Treatment effect evaluation method and system suitable for HBO oxygen cabin

Abstract

The invention discloses a treatment effect evaluation method and system suitable for an HBO oxygen cabin, and relates to the technical field of medical data analysis, and the method comprises the steps: carrying out the recognition linkage of an HIS system through a wrist strap before a patient enters the cabin, collecting the medical advice, diagnosis and multi-modal physiological and subjective scoring data of the patient, and generating an initial treatment vector after standardization; the method comprises the following steps: constructing a curative effect triple based on a medical ontology, generating a semantic map embedding vector by adopting a TransE model, embedding and splicing each treatment data and a map, constructing a time map structure, modeling a patient curative effect evolution trajectory by utilizing a Time-GCN model, outputting a final state representation vector, and calculating a final curative effect score. According to the method, the whole process of curative effect evaluation is intelligentized, structured and traceable, curative effect evolution characteristics can be effectively captured, and a scientific basis is provided for personalized clinical decision making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data analysis, and in particular to a method and system for evaluating the treatment effect of an HBO oxygen cabin. BACKGROUND

[0002] With the wide application of hyperbaric oxygen (HBO) therapy in various chronic diseases, traumatic diseases and nervous system rehabilitation, its importance in clinical practice is increasing. HBO therapy improves microcirculation and metabolic environment by increasing tissue oxygen partial pressure, and has achieved remarkable results in the fields of diabetic foot, cerebral ischemia and post-traumatic cognitive impairment. At present, hyperbaric oxygen cabin systems are widely deployed in hospitals at all levels, and are assisted in managing patient treatment courses through electronic medical records (EMR) and HIS systems. However, the traditional evaluation method mainly relies on the subjective experience of doctors or the comparison of a small number of single physiological indicators before and after treatment, and lacks structured modeling of the treatment process and deep analysis of the evolution of treatment effects, making it difficult to support clinical decision-making needs such as parameter optimization and precise treatment recommendations.

[0003] Although existing research (CN113823402A Disease Treatment Effect Prediction Method Based on Big Data and Machine Learning, CN118675694A Recovery Effect Analysis System for Patients in Hyperbaric Oxygen Cabin Based on Big Data) has introduced some objective indicators such as blood oxygen saturation and VAS score for quantitative evaluation, and some high-end devices have also tried to improve the evaluation dimension by collecting NIRS microcirculation signals, but overall, there are still the following shortcomings: (1) Lack of time modeling mechanism for the whole cycle of treatment effect evolution, resulting in the inability to capture the state evolution in multiple treatment processes; (2) Lack of semantic expression and fusion ability of medical knowledge, making it difficult to effectively integrate multiple sources of data such as medical orders, diagnoses, physiology, and subjective perception; (3) The evaluation results cannot be traced back to the treatment parameter combination, and lack of parameter deviation analysis and intelligent recommendation support. In particular in actual clinical practice, patients have high individual differences in response to the same treatment parameters, and a single score or chart is not enough to support efficient treatment effect tracking and personalized treatment plan matching. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a method and system for evaluating the treatment effect of an HBO oxygen cabin, which solves the problems of "single indicator, lack of process modeling and semantic integration, and inability to trace and optimize parameters" in existing HBO oxygen cabin treatment effect evaluation technology.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The application provides a method for evaluating the treatment effect of an HBO oxygen cabin, which comprises the following steps,

[0008] Before the patient enters the cabin, wristband recognition is performed to link the HIS system, patient medical orders, diagnosis and multi-modal physiological and subjective score data are collected, and an initial treatment vector is generated after standardization;

[0009] Based on a medical ontology, a treatment effect triple is constructed, a semantic graph embedding vector is generated by using a TransE model, each treatment data is spliced with the graph embedding, a time graph structure is constructed, a Time-GCN model is used to model the evolution trajectory of the patient's treatment effect, an end state representation vector is output, and a final treatment effect score is calculated;

[0010] The end state representation vector of the patients in the same cabin is subjected to K-medoids clustering, the optimal treatment effect corresponding parameter combination is analyzed, and a recommended template is generated;

[0011] The individual end state representation is matched with the recommended parameters to score the matching degree, the deviation patients are marked, and individual evaluation reports and ward statistical reports are generated.

