[go: up one dir, main page]

CN110322946B - A computing device for optimal medication granularity based on multi-granularity decision-making model - Google Patents

A computing device for optimal medication granularity based on multi-granularity decision-making model Download PDF

Info

Publication number
CN110322946B
CN110322946B CN201910625179.2A CN201910625179A CN110322946B CN 110322946 B CN110322946 B CN 110322946B CN 201910625179 A CN201910625179 A CN 201910625179A CN 110322946 B CN110322946 B CN 110322946B
Authority
CN
China
Prior art keywords
granularity
medication
layer
decision
decision model
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.)
Active
Application number
CN201910625179.2A
Other languages
Chinese (zh)
Other versions
CN110322946A (en
Inventor
沈夏炯
薛钰
谢毅
沈亚田
孙俊
买宇博
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Henan University
Original Assignee
Henan University
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 Henan University filed Critical Henan University
Priority to CN201910625179.2A priority Critical patent/CN110322946B/en
Publication of CN110322946A publication Critical patent/CN110322946A/en
Application granted granted Critical
Publication of CN110322946B publication Critical patent/CN110322946B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to 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
    • G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Toxicology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention discloses a method for calculating optimal medication granularity based on a multi-granularity decision model, which comprises the following specific steps of: a: constructing a multi-granularity-level medicament granularity decision model by taking different granularity levels for medicament dosage; b: if the cancer patient is stage I or stage II patient, entering step C; if the patients are stage III and stage IV patients, entering step F; c: if the number of the patient data is less than or equal to 5000, entering the step D; e, entering the step E for more than 5000 pieces; d: calculating the global optimal drug granularity by using a coordination method; e: calculating the global optimal medication granularity by using a tree structure method; f: if the number of the patient data is less than or equal to 1000, entering the step G; step H is carried out when the number of the strips is more than 1000; g: calculating local optimal drug granularity by using a serial method; h: and calculating the local optimal medication granularity by using a parallel method. The invention can provide an automatic auxiliary tool for doctors to select anticancer drugs.

Description

一种基于多粒度决策模型的最优用药粒度计算装置A computing device for optimal medication granularity based on multi-granularity decision-making model

技术领域technical field

本发明涉及一种用药粒度选取优化方法，尤其涉及一种基于多粒度决策模型的最优用药粒度的计算方法。The invention relates to a method for selecting and optimizing medication granularity, in particular to a method for calculating optimal medication granularity based on a multi-granularity decision-making model.

背景技术Background technique

目前,随着环境污染的加剧，竞争压力的加重，一些疾病的发病率有增加的趋势，这种情况越来越受到人们的重视。最新统计表明：1810万新增癌症病例中，亚洲约占全球癌症总发病数的50％；960万癌症死亡患者中，亚洲约占全球癌症总死亡数的60％。中国每年有400万的癌症新发病例，很多人都饱受癌症的折磨。At present, with the intensification of environmental pollution and the aggravation of competition pressure, the incidence of some diseases has a tendency to increase, and this situation has been paid more and more attention by people. The latest statistics show that among the 18.1 million new cancer cases, Asia accounts for about 50% of the global total cancer incidence; among the 9.6 million cancer deaths, Asia accounts for about 60% of the global total cancer deaths. There are 4 million new cases of cancer in China every year, and many people suffer from cancer.

在中国新医改的大背景下，智慧医疗正在走进寻常百姓的生活。智慧医疗是以患者数据为中心的医疗服务模式，记录患者在医院的所有行为和诊疗数据。合理的利用收集起来的医疗数据，为诊断病情提供行之有效的解决方案是当前研究的重点任务。In the context of China's new medical reform, smart medical care is entering the lives of ordinary people. Smart medical care is a medical service model centered on patient data, which records all the behaviors and diagnosis and treatment data of patients in the hospital. Rational use of collected medical data to provide effective solutions for diagnosing diseases is the key task of current research.

