WO2023202508A1 - 一种基于认知图谱的全科患者个性化诊疗方案推荐系统 - Google Patents

一种基于认知图谱的全科患者个性化诊疗方案推荐系统 Download PDF

Info

Publication number
WO2023202508A1
WO2023202508A1 PCT/CN2023/088590 CN2023088590W WO2023202508A1 WO 2023202508 A1 WO2023202508 A1 WO 2023202508A1 CN 2023088590 W CN2023088590 W CN 2023088590W WO 2023202508 A1 WO2023202508 A1 WO 2023202508A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
patient
treatment
data
cognitive
Prior art date
Application number
PCT/CN2023/088590
Other languages
English (en)
French (fr)
Inventor
李劲松
田雨
周天舒
刘强华
Original Assignee
浙江大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江大学 filed Critical 浙江大学
Publication of WO2023202508A1 publication Critical patent/WO2023202508A1/zh

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records

Definitions

  • the invention belongs to the field of medical and health information technology, and in particular relates to a recommendation system for personalized diagnosis and treatment plans for general patients based on cognitive maps.
  • general medicine is a comprehensive medical discipline that is oriented to communities and families and integrates clinical medicine, preventive medicine, rehabilitation medicine and humanities and social sciences. It was established in the 1990s and its scope covers various fields. age, gender, various organ systems, and various health issues.
  • the main service area of general medicine is primary health care, with families and communities as the background, mainly dealing with common problems, and a large number of them are health problems in the undifferentiated stage of the disease.
  • General undifferentiated diseases refer to medically unexplained physical symptoms or diseases that are not clearly attributed to a certain system in the early stages of the disease. In other words, patients who visit general practitioners often only show obvious symptoms and may not know the cause of their illness and the system to which the disease belongs.
  • building a personalized diagnosis and treatment plan recommendation system for patients with undifferentiated diseases in general practice can help general practitioners achieve more accurate diagnosis and treatment and early referral, optimize the diagnosis and treatment process, and may also improve the timeliness and efficiency of cancer diagnosis. Improve cancer prognosis.
  • the purpose of this invention is to address the shortcomings of the existing technology and provide a personalized diagnosis and treatment plan recommendation system for general practice patients based on cognitive maps.
  • This system has high scalability, interpretability and generalization, and can solve The following technical issues:
  • This technical solution aims at the problem of undifferentiated diseases in general medicine, and selects a number of common clinical symptoms. These symptoms can be radiated to different types of clinical diseases, and use symptoms as the starting point to provide personalized support for patients’ diagnosis and treatment plans;
  • This technical solution can independently provide clinical decision support for the patient's entire medical treatment process, and can also interact with doctors in real time. Doctors can add new diagnosis or treatment suggestions at any time during the reasoning process.
  • the system provided by the invention works together with doctors. Provide patients with more accurate decision support;
  • This technical solution adopts a combination of data-driven and knowledge-driven methods. It first uses a semi-automatic method to build a general knowledge map, then selects a queue of eligible patients based on common clinical symptoms, and finally uses the idea of cognitive maps to The general knowledge map and selected data sets build a patient personalized diagnosis and treatment plan recommendation system;
  • This technical solution simulates human reasoning ideas for solving problems, uses reasoning methods based on cognitive maps, does not limit the number of input variables, has good scalability, implements a reasoning model that combines domain knowledge, and has high It has good generalization performance and can visually display the reasoning process, which has strong interpretability.
  • This technical solution refers to the human cognitive process and uses reasoning methods based on cognitive maps to achieve the leap from perceptual intelligence to cognitive intelligence, so that the system can truly simulate the diagnosis and treatment ideas of clinicians and provide interpretable information for clinicians. , a highly acceptable clinical decision-making aid tool;
  • the system provided by this technical solution can be used independently in clinical scenarios to provide diagnosis and treatment support for doctors and patients. It also supports real-time interaction between doctors and the system, and timely supplements the system with more comprehensive patient information, thereby providing patients with more accurate information. Diagnosis and treatment recommendations;
  • Figure 1 is a structural diagram of a personalized diagnosis and treatment plan recommendation system for general practice patients based on cognitive maps provided by an embodiment of the present invention
  • Figure 2 is a schematic diagram of constructing a general knowledge graph provided by an embodiment of the present invention.
  • Figure 3 is a schematic diagram of constructing a patient's personalized cognitive map according to an embodiment of the present invention.
  • Figure 4 is a schematic diagram of recommended diagnosis and treatment paths and results provided by the embodiment of the present invention.
  • Embodiments of the present invention provide a system for recommending personalized diagnosis and treatment plans for general practice patients based on cognitive maps.
  • the system includes a data collection module, a data preprocessing module, a data analysis reasoning module and a recommendation result display module. specifically:
  • Data collection module used to extract clinical diagnosis and treatment data of general outpatients from the medical institution database.
  • Data preprocessing module performs preprocessing operations on the data obtained by the data collection module, mainly including structured processing of outpatient electronic medical record text data, data regularization and other operations.
  • Data analysis and reasoning module This part is the core of the system. It first builds a general practice knowledge graph, and then uses the designed variant graph neural network based on the patient's disease, symptoms, medication and other information at the time of treatment and the constructed general practice knowledge graph. and the attention capture module to establish a cognitive map of the patient's personalized disease development trajectory, and then recommend the patient's personalized diagnosis and treatment plan.
  • Recommended result display module It mainly displays the reasoning process of the data analysis and reasoning module visually, providing a clear patient's personalized disease development and treatment process.
  • 22 common clinical symptoms are selected, namely nausea, vomiting, chest pain, low back pain, abdominal pain, blood in the stool, difficulty swallowing, gastrointestinal bleeding, abdominal distension, fecal incontinence, constipation, diarrhea, urinary incontinence, nocturia, and weight loss. , weight loss, frequent urination, urgency, heartburn, melena, dysuria, and gross hematuria. SQL commands were used to extract the symptoms caused by the above symptoms from the medical institution database. All clinical diagnosis and treatment data of patients who visit the outpatient clinic of a medical institution, including the patient's outpatient electronic medical record, personal statistical data, diagnosis data, medication data, operation data, examination result data, etc.
  • the patient's chief complaint, past medical history, current medical history, family history and other information are stored in the outpatient electronic medical record in text form.
  • a rule-based and domain dictionary method is used to extract the patient's symptoms, illness records, medication records, etc. from the text data. information to achieve the purpose of structuring text data.
  • a piece of available data is defined as the collection of all clinical diagnosis and treatment data within a calendar day starting from a patient's visit time. Since the main purpose of this invention is to use the patient's initial disease and symptoms as a starting point to reason about the patient's disease trend and give treatment suggestions based on the general knowledge map, it removes the regular visits to medical institutions to prescribe medicine, The patient data of regular physical examinations and regular re-examinations. In addition, the content of the subsequent general knowledge map constructed is mainly Western medicine, so the data of patients visiting traditional Chinese medicine outpatient clinics must also be removed.
  • the data analysis reasoning module includes a general knowledge graph building module and a cognitive graph-based reasoning system. Specifically:
  • the technical solution proposed by the present invention is based on the domain knowledge graph, so before data training, a general knowledge graph needs to be constructed first.
  • the construction process of the general knowledge graph adopts a combination of manual construction and automatic construction, as shown in Figure 2.
