate evaluation item set C is... (k) Sort the evaluation items in the dataset, and select the top L2 evaluation items to form a set C. (k,1) L2 is less than or equal to C (k) The number of evaluation items in set C; (k,1) Each time, select a number of evaluation items to form a subset, and after N times, form a... ,in Represents set C (k,1) The nth subset, n=1,2,...,N;

[0033] Calculate each of the N subsets Overall utility value:

[0034]

[0035] Where i and j are both subsets The index of the assessment items, i≠j; F n For the nth subset The overall utility value; Representing a subset The sum of the dynamic utility values ​​of all evaluation items; Sim(i,j) represents a subset. The similarity (or redundancy) between any two evaluation items. Representing a subset For the multidimensional somatic symptom vector Y (k) Coverage; These are preset weights;

[0036] Choose the overall utility value F n largest subset As a set C (k,2) .

[0037] Furthermore, step 4, the step of performing the evaluation subspace selection, includes:

[0038] For set C (k,2) Apply the following constraints:

[0039] Set C (k,2) If the number of evaluation items in set C is less than or equal to a preset threshold, then delete set C. (k,2) The number of evaluation items in the evaluation list is less than or equal to a preset threshold.

[0040] Set C (k,2) If the number of core assessment items in the set is greater than or equal to a preset threshold, otherwise move to set C. (k,2) Add core assessment items;

[0041] Set C (k,2) If the mean of the cost parameter of all evaluation items is less than or equal to a preset threshold, then set C is deleted. (k,2) The evaluation item with the highest cost parameter is selected such that the mean of the item cost parameter is less than or equal to a preset threshold.

[0042] The evaluation items that have been answered in the first k-1 rounds are from set C. (k,2) Delete;

[0043] Given any set C of assessment items that have been answered in the first k-1 rounds. (k,2) If the similarity of any evaluation item in set C is greater than or equal to a set threshold, then select from set C. (k,2) The evaluation item has been deleted from the website.

[0044] By analyzing set C (k,2) After applying the above constraints, we obtain set C. (k,3) .

[0045] Furthermore, step 4, the step of performing dynamic orthogonalization selection, includes:

[0046] Set C (k,3)The assessment items are represented as projection vectors, and the similarity between each projection vector is calculated. When the similarity between two projection vectors exceeds a preset threshold, the projection vectors are adjusted by an orthogonalization algorithm, and the corresponding assessment items are updated by the adjusted projection vectors, so that the assessment items of different physical symptoms remain distinguishable.

[0047] By analyzing set C (k,3) After making the above adjustments, we obtain the candidate evaluation item set S. (k) .

[0048] Furthermore, in step 5, before determining the convergence of the iteration, a preset evaluation accuracy threshold, a minimum number of entries, and a maximum number of entries are set.

[0049] Obtain the multi-dimensional somatic symptom vector Y updated in each round (k) and the set of candidate evaluation items S (k) Then, determine whether any of the following convergence conditions are met:

[0050] (1) Multidimensional somatic symptom vector Y (k) The estimation error, standard error, or posterior variance of all physical symptoms are below the assessment accuracy threshold.

[0051] (2) The initial candidate evaluation item set C for two or more consecutive rounds (k) The average information gain of all evaluation items is lower than the preset threshold.

[0052] (3) The number of assessment items answered by cancer patients reaches the maximum number of items;

[0053] (4) The cancer patient has reached the preset time limit for answering the assessment items;

[0054] (5) Candidate evaluation item set S (k) The assessment items correspond to the multidimensional somatic symptom vector Y. (k) The coverage of all physical symptoms has reached the preset threshold;

[0055] (6) Cancer patients voluntarily withdraw from the assessment;

[0056] When any of the above convergence conditions are met, the iteration stops, for example, the iteration stops at the Kth round.

[0057] Furthermore, a symptom adaptive assessment item selection system based on dynamic subspace, used to implement the symptom adaptive assessment item selection method based on dynamic subspace, includes:

[0058] A physical symptom assessment item library, used to store a number of assessment items;

[0059] The terminal is used to provide cancer patients with a way to input initial characteristics and answer assessment items;

[0060] An assessment item generator is used to perform steps 2 to 5 of the symptom adaptive assessment item selection method based on dynamic subspace and obtain the optimal multidimensional somatic symptom vector.