[0012] As a preferred scheme of the method for evaluating the treatment effect of the HBO oxygen cabin, wherein the wristband recognition is performed to link the HIS system before the patient enters the cabin, the patient medical orders, diagnosis and multi-modal physiological and subjective score data are collected, and the initial treatment vector is generated after standardization, which comprises the following steps:

[0013] When the patient arrives at the hyperbaric oxygen cabin treatment area, nursing staff uses a hand-held PDA equipped with NFC function to scan the patient's wristband, realizes fast binding of the patient's identity through a unique hospitalization number ID, establishes a data channel with the HIS system, and automatically pulls the medical order and diagnosis data related to the patient's treatment on the same day;

[0014] The collected data is composed of feature raw data vectors, and the raw vectors are subjected to Z-score standardization processing to form a standardized initial vector X0.

[0015] As a preferred scheme of the method for evaluating the treatment effect of the HBO oxygen cabin, wherein the treatment effect triple is constructed based on a medical ontology, the semantic graph embedding vector is generated by using a TransE model, and the treatment data of each time is spliced with the graph embedding, which comprises the following steps:

[0016] The standardized entity set E is subjected to rule-driven triple generation, and a patient individual triple set is constructed based on the generated triple;

[0017] Extracting all semantic objects involved in the current patient from the triple set T to obtain the entity set and the relationship set, assigning independent embedding vectors to all entities and relationships and initializing them;

[0018] After initialization, graph modeling is performed using the TransE model, and according to the definition of the TransE model, if the semantics of each triple is correct, the embedding vector is set to satisfy the condition;

[0019] Based on the definition, a training objective function is constructed to model the structural consistency of all triples in the embedding space;

[0020] After the target function is constructed, the random gradient descent algorithm is used to train the embedding vector, and the training process is iterated, and the vector parameters are adjusted by batch update in each round, and the vectors are normalized after each round of training;

[0021] After training, the triple embedding results of the current patient semantic graph are uniformly aggregated to form a graph-level overall representation ;

[0022] The graph embedding vector G0 is spliced with the standardized initial vector X0 to form a comprehensive node feature vector F0.

[0023] As a preferred scheme of the HBO oxygen cabin treatment effect evaluation method, the time graph structure is constructed by using a Time-GCN model to model the patient treatment evolution trajectory, and the output terminal state representation vector is collected in time sequence. The node sequence in V is mapped to a time graph structure A=(V,E,T), wherein is the fusion feature vector of the last treatment node in the entire treatment course of the patient, V is a node set, each node F t in the node set represents a treatment node, T is a timestamp set, and E is an edge set.

[0024] For each F t , an edge is formed by connecting adjacent nodes before and after, and a time decay adjacency weight is assigned to each edge ;

[0025] The initial feature F t of each node is used as the initial input representation of the graph neural network , and a two-layer time-enhanced graph convolution network propagation operation is performed on the graph structure. This process aggregates the state vectors of the neighbor nodes and combines the time decay weight to model the historical treatment evolution and extract the semantic representation of the patient state at each time.

[0026] After all the graph convolution propagation is completed, output the node state sequence at each time point in the entire treatment process , wherein represents the final state vector representation at the end of the current patient treatment.

[0027] As a preferred scheme of the HBO chamber treatment effect evaluation method, wherein: the final treatment effect score is calculated using a linear projection scoring function, and is mapped to a scalar treatment effect score .

[0028] As a preferred scheme of the HBO chamber treatment effect evaluation method, wherein: the K-medoids clustering of the same chamber patient treatment effect final state representation vector is performed, the optimal treatment effect corresponding parameter combination is analyzed and the recommended template is generated, which means that all patient state vectors at the end of the treatment in the current chamber are automatically collected to form a clustering sample set, Z-score standardization processing is performed on all vectors to generate a standardized vector set, the optimal cluster number K is determined by the silhouette coefficient method, and K-medoids algorithm is applied on the standardized vector set for clustering modeling according to the optimal cluster number K, to obtain a plurality of patient clusters with similar treatment effects, and each clustering cluster represents a patient population with similar treatment effect curves;

[0029] Further statistics of the treatment parameter combination used in each cluster are collected, and the frequency and average value of the corresponding final state treatment effect score of each combination are calculated, and the treatment configuration parameter group with the highest frequency and optimal effect is selected for each cluster as the recommended optimal treatment strategy for each cluster.

[0030] As a preferred scheme of the HBO chamber treatment effect evaluation method, wherein: the individual final state representation is matched with the recommended parameters, the deviation patients are marked, and the individual evaluation report and the ward statistical report are generated, which means that the corresponding structured parameter vector is extracted according to the optimal treatment parameter combination in the clustering cluster to which the patient belongs, the actual treatment parameters executed by the patient are synchronously obtained, and are mapped into a vector, the deviation degree between the current treatment setting and the optimal scheme is measured by calculating the Euclidean distance between the two, the patient treatment effect quality score is calculated by a lightweight multilayer perceptron model using the treatment effect final state vector, and the patient treatment effect matching score is output based on the treatment effect quality score and the deviation degree by a weighted formula;

[0031] The threshold values D and U are set, and D>U, if the patient treatment effect matching score is greater than the threshold value D, it means that the patient has good treatment effect performance, and the treatment parameters are highly matched with the recommended scheme, and no adjustment is needed.