当前癌症药物种类繁多，例如，肺癌的常用药品有1187种，胃癌的常用药品有754种，结肠癌的常用药品有503种等。间接说明了当前医疗水平有限，对于癌症是没有特效药的，只能尝试多种药物，用药不当不仅病情不会好转，花费大量金钱，还会浪费病人宝贵的时间，耽误病情，错过最佳治疗时机。针对癌症药物的用药粒度选择提出使用多粒度决策模型，如果以往患者用药数据能为当前病人提供有效的用药依据，那对于病人病情的康复意义非凡。Currently, there are many types of cancer drugs. For example, there are 1187 kinds of commonly used drugs for lung cancer, 754 kinds of commonly used drugs for gastric cancer, and 503 kinds of commonly used drugs for colon cancer. It indirectly shows that the current medical level is limited. There is no specific medicine for cancer, and you can only try a variety of drugs. Improper medication will not only not improve the condition, but also cost a lot of money. It will also waste the patient's precious time, delay the disease, and miss the best treatment. opportunity. A multi-granularity decision-making model is proposed for the drug granularity selection of cancer drugs. If the previous patient medication data can provide an effective medication basis for the current patient, it is of great significance to the recovery of the patient's condition.

发明内容SUMMARY OF THE INVENTION

本发明的目的是提供一种基于多粒度决策模型的最优用药粒度的计算方法，能够将癌症患者的用药粒度选择问题转化为多粒度决策模型选取最优粒度问题，并优化最优用药粒度计算方法，根据以往患者用药数据能为当前病人提供有效的用药依据，为医生选取抗癌药物提供自动化的辅助工具。The purpose of the present invention is to provide a method for calculating the optimal granularity of medication based on a multi-granularity decision model, which can transform the problem of selecting the granularity of medication for cancer patients into the problem of selecting the optimal granularity by a multi-granularity decision model, and optimize the calculation of the optimal granularity of medication. The method can provide effective medication basis for current patients and an automated auxiliary tool for doctors to select anticancer drugs according to the previous patient medication data.

本发明采用下述技术方案：The present invention adopts following technical scheme:

一种基于多粒度决策模型的最优用药粒度的计算方法，包括以下具体步骤：A method for calculating an optimal medication granularity based on a multi-granularity decision-making model, comprising the following specific steps:

A：在单粒度层次的用药粒度决策模型的基础上，通过对药物的用药剂量取不同的粒度层次，在癌症患者的用药系统中构建多粒度层次的用药粒度决策模型；然后进入步骤B；A: On the basis of the medication granularity decision-making model at the single granularity level, a multi-granularity medication granularity decision-making model is constructed in the medication system of cancer patients by taking different granularity levels for the drug dosage; then proceed to step B;

B：判断癌症患者病重程度，若癌症患者为Ⅰ期和Ⅱ期患者则选取全局最优用药粒度，进入步骤C；若癌症患者为Ⅲ期和Ⅳ期患者则选取局部最优用药粒度，进入步骤F；B: Determine the severity of the cancer patient. If the cancer patient is in stage I and II, select the global optimal medication granularity and go to step C; if the cancer patient is in stage III and IV, select the local optimal medication granularity and enter step F;

C：根据患者数据的条数进行判断，若患者数据条数小于等于5000条，进入步骤D；若新增患者数据大于5000条，则进入步骤E；其中，患者数据包括对象O、条件属性P和决策属性d；C: Judging according to the number of patient data, if the number of patient data is less than or equal to 5000, go to step D; if the new patient data is greater than 5000, go to step E; wherein, the patient data includes object O, condition attribute P and decision attribute d;

D：使用协调性方法计算全局最优用药粒度；D: Use the coordination method to calculate the global optimal medication granularity;

E：使用树形结构方法计算全局最优用药粒度；E: Use the tree structure method to calculate the global optimal medication granularity;

F：根据患者数据的条数进行判断，如果患者数据的条数小于等于1000条，则进入步骤G；如果患者数据的条数大于1000条，则进入步骤H；F: Judging according to the number of patient data, if the number of patient data is less than or equal to 1000, go to step G; if the number of patient data is greater than 1000, go to step H;

G：利用串行方法计算局部最优用药粒度；G: Use the serial method to calculate the local optimal drug granularity;

H：利用并行方法计算局部最优用药粒度。H: Use parallel methods to calculate the local optimal medication granularity.

