WO2021196239A1 - 跨医疗数据源的网络表示学习算法 - Google Patents
跨医疗数据源的网络表示学习算法 Download PDFInfo
- Publication number
- WO2021196239A1 WO2021196239A1 PCT/CN2020/083377 CN2020083377W WO2021196239A1 WO 2021196239 A1 WO2021196239 A1 WO 2021196239A1 CN 2020083377 W CN2020083377 W CN 2020083377W WO 2021196239 A1 WO2021196239 A1 WO 2021196239A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- network
- layer
- node
- expression
- source
- Prior art date
- Legal status (The legal status 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 status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0499—Feedforward networks
-
- 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
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- 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
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/20—ICT specially adapted for the handling or processing of medical references relating to practices or guidelines
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Definitions
- the present disclosure belongs to the field of medical technology, and in particular relates to a network representation learning algorithm across medical data sources.
- the relationship between nodes in the network can be the treatment relationship between diseases and drugs, the manifestation relationship between diseases and symptoms, and so on.
- the existing network embedding representation algorithms can be divided into two categories: one is the direct inference representation learning algorithm. Given a target network, the direct expression algorithm directly optimizes the expression vector of each node through the attributes of the node and the network relationship, such as DeepWalk and Node2vec. The second is the inductive representation learning algorithm.
- the inductive representation algorithm is often to learn a mapping function. As long as the attributes of the input node and its neighbors are given, the expression vector of the node can be inferred through the mapping function, such as GCN, GraphSAGE and GAT.
- the direct push algorithm because the direct push algorithm directly optimizes the node expression vector in a network composed of data from a certain hospital, for a new network composed of data from another hospital, direct push The formula algorithm cannot directly infer the expression vector of the nodes in the new network. Therefore, the direct push algorithm does not have any available knowledge that can be used for network learning across medical data sources.
- inductive algorithms Although it considers learning a mapping function of node attributes and structural information when modeling, it can naturally perform network inferences across medical data sources, but inductive algorithms It does not take into account that the data distribution between the network and the network is different. The model or knowledge summarized from the medical network of one hospital may not be well applicable to the medical network of another hospital. Therefore, the inductive algorithm is used across the network. The network representation of medical data sources also has certain flaws in the problem of learning.
- the present disclosure proposes a network representation learning algorithm across medical data sources.
- a network representation learning algorithm across medical data sources including:
- S1 Generate medical network data including a source network and a target network, the source network is generated from a medical record of a certain hospital, and the target network is generated from a medical record of another hospital different from the hospital, and the medical network data Including the patient’s medical records, building a network relationship among symptoms, diseases, drugs, and diagnostic methods;
- step S3 Obtain an L-layer neural network from step S2, and calculate the structural characteristics and expression characteristics of the source network and the target network for each layer, and calculate the distance loss between the network characteristics of the source network and the target network;
- S4 Obtain the output of the source network in the L-layer neural network from S3, calculate the loss value according to the classification loss and the distance loss, and update the parameters of the algorithm according to the backpropagation algorithm;
- step S3 an L-layer neural network is obtained from step S2, and the source network and target network are calculated separately for each layer.
- Structural characteristics and expression characteristics, calculating the distance loss between the network characteristics of the source network and the target network include:
- the node feature expression vector of each network obtains the structural feature through a message routing module, and the structural feature obtains the new expression feature vector of the current node through a message aggregation module;
- S32 Calculate the distance loss value between the node features of the current layer from the source network and the target network through the network alignment module across medical data sources;
- step S31 in each layer of the L-layer neural network, the node feature expression vector of each network is obtained through a message routing module Structural features.
- the new expression feature vectors of the current node obtained by the structural features through a message aggregation module include:
- the message routing module of each layer is expressed as:
- Is the structural feature vector of the l-th layer of the node i in the L-layer neural network Is the expression feature vector of the source network and the target network of the l-1th layer in the L-layer neural network
- the expression feature vector of the 0th layer is represented by the original feature vector x i of the node
- a (l)T is the parameter matrix involved in the message routing module of the first layer
- ⁇ is the activation function
- is the direct connection operation of two vectors
- N(v ) Is the set of neighbors directly connected to node v, Is the weight of the message sent from node u to node v;
- the message aggregation module of each layer is expressed as:
- I the parameter matrix involved in the message aggregation module
- I a vector showing the aggregation level of the node.
