A method and system for processing symptom cluster data of a head and neck tumor patient based on sentinel symptoms, equipment and medium

Abstract

A method, system, device, and medium for processing symptom cluster data of head and neck cancer patients based on sentinel symptoms are disclosed. The method includes: obtaining symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal surveys; processing the symptom data using exploratory factor analysis to extract several symptom clusters; performing association rule analysis on the symptom cluster data based on the Apriori algorithm, and determining the sentinel symptoms within each symptom cluster by combining the time of first symptom appearance; and constructing an intervention plan for gastrointestinal symptoms using the sentinel symptoms as intervention targets. This invention achieves accurate determination of sentinel symptoms through longitudinal survey data, combined with the Apriori algorithm and the time of first symptom appearance, and constructs intervention plans based on the MRC complex intervention framework, realizing early symptom identification and providing precise and personalized symptom management support for head and neck cancer patients.

Description

Technical Field

[0001] This invention relates to the field of medical image data processing technology, and more specifically, to a method, system, device, and medium for processing symptom cluster data of head and neck tumor patients with sentinel symptoms. Background Technology

[0002] Head and neck cancer (HNC) is an epithelial malignant tumor originating from the paranasal sinuses, nasal cavity, oral cavity, pharynx, larynx, and salivary glands. Surgery, radiotherapy, chemotherapy, and multimodal combination therapy are the main treatment methods for HNC. Due to the complex anatomical structure and the influence of treatment factors, patients may experience a variety of physiological and psychological discomfort symptoms, such as radiation dermatitis, difficulty chewing, difficulty swallowing, taste disturbances, oral mucositis, dry mouth, pain, fatigue, loss of appetite, anxiety, and depression. These symptoms can persist for years, not only causing malnutrition and delaying the treatment process, but also exacerbating patient suffering, generating negative emotions such as anxiety and fear, and even interrupting the treatment plan, seriously affecting the patient's quality of life and treatment adherence.

[0003] In recent years, domestic and international scholars have employed cross-sectional, longitudinal, and qualitative research designs to explore the composition of symptom clusters in head and neck cancer patients under different treatment modalities, with studies primarily focusing on nasopharyngeal carcinoma, oral cancer, and tongue cancer. Some researchers have proposed the concept of "sentinel symptoms," which are indicative and marker symptoms within a symptom cluster. Identifying sentinel symptoms and implementing targeted single-symptom interventions holds promise as a new direction for comprehensive symptom cluster management. Meanwhile, deep learning-based image recognition technology has made progress in the wound care field, such as using convolutional neural network architectures like U-Net for wound image segmentation training. Furthermore, retrieval-based generative enhancement (RAG) methods, by integrating external knowledge bases with patient interaction information, offer a new technical approach to improving the accuracy and personalization of symptom management recommendations.

[0004] However, existing technologies still have significant shortcomings. On the one hand, head and neck cancer patients present with complex and diverse symptoms, and current clinical symptom identification and management largely rely on offline medical consultations, making it difficult to continuously and efficiently meet the dynamic and individualized nursing needs of patients. On the other hand, the direct application of general-purpose large language models to medical scenarios often leads to hallucination problems, making it difficult to meet actual clinical needs; research on the integration of deep learning and wound care is mainly concentrated abroad, and my country still needs to further deepen its efforts in the precision of wound assessment and its practical application.

[0005] Furthermore, existing symptom cluster studies are mainly descriptive, with relatively few studies exploring intervention measures and mechanisms. They also rarely examine the longitudinal changes in symptom clusters in head and neck cancer patients before, during, and after radiotherapy and chemotherapy. There is a lack of complex intervention programs that target sentinel symptoms within symptom clusters, making it difficult to provide closed-loop management support for clinical practice from symptom identification to precise intervention. Summary of the Invention

[0006] The purpose of this invention is to address the problems in the existing technology, such as the lack of longitudinal research on the dynamic changes of symptom clusters in head and neck cancer patients during radiotherapy and chemotherapy, the lack of objective criteria for sentinel symptoms, and the lack of intelligent and individualized symptom management programs. This invention provides a method and system for processing symptom cluster data of head and neck cancer patients based on sentinel symptoms, enabling dynamic identification of symptom clusters during radiotherapy and chemotherapy, accurate determination of sentinel symptoms, and intelligent symptom intervention recommendations based on this determination.