  • SemMed DB covers rich English triple relationships extracted from the medical literature database PubMed, which needs to be further translated into Chinese form by calling the translation software API.
  • Chinese ICD-10 covers the classification relationships and codes of all diseases. Extracting relevant disease classifications can fully supplement the hierarchical structure of the constructed general knowledge map.
  • Cognitive mapping is a means of realizing cognitive intelligence. Cognitive mapping technology simulates human problem-solving ideas and refers to brain recognition.
  • the cognitive map-based reasoning system proposed by the present invention includes two subsystems. Subsystem 1 is a process of exploring disease development trajectories based on the patient's initial clinical diagnosis and treatment data on the constructed general knowledge map, while subsystem 2 is based on Subsystem 1 explores the patient's multiple potential disease development trajectories to accurately locate the patient's final state.
  • the constructed general knowledge graph is denoted as G, which contains a series of triples
  • Each triplet represents a piece of medical knowledge.
  • Each triplet consists of a relationship r ⁇ R and two entities h, t ⁇ , where the relationship r points from the head entity h to the tail entity t, and ⁇ is the general knowledge graph G.
  • the set of all entities, each entity in the triplet serves as a node in the general knowledge graph G, and R is the set of all relationships in the general knowledge graph G.
  • Entity types include: symptoms, diseases, examination methods, treatment methods, medications, examination results, etc.
  • These triples include relationships between entity types such as symptoms and diseases, diseases and diseases, diseases and examination methods, diseases and treatment methods, etc.
  • n is the number of initial patient data
  • a i is the i-th initial patient data.
  • Patient personalized cognitive mapping Initially, it consists of only n nodes formed by n pieces of patient initial data in the patient's initial clinical diagnosis and treatment data set S.
  • Subsystem 1 first performs global vectorization on the general knowledge graph to achieve semantic interoperability of global nodes in the graph, and then based on the cognitive graph And the global vectorized general knowledge graph, iteratively captures the potential cognitive graph reasoning node set H N .
  • the nodes in the cognitive graph reasoning node set H N may be the patient's disease, the examination that should be done, or the medicine that should be taken. wait;
  • Subsystem 2 gradually supplements the nodes in the cognitive map reasoning node set H N to the cognitive map , and then define the node status update method of the cognitive map, and finally obtain the patient's disease D and treatment method P based on general knowledge map and cognitive map reasoning.
  • subsystem 1 is specifically as follows: a variant network mainly based on Graph Neural Networks (GNN), that is, a variant graph neural network to realize global vectorization of the general knowledge graph, and then imitate traditional attention.
  • GNN Graph Neural Networks
  • the scientific knowledge graph iteratively captures the reasoning nodes required in the reasoning process of the cognitive graph.
  • the global vectorization process of the general knowledge graph is introduced.
  • a learnable embedding vector representation e h is defined, and for each relationship r, a learnable embedding vector representation e r is also defined.
  • m ⁇ M iteration times
  • the node state vector is expressed as and
  • Each triple (h, r, t) represents a piece of medical knowledge and can be vectorized as
  • the vectorization of medical knowledge (h, r, t) related to node h is expressed as follows:
  • the vector representation of the medical knowledge is related to the vector representation of the head and tail entities and relationships in the medical knowledge
  • the status update of node h during the iterative process is related to its own vector representation and the vector representation of all triples where it is located, that is, the neighborhood.
  • the medical information contained in all triples where node h is located is Knowledge is aggregated to obtain a triplet aggregation vector. It can be expressed as:
  • the next state vector of the final node h The update method is:
  • this also expresses the process of transferring information from neighborhood medical knowledge to the current node;
  • the above f a and f b respectively represent two hidden layer functions.
  • the input is the concatenation of all vectors, and the output is consistent with the input single vector dimension.
  • the residual connection method is used to update the state vector of node h to improve the possible loss of information as m increases, and finally select the output result of the last step.
  • the global vector representation of node h is input to the next step.
  • the following introduces the method of capturing the reasoning nodes required in the cognitive graph reasoning process from the global vectorized general knowledge graph.
  • This embodiment proposes an attention capture module to extract the potential disease development trajectory of the current patient from the general knowledge map.
  • This process can control the expansion speed of the cognitive map.
  • the implementation process of the attention capture module is as follows: sample the k nodes with the largest attention values from ai as the inference node set of the current iteration step. where a i is the attention value set of all nodes in the neighborhood node set E i corresponding to ⁇ i . Change the labels of the unvisited no-visited nodes ⁇ i in this traversal to visited, and change does not exist in nodes included in in, generate At the same time, set the labels of these newly included nodes to no-visited.
  • subsystem 2 is as follows: using a variant network of graph neural network GNN to define the node status update method of the cognitive map, and finally obtaining the patient's disease D based on general knowledge map and cognitive map reasoning. and treatment modality P.
  • the cognitive graph reasoning node set H N finally obtained in subsystem 1 and all the relationships of these reasoning nodes in the general knowledge graph G are incorporated into the cognitive graph.
  • the medical knowledge contained in all triples where node v is located is aggregated to obtain a triplet aggregation vector. It can be expressed as:
  • the state vector update method of node v at step t+1 is as follows:
  • step 0 state vector of node v The same as the step 0 state vector of node v
  • the t-th step state vector of node v Aggregation vector of all triples composed of node v and its neighbor nodes in the cognitive map It is also related to the global vector representation of the nodes in the patient's initial clinical diagnosis and treatment data set S.
  • the above f c and f d are also two hidden layer functions, and their parameters can be learned during the training process.
  • the node status update also adopts the form of residual connection.
  • the state vector of each node replaces its global vector representation in the general subject knowledge graph G, so that all nodes of the entire general subject knowledge graph G can be represented by a matrix X.
  • the dimension of the matrix X is J ⁇ Q, where J is The number of nodes in the general knowledge graph, Q is the vector dimension of each node.
  • a two-layer fully connected network is used as the final prediction layer F:
  • prediction is the patient's most likely disease D or the recommended treatment method P given after reasoning
  • the real disease or real treatment method that the patient suffers from is the target.
  • the sigmoid function is used as the activation function
  • the cross-entropy loss function is used to calculate
  • the error between the predicted value prediction and the true value target is finally determined by using the stochastic gradient descent (SGD) algorithm to propagate the gradient of the cross-entropy loss function to complete the optimization of the cognitive map-based reasoning system.
  • SGD stochastic gradient descent
  • the recommendation result explicit module mainly uses the web front-end to combine patients with Existing symptoms, diseases, and possible treatments that may be needed after reasoning are visually displayed.
  • the webpage allows clinicians to input the clinical diagnosis and treatment information of patients during treatment. After clicking on the analysis, the patient's diagnosis and treatment path and plan can be visually displayed.
  • Figure 4 shows the recommended diagnosis and treatment path and results of one embodiment.
  • the dotted box represents the clinical diagnosis and treatment information of the patient.
  • the patient has a history of hypertension, peptic ulcer, gastritis, etc., and has been taking gastric medicine.
  • Domperidone recently developed symptoms of decreased appetite, and came to see a doctor. Later, the system showed the tests that should be done, and gradually ruled out diseases such as irritable bowel syndrome, colorectal cancer, and benign gastric tumors. Finally, the patient was diagnosed with early gastric cancer. , diagnosis and treatment recommendations for gastroscopy and surgical resection are required.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Pathology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