[0061] The medical server is used to output assessment reports based on optimal multi-dimensional somatic symptom vectors for medical staff to view.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] (1) Reduce the number of assessment items and reduce the assessment burden on cancer patients: This invention selects a small number of key assessment items that cancer patients need to answer most from the complete physical symptom assessment item library through dynamic subspace selection, assessment subspace selection and dynamic orthogonalization selection, instead of requiring cancer patients to fill in all assessment items. This can reduce irrelevant assessment items and duplicate assessment items, reduce the assessment burden on cancer patients and improve the assessment completion efficiency.

[0064] (2) Improve the efficiency of multidimensional somatic symptom assessment: This invention does not assess a single somatic symptom, but estimates the state values ​​of multiple somatic symptoms in a multidimensional somatic symptom vector at the same time. The system can select assessment items that are valuable for multiple somatic symptoms based on the current multidimensional somatic symptom vector, thereby improving the efficiency of multidimensional somatic symptom assessment.

[0065] (3) Reduce the redundancy of evaluation items and improve data processing efficiency: This invention incorporates the redundancy of evaluation items into the calculation of dynamic utility value, and limits the number of evaluation items selected in each round by using a sparse constraint-based evaluation subspace selection mechanism. This can reduce the repeated evaluation of similar evaluation items and improve the data processing efficiency of high-dimensional evaluation items.

[0066] (4) Realize individualized assessment path: The present invention dynamically selects the assessment items to be pushed afterward based on the cancer type, treatment method, treatment stage, self-reported physical symptoms, previous assessment results and real-time assessment item answer results of the cancer patient, so that different cancer patients enter different assessment paths and improve the individual adaptability of the assessment.

[0067] (5) Improve the stability of multidimensional somatic symptom classification: This invention reduces the information overlap between assessment items corresponding to different somatic symptoms through a dynamic orthogonal selection mechanism, so that the answer results of assessment items of different somatic symptoms can be well distinguished, which helps to improve the stability of multidimensional somatic symptom classification and risk grading.

[0068] (6) Supports dynamic updates and continuous optimization: The present invention can send newly collected evaluation results back to the evaluation item generator, enabling the system to have continuous learning and optimization capabilities.

[0069] (7) This system is easy to embed into clinical intelligent assessment terminals: This invention can be deployed between the patient end, medical staff end, application server, model server and database server, supporting cancer patients to conduct self-assessment, generate real-time assessment results, medical staff to view assessment reports and update the background model, and is easy to apply in hospitals, follow-up platforms and remote symptom management terminals. Attached Figure Description

[0070] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a flowchart of the method of the present invention.

[0072] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0074] It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance, or suggesting any such actual relationship or order between these entities or operations. Additionally, the terms "connected," "linked," etc., can refer to a direct connection between elements, components, modules, etc., or an indirect connection via other elements, components, modules, etc.

[0075] Example 1:

[0076] This invention is achieved through the following technical solutions, such as... Figure 1As shown, a symptom adaptive assessment item selection method based on dynamic subspace includes the following steps:

[0077] Step 1: Construct a database of physical symptom assessment items.

[0078] The foundation for building a database of physical symptom assessment items includes existing assessment scales, cancer patient reports, clinical symptom records, medical consultation records, and clinical assessment results.

[0079] The constructed somatic symptom assessment item library contains several assessment items, mainly divided into two categories: symptom assessment items and influencing factor items. Among them, the symptom assessment items are used to directly assess the current somatic symptoms of cancer patients; the influencing factor items are used to describe patient characteristics, disease characteristics, and treatment characteristics that may affect the estimation of somatic symptoms and the selection of assessment items.

[0080] Specifically, each symptom assessment item includes, but is not limited to, the following fields: item number, item name, item text, corresponding physical symptom dimension, applicable cancer type, applicable treatment method, applicable treatment stage, scoring range, item discrimination level, item difficulty level, item information content level, item cost parameter, item redundancy coefficient, item clinical weight, whether it is a core assessment item, whether repeated assessment is allowed, and update time.