- step S32 through the cross-medical data source network alignment module, the current layer is calculated between the node features from the source network and the target network.
- the distance loss values include:
- P r , Q r are the structural feature vectors of the source network and the target network with Distribution, Is a distance function used to calculate the structural feature vector with Expected distance
- P a , Q a are the node expression feature vectors of the source network and the target network with Distribution, As a distance function, used to calculate the node expression feature vector with The expected distance.
- the step S33 is to repeat steps S31 to S32 for L times to obtain the node feature vectors of the final source network and the target network and the L-layer accumulation Structural feature distance loss and expression feature distance loss include:
- the cumulative structural feature distance loss of the L layer is:
- the cumulative expression feature distance loss of the L layer is:
- the present disclosure has the following advantages:
- FIG. 1 is a flowchart of an embodiment of the network representation learning algorithm across medical data sources proposed in the present disclosure
- FIG. 2 is a flowchart of another embodiment of the network representation learning algorithm across medical data sources proposed in the present disclosure.
- FIG. 1 is a flowchart of an embodiment of the cross-medical data source network representation learning algorithm proposed in the present disclosure. As shown in FIG. 1, the cross-medical data source network representation learning algorithm:
- S1 Generate medical network data including a source network and a target network, the source network is generated from a medical record of a certain hospital, and the target network is generated from a medical record of another hospital different from the hospital, and the medical network data Including the patient’s medical records, building a network relationship among symptoms, diseases, drugs, and diagnostic methods;
- step S3 Obtain an L-layer neural network from step S2, and calculate the structural characteristics and expression characteristics of the source network and the target network for each layer, and calculate the distance loss between the network characteristics of the source network and the target network;
- S4 Obtain the output of the source network in the L-layer neural network from S3, calculate the loss value according to the classification loss and the distance loss, and update the parameters of the algorithm according to the backpropagation algorithm;
- FIG. 2 is a flowchart of another embodiment of the network representation learning algorithm across medical data sources proposed in the present disclosure.
- an L-layer neural network is obtained from step S2, and
- Each layer calculates the structural characteristics and expression characteristics of the source network and the target network, and the distance loss between the network characteristics of the source network and the target network includes:
- the node feature expression vector of each network obtains the structural feature through a message routing module, and the structural feature obtains the new expression feature vector of the current node through a message aggregation module;
- S32 Calculate the distance loss value between the node features of the current layer from the source network and the target network through the network alignment module across medical data sources;
- step S31 in each layer of the L-layer neural network, the node feature expression vector of each network passes through a message routing module to obtain the structural feature, and the structural feature passes through a message aggregation module to obtain the new expression feature vector of the current node including :
- the message routing module of each layer is expressed as:
- Is the structural feature vector of the l-th layer of the node i in the L-layer neural network Is the expression feature vector of the source network and the target network of the l-1th layer in the L-layer neural network
- the expression feature vector of the 0th layer is represented by the original feature vector x i of the node
- a (l)T is the parameter matrix involved in the message routing module of the first layer
- ⁇ is the activation function
- is the direct connection operation of two vectors
- N(v ) Is the set of neighbors directly connected to node v, Is the weight of the message sent from node u to node v;
- the message aggregation module of each layer is expressed as:
- I the parameter matrix involved in the message aggregation module
- I a vector showing the aggregation level of the node.
- step S32 calculating the distance loss value between the node features of the current layer from the source network and the target network through the network alignment module across medical data sources includes:
- P r , Q r are the structural feature vectors of the source network and the target network with Distribution, Is a distance function used to calculate the structural feature vector with Expected distance
- P a , Q a are the node expression feature vectors of the source network and the target network with Distribution, As a distance function, used to calculate the node expression feature vector with The expected distance.
