CN-121999952-A - Intelligent portrait and accurate follow-up system for cancer pain patients based on dynamic knowledge graph

CN121999952ACN 121999952 ACN121999952 ACN 121999952ACN-121999952-A

Abstract

The invention discloses an intelligent portrait and accurate follow-up visit system of cancer pain patients based on dynamic knowledge patterns, which belongs to the technical field of intersection of medical informatization, artificial intelligence, knowledge patterns and accurate medical treatment, and comprises the following steps of firstly, constructing a double-knowledge-pattern framework comprising a medical knowledge pattern and a follow-up visit problem knowledge pattern, wherein the medical knowledge pattern covers entities and association relations of cancer pain related diseases, medicines, symptoms, treatment schemes and the like, and the follow-up visit problem knowledge pattern establishes a mapping relation between problems and medical entities; the method comprises the steps of accessing multi-source heterogeneous data such as HIS/LIS system data, wearable equipment monitoring data, follow-up feedback data and the like, completing multi-source data fusion after cleaning, standardization and desensitization treatment, and constructing an intelligent portrait of a patient with six dimensions including pathology, pain, treatment, psychology, social and dynamic characteristics based on the fusion data, storing the intelligent portrait in a JSON format, and supporting real-time dynamic updating.

Inventors

YANG HUILI
SHU RAN
ZHAO JIANWEI
LIU DONGYING
HAO JIANLEI
LU YANLING
JIANG JINGYUAN

Assignees

天津市肿瘤医院（天津医科大学肿瘤医院）

Dates

Publication Date: 20260508
Application Date: 20260124

Claims (10)