[0007] This invention provides a method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms, the method comprising: Step S1: Obtain symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal survey; Furthermore, the longitudinal survey used convenience sampling to select research subjects, who were symptom data of head and neck tumor patients who received radiotherapy and chemotherapy in medical institutions. Furthermore, data were collected from the study subjects based on the head and neck cancer-specific module of the Anderson Symptom Assessment Scale and the timeline of first symptom onset, for 3-5 chemotherapy cycles. Step S2: The symptom data is processed using exploratory factor analysis to extract several symptom clusters; Step S3: Based on the Apriori algorithm, perform association rule analysis on the symptom cluster data, and combine the first occurrence time of symptoms to determine the sentinel symptom classification within each symptom cluster; Specifically, step S3 also includes: Step S31: Obtain longitudinal survey data and the timeline of first symptom appearance for each symptom within the symptom cluster; The data was organized into a standardized data matrix, and the behavioral research objects of the data matrix were listed as symptom name, time of first symptom appearance, and symptom score at each time point. Step S32: Use the Apriori algorithm to perform association rule analysis on each symptom within the symptom cluster, and generate a set of association rules between symptoms; Specifically, based on the data matrix organized in step S31, the minimum support is set to 20% and the minimum confidence is set to 50%, and association rule mining is performed on the symptoms in each symptom group to discover the positive association between symptoms; Furthermore, during the algorithm's operation, all symptom data are first scanned to generate frequent 1-itemsets, and then frequent 2-itemsets and frequent 3-itemsets are generated based on the frequent 1-itemsets, until no higher-order frequent itemsets can be generated. Finally, association rules are generated based on the frequent itemsets, and the support, confidence, and lift of each rule are calculated. Step S33: Based on the association rule set, determine the symptoms that meet the conditions as sentinel symptoms; Furthermore, the symptom mentioned is the earliest symptom to appear within the symptom cluster; Furthermore, the symptom is the antecedent in at least one symptom association pair, and satisfies the threshold conditions that the antecedent support is >40%, the confidence is >60%, and the confidence is >the antecedent support in the association analysis of each symptom in the symptom cluster. Step S34: Output sentinel symptom assessment data; Step S4: Using the sentinel symptoms as intervention targets, construct an intervention plan for gastrointestinal symptoms; Furthermore, the proposed intervention scheme for constructing digestive symptoms also includes constructing a knowledge vector base for head and neck cancer health management, using the FAISS library to construct an approximate nearest neighbor vector index for retrieval, and achieving semantic matching and response generation through inverted clustering and group search; Furthermore, based on the generated search results, intervention plans were formulated using the MRC complex intervention framework as the theoretical framework; Furthermore, the MRC Complex Intervention Framework encompasses four stages: assessment, intervention, monitoring, and adjustment.

[0008] This invention also provides a data processing system for symptom clusters in head and neck tumor patients based on sentinel symptoms. The system includes: a data acquisition module for acquiring symptom data of head and neck tumor patients during radiotherapy and chemotherapy through longitudinal surveys; an extraction module for processing the symptom data using exploratory factor analysis to extract several symptom clusters; a data analysis module for performing association rule analysis on the symptom cluster data based on the Apriori algorithm, and determining the sentinel symptoms within each symptom cluster by combining the time of first symptom appearance; an intervention plan construction module for constructing intervention plans for gastrointestinal symptoms using the sentinel symptoms as intervention targets and the MRC complex intervention framework as the theoretical framework; and an intelligent symptom management module including an image recognition unit and a retrieval-based generation enhancement unit for acquiring and recognizing symptom-related image information of head and neck tumor patients, and generating individualized symptom recognition results and intervention recommendations based on the sentinel symptoms and symptom clusters.

[0009] The present invention also provides an electronic device, comprising: a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to perform the above-described method.

[0010] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0011] Through the above technical solution, the present invention can achieve at least the following technical effects: 1. This invention, through longitudinal tracking, identifies stable symptom clusters and their sentinel symptoms during radiotherapy and chemotherapy in patients with head and neck tumors. Using these sentinel symptoms as early intervention targets, and through data processing to develop intervention methods, targeted interventions can be implemented before the symptom clusters fully erupt. This helps achieve the nursing goal of "minimizing symptom burden and optimizing quality of life," thus shifting the focus of symptom management upstream.

[0012] 2. This invention is guided by the UK Medical Research Council (MRC) Complex Interventions Study Framework and targets gastrointestinal symptom clusters in head and neck cancer patients undergoing radiotherapy and chemotherapy. The proposed approach more closely mirrors real-world clinical scenarios, overcoming the limitations of single interventions.