- the step S33 repeating steps S31 to S32 for L times, to obtain the node feature vectors of the final source network and the target network and the accumulated structural feature distance loss and expression feature distance loss of the L layer include:
- the cumulative structural feature distance loss of the L layer is:
- the cumulative expression feature distance loss of the L layer is:
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Bioethics (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种跨医疗数据源的网络表示学习算法,包括:S1,生成包括源网络和目标网络的医疗网络数据;S2,从源网络和目标网络随机采样设定数量的节点;S3,得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失;S4,得到源网络在L层神经网络的输出,并根据分类损失和距离损失计算损失值,根据反向传播算法更新算法的参数;S5,重复步骤S2-S4,直至整个算法收敛,使得算法对于疾病分类的准确率在多个迭代内不再上升。有益效果:考虑了不同医院数据源之间数据分布不一致的问题,通过提取网络的结构信息及节点属性信息、最小化特征距离弥补信息损失,有着广阔的应用空间。
Description
本公开属于医疗技术领域,尤其涉及一种跨医疗数据源的网络表示学习算法。
随着深度学习技术的逐渐成熟,将其应用在医疗场景中逐渐变成可能。通过对于病人病例的建模,再利用深度学习技术可以帮助医生对于病人的病情进行推断以此来促进医疗事业的发展。在医疗场景中,我们的研究对象往往有病人、症状、疾病以及药物等,他们之间存在着复杂的逻辑关系。这种关系直接通过一个简单的深度全连接网络是难以刻画的。同时,考虑到现有的深度学习技术中,网络嵌入表示学习算法可以直观地表示对象之间的连接关系,具有强大的推理能力。因此,我们可以利用网络嵌入表示学习算法的技术,以网络的视角对其进行建模求解。具体来说,我们可以将每一个涉及的对象看成是一个网络中的节点,节点与节点之间有边连接的话,表示他们之间存在着某种关系。比如网络中节点与节点之间的关系可以是疾病和药物的治疗关系、疾病和症状的表现关系等等。通过将复杂的医疗关系抽象成一个网络,再对网络通过网络嵌入表达算法进行信息的提取,从而达到一个病情的推断的作用。
现有的网络嵌入表示算法主要可以分为两类:一是直推式的表示学习算法。给定一个目标网络,直推式的表示算法直接通过节点的属性以及网络关系去优化每个节点的表达向量,比如DeepWalk和Node2vec。二是归纳式的表示学习算法。归纳式的表示算法往往是学习出一个映射函数,只要给定输入节点的属性以及其邻居,就可以通过映射函数推断出节点的表达向量,比如GCN,GraphSAGE和GAT。
在现实的医疗场景中,医疗数据往往是来自于不同的医院。这就导致不同医院的病人数据之间的分布是并不一致的。比如说,同样是感冒这样一个 疾病,对于南方人而言,可能是由于身体受到寒冷引起的寒性感冒,对于北方人还有可能是因为室内的暖气引起的热性感冒。再比如,同样一种疾病,可能有多种相同病理的药物可以治疗,不同医院之间的用药习惯可能也是不同的。正是由于这些现实存在的原因使得不同医院之间的数据分布是不同。这就导致了,在希望尽可能地利用多种数据来帮助模型学习的情况下,如果直接使用多个不同医疗数据源的数据,而不考虑其现实的分布不一致问题的话,很可能学到的知识是有偏差的。
然而,现有的算法都不能够很好地解决这样一种跨医疗数据源的网络表示学习问题。具体来讲:
(1)对于直推式的算法而言,由于直推式算法是直接去优化某个医院数据组成的网络中的节点表达向量,因此对于由另一个医院数据组成的新网络而言,直推式算法无法直接去推断新网络中节点的表达向量。因此直推式算法没有任何可用的知识可以被用来进行跨医疗数据源的网络学习。
(2)对于归纳式的算法而言,尽管其在建模的时候考虑的是学习一个节点属性和结构信息的映射函数,这样可以自然地进行跨医疗数据源的网络推断,但是归纳式的算法并没有考虑到网络与网络之间的数据分布是不同的,从一个医院的医疗网络中归纳出来的模式或者知识可能并不能很好地适用于另一个医院的医疗网络,因此归纳式算法在跨医疗数据源的网络表示学习的问题上也存在着一定的缺陷。
因此,现有技术需要改进。
上述背景技术内容仅用于帮助理解本公开,而并不代表承认或认可所提及的任何内容属于相对于本公开的公知常识的一部分。