本申请公开了一种基于认知图谱的全科患者个性化诊疗方案推荐系统，系统包括数据采集模块、数据预处理模块、数据分析推理模块和推荐结果显示模块，在数据分析推理模块中首先构建全科知识图谱，然后基于患者就诊时的疾病、症状、用药等信息和已构建的全科知识图谱，建立患者个性化的疾病发展轨迹的认知图谱，进而给出患者个性化诊疗方案推荐。本申请使用基于认知图谱的推理方法，使系统能够真正模拟临床医生的诊疗思路，为临床医生提供可解释的、可接受度高的临床辅助决策工具；本申请从症状出发，为患者制定个性化诊疗方案，帮助患者及早发现病因并接受针对性治疗，同时也能够实现危险疾病的早筛，提示患者及时转诊至专科治疗。

Description

一种基于认知图谱的全科患者个性化诊疗方案推荐系统 技术领域
本发明属于医疗健康信息技术领域，尤其涉及一种基于认知图谱的全科患者个性化诊疗方案推荐系统。
背景技术
在我国，全科医学是一个面向社区与家庭，整合临床医学、预防医学、康复医学以及人文社会学科相关内容于一体的综合性医学专业学科，成立于上世纪九十年代，其范围涵盖了各种年龄、性别、各个器官系统以及各类健康问题。全科医学主要服务领域为基层卫生保健，以家庭、社区为背景，处理常见问题为主，并且大量是处于疾病未分化阶段的健康问题。全科未分化疾病是指医学上无法解释的躯体症状或指疾病早期尚未明确归属于某一系统的疾病。也就是说，就诊全科的患者往往只表现出外显的症状，可能并不知道自己患病的原因以及疾病所属的系统。另外，很多疾病往往在早期发展阶段就表现出一些特异性及非特异性症状，比如腹痛几乎会影响所有的腹部癌症，包括结肠癌、前列腺癌、膀胱癌、肾癌，而2020年Lancet Oncology上的一篇文章也指出很多病人的某些症状在癌症的I-III期都已经表现出来。综上，科学地处理未分化疾病，使疾病在早期能够得到诊治，是全科医生应该学会和掌握的技能。然而，虽然我国的全科医生培养体系已经初步形成，培养模式也已基本确立，队伍人数也在不断增加，但是总体上全科医生数量仍然不足、质量有待提高。针对上述情况，构建一个面对全科未分化疾病的患者个性化诊疗方案推荐系统可以帮助全科医生实现更精准的诊疗和及早的转诊，优化诊疗流程，同时可能提高癌症诊断的及时性和改善癌症预后。
目前以从症状出发的、针对全科患者的个性化诊疗方案推荐系统或者辅助诊断系统很少，该领域还处于发展阶段，现有类似技术方案存在的缺点有：
1.现有技术方案很少有以症状为出发点对患者可能患有的不同种类或者不同系统的疾病进行分析的，大多数集中在已经确定的一种或者一类疾病，比如糖尿病、自闭症、传染病等，但在全科门诊中，医生面临的往往是未分化疾病，表现出相同症状的患者最后可能需要转诊到不同的科室进行后续治疗，因此当前的技术方案无法很好地解决目前全科领域遇到的问题；
2.现有技术方案主要用在患者和医生面诊之前的导诊或者预问诊过程以及患者在家中的自我监测，未嵌入到与医生的实时交流活动中，未能真正地应用到临床场景中；
3.现有的技术方案基本采用数据驱动的方法，抽取大量符合条件的患者队列，然后基于患者确诊前的临床数据，使用传统的机器学习或者深度学习的方法进行分析，辅助医生进行临床决策支持，整个过程没有融入临床指南、专家共识等，缺乏知识驱动；
4.现有技术方案往往是基于某一人群的大数据进行模型训练和数据分析，更换人群队列后模型效果会下降很多，即模型的泛化性能差，同时深度学习存在“黑箱”问题，模型不具备可解释性或者可解释性较差。
发明内容
本发明的目的在于针对现有技术的不足，提供一种基于认知图谱的全科患者个性化诊疗方案推荐系统，该系统具备较高的可扩展性、可解释性和泛化性，可解决如下技术问题：
1.本技术方案针对全科未分化疾病问题，选取若干常见临床症状，这些症状能够辐射到临床的不同种类疾病中，以症状为出发点，为患者的诊疗方案提供个性化支持；
2.本技术方案可以独立地为患者的整个就诊过程提供临床决策支持，也可以与医生进行实时互动，医生可以随时在推理过程中加入新的诊断或者治疗建议，本发明提供的系统与医生一同为患者提供更加精准的决策支持；
3.本技术方案采用数据驱动和知识驱动相结合的方式，首先使用半自动的方法构建全科知识图谱，然后根据常见的临床症状，选取符合条件的患者队列，最后基于认知图谱的思想，利用全科知识图谱和选取的数据集构建患者个性化诊疗方案推荐系统；
4.本技术方案模拟人类解决问题的推理思路，使用基于认知图谱的推理方法，不限制输入变量的数目，具备很好的可扩展性，实现结合领域知识进行推理的模式，具备较高的泛化性能，同时可以将推理过程进行可视化展示，具备很强的可解释性。
本发明的目的是通过以下技术方案实现的：
本发明的有益效果是：
1.本技术方案参考人类的认知过程，使用基于认知图谱的推理方法，实现从感知智能向认知智能的跨越，使系统能够真正模拟临床医生的诊疗思路，为临床医生提供可解释的、可接受度高的临床辅助决策工具；
2.本技术方案提供的系统可以在临床场景中独立使用，为医生和患者提供诊疗支持，也支持医生与系统进行实时互动，及时为系统补充更加全面的患者信息，从而为患者提供更加精准的诊疗建议；
3.针对全科未分化疾病问题，本技术方案从症状出发，为患者制定个性化的诊疗方案，帮 助患者及早发现病因并接受针对性治疗，同时也能够实现危险疾病的早筛，提示患者及时转诊至专科治疗。
附图说明
图1为本发明实施例提供的基于认知图谱的全科患者个性化诊疗方案推荐系统结构图；
图2为本发明实施例提供的构建全科知识图谱示意图；
图3为本发明实施例提供的构建患者个性化认知图谱示意图；
图4为本发明实施例提供的推荐诊疗路径及结果示意图。
具体实施方式
为使本发明的上述目的、特征和优点能够更加明显易懂，下面结合附图对本发明的具体实施方式做详细的说明。
在下面的描述中阐述了很多具体细节以便于充分理解本发明，但是本发明还可以采用其它不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本发明内涵的情况下做类似推广，因此本发明不受下面公开的具体实施例的限制。
本发明实施例提供一种基于认知图谱的全科患者个性化诊疗方案推荐系统，如图1所示，该系统包括数据采集模块、数据预处理模块、数据分析推理模块和推荐结果显示模块，具体地：
数据采集模块：用于从医疗机构数据库中抽取全科门诊患者就诊时的临床诊疗数据。