[0081] Each influencing factor entry includes, but is not limited to, the following fields: entry number, entry name, variable type, value range, coding rules, applicable cancer type, applicable treatment method, applicable treatment stage, handling method for missing values, entry clinical weight, and update time.

[0082] Step 2: Collect initial features P from cancer patients, and output a multi-dimensional somatic symptom vector Y based on the initial features P. (1) Based on the initial feature P and the multi-dimensional somatic symptom vector Y (1) Generate a set S of candidate evaluation items (1) .

[0083] Cancer patients fill out initial characteristics P via mobile phones, tablets, web pages, or hospital information systems. The initial characteristics P include, but are not limited to: age, gender, cancer type, time since diagnosis, disease stage, treatment method, treatment stage, current treatment cycle duration, self-reported physical symptoms, past physical symptoms, and past assessment results.

[0084] The initial features P are categorized into three types: continuous variables, categorical variables, and ordinal variables. The initial features P are then numerically processed. Different methods are used to numerically process the continuous, categorical, and ordinal variables so that they can be used as input to the computer model.

[0085] For continuous variables, these may include age, time since diagnosis, and number of days in the current treatment cycle. Based on previous assessment results in the initial feature P, previous physical symptom scores and the interval between the last assessment and the current assessment can be further obtained. Continuous variables can be numerically processed using methods such as maximum / minimum normalization, Z-score standardization, or piecewise normalization.

[0086] For categorical variables, these may include gender, cancer type, treatment method, and treatment stage. Based on the cancer type in the initial feature P, we can further obtain information on whether there are underlying diseases. Based on the treatment method in the initial feature, we can further obtain information on whether the patient received radiotherapy, chemotherapy, immunotherapy, or targeted therapy. Categorical variables can be numerically processed using heat-based coding, label coding, or multi-label coding.

[0087] For ordinal variables, disease staging can be included; based on self-reported physical symptoms in the initial feature P, a physical symptom score can be further obtained; based on past physical symptoms in the initial feature P, a physical symptom severity level can be further obtained. Ordinal variables can be numerically processed by mapping them to numerical values ​​in an ordinal order.

[0088] The system outputs a multi-dimensional vector of somatic symptoms based on the initial features P of cancer patients. ,in This represents the somatic symptoms of the d-th dimension, where d = 1, 2, ..., D, and D is the value of Y. (1) The dimensions of somatic symptoms, somatic symptoms The state estimate is The multidimensional somatic symptom vector Y (1) It is formed based on the initial characteristics P of cancer patients, and therefore cannot accurately reflect the actual physical symptoms of cancer patients. However, the multi-dimensional physical symptom vector Y output in this step is used when selecting assessment items in the first round. (1) Make a selection.

[0089] Based on the initial characteristics P and multidimensional somatic symptom vector Y of cancer patients (1) An initial set of candidate assessment items C is selected from the somatic symptom assessment item database. (1) The filtering rules include:

[0090] (1) The applicable cancer types for the assessment items match the cancer types in the initial characteristics;

[0091] (2) The applicable treatment stage of the assessment item matches the treatment stage in the initial feature P;

[0092] (3) The physical symptoms dimension to which the assessment item belongs is related to the self-reported physical symptoms in the initial feature P;

[0093] (4) The information content level of the evaluation item is higher than the preset threshold;

[0094] (5) The cost parameter of the evaluation item is lower than the preset threshold;

[0095] (6) Each self-reported somatic symptom in the initial feature P retains at least one assessment item belonging to the same somatic symptom dimension.

[0096] From the initial set of candidate evaluation items C (1) Select a subset of assessment items to generate a candidate assessment item set S. (1) The candidate evaluation item set S (1) As the first round of assessment items for cancer patients. Specifically, based on the initial candidate assessment item set C (1) For each assessment item, the item information level, item cost parameter, item redundancy coefficient, item clinical weight, and whether it is a core assessment item are comprehensively scored and ranked. The top L1 assessment items are selected to generate a candidate assessment item set S. (1) L1 is less than or equal to C (1) The number of evaluation items in the test, and the set of candidate evaluation items S. (1) Pushed to cancer patients.