发明内容
为解决上述技术问题,本公开提出了一种跨医疗数据源的网络表示学习算法。
基于本公开实施例的一个方面,公开一种跨医疗数据源的网络表示学习算法,包括:
S1,生成包括源网络和目标网络的医疗网络数据,所述源网络由某一个医院的就诊记录生成,所述目标网络由与该医院不同的另一个医院的就诊记录生成,所述医疗网络数据包括病人的就诊记录信息,构建症状、疾病、药物、诊断方法之间的网络关系;
S2,分别从源网络和目标网络随机采样设定数量的节点,采集节点的数量与所述医疗网络的度数相关;
S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失;
S4,从S3中得到源网络在L层神经网络的输出,并根据分类损失和距离损失计算损失值,根据反向传播算法更新算法的参数;
S5,重复步骤S2-S4,直至整个算法收敛,使得算法对于疾病分类的准确率在多个迭代内不再上升。
基于本公开的跨医疗数据源的网络表示学习算法的另一个实施例中,所述步骤S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失包括:
S30,将源网络和目标网络的节点特征输入到L层的神经网络中;
S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量;
S32,通过跨医疗数据源的网络对齐模块,计算当前层来自源网络和目标网络的节点特征之间的距离损失值;
S33,重复步骤S31至S32进行L次,得到最终源网络和目标网络的节 点特征向量和L层累积的结构特征距离损失和表达特征距离损失。
基于本公开的跨医疗数据源的网络表示学习算法的另一个实施例中,所述步骤S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量包括:
每一层的消息路由模块表示为:
式中,
为节点i在L层神经网络中第l层的结构特征向量,
为L层神经网络中第l-1层的源网络和目标网络的表达特征向量,第0层的表达特征向量由节点的原始特征向量x
i表示,
为第l层的消息路由模块涉及的参数矩阵,a
(l)T为第l层的消息路由模块涉及的参数矩阵,σ为激活函数,||为两个向量的直接相连操作,N(v)为节点v直接相连的邻居集合,
为节点u传向节点v的消息权重;
每一层的消息聚合模块表示为:
基于本公开的跨医疗数据源的网络表示学习算法的另一个实施例中,所述步骤S32,通过跨医疗数据源的网络对齐模块,计算当前层来自源网络和目标网络的节点特征之间的距离损失值包括:
每一层的结构特征距离为:
每一层的表达特征距离损失为:
基于本公开的跨医疗数据源的网络表示学习算法的另一个实施例中,所述步骤S33,重复步骤S31至S32进行L次,得到最终源网络和目标网络的节点特征向量和L层累积的结构特征距离损失和表达特征距离损失包括:
L层累积的结构特征距离损失为:
L层累积的表达特征距离损失为:
与现有技术相比,本公开具有如下优点:
采用本公开的跨医疗数据源的网络表示学习算法,考虑了不同医院数据源之间的数据分布不一致问题,通过最小化特征距离来弥补由于不一致导致的信息损失,从而能够充分利用多源的医疗数据,在医疗领域有着广阔的应用空间。
图1为本公开提出的跨医疗数据源的网络表示学习算法的一个实施例的流程图;
图2为本公开提出的跨医疗数据源的网络表示学习算法的另一个实施例的流程图。
为使本公开实施例的目的、技术方案和优点更加清楚,下面将结合本公开实施例中的附图,对本公开实施例中的技术方案进行清楚、完整的描述。显然,所描述的实施例只是本公开一部分实施例,而不是全部的实施例。基于本公开中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本公开保护的范围。
下面结合附图和实施例对本公开提供的一种跨医疗数据源的网络表示学习算法进行更详细地说明。
图1为本公开提出的跨医疗数据源的网络表示学习算法的一个实施例的流程图,如图1所示,所述跨医疗数据源的网络表示学习算法:
S1,生成包括源网络和目标网络的医疗网络数据,所述源网络由某一个医院的就诊记录生成,所述目标网络由与该医院不同的另一个医院的就诊记录生成,所述医疗网络数据包括病人的就诊记录信息,构建症状、疾病、药物、诊断方法之间的网络关系;
S2,分别从源网络和目标网络随机采样设定数量的节点,采集节点的数量与所述医疗网络的度数相关,以疾病诊断为例,需要从每个疾病节点出发迭代L次采集其邻居节点,将采集到的节点对应的节点属性作为算法的输入数据;
S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失;
S4,从S3中得到源网络在L层神经网络的输出,并根据分类损失和距离损失计算损失值,根据反向传播算法更新算法的参数;
S5,重复步骤S2-S4,直至整个算法收敛,使得算法对于疾病分类的准确率在多个迭代内不再上升。