所述的步骤A中：In the described step A:

单粒度层次的用药粒度决策模型是一个二元组S＝(O，P∪{d})；其中，O为对象，是癌症患者的集合，用O＝{x1，x2，…，xn}表示，其中x1为患者1，x2为患者2，以此类推,xn为患者n；P为条件属性，包括患者的基本情况和用药量，P＝{a1，a2，…，an}，分别代表不同的条件属性，决策属性d表示药物治疗效果，用Y或N表示，Y表示治疗有效，N表示治疗无效；The single-granularity-level medication granularity decision-making model is a two-tuple S=(O, P∪{d}); among them, O is the object, which is the set of cancer patients, and O={x 1 , x 2 , ..., x n } represents, where x 1 is patient 1, x 2 is patient 2, and so on, x n is patient n; P is a condition attribute, including the patient's basic condition and dosage, P={a 1 , a 2 , ..., a n }, represent different condition attributes respectively, the decision attribute d represents the effect of drug treatment, which is represented by Y or N, Y represents effective treatment, and N represents ineffective treatment;

在单粒度层次的用药粒度决策模型S＝(O，P∪{d})中，通过对条件属性集合P中药物用量取不同的观测值得到多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})，其中k表示层数，包含了所能构造的所有粒度层；多粒度层次的用药粒度决策模型所构造的粒度层次数是I，由细粒度向粗粒度逐层构造，对象O＝{x1，x2，…，xn}，每一层的属性集记为

Figure GDA0003697453700000021
1≤k≤I，
Figure GDA0003697453700000022
分别表示第k层用药粒度决策模型中的第1个条件属性、第2个条件属性、…、第m个条件属性。In the single-granularity-level medication granularity decision-making model S=(O, P∪{d}), the multi-granularity-level medication granularity decision-making model Sk = (O , P k ∪{d}), where k represents the number of layers, including all the granularity layers that can be constructed; the number of granularity layers constructed by the multi-granularity level of medication granularity decision model is I, from fine-grained to coarse-grained layer by layer Construct, object O={x 1 , x 2 ,..., x n }, the attribute set of each layer is denoted as
Figure GDA0003697453700000021
1≤k≤I,
Figure GDA0003697453700000022
respectively represent the first condition attribute, the second condition attribute, ... and the mth condition attribute in the k-th layer of medication granularity decision-making model.

所述的步骤A中所构建的多粒度层次的用药粒度决策模型中，条件属性P包括患者的基本情况和用药量，P＝{a1，a2，…，a10}，依次为病重程度、性别、年龄、药物A、药物B、药物C、药物D、药物E、药物F和药物G；多粒度层次的用药粒度决策模型的层数k为3，将患者的药量用片或粒表示时，得到第一层用药粒度决策模型S1＝(O，P1∪{d})；当患者的药物用量用盒或者瓶表示时，得到第二层用药粒度决策模型S2＝(O，P2∪{d})；当患者的药物用量用疗程表示时，得到第三层用药粒度决策模型S3＝(O，P3∪{d})。In the multi-granularity level medication granularity decision-making model constructed in step A, the condition attribute P includes the patient's basic condition and medication dosage, P={a 1 , a 2 , ..., a 10 }, followed by serious illness Degree, gender, age, drug A, drug B, drug C, drug D, drug E, drug F and drug G; the number of layers k of the multi-granularity-level drug granularity decision model is 3, and the patient's drug dose is divided into tablets or drugs. When represented by granules, the first-level drug granularity decision model S 1 =(O, P 1 ∪{d}) is obtained; when the patient’s drug dosage is represented by boxes or bottles, the second-level drug granularity decision model S 2 =( O, P 2 ∪{d}); when the patient's drug dosage is represented by a course of treatment, the third-level drug granularity decision model S 3 =(O, P 3 ∪{d}) is obtained.

所述的步骤D包括以下具体步骤：Described step D includes the following specific steps:

D1：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按条件属性进行划分，设对象x1的每个条件属性

Figure GDA0003697453700000023
和对象x2的每个条件属性
Figure GDA0003697453700000031
一一对应相同，则把x1和x2划分为一类，以此类推，记录划分结果并得到每一层划分结果的集合，记为RO；D1: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to conditional attributes, and each conditional attribute of object x 1 is set
Figure GDA0003697453700000023
and each condition property of object x 2
Figure GDA0003697453700000031
The one-to-one correspondence is the same, then divide x 1 and x 2 into one class, and so on, record the division result and obtain a set of division results for each layer, denoted as R O ;