数据预处理模块：对数据采集模块获取的数据进行预处理操作，主要包括对门诊电子病历文本数据结构化处理、数据规整等操作。
数据分析推理模块：这一部分是该系统的核心，首先构建全科知识图谱，然后基于患者就诊时的疾病、症状、用药等信息和已构建的全科知识图谱，利用设计的变体图神经网络和注意力捕捉模块建立患者个性化的疾病发展轨迹的认知图谱，进而给出患者的个性化诊疗方案推荐。
推荐结果显示模块：主要对数据分析推理模块的推理过程进行可视化展示，提供清晰的患者个性化疾病发展和治疗过程。
下述说明进一步给出了符合本申请要求的基于认知图谱的全科患者个性化诊疗方案推荐系统各模块实现的部分实施例。
一、数据采集模块
本实施例中选取临床上常见的22种症状，分别是恶心、呕吐、胸痛、腰痛、腹痛、便血、吞咽困难、消化道出血、腹胀、大便失禁、便秘、腹泻、尿失禁、夜尿、消瘦、体重减轻、尿频、尿急、烧心、黑便、排尿困难和肉眼血尿，采用SQL命令从医疗机构数据库中抽取因上述症状 到医疗机构门诊就诊的患者全部临床诊疗数据，包括患者的门诊电子病历、个人统计学数据、诊断数据、用药数据、操作数据、检查结果数据等。
二、数据预处理模块
患者就诊的主诉、既往病史、现病史、家族史等信息均以文本形式存储于门诊电子病历中，采用基于规则和领域词典的方法从文本数据中提取患者的症状、患病记录、用药记录等信息，从而达到文本数据结构化的目的。
针对含有数值的检查结果数据类型，比如血常规、尿常规所涵盖的各种检查指标，只保留指标结果为阳性的数据。由于本技术方案采用认知图谱的方法，每条数据的输入变量数目无需固定，所以不涉及缺失数据插补操作。
一条可利用的数据定义为以患者某次就诊时间为起点，一个自然日内的所有临床诊疗数据的集合。由于本发明的主要目的是基于全科知识图谱，以患者的初始疾病、症状为起点，推理患者的疾病走向并给出治疗建议，因此去除数据采集模块中抽取到的定期到医疗机构开药、定期体检以及定期复查的患者数据，另外后续构建的全科知识图谱内容以西医为主，因此也要去除就诊中医门诊的患者数据。
三、数据分析推理模块
数据分析推理模块包括全科知识图谱构建模块和基于认知图谱的推理系统，具体地：
3.1全科知识图谱构建模块
本发明提出的技术方案以领域知识图谱为基础，因此数据训练前首先需要构建全科知识图谱。全科知识图谱的构建过程采用手动构建和自动构建相结合的方式，如图2所示。
(a)本实施例中，临时聘请10名临床医生，完成与数据采集模块抽取的症状、疾病相关知识的全科知识图谱的手动构建部分。具体来说，医生根据数据采集模块抽取的症状和疾病，收集相关临床指南、专家共识、专业教材以及uptodate网站上的内容等，并对所收集文本内容中的临床实体和关系进行手动提取。为了使提取结构形式标准化，需要手动设计一份文本提取模板。
(b)知识图谱自动构建部分涵盖从SemMed DB以及中文ICD-10中提取到的相关临床术语及关系。SemMed DB涵盖了从医学文献数据库PubMed中提取到的丰富的英文三元组关系，需要进一步调用翻译软件API将其翻译为中文形式。中文ICD-10中涵盖了全部疾病的分类关系和代码，提取相关疾病分类可以对构建的全科知识图谱的层次结构进行充分的补充。
3.2基于认知图谱的推理系统
认知图谱是实现认知智能的一种手段，认知图谱技术模拟人类解决问题的思路，参考了脑认 知科学领域提出的双系统理论。该双系统理论认为人的认知过程包括两个系统：系统1是基于直觉的、无知觉的思考系统，是感知的过程，系统2是显式的，有逻辑的，需要意识控制，是人类高级智能的体现。本发明提出的基于认知图谱的推理系统，包括两个子系统，子系统1是基于患者初始临床诊疗数据在已构建的全科知识图谱上进行疾病发展轨迹探索的过程，而子系统2是基于子系统1探索的患者多条潜在疾病发展轨迹进行患者最终状态精准定位的过程。
如图3所示，已构建的全科知识图谱记作G，其包含一系列的三元组每个三元组代表一条医学知识，每个三元组由一个关系r∈R和两个实体h,t∈ε组成，其中关系r由头实体h指向尾实体t，ε为全科知识图谱G的所有实体集合，三元组中的每个实体作为全科知识图谱G中的一个节点，R为全科知识图谱G的所有关系集合。实体类型包括：症状、疾病、检查手段、治疗方式、用药、检查结果等。这些三元组包含了症状与疾病、疾病与疾病、疾病与检查手段、疾病与治疗方式等实体类型之间的关系。本技术方案的任务是基于患者初始临床诊疗数据集S，推理得到患者的疾病D和治疗方式P，并给出中间的推理过程，其中初始临床诊疗数据集S＝{a1,a2,…,ai,…,an}，该初始临床诊疗数据集按照患者初始就诊时的每条症状、病史、用药数据产生的时间顺序进行排序，n为患者初始数据条数，ai为第i条患者初始数据。
患者个性化认知图谱最初仅由患者初始临床诊疗数据集S中的n条患者初始数据形成的n个节点组成。
子系统1首先对全科知识图谱进行全局向量化，实现图谱全局节点的语义互通，然后基于认知图谱和全局向量化后的全科知识图谱，迭代捕捉潜在的认知图谱推理节点集HN，认知图谱推理节点集HN中的节点可能是患者的疾病、应做的检查或者应服用的药物等；
子系统2将认知图谱推理节点集HN中的节点逐步补充到认知图谱中，然后定义认知图谱的节点状态更新方式，最终基于全科知识图谱和认知图谱推理得到患者的疾病D和治疗方式P。
在一个实施例中，子系统1的实现具体为：主要基于图神经网络(Graph Neural Networks，GNN)的变体网络即变体图神经网络实现全科知识图谱的全局向量化，然后模仿传统注意力思想，设计一个患者潜在疾病发展轨迹的注意力捕捉模块，该模块基于初始临床诊疗数据集S＝{a1,a2,…,ai,…,an}从全局向量化后的全科知识图谱中迭代捕捉认知图谱推理过程中所需的推理节点。
首先介绍全科知识图谱的全局向量化过程。针对全科知识图谱的每个节点h，定义一个可学习的嵌入向量表示eh，针对每个关系r，也定义一个可学习的嵌入向量表示er。全科知识图谱的全局向量化过程需要进行M步迭代(M人为设定，本实施例中设定M＝3)，第m(m≤M)步迭 代时，节点状态向量表示为而每一个三元组(h,r,t)代表一条医学知识，在第m步迭代时可被向量化为在全科知识图谱的全局向量化过程中，与节点h相关的医学知识(h,r,t)的向量化表示如下：
其中分别为第m步迭代时节点h,t的节点状态向量；该医学知识的向量化表示与该医学知识中的头、尾实体和关系的向量表示有关；