图2为本公开提出的跨医疗数据源的网络表示学习算法的另一个实施例的流程图,如图2所示,所述步骤S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失包括:
S30,将源网络和目标网络的节点特征输入到L层的神经网络中;
S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量;
S32,通过跨医疗数据源的网络对齐模块,计算当前层来自源网络和目标网络的节点特征之间的距离损失值;
S33,重复步骤S31至S32进行L次,得到最终源网络和目标网络的节点特征向量和L层累积的结构特征距离损失和表达特征距离损失。
所述步骤S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量包括:
每一层的消息路由模块表示为:
式中,
为节点i在L层神经网络中第l层的结构特征向量,
为L层神经网络中第l-1层的源网络和目标网络的表达特征向量,第0层的表达特征向量由节点的原始特征向量x
i表示,
为第l层的消息路由模块涉及的参数矩阵,a
(l)T为第l层的消息路由模块涉及的参数矩阵,σ为激活函数,||为两个向量的直接相连操作,N(v)为节点v直接相连的邻居集合,
为节点u传向节点v的消息权重;
每一层的消息聚合模块表示为:
所述步骤S32,通过跨医疗数据源的网络对齐模块,计算当前层来自源网络和目标网络的节点特征之间的距离损失值包括:
每一层的结构特征距离为:
每一层的表达特征距离损失为:
所述步骤S33,重复步骤S31至S32进行L次,得到最终源网络和目标网络的节点特征向量和L层累积的结构特征距离损失和表达特征距离损失包括:
L层累积的结构特征距离损失为:
L层累积的表达特征距离损失为:
对于本领域技术人员而言,显然本公开实施例不限于上述示范性实施例的细节,而且在不背离本公开实施例的精神或基本特征的情况下,能够以其 他的具体形式实现本公开实施例。因此,无论从哪一点来看,均应将实施例看作是示范性的,而且是非限制性的,本公开实施例的范围由所附权利要求而不是上述说明限定,因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本公开实施例内。不应将权利要求中的任何附图标记视为限制所涉及的权利要求。此外,显然“包括”一词不排除其他单元或步骤,单数不排除复数。系统、装置或终端权利要求中陈述的多个单元、模块或装置也可以由同一个单元、模块或装置通过软件或者硬件来实现。第一,第二等词语用来表示名称,而并不表示任何特定的顺序。
最后应说明的是,以上实施方式仅用以说明本公开实施例的技术方案而非限制,尽管参照以上较佳实施方式对本公开实施例进行了详细说明,本领域的普通技术人员应当理解,可以对本公开实施例的技术方案进行修改或等同替换都不应脱离本公开实施例的技术方案的精神和范围。
Claims (5)
- 一种跨医疗数据源的网络表示学习算法,其特征在于,包括:S1,生成包括源网络和目标网络的医疗网络数据,所述源网络由某一个医院的就诊记录生成,所述目标网络由与该医院不同的另一个医院的就诊记录生成,所述医疗网络数据包括病人的就诊记录信息,构建症状、疾病、药物、诊断方法之间的网络关系;S2,分别从源网络和目标网络随机采样设定数量的节点,采集节点的数量与所述医疗网络的度数相关;S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失;S4,从S3中得到源网络在L层神经网络的输出,并根据分类损失和距离损失计算损失值,根据反向传播算法更新算法的参数;S5,重复步骤S2-S4,直至整个算法收敛,使得算法对于疾病分类的准确率在多个迭代内不再上升。
- 根据权利要求1所述的跨医疗数据源的网络表示学习算法,其特征在于,所述步骤S3,从步骤S2中得到一个L层的神经网络,并对每一层分别计算源网络和目标网络的结构特征和表达特征,计算源网络和目标网络的网络特征之间的距离损失包括:S30,将源网络和目标网络的节点特征输入到L层的神经网络中;S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量;S32,通过跨医疗数据源的网络对齐模块,计算当前层来自源网络和目标网络的节点特征之间的距离损失值;S33,重复步骤S31至S32进行L次,得到最终源网络和目标网络的节 点特征向量和L层累积的结构特征距离损失和表达特征距离损失。
- 根据权利要求2所述的跨医疗数据源的网络表示学习算法,其特征在于,所述步骤S31,在L层神经网络的每一层中,每个网络的节点特征表达向量经过一个消息路由模块得到结构特征,结构特征经过一个消息聚合模块得到当前节点的新的表达特征向量包括:每一层的消息路由模块表示为:式中, 为节点i在L层神经网络中第l层的结构特征向量, 为L层神经网络中第l-1层的源网络和目标网络的表达特征向量,第0层的表达特征向量由节点的原始特征向量x i表示, 为第l层的消息路由模块涉及的参数矩阵,a (l)T为第l层的消息路由模块涉及的参数矩阵,σ为激活函数,||为两个向量的直接相连操作,N(v)为节点v直接相连的邻居集合, 为节点u传向节点v的消息权重;每一层的消息聚合模块表示为:
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202080005552.5A CN113767401B (zh) | 2020-04-03 | 2020-04-03 | 跨医疗数据源的网络表示学习方法 |
| PCT/CN2020/083377 WO2021196239A1 (zh) | 2020-04-03 | 2020-04-03 | 跨医疗数据源的网络表示学习算法 |
| PCT/CN2021/085611 WO2021197491A1 (zh) | 2020-04-03 | 2021-04-06 | 跨医疗数据源的网络表示学习算法 |
| CN202180006088.6A CN114730638A (zh) | 2020-04-03 | 2021-04-06 | 跨医疗数据源的网络表示学习算法 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2020/083377 WO2021196239A1 (zh) | 2020-04-03 | 2020-04-03 | 跨医疗数据源的网络表示学习算法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021196239A1 true WO2021196239A1 (zh) | 2021-10-07 |
Family
ID=77927797
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2020/083377 Ceased WO2021196239A1 (zh) | 2020-04-03 | 2020-04-03 | 跨医疗数据源的网络表示学习算法 |
| PCT/CN2021/085611 Ceased WO2021197491A1 (zh) | 2020-04-03 | 2021-04-06 | 跨医疗数据源的网络表示学习算法 |
Family Applications After (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2021/085611 Ceased WO2021197491A1 (zh) | 2020-04-03 | 2021-04-06 | 跨医疗数据源的网络表示学习算法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (2) | CN113767401B (zh) |
| WO (2) | WO2021196239A1 (zh) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114357008A (zh) * | 2021-12-16 | 2022-04-15 | 上海金仕达卫宁软件科技有限公司 | 医疗行为一致性识别模型建立方法及风险识别方法 |
| CN115018662A (zh) * | 2022-06-10 | 2022-09-06 | 扬州大学 | 一种基于深度学习的多源定位方法 |
| CN115600576A (zh) * | 2022-10-09 | 2023-01-13 | 浪潮云信息技术股份公司(Cn) | 一种基于双层迭代直推式学习的地址单元解析方法和装置 |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113767401B (zh) * | 2020-04-03 | 2025-01-03 | 清华大学 | 跨医疗数据源的网络表示学习方法 |
| CN116597971B (zh) * | 2023-07-18 | 2023-09-19 | 山东新睿信息科技有限公司 | 基于数字孪生的医院空间优化模拟方法及系统 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108399431A (zh) * | 2018-02-28 | 2018-08-14 | 国信优易数据有限公司 | 分类模型训练方法以及分类方法 |
| CN109273062A (zh) * | 2018-08-09 | 2019-01-25 | 北京爱医声科技有限公司 | Icd智能辅助编码系统 |
| CN109636788A (zh) * | 2018-12-11 | 2019-04-16 | 中国石油大学(华东) | 一种基于深度神经网络的ct图像胆结石智能检测模型 |
| US20190171714A1 (en) * | 2008-03-21 | 2019-06-06 | Safermed, LLC d/b/a SaferMD, LLC | Artificial Intelligence Quality Measures Data Extractor |