[0013] 3. This invention utilizes convolutional neural networks (such as U-Net architecture) to automatically segment and evaluate wound images of oropharyngeal radiation dermatitis, oral mucositis, etc., improving the objectivity and accuracy of wound assessment. Furthermore, by integrating symptom management knowledge base and individualized patient interaction information through a retrieval-based generative enhancement method, it effectively overcomes the problem of hallucinations caused by general large language models in medical scenarios, significantly improving the accuracy and intelligence level of symptom recognition and intervention recommendations.

[0014] 4. This invention uses a longitudinal survey method to reveal the evolution of symptom clusters as the treatment progresses, providing a scientific basis for phased and dynamic adjustment of intervention strategies in clinical practice, and laying a theoretical foundation for subsequent complex intervention studies based on sentinel symptoms. Attached Figure Description

[0015] Figure 1 This is a flowchart of a data processing method according to Embodiment 1 of the present invention. Detailed Implementation

[0016] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the following embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0017] Example 1 This embodiment provides a method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms, which is used for dynamic monitoring of symptoms and identification of sentinel symptoms in head and neck tumor patients during radiotherapy and chemotherapy, so as to achieve early warning and precise intervention of symptom clusters.

[0018] like Figure 1 As shown, the data processing method in this embodiment includes the following steps: Step S1: Obtain symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal survey; Furthermore, the longitudinal survey used convenience sampling to select study subjects who were head and neck cancer patients who received radiotherapy and chemotherapy in the oncology and head and neck surgery departments of medical institutions. The subjects covered common head and neck cancer types such as laryngeal cancer, nasopharyngeal cancer, oral cancer, and hypopharyngeal cancer to ensure the representativeness of the study subjects. The inclusion criteria for the study participants were: ① Age ≥18 years and ≤75 years, covering patients from young to old age to reduce the influence of age on symptom presentation; ② A clear pathological diagnosis of head and neck cancer, confirmed by pathological biopsy or postoperative pathology, excluding suspected cases; ③ No mental or psychological illnesses (such as schizophrenia, major depressive disorder, etc.), and a score below the cutoff value on the Anxiety and Depression Scale (SAS, SDS); ④ Able to communicate normally, have the ability to read and understand text without difficulty, and be able to complete the questionnaire independently or with the assistance of family members, excluding patients with confusion or cognitive impairment.

[0019] Exclusion criteria for study participants: ① Patients diagnosed with other primary cancers (such as lung cancer, gastric cancer, etc.) to avoid interference from other cancer symptoms in the assessment of symptoms related to radiotherapy and chemotherapy for head and neck tumors; ② Severe heart, liver, kidney, or digestive system diseases (such as heart failure, cirrhosis, renal failure, severe gastric ulcers, etc.), as these diseases themselves may cause symptoms similar to the side effects of radiotherapy and chemotherapy, affecting the accuracy of the data; ③ Patients who interrupted treatment, were transferred to other hospitals, or died during radiotherapy and chemotherapy due to severe adverse reactions, to ensure the continuity of data; ④ Patients who refused to participate in the study or were unable to cooperate in completing the entire follow-up.