[0032] If the patient treatment effect matching score is less than the threshold value D and greater than or equal to the threshold value U, it means that the patient has certain deviation in treatment effect, and whether to adjust needs to be determined in combination with the parameter deviation and the doctor's judgment.

[0033] If the patient efficacy matching score is less than the threshold value U, the patient will be automatically marked as "optimization object", and the medical staff will be prompted to pay attention;

[0034] An individualized assessment report is generated for each patient, and batch statistical tables are automatically summarized according to the ward dimension, covering efficacy grade distribution, parameter deviation heat map, efficacy trend change curve, and pre / post-score ranking. The individual report is returned to the electronic medical record system in the form of PDF, and the ward report is simultaneously exported to Excel and provided for management terminal review.

[0035] In a second aspect, the present application provides an HBO oxygen cabin treatment effect evaluation system, comprising,

[0036] A patient information acquisition module is configured to automatically acquire medical orders, diagnoses and historical data, and multi-modal indicators through wristband recognition and linkage with the HIS system, and to standardize the generation of treatment initial vectors;

[0037] A medical semantic modeling module is configured to construct efficacy triples based on standardized data and medical ontology, generate semantic graph embedding vectors using the TransE model, and splice the initial vectors to form node features for each treatment;

[0038] A treatment modeling module is configured to construct a time graph structure according to the time sequence of all treatment node features, model the patient treatment evolution trajectory using the Time-GCN model, and output a terminal state representation vector;

[0039] An efficacy scoring module is configured to perform efficacy scoring on the terminal state vector using a linear scoring function, and obtain a structured representation reflecting the overall treatment effect;

[0040] A parameter recommendation module is configured to perform K-medoids clustering analysis on the terminal state vectors of patients in the same cabin, extract optimal treatment parameter groups for each type of group, and generate a recommendation template;

[0041] A matching evaluation module is configured to calculate the matching degree of the individual's current treatment scheme and the recommended parameters, automatically generate individualized assessment reports and ward efficacy statistical results, and support doctor decision adjustment.

[0042] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the HBO oxygen cabin treatment effect evaluation method according to the first aspect of the present application.

[0043] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the method for evaluating the therapeutic effect of HBO oxygen cabin according to the first aspect of the present application.

[0044] The present application has the beneficial effects that: the present application integrates key technologies such as wristband recognition and HIS data linkage, multi-modal data standardized modeling, medical semantic graph embedding based on TransE, Time-GCN course modeling and K-medoids efficacy clustering recommendation, realizes the full-process intelligentization, structurization and traceability of efficacy evaluation. By constructing the time graph structure of the individual multiple treatment process of the patient and combining semantic triple modeling, the method can effectively capture the efficacy evolution characteristics; further, the optimal parameter combination is extracted by clustering the group terminal state, and the deviation is identified and the report is generated by matching the current individual state, which provides a scientific basis for personalized clinical decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 The flowchart of the method for evaluating the therapeutic effect of HBO oxygen cabin in embodiment 1.

[0047] Figure 2 The structure diagram of the system for evaluating the therapeutic effect of HBO oxygen cabin in embodiment 1. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0049] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0050] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or selective embodiment that excludes other embodiments.

[0051] Embodiment 1, Reference Figure 1 and Figure 2 As a first embodiment of the present application, the embodiment provides a method for evaluating the therapeutic effect of an HBO oxygen cabin, comprising the following steps:

[0052] S1, before the patient enters the cabin, through wristband recognition linkage HIS system, collecting patient orders, diagnosis and multi-modal physiological and subjective score data, and generating an initial treatment vector after standardization;

[0053] Specifically, before the patient enters the cabin, through wristband recognition linkage HIS system, collecting patient orders, diagnosis and multi-modal physiological and subjective score data, and generating an initial treatment vector after standardization includes:

[0054] When the patient arrives at the hyperbaric oxygen cabin treatment area, the nursing staff uses a handheld PDA equipped with NFC function to scan the patient's wristband, realizes the rapid binding of the patient's identity through the unique hospitalization number ID, and establishes a data channel with the HIS system, automatically pulling the patient's orders and diagnosis data related to the patient's treatment on the same day, including:

[0055] Current oxygen cabin treatment parameters: treatment pressure (unit: ATA), oxygen inhalation time (unit: minutes), treatment course number;

[0056] Diagnosis code: standard structured representation based on ICD-10;

[0057] History of treatment records: past treatment effect scores, whether interrupted, treatment abnormality markers, etc.