[0097] Step 3: Cancer patients answer the candidate assessment item set S (1) The first round of evaluation results R were subsequently obtained. (1) Based on the results of the first round of evaluation R (1) Obtain updated feature matrices and multidimensional somatic symptom vectors.

[0098] Cancer patients answer the set of candidate assessment items S (1) The first round of evaluation results R were subsequently obtained. (1) Then, the results of the first round of evaluations will be R. (1) Combined with the initial features, the feature matrix X for the next round is obtained. (k+1) =[P,R (1) ,R (2) ,...,R (k) [], P represents the initial feature, and the superscript (k) represents the current round. In this round, k=1. When in the kth round, the updated feature matrix X (k+1) The evaluation only includes the initial feature P and the results R from the first k rounds. (1) ~R (k) .

[0099] Using the feature matrix X of the next round (k+1) Update the multidimensional somatic symptom vector Y (k+1) The updated multidimensional somatic symptom vector is obtained. It should be noted that the dimension D in the multidimensional somatic symptom vector is not necessarily equal in different rounds.

[0100] Step 4, starting from the second round of evaluation, based on the results of the previous round of evaluation R (k-1) Update the feature matrix X for this round. (k) and multidimensional somatic symptom vector Y (k) And based on the feature matrix X of this round (k) and multidimensional somatic symptom vector Y (k) Generate an initial set of candidate evaluation items C (k) For the initial candidate evaluation item set C (k) The dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection are performed sequentially to generate a candidate evaluation item set S. (k) .

[0101] It should be noted that in this scheme, the first round of candidate evaluation item set S is generated. (1) At that time, based only on the initial feature P and the multi-dimensional somatic symptom vector Y (1) Simply generate the set S of candidate assessment items for cancer patients to answer. (1) Afterwards, the first round of evaluation results R were obtained. (1) Then we move on to the second round, in which we first assess the results R from the previous round. (1) The feature matrix X is updated to obtain (2) =[P,R (1) ] and multidimensional somatic symptom vector Y (2) Then use X (2) and Y (2) Generate an initial set of candidate evaluation items C (2) Then C (2) The operations of dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection are performed sequentially to obtain the candidate evaluation item set S for the second round. (2) Finally, S (2) The results of a second round of assessments were sent to cancer patients after they answered the questions. (2) This is equivalent to each round starting from the second round, the difference from the first round being that C is generated. (k) Next, we need to address C. (k) S can only be obtained by sequentially performing the operations of dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection. (k) And in the first round, C was obtained. (1) Then, a simple comprehensive score ranking is all that is needed to obtain S. (1) .

[0102] The following sections will explain the computational processes for dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection:

[0103] (1) Dynamic subspace selection.

[0104] First, calculate the initial candidate evaluation item set C. (k) The dynamic utility value of each evaluation item:

[0105]

[0106] Where j is the initial set of candidate evaluation items C (k) Index of the j-th assessment item; U j Let I be the dynamic utility value of the j-th evaluation item; j A represents the information gain of the j-th evaluation item; j For the j-th assessment item, the multidimensional somatic symptom vector Y (k) Coverage; M j D represents the degree of matching between the j-th assessment item and the current round feature matrix of cancer patients; j The redundancy between the j-th assessment item and the assessment items already answered by the cancer patient; B j Let α be the item cost parameter for the j-th evaluation item; α1, α2, α3, α4, and α5 are the preset weights.

[0107] Based on the dynamic utility value, the initial candidate evaluation item set C is... (k) Sort the evaluation items in the dataset, and select the top L2 evaluation items to form a set C. (k,1) L2 is less than or equal to C (k) The number of evaluation items in set C. (k,1) Each time, select a number of evaluation items to form a subset, and after N times, form a... ,in Represents set C (k,1) The nth subset, n=1,2,...,N.