[0020] After the study subjects were determined, the sample size was estimated using a table based on a single-group repeated measures design. Each patient was measured four times to ensure that the changes in symptoms at different treatment stages were captured. Using an average correlation coefficient of r=0.3, f=0.12 (weak effect), and α=0.01, and ensuring 1-β=0.8, the required sample size was estimated to be 273 cases using the sample size calculation formula. Considering the possibility of loss to follow-up in longitudinal studies, an estimated loss to follow-up rate of 20% was adopted, and the final required sample size was determined to be 342 cases. The research tools for the research subjects consist of two parts: a demographic questionnaire and a disease characteristics questionnaire. The first part contains demographic information such as age, gender, education level, and medical payment method, and the second part contains disease-related information such as pathological type, tumor stage, chemotherapy regimen, and chemotherapy cycle. The first part contains demographic information, including but not limited to: 1. Age, recorded in segments: 18-44 years old, 45-59 years old, 60-75 years old; 2. Gender: Male or Female; 3. Education level: Primary school or below, junior high school, senior high school / technical secondary school, junior college or above; 4. Medical payment methods: Employee medical insurance, urban and rural resident medical insurance, out-of-pocket payment, and others; 5. Marital status: Married, Single, Divorced / Widowed; 6. Occupational status: Employed, Retired, Unemployed; The second part contains disease-related information, including but not limited to: 1. Pathological types: squamous cell carcinoma, adenocarcinoma, others; 2. Tumor staging: Stage I, Stage II, Stage III, Stage IV; 3. Chemotherapy cycle; 4. Whether surgical treatment is required; Furthermore, data were collected from the study subjects based on the head and neck cancer-specific module of the Anderson Symptom Assessment Scale and the timeline of first symptom onset, for 3-5 chemotherapy cycles. The timeline of the first appearance of symptoms was a self-made questionnaire with the same symptom items as the head and neck module of the Anderson Symptom Assessment Scale, including a total of 19 symptoms. Patients recorded the time of the first appearance of each symptom (from the time of chemotherapy drug infusion) on the questionnaire in hours. Select research subjects who meet the inclusion and exclusion criteria, inform them of the research purpose and methods, and obtain written informed consent from the research subjects. Use standardized instructions to guide research subjects in completing the questionnaire. If research subjects encounter difficulties in completing the questionnaire, they should communicate carefully with the researchers and confirm that there are no errors before the researchers can complete the questionnaire on their behalf or the patients' family members can assist in completing it. After all questionnaires are completed, the researchers check them for any errors or omissions, verify with the patients, and then archive the questionnaire numbers.

[0021] Step S2: The symptom data is processed using exploratory factor analysis to extract several symptom clusters; SPSS Statistics 22.0 and SPSS Modeler 18.0 statistical software were used to analyze the data. Before data analysis, data cleaning was performed to remove outliers. The Z-test was used, and values ​​with |Z| > 3 were identified as outliers. Missing data with a missing rate of < 5% were imputed using the mean / median, while samples with a missing rate of ≥ 5% were removed to ensure data validity. Normality and homogeneity of variance tests were also performed to provide a basis for the selection of subsequent statistical methods. In quantitative data, data that conform to a normal distribution are expressed as mean ± standard deviation (SD). Data such as patient age and radiotherapy dose are expressed as mean ± standard deviation (±s); non-normally distributed data are expressed as median and interquartile range [M(P25, P75)], such as symptom scores and time of first symptom onset; count data are expressed as frequency and proportion for descriptive statistics, such as the number and proportion of patients with different pathological types and the incidence of different symptoms. Repeated measures ANOVA was used to compare the mean values ​​of multiple sets of measurement data to compare the differences in symptom scores of patients at different follow-up time points. If the difference was statistically significant (p < 0.05), the LSD method was used for pairwise comparisons. For symptom clusters based on severity, exploratory factor analysis (EFA) was used. Specifically, principal component analysis was used to extract factors, and the factor rotation method was used to rotate the factors to make the factor loadings clearer and facilitate the classification of symptom clusters. To ensure sufficient variability in the data to support exploratory factor analysis, only symptoms with an incidence rate greater than 20% were included in the analysis. Furthermore, the factor selection criteria are as follows: the factor eigenvalue is greater than 1 (Kaiser criterion) to ensure that the extracted factors have a certain explanatory power; the symptom factor loading is greater than 0.5 to ensure a strong correlation between symptoms and corresponding factors; and the symptoms have loadings on at least two factors to facilitate the subsequent integration and analysis of symptom clusters; a p-value of less than 0.05 is considered statistically significant to ensure the reliability of the analysis results. Exploratory factor analysis yielded four symptom clusters, with a cumulative variance contribution rate of 72.36%, indicating that the extracted symptom clusters can effectively explain the original symptom data. These are as follows: ① Gastrointestinal symptom cluster, such as nausea, vomiting, and loss of appetite; ② Oral and pharyngeal symptom cluster, such as dry mouth, difficulty swallowing, oral ulcers, and throat discomfort; ③ Systemic symptom cluster, such as fatigue, weakness, and dizziness; ④ Pain-related symptom cluster, such as pain and altered taste. Step S3: Based on the Apriori algorithm, perform association rule analysis on the symptom cluster data, and combine the first occurrence time of symptoms to determine the sentinel symptom classification within each symptom cluster; Specifically, step S3 also includes: Step S31: Obtain longitudinal survey data and the timeline of first symptom appearance for each symptom within the symptom cluster; The data is organized into a standardized data matrix. The behavioral research objects of the data matrix are listed as symptom name, first appearance time of symptom, and symptom score at each time point to ensure that the data format is consistent and facilitates Apriori algorithm analysis. Step S32: Use the Apriori algorithm to perform association rule analysis on each symptom within the symptom cluster, and generate a set of association rules between symptoms; Specifically, the Apriori algorithm is implemented using Python; The steps are as follows: Based on the data matrix organized in step S31, set the minimum support to 20% and the minimum confidence to 50%, and perform association rule mining on the symptoms in each symptom cluster to discover positive associations between symptoms; during the algorithm operation, first scan all symptom data to generate frequent 1-itemsets, then generate frequent 2-itemsets and frequent 3-itemsets based on the frequent 1-itemsets, until no higher-order frequent itemsets can be generated. Finally, generate association rules based on the frequent itemsets, and calculate the support, confidence, and lift of each rule. Among them, lift > 1 indicates that there is a positive association between two symptoms. The larger the lift, the stronger the association. Step S33: Based on the association rule set, determine the symptoms that meet the conditions as sentinel symptoms; Furthermore, the symptom mentioned is the earliest symptom to appear within the symptom cluster; Furthermore, the symptom is the antecedent in at least one symptom association pair, and satisfies the threshold conditions that the antecedent support is >40%, the confidence is >60%, and the confidence is >the antecedent support in the association analysis of each symptom in the symptom cluster. Step S34: Output sentinel symptom assessment data; Specifically, the sentinel symptoms, association rule parameters (support, confidence, lift), and statistical results of the first occurrence time of symptoms for each symptom cluster are compiled into a standardized report and output to the data management system. The report clearly marks the criteria for determining each sentinel symptom, its scope of application, and the warning threshold, providing clear data support for the construction of subsequent intervention plans. Step S4: Using the sentinel symptoms as intervention targets, construct an intervention plan for gastrointestinal symptoms; Furthermore, the proposed intervention scheme for constructing digestive symptoms also includes constructing a knowledge vector base for head and neck cancer health management, using the FAISS library to construct an approximate nearest neighbor vector index for retrieval, and achieving semantic matching and response generation through inverted clustering and group search; Among them, the present invention selects a large language model (GLM) with a multi-head self-attention mechanism and a sequence-to-sequence architecture as the base model, and trains it by optimizing the autoregressive fill-in-the-blank objective. Let \(Z_m\) be the set of all possible permutations of the index sequence \([1, 2, \ldots, m]\) of length \(m\). Multiple text fragments \([S_1, \ldots, S_m]\) are sampled from the input text \(x\), where \(S_{z < i}\) represents \([S_{z1}, \ldots, S_{zi - 1}]\). The input \(x\) is divided into two parts: one part is the corrupted text \(x_{corrupt}\), and the other part is the text containing the masked span, and the pre-training objective is defined as the following formula:

[0022] Specifically, the present invention adopts the upgraded version ChatGLM3 - 6B model of the GLM series. This model is optimized based on a mixed objective function, and the inference speed is improved by about 42% compared with the previous generation. At the same time, by adopting INT4 quantization technology, it supports an 8K context length and can run efficiently under the condition of 6 GB video memory, which is suitable for building a specialist intelligent question-answering model in resource-constrained environments.

[0023] In terms of text data processing, the present invention uses the TextSplitter tool to perform chunking on the text data. Texts related to head and neck cancer health management, including guidelines for the management of radiotherapy and chemotherapy side effects, specifications for the intervention of digestive tract symptoms, patient care manuals, etc., are chunked according to semantic logic, and the length of each chunk is controlled within 512 characters to avoid semantic loss caused by overly long texts; At the same time, the m3e-base model is used to generate sentence vectors, and the chunked text is converted into high-dimensional vectors, so as to build a head and neck cancer health management knowledge vector library. The vector library includes multiple modules such as the intervention of digestive tract symptoms, radiotherapy and chemotherapy care, diet guidance, and psychological intervention, so as to ensure the comprehensiveness of knowledge coverage and the accuracy of retrieval; The present invention adopts the retrieval-augmented generation (RAG) technology. First, question-answer pairs related to the user's question are retrieved from the head and neck cancer health management knowledge vector library. Then, the retrieval results are input into the large language model for generation enhancement and accuracy verification; Among them, to improve the efficiency of large-scale data retrieval, the present invention uses the FAISS library to build an approximate nearest neighbor vector index. Through the inverted index clustering and grouped search strategies, fast and accurate semantic matching and response generation are achieved.