[0058] After identification and binding, the integrated multi-modal sensing channel (such as SpO2 sensor, near-infrared spectrometer NIRS, blood test interface) is called to complete the initial physiological index collection of the patient before entering the cabin, including:

[0059] Blood oxygen saturation, C-reactive protein, local tissue perfusion index, VAS (visual analog score, 0-10), MoCA (cognitive function scale, 0-30);

[0060] The collected data is composed of a feature original data vector, and the original vector is subjected to Z-score standardization processing to form a standardized initial vector X0.

[0061] The automatic binding of patient identity is realized through the identification of NFC wristband, and the HIS system is linked to efficiently extract medical order and diagnosis data, which significantly improves the accuracy and efficiency of information collection. Combined with multi-modal sensor data collection of objective physiological indicators and subjective score data, the patient's pre-treatment state is comprehensively reflected. Through Z-score standardization, the dimensions of each feature are unified to ensure the numerical consistency and stability of subsequent modeling. This process effectively constructs an initial treatment vector with complete structure, rich expression and standardized format, laying a high-quality data foundation for subsequent semantic graph modeling and efficacy prediction, and significantly improving the scientificity and intelligent level of HBO treatment effect evaluation.

[0062] S2, based on medical ontology, constructing efficacy triples, using TransE model to generate semantic graph embedding vectors, splicing each treatment data with graph embedding, constructing time graph structure, using Time-GCN model to model patient efficacy evolution trajectory, outputting final state representation vector and calculating final efficacy score;

[0063] Specifically, based on medical ontology, constructing efficacy triples, using TransE model to generate semantic graph embedding vectors, splicing each treatment data with graph embedding, extracting all feature names and feature directions from the standardized initial vector, performing first-order rate calculation on each index field (based on historical mean and current value comparison), calling medical ontology library (SNOMED CT) or local mapping table for entity standardization, establishing index entity set E, and recording the source of each entity medical index;

[0064] Performing rule-driven triple generation on the standardized entity set E, using clinical semantic rules as follows:

[0065] If the index is a pathological index that decreases (such as CRP, blood sugar): construct (entity, representation, functional recovery);

[0066] If the index is a physiological / cognitive index that increases (such as NIRS, MoCA): construct (entity, promote, functional enhancement);

[0067] Example triple generation:

[0068] (CRP decreases, representation, inflammation relief);

[0069] (NIRS increases, promotes, microcirculation improvement);

[0070] (MoCA increases, reflects, cognitive recovery);

[0071] Based on the generated triples, constructing patient individual triple set;

[0072] The entity set and the relation set are obtained by extracting all semantic objects related to the current patient from the triple set T, and the independent embedding vectors are assigned to all entities and relations and are initialized by using uniform distribution sampling and L2 regularization normalization.

[0073] After the initialization, the TransE model is used for graph modeling, and according to the definition of the TransE model, if the semantics of each triple is correct, the embedding vectors thereof should satisfy the following linear approximation:

[0074]

[0075] In the formula, h is the embedding vector of the head entity in the triple, represents a certain physiological or semantic state, r is the embedding vector of the semantic relation in the triple, and t is the embedding vector of the tail entity in the triple.

[0076] That is, the linear sum of the head entity vector and the relation vector should be as close to the tail entity vector as possible.

[0077] Based on the definition, a training target function is constructed to model the structural consistency of all triples in the embedding space:

[0078]

[0079] In the formula, is the total loss value of the embedding model, is a semantic relation path in the semantic graph of the patient, represents “index change – semantic relation – improvement of curative effect”, and T is the set of all triples in the semantic graph of the current patient.

[0080] After the target function is constructed, the random gradient descent algorithm is used to train the embedding vectors, and the training process is iterated. In each round, the vector parameters are adjusted by batch updating, and all vectors are normalized after each round of training. In this embodiment, the entire training process is set to 100 rounds, the learning rate is 0.01, and the batch size is 16, so as to ensure that the semantic embedding converges and has discriminative ability.

[0081] After the training is completed, the triple embedding results of the semantic graph of the current patient are uniformly aggregated to form a whole representation at the graph level :

[0082]

[0083] In the formula, m is the number of triples in the graph of the current patient, , and is the embedding vector of the head entity, the relation, and the tail entity in the i-th triple.

[0084] The vector contains deep relationship information between physiological changes and treatment effects in the semantic structure of the patient, and has strong semantic expression ability.