D2：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按决策属性进行划分，设对象x1、x2和x3的药物治疗有效，而对象x4和x5的药物治疗无效，则按照决策属性进行划分，将x1、x2和x3划分为一类，将x4和x5划分为一类，依次类推，记录划分结果，得到所有对象在每一层上划分的集合，记为Rr；D2: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), divide each object according to the decision attribute, and set the objects x 1 , x 2 and x 3 The drug treatment is effective, but the drug treatment of the objects x 4 and x 5 is invalid, then it is divided according to the decision attribute, and x 1 , x 2 and x 3 are divided into one class, x 4 and x 5 are divided into one class, and then By analogy, record the division result, and get the set of all objects divided on each layer, denoted as R r ;

D3：比较划分结果RO和划分结果Rr，如果

Figure GDA0003697453700000034
则判断此层决策模型是协调的；如果
Figure GDA0003697453700000036
则判断此层决策模型是不协调；首次出现不协调的粒度层数的上一层即为全局最优用药粒度；RO表示对象按照属性划分所得到的集合，Rr表示对象按照药物治疗效果划分所得到的集合。D3: Compare the division result R O and the division result R r , if
Figure GDA0003697453700000034
Then it is judged that the decision model of this layer is coordinated; if
Figure GDA0003697453700000036
Then it is judged that the decision-making model at this layer is incongruous; the upper layer of the granularity layer where inconsistency occurs for the first time is the global optimal drug granularity; RO represents the set obtained by dividing the object according to the attribute, and R r represents the object according to the drug treatment effect. Divide the resulting set.

所述的步骤E包括以下具体步骤：Described step E includes following concrete steps:

用树形结构求解全局最优粒度，将步骤A得到的多粒度层次的用药粒度决策模型转换成树形存储，多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中，多粒度层次的用药粒度决策模型的每一层决策模型都是由多个树组成的森林，树的每一层表示每一个条件属性，叶子结点的下方存储的是对象O和决策属性d，根据同一叶子结点下方决策属性d的一致性判断此层决策模型是否协调，首次出现不协调的粒度层数的上一层即为全局最优用药粒度。Use the tree structure to solve the global optimal granularity, convert the multi-granularity level of medication granularity decision model obtained in step A into tree storage, and the multi-granularity level of medication granularity decision model S k =(O, P k ∪{d}) In the multi-granularity level of drug granularity decision-making model, each layer of decision-making model is a forest composed of multiple trees. Each layer of the tree represents each condition attribute, and the object O and decision-making attributes are stored below the leaf node. d. According to the consistency of the decision attribute d under the same leaf node, it is judged whether the decision model of this layer is coordinated, and the upper layer of the granularity layer where the inconsistency occurs for the first time is the global optimal medication granularity.

所述的步骤G包括以下具体步骤：Described step G includes following specific steps:

G1：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按条件属性进行划分，假设对象x1的每个属性

Figure GDA0003697453700000032
和对象x2的每个属性
Figure GDA0003697453700000033
一一对应相同，则把x1和x2划分为一类，以此类推，记录划分结果并得到每个对象划分结果的集合，记为
Figure GDA0003697453700000035
即对象x在第k层用药粒度决策模型上按条件属性P划分的结果；G1: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to conditional attributes, assuming that each attribute of the object x 1
Figure GDA0003697453700000032
and each property of object x 2
Figure GDA0003697453700000033
The one-to-one correspondence is the same, then divide x 1 and x 2 into one class, and so on, record the division result and get the set of division results for each object, denoted as
Figure GDA0003697453700000035
That is, the result of object x being divided by conditional attribute P on the k-th layer of medication granularity decision model;

G2：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{x})中的每一层上，把各个对象按决策属性进行划分，假设对象x1和x2的药物治疗有效，而对象x3，x4和x5的药物治疗无效，则按照决策属性进行划分，将x1和x2划分为一类，将x3，x4和x5划分为一类，依次类推，记录划分结果并得到各个对象所在的集合，记为[x]d，即对象x按决策属性d划分的结果；G2: On each layer of the multi-granularity-level medication granularity decision model Sk = (O, P k ∪{x}), each object is divided according to decision attributes, assuming that the drug treatment of objects x 1 and x 2 Effective, and the drug treatment of objects x 3 , x 4 and x 5 is ineffective, then divide according to the decision attribute, divide x 1 and x 2 into one class, divide x 3 , x 4 and x 5 into one class, in turn By analogy, record the division result and get the set where each object is located, denoted as [x] d , that is, the result of the division of object x according to the decision attribute d;