[0108] Calculate each of the N subsets Overall utility value:

[0109]

[0110] Where i and j are both subsets The index of the assessment items, i≠j; F n For the nth subset The overall utility value; Representing a subset The sum of the dynamic utility values ​​of all evaluation items; Sim(i,j) represents a subset. The similarity (or redundancy) between any two evaluation items. Representing a subset For the multidimensional somatic symptom vector Y (k) Coverage; These are preset weights.

[0111] Choose the overall utility value F n largest subset As a set C (k,2) Perform evaluation subspace selection.

[0112] (2) Evaluation subspace selection.

[0113] For set C (k,2) Apply the following constraints:

[0114] Set C (k,2) If the number of evaluation items in set C is less than or equal to a preset threshold, then delete set C. (k,2) The number of evaluation items in the evaluation list is less than or equal to a preset threshold.

[0115] Set C (k,2) If the number of core assessment items in the set is greater than or equal to a preset threshold, otherwise move to set C. (k,2) Add core assessment items;

[0116] Set C (k,2) If the mean of the cost parameter of all evaluation items is less than or equal to a preset threshold, then set C is deleted. (k,2) The evaluation item with the highest cost parameter is selected such that the mean of the item cost parameter is less than or equal to a preset threshold.

[0117] The evaluation items that have been answered in the first k-1 rounds are from set C. (k,2) Delete;

[0118] Given any set C of assessment items that have been answered in the first k-1 rounds. (k,2) If the similarity of any evaluation item in set C is greater than or equal to a set threshold, then select from set C. (k,2) The evaluation item has been deleted.

[0119] By analyzing set C (k,2) After applying the above constraints, we obtain set C. (k,3) Perform dynamic orthogonalization selection.

[0120] (3) Dynamic orthogonalization selection.

[0121] To reduce the number of assessment items corresponding to different physical symptoms in set C (k,3) Overlapping information in set C (k,3) Perform dynamic orthogonalization selection. Specifically, select set C. (k,3)The assessment items are represented as projection vectors, and the similarity between each projection vector is calculated. When the similarity between two projection vectors exceeds a preset threshold, the projection vectors are adjusted using an orthogonalization algorithm. The corresponding assessment items are then updated using the adjusted projection vectors to maintain differentiation between assessment items for different bodily symptoms. For example, assessment items for nausea and vomiting and decreased appetite may have some similarity, but to better distinguish between them, their corresponding assessment items are adjusted using an orthogonalization algorithm.

[0122] Set C (k,3) After dynamic orthogonalization selection, the candidate evaluation item set S for this round is generated. (k) and S (k) Pushed to cancer patients.

[0123] Step 5: Cancer patients answer the candidate assessment item set S (k) The evaluation result R was subsequently obtained. (k) Return to step 4 and continue iteratively until convergence is achieved, yielding the optimal multidimensional somatic symptom vector Y. (K) Output the evaluation report.

[0124] Cancer patients answer the set of candidate assessment items S (k) The evaluation result R was subsequently obtained. (k) Return to step 4 and proceed based on the evaluation result R. (k) The feature matrix X is updated to obtain (k+1) =[P,R (1) ,R (2) ,...,R (k) ] and multidimensional somatic symptom vector Y (k+1) Then use X (k+1) and Y (k+1) Generate an initial set of candidate evaluation items C (k+1) Then C (k+1) Performing dynamic subspace selection yields C (k+1,2) , for C (k+1,2) C is obtained by performing the evaluation subspace selection. (k+1,3) , for C (k+1,3) Performing dynamic orthogonalization selection yields S (k+1) Finally, S (k+1) The results of the assessment (R) were sent to cancer patients after they answered the questions. (k+1) Each round repeats in turn, which will not be described in detail here.

[0125] To determine whether iterative convergence is satisfied, it is necessary to preset an evaluation accuracy threshold, a minimum number of entries, and a maximum number of entries. The evaluation accuracy threshold is used to evaluate the multi-dimensional somatic symptom vector Y. (k) State estimates for each somatic symptom Whether the assessment has stabilized, for example, whether the estimation error, standard error, or posterior variance of physical symptoms is below the assessment accuracy threshold. The minimum number of items ensures that at least basic assessment information for cancer patients is obtained. The maximum number of items limits the assessment burden on cancer patients and prevents the system from repeatedly adding similar assessment items.