1. Dynamic knowledge graph-based intelligent portrait and accurate follow-up system for cancer pain patients is characterized by comprising the following steps: Step one, constructing a dual-knowledge-graph framework comprising a medical knowledge graph and a follow-up problem knowledge graph, wherein the medical knowledge graph covers entities and association relations of cancer pain related diseases, medicines, symptoms, treatment schemes and the like, and the follow-up problem knowledge graph establishes a mapping relation between a problem and a medical entity; accessing multi-source heterogeneous data such as HIS/LIS system data, wearable equipment monitoring data, follow-up feedback data and the like, and completing multi-source data fusion after cleaning, standardization and desensitization treatment; thirdly, constructing a patient intelligent portrait containing six dimensions of pathology, pain, treatment, psychology, society and dynamic characteristics based on the fusion data, and storing and supporting real-time dynamic update by adopting a JSON format; Step four, matching and reasoning by using the intelligent portraits and the double knowledge maps of the patient, and evaluating the pain risk level of the patient through a multi-factor risk scoring algorithm; Step five, dynamically generating an accurate follow-up strategy comprising follow-up frequency, follow-up channels and personalized questionnaire contents according to the risk level and portrait characteristics, wherein the personalized questionnaire is dynamically generated based on a follow-up problem knowledge graph; Step six, adopting an LSTM long-term memory network model, inputting time sequence data such as pain scores, medication time points, activity and the like in the past 7 days, predicting the occurrence probability of the future 24-72 hours of internal burst pain, and triggering three-level early warning of yellow, orange and red; Step seven, carrying out emotion analysis and risk identification by fusing text, voice and behavior multi-modal data, and realizing intelligent grading treatment of common, medium-grade and important three-grade responses; Step eight, automatically generating a multidisciplinary consultation suggestion list and pushing the multidisciplinary consultation suggestion list to a related department terminal when the conditions of the psychological score of a patient, such as a super threshold value or poor pain control, are monitored; Step nine, collecting follow-up feedback data and intervention effects, which are used for retraining a predictive model, and simultaneously carrying out incremental updating and optimization on the double-knowledge-graph to realize closed loop iteration; and step ten, in the home peace therapy scene, the remote nursing guidance and the moribund period care support are provided by combining the noninvasive continuous monitoring data of the wearable equipment.
2. The intelligent portrait and accurate follow-up system for cancer pain patients based on dynamic knowledge graph according to claim 1, wherein when constructing dynamic medical knowledge graph in step one, adopting BiLSTM-CRF model to identify 200+ disease entities, 500+ drug entities, 300+ symptom entities and 1000+ treatment plan entities from multi-source data such as clinical guideline, drug instruction and medical record, extracting complex relations such as treatment, causal, tabu and time sequence, and calculating fusion confidence through weighted fusion algorithm, the formula is fusion confidence = alpha-guide authority + beta-literature evidence intensity + gamma, real world data + delta-timeliness, wherein alpha = 0.4, beta = 0.3, gamma = 0.2, delta = 0.1.
3. The intelligent representation and accurate follow-up system for cancer pain patients based on dynamic knowledge graph according to claim 2, wherein the calculation formula of the multi-factor risk scoring algorithm in the fourth step is S=w1+mSrS+w2, deltaTrend+w3, freq_ BTcP +w4, score_Psy+w5, (100-company) +w6+ Progression, wherein NRS is pain Score, deltaTrend is pain Trend slope, freq_ BTcP is outbreak pain frequency, score_Psy is psychological Score, company is follow-up Compliance rate, progression is illness progress, and patients are classified into three classes of 0-30 minutes low risk, 31-60 minutes medium risk, 61-100 minutes high risk according to Score S.
4. The intelligent portrait and accurate follow-up system for cancer pain patients based on dynamic knowledge graph according to claim 3 is characterized in that the dynamic follow-up strategy in the fifth step is specifically implemented in such a way that 1 active follow-up is carried out every 2 weeks for low risk patients, life quality is focused, 1 active follow-up is carried out every week for stroke patients, drug adjustment effect is focused, 1 follow-up is carried out every 3 days for high risk patients or APP is punched and blocked every day, bursting pain and drug side effects are closely monitored, follow-up channels are selected based on decision trees, advanced or low school patients take precedence on artificial telephones, middle-aged or high compliance patients take precedence on APP push and WeChat reminding, and critical values such as NRS >8 trigger short message and telephone double alarm.
5. The intelligent portrayal and accurate follow-up system for cancer pain patients based on dynamic knowledge graph according to claim 4, wherein the multi-mode emotion analysis in the seventh step adopts PyTorch + Transformers framework, realizes the judgment of the emotion state of the patient through text emotion recognition, voice feature extraction and behavior pattern analysis, generates personalized response dialects by combining a co-emotion interaction strategy, and automatically screens out invalid follow-up problems based on a 90% consistency threshold.
6. The intelligent portrait and accurate follow-up visit system for cancer pain patients based on dynamic knowledge graph is characterized by comprising a data layer module, a data processing module and a data processing module, wherein the data layer module is used for accessing HIS/LIS/PACS system data, wearable equipment data and follow-up visit data, executing data cleaning, standardization and desensitization processing and meeting the requirements of medical data and other security requirements; The double-knowledge-graph construction module adopts a Neo4j database for storage, and constructs a medical knowledge graph and a follow-up problem knowledge graph through a self-grinding graph construction algorithm, so that real-time individual graph updating and batch general graph optimization are supported; the patient portrait construction module generates a digital portrait of the patient with six dimensions, stores the digital portrait in a PostgreSQL database, and associates historical data with portrait labels; the follow-up strategy generation module dynamically outputs follow-up frequency, channels and personalized questionnaire schemes based on a risk scoring algorithm, and follow-up records are stored in a MongoDB database; the prediction and early warning module adopts an LSTM neural network model to realize the prediction of the burst pain and is provided with a three-level early warning mechanism; the multi-mode AI analysis module is used for carrying out emotion calculation and risk classification judgment by fusing text, voice and behavior data; The system comprises a multidisciplinary collaboration module, a closed loop optimization module, a model iteration driving module, a double knowledge graph evolution module, a multi-disciplinary collaboration module, a simulation model analysis module and a simulation model analysis module, wherein the multidisciplinary collaboration module automatically starts an MDT flow when a triggering condition is met and pushes consultation advice sheets; And the interaction layer module is used for providing a patient end applet, a doctor end Web workbench, an individual case manager end management background and a nurse workbench multi-terminal interface.
7. The intelligent portrayal and accurate follow-up system for cancer pain patients based on dynamic knowledge graph as set forth in claim 6, wherein the medical knowledge graph in the dual knowledge graph construction module covers entity association relations such as cancer types, pain mechanisms, treatment schemes, drug actions, complications and the like, the follow-up problem knowledge graph builds a targeted follow-up problem base based on key nodes of the medical knowledge graph, builds a mapping relation between problems and medical entities, and integrates a disease development path prediction dimension to build a time sequence relation model.
8. The intelligent portrayal and accurate follow-up system for cancer pain patients based on dynamic knowledge graph of claim 7, wherein the LSTM model of the predictive early warning module comprises an input layer, a double-layer hidden layer and an output layer, the input features at least comprise pain scoring sequences of 7 days in the past, medication time points, activity and weather air pressure data, probability distribution of breakthrough pain within 24 to 72 hours in the future is output, and target predictive accuracy is >75%.
9. The intelligent portrayal and accurate follow-up system for cancer pain patients based on dynamic knowledge graph according to claim 8, wherein the multidisciplinary collaboration module supports cross-department collaboration of oncology, pain department, psychology department and the like, and when detecting that PHQ-9>15 of the patients or bone metastasis pain medicine is poorly controlled and the like, consultation application forms containing the abstract of the key portrayal of the patients are automatically generated and pushed to doctor workstations of relevant departments.
10. The intelligent portrait and accurate follow-up visit system for cancer pain patients based on dynamic knowledge graph as claimed in claim 9, wherein the small program at the patient end of the interaction layer module supports pain score reporting, medication recording, follow-up visit feedback and trouble teaching interaction functions, the nurse workstation has risk grading display, emergency treatment triggering and manual intervention recording functions, the management background can realize data analysis, system configuration and effect monitoring, and data at each end are synchronous in real time and support retrospective query.