[0085] The atlas embedding vector G0 is spliced with the standardized initial vector X0 to form a comprehensive node feature vector F0.

[0086] Through the medical ontology and rule triple generation mechanism, different data sources (such as physiological measurement and chief complaint score) are ensured to have unified interpretation at the semantic level. The atlas embedding not only preserves the quantitative information of the original indicators, but also introduces clinical relationship structure, improving the expression ability of complex treatment mechanism. The TransE embedding mechanism supports the introduction of more semantic dimensions and external knowledge bases, which is beneficial to the continuous optimization of the efficacy modeling framework. The spliced vector F0 is the core feature of constructing the time series graph, which has a good foundation for jointly modeling the patient evolution trajectory.

[0087] Further, the time graph structure is constructed, and the Time-GCN model is used to model the patient efficacy evolution trajectory, and the final state representation vector is output. The comprehensive node feature vector F0 is taken as the first time point node, in which the node sequence in V is mapped to the time graph structure A=(V,E,T), where is the fusion feature vector of the last treatment node in the entire treatment course of the patient, V is the node set, each node F t in the node set represents a treatment node, T is the time stamp set, each node records the actual treatment time, E is the edge set, and a fixed time window strategy is used to construct adjacent edges to ensure that only nodes close in time are connected;

[0088] For each F t , connect to the adjacent front and rear nodes to form edges and assign time decay adjacency weights to each edge :

[0089]

[0090] In the formula, is the time decay adjacency weight of node j to node t, t is the current treatment node time step, and j is the time adjacent node of the current node t;

[0091] The initial feature F t of each node is taken as the initial input representation of the graph neural network , and a two-layer time-enhanced graph convolution network (Time-GCN) propagation operation is performed on the graph structure. This process realizes the modeling of historical efficacy evolution by aggregating the state vectors of adjacent nodes and combining the time decay weight, extracts the semantic representation of the patient state at each time, and each layer of propagation is defined by the following formula:

[0092]

[0093] wherein, is the trainable weight matrix of the l-th layer, which is used to map the state of the neighboring nodes into the current layer space, is the set of time neighboring nodes of node t, is the bias vector of the l-th layer, is the updated hidden state of node t in the l+1-th layer, is the hidden state vector of the neighboring node j in the l-th layer, and σ() is a nonlinear activation function, ReLU function is used in the embodiment;

[0094] After all the graph convolution propagation is completed, the node state sequence of each time point in the entire treatment process is output wherein represents the terminal state vector at the end of the current patient treatment. The terminal state vector not only contains the current treatment effect, but also reflects the evolution trajectory and historical treatment accumulation of the entire treatment.

[0095] The graph structured expression and dynamic modeling of the multiple treatment information of the patient are realized, so that the system can not only identify the single treatment effect, but also master the evolution trend across time, has the ability of treatment trajectory comparison, abnormal state recognition and future treatment trend prediction. In addition, by introducing the time decay mechanism and the time window restriction, the far cause interference is effectively inhibited, and the model learning result is more suitable for the real medical causal process, which provides strong modeling support for subsequent personalized treatment evaluation.

[0096] Further, the final treatment score is calculated by using a linear projection score function to map to a scalar treatment score :

[0097]

[0098] wherein, is the i-th dimension value of the terminal state vector of the patient, is the mean value of the i-th dimension of all patients in the training set, is the standard deviation of the i-th dimension in the training set, is the score weight corresponding to the i-th dimension, reflecting its importance to the treatment effect, which is obtained by using linear regression method with labeled training data, b is the bias term, which is used to adjust the global average treatment score, and d is the dimension number of the treatment terminal state vector .

[0099] The complex efficacy results are visually presented by compressing the multi-dimensional efficacy state vector into a single score through a linear function, which is convenient for doctors to evaluate and rank. The weight is fitted from real labeled data, which has statistical significance and clinical interpretability, and does not rely on hard-coded rules, and can be iteratively optimized as data accumulates. The score result can not only be used for individual efficacy evaluation, but also be used in combination with cluster analysis to construct the optimal parameter template and evaluate the deviation degree of the current treatment course, providing fine-tuning decision basis for doctors. The scoring function has good generalization performance and is suitable for efficacy quantification of different diseases, treatment parameter combinations or stages, and can also be extended to a nonlinear scoring model (such as MLP) to further improve the expression ability.