[0024] Based on the generated retrieval results, an intervention plan is formulated with the MRC complex intervention framework as the theoretical framework; The above framework covers four links: evaluation, intervention, monitoring and adjustment. The specific intervention plan is as follows: 1. Early Warning and Assessment Phase: A symptom monitoring mechanism is established, focusing on sentinel symptoms. Starting 6 hours after chemotherapy, patients record the occurrence of related discomfort symptoms and symptom scores every 3 hours. If discomfort symptoms occur (score ≥ 1 point), the intervention process is initiated immediately. Simultaneously, combining the patient's demographic information, disease characteristics, and past symptom history, a risk assessment scale is used to classify the patient's risk of gastrointestinal symptoms (low risk, medium risk, high risk), providing a basis for personalized intervention. 2. Precision Intervention Phase: Based on risk stratification and the patient's specific condition, tiered intervention measures are developed, specifically as follows: ① Low-risk patients (score 1-3): Non-pharmacological interventions are used, including dietary interventions, psychological interventions, and postural interventions; ② For patients with medium risk (score of 4-6): In addition to non-pharmacological intervention, oral antiemetics should be added, and the dosage should be adjusted according to the patient's weight; ③ High-risk patients (score 7-10): Intravenous infusion of antiemetic drugs, combined with gastrointestinal mucosal protectants, and adjustment of chemotherapy drug dosage or infusion rate as necessary. At the same time, nutritional support should be strengthened to avoid dehydration and electrolyte imbalance. 3. Dynamic monitoring phase: After the intervention is initiated, the patient's nausea symptom score and the occurrence of other gastrointestinal symptoms are monitored every 2 hours, and the trend of symptom changes is recorded; at the same time, through the intelligent question and answer system, the patient can provide feedback on their symptoms at any time, and the system will adjust the search results and intervention suggestions in real time based on the feedback to ensure the targeted nature of the intervention measures.

[0025] 4. Treatment Plan Adjustment Phase: After each chemotherapy cycle, the intervention effect is evaluated. If the symptom relief rate is <70% or the patient satisfaction rate is <80%, the intervention plan is optimized based on the changes in the patient's symptoms and the adjustment of the treatment plan. This may include adjusting the drug dosage or changing the intervention measures to ensure the effectiveness and suitability of the intervention plan.

[0026] Example 2 This embodiment provides a symptom cluster data processing system for head and neck tumor patients based on sentinel symptoms. The system includes: Data acquisition module: used to obtain symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal surveys; Extraction module: used to process the symptom data using exploratory factor analysis to extract several symptom clusters; Data analysis module: used to perform association rule analysis on the symptom cluster data based on the Apriori algorithm, and determine the sentinel symptoms in each symptom cluster by combining the first occurrence time of the symptoms; Intervention program construction module: used to construct intervention programs for gastrointestinal symptoms using the sentinel symptoms as intervention targets and the MRC complex intervention framework as the theoretical framework; The intelligent symptom management module includes an image recognition unit and a retrieval-based generation enhancement unit, which are used to acquire and identify symptom-related image information of patients with head and neck tumors, and generate individualized symptom identification results and intervention recommendations based on the sentinel symptoms and symptom clusters.

[0027] The system in this embodiment can be deployed in a variety of application scenarios: 1. In hospital clinical information system deployment scenarios, the system can be integrated into the hospital's electronic medical record system or nursing information platform, serving as a clinical decision support module for medical staff. Doctors or nurses can access patients' symptom assessment data in real time, and the system automatically identifies sentinel symptoms and pushes individualized intervention plans to assist clinicians in carrying out precise symptom management.

[0028] 2. In mobile terminal application deployment scenarios, patients can independently upload symptom descriptions or images of affected areas via mobile devices. The system will return symptom recognition results and management suggestions in real time, enabling remote symptom monitoring and self-management in out-of-hospital settings.

[0029] In summary, the embodiments of the present invention have the following advantages over the prior art: 1. This invention utilizes structured, semi-structured, and unstructured data, and employs a bottom-up approach to construct a symptom management knowledge graph. By combining the Neo4j graph database with vector indexing tools, it achieves efficient construction of a symptom management knowledge database for head and neck cancer patients, overcoming the shortcomings of isolated knowledge representation and low retrieval efficiency in existing technologies.