节点h在迭代过程中的状态更新与其本身的向量表示和其所在的全部三元组即邻域的向量表示均有关系，在全科知识图谱中对节点h所在的全部三元组包含的医学知识进行聚合，得到三元组聚合向量可表示为：
其中为全科知识图谱中节点h所在的全部三元组的数量；通过将相关三元组的向量信息进行聚合再除以邻域数目均方根来计算；
最终节点h的下一步状态向量的更新方式为：
与节点初始状态向量eh、当前迭代步数的状态向量以及当前节点和其邻域节点组成的三元组聚合向量有关，这也表达了邻域医学知识向当前节点传递信息的过程；
上述的fa、fb分别表示两个隐藏层函数，输入为全部向量的拼接，输出与输入的单个向量维度一致。另外采用残差连接的方式进行节点h的状态向量更新，以改善随着m的增加可能出现信息丢失的情况，最终选择最后一步的输出结果作为节点h的全局向量表示输入到接下来的步骤中。
下面介绍从全局向量化后的全科知识图谱中捕捉认知图谱推理过程中所需推理节点的方法。
认知图谱最初由患者初始临床诊疗数据集S＝{a1,a2,…,ai,…,an}构成，定义一个节点访问集合Vvisit，初始节点访问集合并将中的每个节点都标记为未访问no-visited，设置最大迭代步数为L(L人为设定，本实施例中设定L＝6)。当迭代步数为l(l≤L)时，遍历当前迭代步数的节点访问集合中的每个未访问no-visited节点从全科知识图谱中查询νi的一步(one-step)邻域节点集Ei。虽然我们只关注νi的一步邻域节点，但随着迭代步数的增加，所遍历节点的邻域节点集会逐渐膨胀，为了减少计算量，需要控制遍历节点的数量。
本实施例提出一个注意力捕捉模块，用于从全科知识图谱中抽取当前患者潜在疾病发展轨迹， 该过程可以控制认知图谱的扩展速度，该注意力捕捉模块的实施过程如下：从ai中采样注意力值最大的k个节点作为当前迭代步数的推理节点集其中ai为νi对应的邻域节点集Ei中的所有节点的注意力值集合，将此次遍历的未访问no-visited节点νi的标签均改为访问visited，并将中的未存在于的节点纳入中，生成同时将这些新纳入的节点的标签设置为未访问no-visited。另外，当l＝1时，首先需要从全科知识图谱中查询中全部节点之间的所有关系并补充到认知图谱中，再进行上述遍历节点访问集合中未访问节点的操作，更新节点访问集合和认知图谱中的推理节点。当迭代步数到达L或者认知图谱推理节点集中的推理节点不再增加时，迭代过程结束，将得到的认知图谱推理节点集记作HN。
在一个实施例中，子系统2的实现具体为：利用图神经网络GNN的变体网络定义认知图谱的节点状态更新方式，并且最终基于全科知识图谱和认知图谱推理得到患者的疾病D和治疗方式P。
首先将子系统1中最终得到的认知图谱推理节点集HN以及这些推理节点在全科知识图谱G中的全部关系纳入到认知图谱中，认知图谱全部节点集合记作V，其中V＝S+HN，S＝{a1,a2,…,ai,…,an}，然后定义认知图谱节点状态更新所需迭代步数T(T人为设定，本实施例中设定T＝6)。认知图谱中的节点v(v∈V)的初始向量表示认知图谱中的节点v在第t+1(t＜T)步的节点状态的更新与其在第0步和第t步的节点状态以及其在认知图谱中所在的三元组的向量表示有关。迭代步数为t时，三元组(v,r,v′)蕴含的医学知识可被向量化为其计算方式表示如下：
其中代表了患者初始临床诊疗数据集S中节点的全局向量表示，由子系统1给出，由于S中的节点所代表的症状、疾病或者用药等信息是患者真实产生的临床诊疗数据，是推理的全部起点，所以将其纳入每一步更新的计算中；
在认知图谱中对节点v所在的全部三元组包含的医学知识进行聚合，得到三元组聚合向量可表示为：
其中为认知图谱中节点v所在的全部三元组的数量；与子系统1类似，通过将相关三元组的向量信息进行聚合再除以邻域数目均方根来计算；
节点v的t+1步的状态向量更新方式如下：
同样与节点v的第0步状态向量节点v的第t步状态向量节点v与其在认知图谱的邻域节点组成的所有三元组聚合向量以及患者初始临床诊疗数据集S中节点的全局向量表示有关。
上述fc、fd同样为两个隐藏层函数，其参数可在训练过程中学习得到，节点状态更新时同样采用了残差连接的形式。
由于认知图谱的所有节点都来自于全科知识图谱G，使用此时更新完成后的认知图谱的每一个节点的状态向量取代其在全科知识图谱G中的全局向量表示，这样整个全科知识图谱G的全部节点可以用一个矩阵X表示，矩阵X的维度为J×Q，其中J为全科知识图谱中的节点数目，Q为每个节点的向量维度。采用一个两层的全连接网络作为最终的预测层F：
其中，prediction为经过推理后给出的患者最可能的疾病D或者建议的治疗方式P，而患者所患的真实疾病或者真实治疗方式为target，使用sigmoid函数作为激活函数，使用交叉熵损失函数计算预测值prediction和真实值target之间的误差，最后使用随机梯度下降(stochastic gradient descent,SGD)算法传播交叉熵损失函数梯度以完成对基于认知图谱的推理系统的优化。
四、推荐结果显式模块
基于认知图谱的全科患者个性化诊疗方案推荐系统的最大优点是具备可解释性，能够模拟临床医生的诊疗思路，给出清晰的推理过程，因此推荐结果显式模块主要利用网页前端将患者已有的症状、疾病以及经过推理后可能需要的治疗方式和可能患有的疾病进行可视化展示。网页支持临床医生输入患者就诊时的临床诊疗信息，点击分析后，可对患者的诊疗路径和方案进行可视化展示。如图4所示为一实施例的推荐诊疗路径及结果，图4中，虚线框中代表患者就诊时的临床诊疗信息，该患者有高血压、消化性溃疡、胃炎等病史，一直服用胃药多潘立酮，最近出现食欲下降的症状，遂来就医，后面展示了系统提示应做的检查，并逐步排除了肠易激综合征、结直肠癌和胃良性肿瘤等疾病，最终给出患者为早期胃癌，需要做胃镜手术切除的诊断和治疗建议。
以上所述仅是本发明的优选实施方式，虽然本发明已以较佳实施例披露如上，然而并非用以限定本发明。任何熟悉本领域的技术人员，在不脱离本发明技术方案范围情况下，都可利用上述揭示的方法和技术内容对本发明技术方案做出许多可能的变动和修饰，或修改为等同变化的等效实施例。因此，凡是未脱离本发明技术方案的内容，依据本发明的技术实质对以上实施例所做的任何的简单修改、等同变化及修饰，均仍属于本发明技术方案保护的范围内。