| CN109920547A (zh) * | 2019-03-05 | 2019-06-21 | 北京工业大学 | 一种基于电子病历数据挖掘的糖尿病预测模型构建方法 |
| CN110532436A (zh) * | 2019-07-17 | 2019-12-03 | 中国人民解放军战略支援部队信息工程大学 | 基于社区结构的跨社交网络用户身份识别方法 |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120084092A1 (en) * | 2010-10-04 | 2012-04-05 | Kozuch Michael J | Method and apparatus for a comprehensive dynamic personal health record system |
| 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 |
| US20190279767A1 (en) * | 2018-03-06 | 2019-09-12 | James Stewart Bates | Systems and methods for creating an expert-trained data model |
| US11537719B2 (en) * | 2018-05-18 | 2022-12-27 | Deepmind Technologies Limited | Deep neural network system for similarity-based graph representations |
| CN108898218A (zh) * | 2018-05-24 | 2018-11-27 | 阿里巴巴集团控股有限公司 | 一种神经网络模型的训练方法、装置、及计算机设备 |
| CN109036553B (zh) * | 2018-08-01 | 2022-03-29 | 北京理工大学 | 一种基于自动抽取医疗专家知识的疾病预测方法 |
| CN109635121A (zh) * | 2018-11-07 | 2019-04-16 | 平安科技(深圳)有限公司 | 医疗知识图谱创建方法及相关装置 |
| CN109559822A (zh) * | 2018-11-12 | 2019-04-02 | 平安科技(深圳)有限公司 | 智能初诊方法、装置、计算机设备及存储介质 |
| CN110299209B (zh) * | 2019-06-25 | 2022-05-20 | 北京百度网讯科技有限公司 | 相似病历查找方法、装置、设备及可读存储介质 |
| CN113767401B (zh) * | 2020-04-03 | 2025-01-03 | 清华大学 | 跨医疗数据源的网络表示学习方法 |
-
2020
- 2020-04-03 CN CN202080005552.5A patent/CN113767401B/zh active Active
- 2020-04-03 WO PCT/CN2020/083377 patent/WO2021196239A1/zh not_active Ceased
-
2021
- 2021-04-06 WO PCT/CN2021/085611 patent/WO2021197491A1/zh not_active Ceased
- 2021-04-06 CN CN202180006088.6A patent/CN114730638A/zh active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190171714A1 (en) * | 2008-03-21 | 2019-06-06 | Safermed, LLC d/b/a SaferMD, LLC | Artificial Intelligence Quality Measures Data Extractor |
| CN108399431A (zh) * | 2018-02-28 | 2018-08-14 | 国信优易数据有限公司 | 分类模型训练方法以及分类方法 |
| CN109273062A (zh) * | 2018-08-09 | 2019-01-25 | 北京爱医声科技有限公司 | Icd智能辅助编码系统 |
| CN109636788A (zh) * | 2018-12-11 | 2019-04-16 | 中国石油大学(华东) | 一种基于深度神经网络的ct图像胆结石智能检测模型 |
| CN109920547A (zh) * | 2019-03-05 | 2019-06-21 | 北京工业大学 | 一种基于电子病历数据挖掘的糖尿病预测模型构建方法 |