G3：比较步骤2中各个对象x所在集合[x]d是否包含步骤1中各个对象x所在集合

Figure GDA0003697453700000041
若
Figure GDA0003697453700000042
则判断该对象x所在层的决策模型是协调的；若
Figure GDA0003697453700000043
则判断该对象x所在层的决策模型是不协调；首次出现不协调的粒度层数的上一层即为对象x的局部最优用药粒度，即
Figure GDA0003697453700000044
且
Figure GDA0003697453700000045
时，第k层粒度是关于对象x的局部最优用药粒度；并由此得到所有对象的局部最优用药粒度的集合。G3: Compare whether the set [x] d of each object x in step 2 contains the set of each object x in step 1
Figure GDA0003697453700000041
like
Figure GDA0003697453700000042
Then it is judged that the decision model of the layer where the object x is located is coordinated; if
Figure GDA0003697453700000043
Then it is judged that the decision model of the layer where the object x is located is incongruous; the upper layer of the granularity level where inconsistency first appears is the local optimal drug granularity of the object x, that is,
Figure GDA0003697453700000044
and
Figure GDA0003697453700000045
When , the granularity of the kth layer is the local optimal medication granularity of object x; and thus the set of local optimal medication granularity of all objects is obtained.

所述的步骤H中，根据所使用的计算机的处理器的内核数量N，将若干条患者数据平均划分为N组，然后所使用的计算机的处理器的N个内核分别按照步骤G所述的方法，分别计算对应的一组患者数据，最终得到所有对象的局部最优用药粒度的集合。In the described step H, according to the number of cores N of the processor of the computer used, several pieces of patient data are evenly divided into N groups, and then the N cores of the processor of the computer used are respectively as described in step G. method, respectively calculate the corresponding group of patient data, and finally obtain a set of local optimal drug granularity for all subjects.

本发明通过利用多粒度决策模型来解决癌症患者的用药选取问题；针对不同情况，给出了计算全局最优用药粒度的两种方法，通过分析癌症患者病重程度，在为Ⅰ期和Ⅱ期患者所建立的多粒度决策模型中，选用全局最优用药粒度；在为Ⅲ期和Ⅳ期患者所建立的多粒度决策模型中，选用局部最优用药粒度；同时，本发明还在新增患者数据条数大于5000条的情况下，选取树形结构求解全局最优用药粒度的方法，提升了时间效率；对选取局部最优用药粒度的过程做了并行，解决了时间过长的问题，节省了将近一半的时间。The invention solves the problem of drug selection of cancer patients by using a multi-granularity decision-making model; according to different situations, two methods for calculating the global optimal drug granularity are provided. In the multi-granularity decision-making model established by the patient, the global optimal medication granularity is selected; in the multi-granularity decision-making model established for stage III and IV patients, the local optimal medication granularity is selected; at the same time, the present invention also adds new patients. When the number of data items is greater than 5000, the tree structure is selected to solve the global optimal medication granularity method, which improves the time efficiency; the process of selecting the local optimal medication granularity is parallelized, which solves the problem of excessive time and saves money. nearly half the time.

对于抗癌药物的选取，医生一般基于现有的经验，存在一定的误诊率，不仅造成病人时间和金钱的浪费，还会延误病情。本文基于多粒度决策模型，提出了两种选取全局最优用药粒度的方法，并给出了具体算法，并行了选取局部最优粒度的过程。结果表明：新增患者数据量大于5000的情况下，树形结构选择全局最优粒度的方法在时间性能上较协调性的方法有很大的优势，并行选取局部最优粒度所用时间大约为串行所用时间的一半，为医生选取抗癌药物提供了自动化的辅助工具。For the selection of anti-cancer drugs, doctors generally based on the existing experience, there is a certain rate of misdiagnosis, which not only causes a waste of patients' time and money, but also delays the treatment. Based on the multi-granularity decision-making model, this paper proposes two methods for selecting the global optimal medication granularity, and gives a specific algorithm to parallelize the process of selecting the local optimal granularity. The results show that when the amount of new patient data is greater than 5000, the method of selecting the global optimal granularity in the tree structure has a great advantage over the coordinated method in terms of time performance, and the time required to select the local optimal granularity in parallel is approximately It takes half of the time spent on the line, providing an automated aid for doctors to choose anti-cancer drugs.