[0030] 2. By combining large language models with advanced natural language processing techniques, a complete process from user query parsing to accurate symptom management recommendations was established, realizing a head and neck cancer patient symptom management system driven by image recognition and retrieval-based generation enhancement. This effectively solves the problems of general-purpose large language models being prone to producing hallucinations and lacking targeted recommendations in medical scenarios.

[0031] 3. By integrating the symbolic reasoning capabilities of knowledge graphs with the semantic understanding and generation capabilities of large language models, intelligent, precise, and personalized management of complex symptoms in head and neck cancer patients has been achieved. This provides patients with timely and accurate intervention suggestions, significantly reduces their symptom burden, improves their quality of life and treatment adherence, and promotes the digital transformation of oncology nursing management.

[0032] The above embodiments are merely preferred embodiments of the present invention and do not constitute a limitation on the scope of protection of the present invention. Those skilled in the art should understand that various modifications, combinations, or substitutions can be made to the above embodiments without departing from the concept of the present invention. For example, reasonable adjustments can be made to video acquisition parameters, the number of keyframes, parameterized face model parameters, the number of Gaussian mixture model components, the multi-task learning network structure, the classification threshold, and the training data of the clinical decision-making model. All resulting technical solutions fall within the scope of protection claimed by the present invention.

Claims

1. A method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms, characterized in that, Includes the following steps: Step S1: Obtain symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal survey; Step S2: The symptom data is processed using exploratory factor analysis to extract several symptom clusters; Step S3: Based on the Apriori algorithm, perform association rule analysis on the symptom cluster data, and combine the first occurrence time of symptoms to determine the sentinel symptom classification within each symptom cluster; Step S4: Using the sentinel symptoms as intervention targets, construct an intervention plan for gastrointestinal symptoms.

2. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 1, characterized in that: The longitudinal survey described in step S1 uses convenience sampling to select research subjects, who are symptom data of head and neck tumor patients who receive radiotherapy and chemotherapy in medical institutions.

3. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 2, characterized in that: Data were collected from the study subjects based on the head and neck cancer-specific module of the Anderson Symptom Assessment Scale and the timeline of first symptom appearance, spanning 3-5 chemotherapy cycles.

4. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 1, characterized in that, Step S3 includes: Step S31: Obtain longitudinal survey data and the timeline of first symptom appearance for each symptom within the symptom cluster; Step S32: Use the Apriori algorithm to perform association rule analysis on each symptom within the symptom cluster, and generate a set of association rules between symptoms; Step S33: Based on the association rule set, determine the symptoms that meet the conditions as sentinel symptoms; Step S34: Output sentinel symptom assessment data.

5. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 4, characterized in that, The conditions in step S33 include: The symptoms mentioned are those that first appeared earliest within the symptom cluster; The symptom is the antecedent in at least one symptom association pair, and satisfies the threshold conditions that the antecedent support is >40%, the confidence is >60%, and the confidence is >the antecedent support in the association analysis of each symptom in the symptom cluster.

6. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 1, characterized in that, Step S4 also includes: A knowledge vector base for head and neck cancer health management is constructed. The retrieval uses the FAISS library to construct an approximate nearest neighbor vector index. Semantic matching and response generation are achieved through inverted clustering and group search.

7. The method for processing symptom cluster data of head and neck tumor patients based on sentinel symptoms according to claim 6, characterized in that: Based on the generated search results, intervention plans were developed using the MRC complex intervention framework as the theoretical framework.

8. A symptom cluster data processing system for head and neck tumor patients based on sentinel symptoms, characterized in that, The system includes: Data acquisition module: used to obtain symptom data of head and neck cancer patients during radiotherapy and chemotherapy through longitudinal surveys; Extraction module: used to process the symptom data using exploratory factor analysis to extract several symptom clusters; Data analysis module: used to perform association rule analysis on the symptom cluster data based on the Apriori algorithm, and determine the sentinel symptoms in each symptom cluster by combining the first occurrence time of the symptoms; Intervention program construction module: used to construct intervention programs for gastrointestinal symptoms using the sentinel symptoms as intervention targets and the MRC complex intervention framework as the theoretical framework; The intelligent symptom management module includes an image recognition unit and a retrieval-based generation enhancement unit, which are used to acquire and identify symptom-related image information of patients with head and neck tumors, and generate individualized symptom identification results and intervention recommendations based on the sentinel symptoms and symptom clusters.

9. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store a computer program, the processor being used to invoke and run the computer program stored in the memory to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images