[0100] S3, K-medoids clustering is performed on the terminal state representation vector of the patients in the same cabin to analyze the optimal parameter combination corresponding to the optimal efficacy and generate a recommended template;

[0101] Specifically, the K-medoids clustering is performed on the terminal state representation vector of the patients in the same cabin to analyze the optimal parameter combination corresponding to the optimal efficacy and generate a recommended template, which automatically collects all the state vectors of the patients in the current cabin at the end of the treatment course, forms a clustering sample set, performs Z-score standardization processing on all vectors to generate a standardized vector set, determines the optimal clustering number K using the silhouette coefficient method, and applies the K-medoids algorithm to the standardized vector set based on K to perform clustering modeling, obtaining multiple patient clusters with similar efficacy, each clustering cluster representing a patient population with similar efficacy curve performance.

[0102] Further, the treatment parameter combinations (such as oxygen pressure, oxygen inhalation time, treatment interval, etc.) used in each cluster are further counted, and the frequency of each combination and the average value of the corresponding terminal state efficacy score are calculated, to select the highest frequency and optimal treatment configuration parameter group for each cluster as the recommended optimal treatment strategy for each cluster.

[0103] By deducing the optimal parameter combination from the "efficacy performance similar group", the traditional template setting method based on diagnostic labels is broken, and the precise strategy generation based on efficacy distribution is realized. The implicit mapping relationship between treatment parameter combinations and efficacy is automatically counted and analyzed to provide parameter adjustment suggestions for doctors and shorten the scheme iteration cycle. K-medoids is superior to K-means in handling boundary samples and small class groups, and is suitable for situations where there is heterogeneity and irregular distribution in clinical samples. The recommended parameter template is derived from real case data statistics, has high interpretability, and is convenient for doctors to further adjust and deploy based on experience.

[0104] S4, the individual terminal state representation is matched with the recommended parameters for scoring, the deviation patients are marked, and the individual evaluation report and ward statistical report are generated;

[0105] Specifically, the individual end-state representation is matched with the recommended parameters to score the matching degree, the deviation patients are marked, and the individual evaluation report and ward statistical report are generated. According to the optimal treatment parameter combination in the cluster to which the patient belongs, the corresponding structured parameter vector is extracted, the actual treatment parameters executed by the patient are synchronously acquired and mapped into a vector, the deviation degree between the current treatment setting and the optimal scheme is measured by calculating the Euclidean distance between the two, the patient efficacy quality score is calculated by a lightweight multi-layer perceptron (MLP) model using the efficacy end-state vector, and the patient efficacy matching score is output based on the efficacy quality score and the deviation degree by a weighting formula;

[0106] The threshold values D and U are set based on experience, and D>U. If the patient efficacy matching score is greater than the threshold value D, it indicates that the current efficacy of the patient is good, and the treatment parameters are highly matched with the recommended scheme, and no adjustment is needed.

[0107] If the patient efficacy matching score is less than the threshold value D and greater than or equal to the threshold value U, it indicates that there is a certain deviation in the patient's efficacy, and whether to adjust needs to be determined in combination with the parameter deviation and the doctor's judgment.

[0108] If the patient efficacy matching score is less than the threshold value U, the patient will be automatically marked as a "optimization object" and the medical staff will be prompted to pay attention.

[0109] An individualized evaluation report is generated for each patient, containing basic information, comparison of current parameters and recommended parameters, efficacy matching score, grade explanation, and parameter adjustment suggestion prompt. A batch statistical table is automatically summarized according to the ward dimension, covering efficacy grade distribution, parameter deviation heat map, efficacy trend change curve, and score before / after ranking. The individual report is returned to the electronic medical record system in the form of PDF, and the ward report is simultaneously exported to Excel and provided for management terminal reference.

[0110] By fusing the structured comparison of the efficacy end-state vector and the recommended parameters, the matching degree between the individual treatment effect of the patient and the optimal treatment strategy is realized, and the ability to accurately identify deviation patients is possessed. A lightweight multi-layer perceptron model is introduced to score the efficacy quality, and a fusion scoring system is constructed by combining the parameter deviation degree, which can dynamically evaluate the patient efficacy performance and the degree of scheme adaptation. Through a threshold grading mechanism, hierarchical intervention strategies are clearly defined, and individual and ward statistical information are output in the form of visual reports, effectively assisting doctors in quickly identifying optimization objects and improving the efficiency of individualized management of oxygen cabin treatment and the quality of group intervention decision-making.

[0111] The embodiment also provides an HBO oxygen cabin treatment effect evaluation system, comprising:

[0112] A patient information acquisition module is configured to automatically acquire medical orders, diagnoses, and historical data, and multi-modal indicators by wristband recognition, and to standardize the treatment initial vector.