[0126] Obtain the multi-dimensional somatic symptom vector Y updated in each round (k) and the set of candidate evaluation items S (k) Then, determine whether any of the following convergence conditions are met:

[0127] (1) Multidimensional somatic symptom vector Y (k) The estimation error, standard error, or posterior variance of all physical symptoms are below the assessment accuracy threshold.

[0128] (2) The initial candidate evaluation item set C for two or more consecutive rounds (k) The average information gain of all evaluation items is lower than the preset threshold.

[0129] (3) The number of assessment items answered by cancer patients reaches the maximum number of items;

[0130] (4) The cancer patient has reached the preset time limit for answering the assessment items;

[0131] (5) Candidate evaluation item set S (k) The assessment items correspond to the multidimensional somatic symptom vector Y. (k) The coverage of all physical symptoms has reached the preset threshold;

[0132] (6) Cancer patients voluntarily withdraw from the assessment.

[0133] When any of the above convergence conditions is met, the iteration stops. For example, if the iteration stops at the Kth round, then the cancer patient answers the set S of candidate assessment items for the Kth round. (K) The evaluation results R (K) This represents the optimal assessment result, and the multi-dimensional somatic symptom vector Y in the Kth round is... (K) This is also the optimal choice.

[0134] Furthermore, based on the evaluation results of round K, R (K) Risk grading is generated, which is used to convert the final assessment results of cancer patients into low-risk, medium-risk, or high-risk categories.

[0135] Finally, an assessment report is output, which includes, but is not limited to, a multi-dimensional somatic symptom vector Y. (K) The scores for each physical symptom, the risk level of each physical symptom, the ranking of physical symptoms, the number of completed assessment items, and the total assessment time.

[0136] It should be noted that, assuming the cancer patient completes K rounds of responses, but the set of candidate assessment items for each round is S... (k) The evaluation items in the set are all unique, and the set of candidate evaluation items S in each round is also unique. (k) The number of assessment items may vary. It's easy to understand that different cancer patients may need to complete different rounds of responses, and the assessment items they answer may also differ, because different cancer patients have different initial characteristics. For example, if a cancer patient inputs initial characteristics, the system will execute steps 2 and 3 of this scheme, outputting a set S of candidate assessment items. (1) There are 3 evaluation items. Then the system starts executing step 4 of this solution, outputting a set S of candidate evaluation items. (2) There are 5 evaluation items. The system then repeats step 4 of this solution, outputting a set S of candidate evaluation items. (3) There are four assessment items, which are iterated in a loop until convergence. Therefore, from the perspective of cancer patients, they will not know that they are answering the assessment item in the kth round; it is all done by the algorithm running in the background. It can be seen that this solution can update the subsequent assessment items pushed in real time based on the assessment results of cancer patients' current answers, making the assessment item push more in line with the actual physical symptoms of cancer patients without increasing the assessment burden on cancer patients, and the final assessment report is more in line with the actual situation.

[0137] Based on the above methods, this scheme also proposes a symptom adaptive assessment item selection system based on dynamic subspace, such as... Figure 2 As shown, it includes:

[0138] A physical symptom assessment item library, used to store a number of assessment items.

[0139] The terminal is used to provide cancer patients with a way to input initial characteristics and answer assessment items.

[0140] An assessment item generator is used to perform steps 2 to 5 of the method and obtain the optimal multidimensional somatic symptom vector.

[0141] The medical server is used to output assessment reports based on optimal multi-dimensional somatic symptom vectors for medical staff to view.