Claims (7)

  1. 一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，包括：
    数据采集模块：用于从医疗机构数据库中抽取全科门诊患者就诊时的临床诊疗数据；
    数据预处理模块：对数据采集模块获取的数据进行数据预处理操作，包括对门诊电子病历文本数据结构化处理及数据规整；
    数据分析推理模块：构建全科知识图谱，基于患者就诊时的疾病、症状、用药信息和已构建的全科知识图谱，利用变体图神经网络和注意力捕捉模块建立患者个性化的疾病发展轨迹的认知图谱，给出患者个性化诊疗方案推荐；
    所述数据分析推理模块包括全科知识图谱构建模块和基于认知图谱的推理系统；所述基于认知图谱的推理系统包括两个子系统：子系统1用于基于患者初始临床诊疗数据集在已构建的全科知识图谱上进行疾病发展轨迹探索；子系统2用于基于子系统1探索的患者多条潜在疾病发展轨迹进行患者最终状态精准定位；
    所述子系统1包括：对全科知识图谱进行全局向量化；患者个性化的认知图谱最初由患者初始临床诊疗数据集中的若干患者初始数据形成的节点构成，基于认知图谱和全局向量化后的全科知识图谱，迭代捕捉潜在的认知图谱推理节点集；具体为：
    定义节点访问集合Vvisit，初始节点访问集合S为患者初始临床诊疗数据集，将中的每个节点均标记为未访问，设置最大迭代步数L，当迭代步数为l，l≤L时，遍历当前迭代步数的节点访问集合中的每个未访问节点vi，从全科知识图谱中查询vi的一步邻域节点集Ei；
    通过注意力捕捉模块从全科知识图谱中抽取当前患者潜在疾病发展轨迹，包括：从vi对应的邻域节点集Ei中的所有节点的注意力值集合中采样注意力值最大的k个节点作为当前迭代步数的推理节点集将此次遍历的未访问节点vi的标签修改为访问，并将中的未存在于的节点纳入中，生成同时将新纳入的节点的标签设置为未访问；
    当l＝1时，从全科知识图谱中查询中全部节点之间的所有关系并补充到认知图谱中，再进行遍历节点访问集合中未访问节点的操作，更新节点访问集合和认知图谱中的推理节点；
    当迭代步数到达L或者认知图谱推理节点集中的推理节点不再增加时，迭代过程结束，得 到认知图谱推理节点集；
    所述子系统2包括：将认知图谱推理节点集中的节点逐步补充到认知图谱中，定义认知图谱的节点状态更新方式，基于全科知识图谱和认知图谱推理得到患者的疾病和治疗方式；
    推荐结果显示模块：对数据分析推理模块的推理过程进行可视化展示，提供患者个性化疾病发展和治疗过程。
  2. 根据权利要求1所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述数据预处理模块中，采用基于规则和领域词典的方法从门诊电子病历文本数据中提取患者的症状、患病记录及用药记录，实现文本数据结构化；
    针对含有数值的检查结果数据类型，仅保留指标结果为阳性的数据；去除定期到医疗机构开药、定期体检以及定期复查的患者数据；去除就诊中医门诊的患者数据。
  3. 根据权利要求1所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述数据分析推理模块中，所述全科知识图谱采用手动构建和自动构建相结合的方式；通过临床医生完成与数据采集模块抽取的症状、疾病相关知识的全科知识图谱手动构建部分；全科知识图谱自动构建部分涵盖从SemMed DB以及中文ICD-10中提取到的相关临床术语及关系。
  4. 根据权利要求1所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述子系统1中，基于变体图神经网络实现全科知识图谱的全局向量化，具体为：
    全科知识图谱包含一系列三元组(h，r，t)，每个三元组由一个关系r和两个实体h，t组成，每个实体作为全科知识图谱中的一个节点，每个三元组代表一条医学知识；
    为全科知识图谱的每个节点h定义一个可学习的嵌入向量表示eh，为每个关系r定义一个可学习的嵌入向量表示er；设置全局向量化过程迭代步数M，第m，m≤M步迭代时，节点h的状态向量表示为与节点h相关的医学知识(h，r，t)的向量化表示
    在全科知识图谱中对节点h所在的全部三元组包含的医学知识进行聚合，得到三元组聚合向量其中为全科知识图谱中节点h所在的全部三元组的数量；
    节点h的下一步状态向量的更新方式为
    fa和fb为两个隐藏层函数，输入为全部向量的拼接，输出与输入的单个向量维度一致；
    将最后一步的输出结果作为节点h的全局向量表示。
  5. 根据权利要求4所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述子系统2中，利用变体图神经网络定义认知图谱的节点状态更新方式，基于全科知识图谱和认知图谱推理得到患者的疾病和治疗方式，具体为：
    将子系统1中得到的认知图谱推理节点集HN以及推理节点在全科知识图谱中的全部关系纳入到认知图谱中，认知图谱全部节点集合记作V，V＝S+HN，初始临床诊疗数据集S＝{a1，a2，...，ai，...，an}，n为患者初始数据条数，ai为第i条患者初始数据；
    设置认知图谱节点状态更新所需迭代步数T，认知图谱中的节点v，v∈V的初始状态向量表示第t，t＜T步迭代时，三元组(v，r，v′)蕴含的医学知识的向量化表示其中为患者初始临床诊疗数据集S中节点的全局向量表示；
    在认知图谱中对节点v所在的全部三元组包含的医学知识进行聚合，得到三元组聚合向量其中为认知图谱中节点v所在的全部三元组的数量；
    节点v的t+1步的状态向量的更新方式为
    fc和fd为两个隐藏层函数。
  6. 根据权利要求5所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述子系统2中，将节点状态更新完成的认知图谱的每个节点的状态向量取代节点在全科知识图谱中的全局向量表示，将全科知识图谱的全部节点用矩阵X表示，矩阵X的维度为J×Q，其中J为全科知识图谱中的节点数目，Q为每个节点的向量维度；采用一个两层的全连接网络作为最终的预测层F：
    其中，prediction为经过推理后给出的患者最可能的疾病D或者建议的治疗方式P，患者所患的真实疾病或者真实治疗方式记为target，使用sigmoid函数作为激活函数，使用交叉熵损失函数计算prediction和target之间的误差，最后使用随机梯度下降算法传播交叉熵损失函数梯度以完成对基于认知图谱的推理系统的优化。
  7. 根据权利要求1-6中任一项所述的一种基于认知图谱的全科患者个性化诊疗方案推荐系统，其特征在于，所述推荐结果显式模块中，利用网页前端将患者已有的症状、疾病以及经过推 理后可能需要的治疗方式和可能患有的疾病进行可视化展示，并支持临床医生输入患者就诊时的临床诊疗信息，对患者的诊疗路径和方案进行可视化展示。
PCT/CN2023/088590 2022-04-18 2023-04-17 一种基于认知图谱的全科患者个性化诊疗方案推荐系统 WO2023202508A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210403938.2 2022-04-18
CN202210403938.2A CN114496234B (zh) 2022-04-18 2022-04-18 一种基于认知图谱的全科患者个性化诊疗方案推荐系统