| CN110532436A (zh) * | 2019-07-17 | 2019-12-03 | 中国人民解放军战略支援部队信息工程大学 | 基于社区结构的跨社交网络用户身份识别方法 |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114357008A (zh) * | 2021-12-16 | 2022-04-15 | 上海金仕达卫宁软件科技有限公司 | 医疗行为一致性识别模型建立方法及风险识别方法 |
| CN115018662A (zh) * | 2022-06-10 | 2022-09-06 | 扬州大学 | 一种基于深度学习的多源定位方法 |
| CN115600576A (zh) * | 2022-10-09 | 2023-01-13 | 浪潮云信息技术股份公司(Cn) | 一种基于双层迭代直推式学习的地址单元解析方法和装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021197491A1 (zh) | 2021-10-07 |
| CN113767401A (zh) | 2021-12-07 |
| CN113767401B (zh) | 2025-01-03 |
| CN114730638A (zh) | 2022-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021197491A1 (zh) | 跨医疗数据源的网络表示学习算法 | |
| US20230039182A1 (en) | Method, apparatus, computer device, storage medium, and program product for processing data | |
| US11669729B2 (en) | Model training method and apparatus | |
| WO2023000574A1 (zh) | 一种模型训练方法、装置、设备及可读存储介质 | |
| CN108461151B (zh) | 一种知识图谱的逻辑增强方法及装置 | |
| WO2021159742A1 (zh) | 图像分割方法、装置和存储介质 | |
| WO2021151351A1 (zh) | 一种数据处理方法、装置、计算机设备及存储介质 | |
| CN114639483B (zh) | 一种基于图神经网络的电子病历检索方法及装置 | |
| CN112925857A (zh) | 基于谓语类型预测关联的数字信息驱动的系统和方法 | |
| CN116383413B (zh) | 基于医疗数据提取的知识图谱更新方法和系统 | |
| CN113362931B (zh) | 基于知识图谱的归因分析方法、装置和系统 | |
| CN113012803A (zh) | 计算机设备、系统、可读存储介质及医学数据分析方法 | |
| WO2022094884A1 (zh) | 一种面向决策树的横向联邦学习方法 | |
| CN110600121A (zh) | 一种基于知识图谱病因初步诊断方法 | |
| WO2021159814A1 (zh) | 文本数据的错误检测方法、装置、终端设备及存储介质 | |
| CN120822587A (zh) | 专病知识谱图构建分析系统 | |
| CN113658713A (zh) | 传染趋势预测方法、装置、设备及存储介质 | |
| CN118553441A (zh) | 一种基于效果评估的个性化妇科护理推荐方法 | |
| CN119694484A (zh) | 一种护理决策支持系统及护理决策支持方法 | |
| CN112802573B (zh) | 药品包推荐方法、装置、计算机系统及可读存储介质 | |
| WO2021051496A1 (zh) | 诊断结果识别、模型训练的方法、计算机设备及存储介质 | |
| CN109564782A (zh) | 基于医院人口统计的电子临床决策支持设备 | |
| CN113299361A (zh) | 一种基于病人临床表现的档案构建方法及系统 | |
| CN120806048B (zh) | 医学去偏见大语言模型训练方法、装置、电子设备和介质 | |
| CN114883001B (zh) | 基于异质信息网络的疾病预测系统 |
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: 20928359 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20928359 Country of ref document: EP Kind code of ref document: A1 |
|
| WWG | Wipo information: grant in national office |
Ref document number: 202080005552.5 Country of ref document: CN |