[0113] a medical semantic modeling module, configured to construct efficacy triples based on the standardized data and the medical ontology, generate semantic graph embedding vectors by using a TransE model, and splice the initial vectors to form node features of each treatment;

[0114] a treatment course modeling module, configured to construct a time graph structure in time sequence according to all the treatment node features, model a patient treatment course evolution track by using a Time-GCN model, and output a terminal state representation vector;

[0115] an efficacy scoring module, configured to perform efficacy scoring on the terminal state vector by using a linear scoring function, and obtain a structured representation reflecting overall treatment effect;

[0116] a parameter recommendation module, configured to perform K-medoids clustering analysis on the terminal state vectors of patients in the same cabin, extract optimal treatment parameter groups of various groups, and generate a recommendation template;

[0117] a matching evaluation module, configured to calculate a matching degree between a current treatment scheme of an individual and recommended parameters, automatically generate an individualized evaluation report and ward efficacy statistical results, and support doctor decision adjustment.

[0118] The embodiment also provides a computer device suitable for the HBO oxygen cabin treatment effect evaluation method, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the HBO oxygen cabin treatment effect evaluation method suitable for the HBO oxygen cabin treatment effect evaluation method.

[0119] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to perform wired or wireless communication with external terminals. The wireless communication can be achieved by WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0120] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for evaluating the treatment effect of HBO oxygen cabin according to the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0121] It should be noted that the above embodiment is only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for assessing the effectiveness of HBO oxygen chamber treatment, characterized in that: The method comprises the following steps: Before the patient enters the cabin, the wristband is used to identify the HIS system, collect the patient's medical order, diagnosis, and multi-modal physiological and subjective score data, and generate an initial treatment vector after standardization; Based on the medical ontology, a treatment effect triple is constructed, a semantic graph embedding vector is generated using the TransE model, the treatment data of each time is spliced with the graph embedding, a time graph structure is constructed, the Time-GCN model is used to model the patient's treatment effect evolution trajectory, and the final state representation vector is output and the final treatment effect score is calculated; The final state representation vector of the patients in the same cabin is clustered by K-medoids, the optimal treatment effect corresponding parameter combination is analyzed, and a recommended template is generated; The individual final state representation and the recommended parameters are matched and scored, the deviation patients are marked, and individual evaluation reports and ward statistical reports are generated.

2. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 1, characterized in that: The method comprises the following steps: When the patient arrives at the hyperbaric oxygen cabin treatment area, the nursing staff uses a handheld PDA equipped with NFC function to scan the patient's wristband, realizes the rapid binding of the patient's identity through the unique hospitalization number ID, establishes a data channel with the HIS system, and automatically pulls the medical order and diagnosis data related to the patient's treatment on the same day; The collected data is composed into a feature raw data vector, and the raw vector is subjected to Z-score standardization processing to form a standardized initial vector X0.

3. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 2, characterized in that: The method comprises the following steps: From the standardized initial vector, all feature names and feature directions are extracted, the first-order change rate calculation is performed on each index field, the medical ontology library or local mapping table is called for entity standardization, the index entity set E is established, and the medical index source of each entity is recorded; Rule-driven triple generation is performed on the standardized entity set E, and a patient individual triple set is constructed based on the generated triples; From the triple set T, all semantic objects involved in the current patient are extracted to obtain the entity set and the relationship set, and independent embedding vectors are assigned to all entities and relationships and initialized; After initialization, the TransE model is used for graph modeling, according to the definition of the TransE model, if the semantic of each triple is correct, the embedding vector satisfies the condition; Based on the definition, a training target function is constructed, and the structural consistency of all triples in the embedding space is modeled; After the training is completed, the triple embedding results of the current patient semantic graph are uniformly aggregated to form a graph-level overall representation ; After the target function is constructed, the random gradient descent algorithm is used to train the embedding vector, and the training process is iterated, the vector parameters are adjusted by batch update in each round, and the vectors are normalized after each round of training; The graph embedding vector G0 and the standardized initial vector X0 are subjected to vector splicing operation to form a comprehensive node feature vector F0.

4. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 3, characterized in that: The construction time graph structure models the patient treatment evolution trajectory by using a Time-GCN model, and outputs a terminal state representation vector indicating that the patient treatment node features are collected in time sequence and constitute an ordered set The node sequence in V is mapped to a time graph structure A=(V,E,T), wherein is the fusion feature vector of the last treatment node in the entire treatment course of the patient, V is a node set, each node F t represents a treatment node, T is a timestamp set, and E is an edge set; For each F t , connect to adjacent nodes before and after to form edges and assign time-decaying adjacency weights to each edge ; The initial feature F of each node is obtained t as the initial input representation of the graph neural network and performs a two-layer time-enhanced graph convolution network propagation operation on the graph structure, which realizes the modeling of the evolution of historical curative effects by aggregating the state vectors of neighboring nodes and combining time decay weights, and extracts the semantic representation of the patient state at each time point. After all graph convolution propagations are done, output the sequence of node states at each time point throughout the entire course of treatment wherein denotes the terminal state vector representation at the end of the current patient course.

5. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 4, characterized in that: The computing the final efficacy score refers to mapping to a scalar efficacy score using a linear projection scoring function .

6. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 5, wherein: The K-medoids clustering of the treatment effect end state representation vector of the patient in the same cabin is performed, the optimal treatment effect corresponding parameter combination is analyzed, and a recommended template is generated. The state vector of all patients in the current cabin at the end of the treatment is automatically collected, a clustering sample set is formed, Z-score standardization processing is performed on all vectors to generate a standardized vector set, the optimal cluster number K is determined by using the silhouette coefficient method, and K-medoids algorithm is applied on the standardized vector set for clustering modeling according to the K, so that a plurality of patient clusters with similar treatment effects are obtained, and each clustering cluster represents a patient population with similar treatment effect curve performance; Further, the treatment parameter combination used in each cluster is counted, the frequency of each combination and the average value of the corresponding end state treatment effect score are calculated, the treatment configuration parameter group with the highest frequency and the optimal effect is selected for each cluster as the optimal treatment strategy recommended for each cluster.

7. The method for assessing the therapeutic effect of HBO oxygen cabin according to claim 6, characterized in that: The individual end state representation is matched with the recommended parameters, the deviation patients are marked, and the individual evaluation report and the ward statistical report are generated. According to the optimal treatment parameter combination in the clustering cluster to which the patient belongs, the corresponding structured parameter vector is extracted, the treatment parameters actually executed by the patient are synchronously acquired and mapped into a vector, the deviation degree between the current treatment setting and the optimal scheme is measured by calculating the Euclidean distance between the two, the patient treatment effect quality score is calculated by using a lightweight multi-layer perceptron model through the treatment effect end state vector, and the patient treatment effect matching score is output based on the treatment effect quality score and the deviation degree through a weighted formula; Threshold values D and U are set, and D>U, if the patient treatment effect matching score is greater than the threshold value D, it indicates that the patient has good current treatment effect performance, and the treatment parameters are highly matched with the recommended scheme, and no adjustment is needed; If the patient treatment effect matching score is less than the threshold value D and greater than or equal to the threshold value U, it indicates that the patient treatment effect has a certain deviation, and whether to adjust needs to be determined in combination with the parameter deviation and the doctor's judgment; If the patient treatment effect matching score is less than the threshold value U, the patient will be automatically marked as a “optimization object”, and medical staff are prompted to pay attention to it; An individualized evaluation report is generated for each patient, and a batch statistical table is automatically summarized according to the ward dimension, covering the treatment effect grade distribution, the parameter deviation heat map, the treatment effect trend change curve and the ranking before and after scoring. The individual report is returned to the electronic medical record system in the form of PDF, and the ward report is synchronously exported to Excel and provided for management terminal to check.

8. A system for assessing the therapeutic effect of an HBO oxygen cabin, based on the method for assessing the therapeutic effect of an HBO oxygen cabin according to any one of claims 1 to 7, characterized in that: The patient information acquisition module is used to automatically acquire medical orders, diagnoses and historical data and multi-modal indexes through wristband recognition and linkage with the HIS system, and to standardize the treatment initial vector; The medical semantic modeling module is used to construct a treatment effect triple based on standardized data and medical ontology, generate a semantic graph embedding vector by using a TransE model, and splice the initial vector to form the node features of each treatment; The treatment course modeling module is used to construct a time graph structure according to the sequence of all treatment node features, model the patient treatment evolution track by using a Time-GCN model, and output an end state representation vector; The treatment effect scoring module is used to perform treatment scoring on the end state vector by using a linear scoring function, so as to obtain a structured representation reflecting the overall treatment effect. ​ A parameter recommendation module is configured to perform K-medoids clustering analysis on the terminal state vectors of the patients in the same cabin, extract optimal treatment parameter groups of each cluster, and generate a recommendation template. A matching evaluation module is configured to calculate the matching degree between the current treatment scheme of the individual and the recommended parameters, automatically generate an individualized evaluation report and ward efficacy statistical results, and support the doctor in decision adjustment. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the HBO oxygen cabin treatment effect evaluation method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the HBO oxygen cabin treatment effect evaluation method according to any one of claims 1-7.

Images