Publications (1)

Publication Number Publication Date
WO2023202508A1 true WO2023202508A1 (zh) 2023-10-26

Family

ID=81489383

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/088590 WO2023202508A1 (zh) 2022-04-18 2023-04-17 一种基于认知图谱的全科患者个性化诊疗方案推荐系统

Country Status (2)

Country Link
CN (1) CN114496234B (zh)
WO (1) WO2023202508A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117524405A (zh) * 2024-01-05 2024-02-06 长春中医药大学 一种基于云计算的妇科护理方法智能选择系统
CN117649949A (zh) * 2024-01-29 2024-03-05 浙江大学 基于强化学习的临床思维数据生成系统及方法
CN118016316A (zh) * 2024-04-10 2024-05-10 健数(长春)科技有限公司 知识图谱结合血常规检验数据的疾病筛率提升方法和系统

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114496234B (zh) * 2022-04-18 2022-07-19 浙江大学 一种基于认知图谱的全科患者个性化诊疗方案推荐系统
CN114969557B (zh) * 2022-07-29 2022-11-08 之江实验室 一种基于多来源信息融合的宣教推送方法和系统
CN115359870B (zh) * 2022-10-20 2023-03-24 之江实验室 一种基于层次图神经网络的疾病诊疗过程异常识别系统
CN115658877B (zh) * 2022-12-27 2023-03-21 神州医疗科技股份有限公司 基于强化学习的药物推荐方法、装置、电子设备和介质
CN115862848B (zh) * 2023-02-15 2023-05-30 之江实验室 基于临床数据筛选和医学知识图谱的疾病预测系统和装置
CN115858876B (zh) * 2023-02-21 2023-05-26 杭州华卓信息科技有限公司 一种基于疾病知识图谱的随访内容智能推送方法及系统
CN116187448B (zh) * 2023-04-25 2023-08-01 之江实验室 一种信息展示的方法、装置、存储介质及电子设备
CN116682553B (zh) * 2023-08-02 2023-11-03 浙江大学 一种融合知识与患者表示的诊断推荐系统
CN117012374B (zh) * 2023-10-07 2024-01-26 之江实验室 一种融合事件图谱和深度强化学习的医疗随访系统及方法
CN117012375B (zh) * 2023-10-07 2024-03-26 之江实验室 一种基于患者拓扑特征相似性的临床决策支持方法和系统
CN117153378B (zh) * 2023-10-31 2024-03-01 北京博晖创新生物技术集团股份有限公司 一种导诊方法、装置、电子设备及存储介质
CN117973947B (zh) * 2024-04-01 2024-06-14 国网山东省电力公司宁津县供电公司 一种配电网工程施工工艺标准化验收方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021151325A1 (zh) * 2020-09-09 2021-08-05 平安科技（深圳）有限公司 基于医疗知识图谱的分诊模型训练方法、装置及设备
CN113808693A (zh) * 2021-09-10 2021-12-17 浙江科技学院 一种基于图神经网络和注意力机制的药物推荐方法
CN113871003A (zh) * 2021-12-01 2021-12-31 浙江大学 一种基于含因果性医学知识图谱的疾病辅助鉴别诊断系统
WO2022072785A1 (en) * 2020-10-01 2022-04-07 University Of Massachusetts A neural graph model for automated clinical assessment generation
CN114496234A (zh) * 2022-04-18 2022-05-13 浙江大学 一种基于认知图谱的全科患者个性化诊疗方案推荐系统