Description

Intelligent portrait and accurate follow-up system for cancer pain patients based on dynamic knowledge graph Technical Field The invention relates to the crossing fields of medical informatization, artificial intelligence, knowledge graph and accurate medical treatment, in particular to an intelligent portrait and accurate follow-up system for cancer pain patients based on a dynamic knowledge graph. Background Cancer pain (CANCERPAIN, abbreviated cancer pain) is one of the most common and intolerable symptoms of cancer patients, severely affecting the quality of life, sleep and mood of the patient, and possibly even causing suicidal tendencies to the patient. The World Health Organization (WHO) has counted about 25% of pain occurring in cancer patients at first diagnosis, and up to 70% -80% of pain occurring in cancer patients at advanced stages, with about 1/3 of the patients suffering from severe pain. Although the WHO established "three-step analgesic principle" has been spreading for many years, the phenomenon of insufficient cancer pain control in actual clinical practice is still common, especially breakthrough pain (BreakthroughCancerPain, BTcP), which occurs at a rate of up to 40% -80%, is abrupt, has a short duration (usually less than 30 minutes), is strong, and is difficult to cope with in the traditional on-time mode of administration. The current cancer pain management and follow-up system has a plurality of core technical defects: the existing assessment system mainly depends on single dimension indexes such as a digital scoring method (NRS) and the like, only pays attention to the physical strength of pain, ignores the comprehensive influence of multidimensional factors such as pain property, psychological state, social support and the like on pain perception, and causes difficulty in realizing 'physical and psychological treatment'; the follow-up strategy is standardized (one-step) in that most follow-up systems adopt fixed time intervals and standardized questionnaire templates, and the same follow-up frequency is adopted for patients in the stable period and the severe fluctuation period of illness state, so that medical resources are wasted, the illness state of high-risk patients cannot be intervened in time, and a personalized dynamic adjustment mechanism is lacked; Knowledge fragmentation and island effect, namely clinical guidelines, drug specifications, literature research and expert experience related to cancer pain treatment are dispersed in different carriers, the existing electronic medical record system (EMR) is only used as a data recording tool, systematic integration and structural association of the fragmentation knowledge cannot be realized, and a doctor can not quickly obtain comprehensive basis when making a scheme; The prediction capability is lost, the existing system focuses on the history data record and review, lacks the prediction capability based on time sequence data, cannot effectively predict the occurrence rule of the burst pain, the pain deterioration risk trend and the occurrence probability of the drug side effect, and is in a 'passive fire fighting' state all the time; the follow-up mainly depends on manual telephone, the efficiency is low, the coverage is narrow, the patient lacks the effective communication channel outside the hospital, the medication compliance is only 30% -40%, the frequent self-decrement or stopping medication causes pain to repeatedly occur; multidisciplinary collaboration (MDT) is difficult, cancer pain management requires multidisciplinary collaboration of oncology, pain, psychology and the like, but the existing system lacks intelligent triggering and collaborative workflow across departments, data cannot be effectively shared, and diagnosis and treatment schemes are split; the pain point management of the outpatient service scene is outstanding, the pain removal outpatient service faces the problems that the mobility of the patient is high, the traditional follow-up coverage rate is only 9.55%, the patient loss rate is up to 65% due to the out-of-hospital blind area, and the like, and the difficulty of cancer pain management is further increased. From the technical application, the traditional electronic medical record system data mainly comprise unstructured texts and are difficult to deeply mine and analyze, the traditional mobile medical follow-up APP only provides simple questionnaire filling and suffering teaching functions, lacks powerful intelligent decision engine support, is mainly built on a static knowledge base (such as disease-symptom query) although the knowledge graph technology is applied to the medical field, lacks a dynamic knowledge graph aiming at the cancer pain field and comprising a time sequence evolution relation, and cannot simulate the dynamic process of disease progress and treatment reaction. These problems together lead to difficulty in achieving accurate, personalized and full cycle coverage of pain