[0142] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A symptom adaptive assessment item selection method based on dynamic subspace, characterized in that, Includes the following steps: Step 1: Construct a database of physical symptom assessment items; Step 2: Collect initial features P from cancer patients, and output a multi-dimensional somatic symptom vector Y based on the initial features P. (1) Based on the initial feature P and the multi-dimensional somatic symptom vector Y (1) Generate a set S of candidate evaluation items (1) ; Step 3: Cancer patients answer the candidate assessment item set S (1) The first round of evaluation results R were subsequently obtained. (1) Based on the results of the first round of evaluation R (1) Obtain updated feature matrices and multi-dimensional somatic symptom vectors; Step 4, starting from the second round of evaluation, based on the results of the previous round of evaluation R (k-1) Update the feature matrix X for this round. (k) and multidimensional somatic symptom vector Y (k) And based on the feature matrix X of this round (k) and multidimensional somatic symptom vector Y (k) Generate an initial set of candidate evaluation items C (k) For the initial candidate evaluation item set C (k) The dynamic subspace selection, evaluation subspace selection, and dynamic orthogonalization selection are performed sequentially to generate a candidate evaluation item set S. (k) ; Step 5: Cancer patients answer the candidate assessment item set S (k) The evaluation result R was subsequently obtained. (k) Return to step 4 and continue iteratively until convergence is achieved, yielding the optimal multidimensional somatic symptom vector Y. (K) Output the evaluation report.

2. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 1, characterized in that, The somatic symptom assessment item library contains several assessment items, which are divided into two categories: symptom assessment items and influencing factor items. Each symptom assessment item includes the item number, item name, item text, the physical symptom dimension to which it belongs, applicable cancer type, applicable treatment method, applicable treatment stage, scoring range, item discrimination level, item difficulty level, item information content level, item cost parameter, item redundancy coefficient, item clinical weight, whether it is a core assessment item, whether repeated assessment is allowed, and update time. Each influencing factor entry includes entry number, entry name, variable type, value range, coding rules, applicable cancer type, applicable treatment method, applicable treatment stage, handling method for missing values, entry clinical weight, and update time.

3. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 1, characterized in that, In step 2, the initial characteristics include age, gender, cancer type, time since diagnosis, disease stage, treatment method, treatment stage, current treatment cycle days, self-reported physical symptoms, past physical symptoms, and past assessment results.

4. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 1, characterized in that, In step 2, the initial feature P and the multidimensional somatic symptom vector Y are used as the basis. (1) Generate a set S of candidate evaluation items (1) The steps include: Based on the initial characteristics P and multidimensional somatic symptom vector Y of cancer patients (1) An initial set of candidate assessment items C is selected from the somatic symptom assessment item database. (1) The filtering rules include: (1) The applicable cancer types for the assessment items match the cancer types in the initial characteristics; (2) The applicable treatment stage of the assessment item matches the treatment stage in the initial feature P; (3) The physical symptoms dimension to which the assessment item belongs is related to the self-reported physical symptoms in the initial feature P; (4) The information content level of the evaluation item is higher than the preset threshold; (5) The cost parameter of the evaluation item is lower than the preset threshold; (6) Each self-reported somatic symptom in the initial feature P retains at least one assessment item belonging to the same somatic symptom dimension; Based on the initial candidate evaluation item set C (1) For each assessment item, its information content level, cost parameter, redundancy coefficient, clinical weight, and whether it is a core assessment item are comprehensively scored and ranked. The top L1 assessment items are selected to generate a candidate assessment item set S. (1) L1 is less than or equal to C (1) The number of evaluation items.

5. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 1, characterized in that, The updated feature matrix X in step 4 (k) For: X (k) =[P,R (1) ,...,R (k-1) ], where P is the initial feature, R (1) For the first round of assessment results, R (k-1) This is the result of the (k-1)th round of assessment.

6. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 1, characterized in that, Step 4, the step of performing dynamic subspace selection, includes: Calculate the initial candidate evaluation item set C (k) The dynamic utility value of each evaluation item: Where j is the initial set of candidate evaluation items C (k) Index of the j-th assessment item; U j Let I be the dynamic utility value of the j-th evaluation item; j A represents the information gain of the j-th evaluation item; j For the j-th assessment item, the multidimensional somatic symptom vector Y (k) Coverage; M j D represents the degree of matching between the j-th assessment item and the current round feature matrix of cancer patients; j The redundancy between the j-th assessment item and the assessment items already answered by the cancer patient; B j Let be the item cost parameter for the j-th evaluation item; α1, α2, α3, α4, and α5 are the preset weights; Based on the dynamic utility value, the initial candidate evaluation item set C is... (k) Sort the evaluation items in the dataset, and select the top L2 evaluation items to form a set C. (k,1) L2 is less than or equal to C (k) The number of evaluation items in set C; (k,1) Each time, select a number of evaluation items to form a subset, and after N times, form a... ,in Represents set C (k,1) The nth subset, n=1,2,...,N; Calculate each of the N subsets Overall utility value: Where i and j are both subsets The index of the assessment items, i≠j; F n For the nth subset The overall utility value; Representing a subset The sum of the dynamic utility values ​​of all evaluation items; Sim(i,j) represents a subset. The similarity (or redundancy) between any two evaluation items. Representing a subset For the multidimensional somatic symptom vector Y (k) Coverage; These are preset weights; Choose the overall utility value F n largest subset As a set C (k,2) .

7. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 6, characterized in that, Step 4, the step of selecting the evaluation subspace, includes: For set C (k,2) Apply the following constraints: Set C (k,2) If the number of evaluation items in set C is less than or equal to a preset threshold, then delete set C. (k,2) The number of evaluation items in the evaluation list is less than or equal to a preset threshold. Set C (k,2) If the number of core assessment items in the set is greater than or equal to a preset threshold, otherwise move to set C. (k,2) Add core assessment items; Set C (k,2) If the mean of the cost parameter of all evaluation items is less than or equal to a preset threshold, then set C is deleted. (k,2) The evaluation item with the highest cost parameter is selected such that the mean of the item cost parameter is less than or equal to a preset threshold. The evaluation items that have been answered in the first k-1 rounds are from set C. (k,2) Delete; Given any set C of assessment items that have been answered in the first k-1 rounds. (k,2) If the similarity of any evaluation item in set C is greater than or equal to a set threshold, then select from set C. (k,2) The evaluation item has been deleted from the website. By analyzing set C (k,2) After applying the above constraints, we obtain set C. (k,3) .

8. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 7, characterized in that, Step 4, the step of performing dynamic orthogonalization selection, includes: Set C (k,3) The assessment items are represented as projection vectors, and the similarity between each projection vector is calculated. When the similarity between two projection vectors exceeds a preset threshold, the projection vectors are adjusted by an orthogonalization algorithm, and the corresponding assessment items are updated by the adjusted projection vectors, so that the assessment items of different physical symptoms remain distinguishable. By analyzing set C (k,3) After making the above adjustments, we obtain the candidate evaluation item set S. (k) .

9. The symptom adaptive assessment item selection method based on dynamic subspace according to claim 8, characterized in that, In step 5, before determining the convergence of the iteration, a preset evaluation accuracy threshold, minimum number of entries, and maximum number of entries are set. Obtain the multi-dimensional somatic symptom vector Y updated in each round (k) and the set of candidate evaluation items S (k) Then, determine whether any of the following convergence conditions are met: (1) Multidimensional somatic symptom vector Y (k) The estimation error, standard error, or posterior variance of all physical symptoms are below the assessment accuracy threshold. (2) The initial candidate evaluation item set C for two or more consecutive rounds (k) The average information gain of all evaluation items is lower than the preset threshold. (3) The number of assessment items answered by cancer patients reaches the maximum number of items; (4) The cancer patient has reached the preset time limit for answering the assessment items; (5) Candidate evaluation item set S (k) The assessment items correspond to the multidimensional somatic symptom vector Y. (k) The coverage of all physical symptoms has reached the preset threshold; (6) Cancer patients voluntarily withdraw from the assessment; When any of the above convergence conditions are met, the iteration stops, for example, the iteration stops at the Kth round.

10. A symptom adaptive assessment item selection system based on dynamic subspace, used to implement the symptom adaptive assessment item selection method based on dynamic subspace as described in any one of claims 1-9, characterized in that, include: A physical symptom assessment item library, used to store a number of assessment items; The terminal is used to provide cancer patients with a way to input initial characteristics and answer assessment items; An assessment item generator is used to perform steps 2 to 5 of the symptom adaptive assessment item selection method based on dynamic subspace and obtain the optimal multidimensional somatic symptom vector. The medical server is used to output assessment reports based on optimal multi-dimensional somatic symptom vectors for medical staff to view.

Images