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999004043A1 (en) * 1997-07-14 1999-01-28 Abbott Laboratories Telemedicine
US6687685B1 (en) * 2000-04-07 2004-02-03 Dr. Red Duke, Inc. Automated medical decision making utilizing bayesian network knowledge domain modeling
US20050108052A1 (en) * 2003-11-03 2005-05-19 Omaboe Nortey J. Proces for diagnosic system and method applying artificial intelligence techniques to a patient medical record and that combines customer relationship management (CRM) and enterprise resource planning (ERP) software in a revolutionary way to provide a unique-and uniquely powerful and easy-to-use-tool to manage veterinary or human medical clinics and hospitals
CN104573350A (zh) * 2014-12-26 2015-04-29 深圳市前海安测信息技术有限公司 基于网络医院的全科医生辅助诊疗系统及方法
CN108461151B (zh) * 2017-12-15 2021-06-15 北京大学深圳研究生院 一种知识图谱的逻辑增强方法及装置
KR20190135908A (ko) * 2019-02-01 2019-12-09 (주)제이엘케이인스펙션 인공지능 기반 치매 진단 방법 및 장치
CN109920540A (zh) * 2019-03-14 2019-06-21 宁波中云创科信息技术有限公司 辅助诊疗决策系统的构建方法、装置及计算机设备
US20210169355A1 (en) * 2019-12-06 2021-06-10 Savan Patel Systems & Methods for Vascular Disease Prediction, Indication, or Diagnosis
CN111767410B (zh) * 2020-06-30 2023-05-30 深圳平安智慧医健科技有限公司 临床医疗知识图谱的构建方法、装置、设备及存储介质
CN112242187B (zh) * 2020-10-26 2023-06-27 平安科技（深圳）有限公司 基于知识图谱表征学习的医疗方案推荐系统及方法
CN112102937B (zh) * 2020-11-13 2021-02-12 之江实验室 一种慢性病辅助决策的患者数据可视化方法及系统
CN112463987A (zh) * 2020-12-09 2021-03-09 中国园林博物馆北京筹备办公室 一种中国古典园林知识图谱补全与认知推理方法
CN113010663A (zh) * 2021-04-26 2021-06-22 东华大学 一种基于工业认知图谱的自适应推理问答方法和系统
CN113889259A (zh) * 2021-09-06 2022-01-04 浙江工业大学 一种知识图谱辅助下的自动诊断对话系统
CN113779220B (zh) * 2021-09-13 2023-06-23 内蒙古工业大学 一种基于三通道认知图谱和图注意力网络的蒙语多跳问答方法
CN113990495B (zh) * 2021-12-27 2022-04-29 之江实验室 一种基于图神经网络的疾病诊断预测系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021151325A1 (zh) * 2020-09-09 2021-08-05 平安科技（深圳）有限公司 基于医疗知识图谱的分诊模型训练方法、装置及设备
WO2022072785A1 (en) * 2020-10-01 2022-04-07 University Of Massachusetts A neural graph model for automated clinical assessment generation
CN113808693A (zh) * 2021-09-10 2021-12-17 浙江科技学院 一种基于图神经网络和注意力机制的药物推荐方法
CN113871003A (zh) * 2021-12-01 2021-12-31 浙江大学 一种基于含因果性医学知识图谱的疾病辅助鉴别诊断系统
CN114496234A (zh) * 2022-04-18 2022-05-13 浙江大学 一种基于认知图谱的全科患者个性化诊疗方案推荐系统

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117524405A (zh) * 2024-01-05 2024-02-06 长春中医药大学 一种基于云计算的妇科护理方法智能选择系统
CN117524405B (zh) * 2024-01-05 2024-03-26 长春中医药大学 一种基于云计算的妇科护理方法智能选择系统
CN117649949A (zh) * 2024-01-29 2024-03-05 浙江大学 基于强化学习的临床思维数据生成系统及方法
CN117649949B (zh) * 2024-01-29 2024-04-30 浙江大学 基于强化学习的临床思维数据生成系统及方法
CN118016316A (zh) * 2024-04-10 2024-05-10 健数(长春)科技有限公司 知识图谱结合血常规检验数据的疾病筛率提升方法和系统
CN118016316B (zh) * 2024-04-10 2024-06-04 健数(长春)科技有限公司 知识图谱结合血常规检验数据的疾病筛率提升方法和系统

Also Published As

Publication number Publication date
CN114496234B (zh) 2022-07-19
CN114496234A (zh) 2022-05-13

Similar Documents

Publication Publication Date Title
WO2023202508A1 (zh) 一种基于认知图谱的全科患者个性化诊疗方案推荐系统
Liang et al. Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence
Su et al. Deep learning in mental health outcome research: a scoping review
Shickel et al. Deep EHR: a survey of recent advances in deep learning techniques for electronic health record (EHR) analysis
CN113015977A (zh) 使用自然语言处理的对疾病和病症的基于深度学习的诊断和转诊
Lee et al. Machine learning in relation to emergency medicine clinical and operational scenarios: an overview
Madhavan et al. Res-CovNet: an internet of medical health things driven COVID-19 framework using transfer learning
CN111492437A (zh) 支持医学决策的方法和系统
CN107247881A (zh) 一种多模态智能分析方法及系统
CN112466462B (zh) 一种基于图深度学习的emr信息关联及演化方法
WO2023207795A1 (zh) 一种基于医疗健康的数字孪生建立方法、装置和存储介质
Huang et al. Medical knowledge-based network for patient-oriented visual question answering
CN117316466A (zh) 一种基于知识图谱与自然语言处理技术的临床决策方法、系统及设备
Davazdahemami et al. A deep learning approach for predicting early bounce-backs to the emergency departments
WO2019132686A1 (ru) Способ формирования математических моделей пациента с использованием технологий искусственного интеллекта
Hansen et al. Assigning diagnosis codes using medication history
Leng et al. Bi-level artificial intelligence model for risk classification of acute respiratory diseases based on Chinese clinical data
Sampath et al. Ensemble Nonlinear Machine Learning Model for Chronic Kidney Diseases Prediction
An Construction and application of Chinese breast cancer knowledge graph based on multi-source heterogeneous data
Jia et al. DKDR: An approach of knowledge graph and deep reinforcement learning for disease diagnosis
Saha et al. LM-DNN: pre-trained DNN with LSTM and cross fold validation for detecting viral pneumonia from chest CT
CN116543917A (zh) 一种针对异构时间序列数据的信息挖掘方法
Ming et al. AI assisted clinical diagnosis & treatment and development strategy
Song et al. Research of medical aided diagnosis system based on temporal knowledge graph
CN116168828A (zh) 基于知识图谱和深度学习的疾病预测方法、装置及计算机设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23791163

Country of ref document: EP

Kind code of ref document: A1