附图说明Description of drawings

图1为本发明生成的第一层用药粒度决策模型示意图；Fig. 1 is the schematic diagram of the first-layer medication granularity decision model generated by the present invention;

图2为本发明生成的第二层用药粒度决策模型示意图；Fig. 2 is the schematic diagram of the second-layer medication granularity decision model generated by the present invention;

图3为本发明生成的第三层用药粒度决策模型示意图；Fig. 3 is the schematic diagram of the third-layer medication granularity decision model generated by the present invention;

图4为本发明提出的病患数据决策说明示意图；4 is a schematic diagram illustrating the patient data decision-making proposed by the present invention;

图5为本发明构造的树形结构说明示意图；5 is a schematic diagram illustrating the tree structure constructed by the present invention;

图6为分别使用树形结构方法和协调性方法计算系统最优粒度时间对比示意图；FIG. 6 is a schematic diagram showing the comparison of optimal granularity time of the computing system using the tree structure method and the coordination method respectively;

图7为分别使用串行方法和并行方法计算局部最优粒度时间对比示意图；FIG. 7 is a schematic diagram of the time comparison of calculating the local optimal granularity using the serial method and the parallel method respectively;

图8为使用串行方法和并行方法计算局部最优粒度时间对比折线图；Figure 8 is a line graph showing the time comparison of calculating the local optimal granularity using the serial method and the parallel method;

图9为本发明的流程示意图。FIG. 9 is a schematic flowchart of the present invention.

具体实施方式Detailed ways

以下结合附图和实施例对本发明作以详细的描述：Below in conjunction with accompanying drawing and embodiment, the present invention is described in detail:

如图1至图9所示，本发明所述的基于多粒度决策模型的最优用药粒度的计算方法，包括以下步骤：As shown in Fig. 1 to Fig. 9, the method for calculating the optimal medication granularity based on the multi-granularity decision-making model of the present invention includes the following steps:

A：在单粒度层次的用药粒度决策模型的基础上，通过对药物的用药剂量取不同的粒度层次，在癌症患者的用药系统中构建多粒度层次的用药粒度决策模型；然后进入步骤B；A: On the basis of the medication granularity decision-making model at the single granularity level, a multi-granularity medication granularity decision-making model is constructed in the medication system of cancer patients by taking different granularity levels for the drug dosage; then proceed to step B;

由于市面上的抗癌药物制剂大约有一千多种，而医生对于药物的选取只能基于现有的知识与经验，缺乏自动化的辅助工具。癌症患者的用药粒度选择问题可以抽象为多粒度决策模型选取最优粒度问题，对癌症患者的药物用量取不同的观测值就可以得到不同粒度层次的用药粒度决策模型。Since there are about a thousand anticancer drug preparations on the market, doctors can only select drugs based on existing knowledge and experience, and lack automated auxiliary tools. The problem of drug granularity selection for cancer patients can be abstracted as a multi-granularity decision-making model to select the optimal granularity problem. Taking different observation values of the drug dosage of cancer patients can obtain drug granularity decision models with different granularity levels.

本发明中所构建的单粒度层次的用药粒度决策模型是一个二元组S＝(O，P∪{d})；其中，O为对象，是癌症患者的集合，可用O＝{x1，x2，…，xn}表示，其中x1为患者1，x2为患者2，以此类推；P＝{a1，a2，…，an}，分别代表不同的条件属性，决策属性d表示药物治疗效果，用Y或N表示，Y表示治疗有效，N表示治疗无效。The single-granularity-level medication granularity decision-making model constructed in the present invention is a two-tuple S=(O, P∪{d}); wherein, O is the object, which is a collection of cancer patients, and O={x 1 , x 2 ,..., x n }, where x 1 is patient 1, x 2 is patient 2, and so on; P={a 1 , a 2 ,..., a n }, representing different conditional attributes, decision-making The attribute d represents the effect of drug treatment, which is represented by Y or N, where Y indicates that the treatment is effective, and N indicates that the treatment is ineffective.

在单粒度层次的用药粒度决策模型S＝(O，P∪{d})中，通过对条件属性集合P中药物用量取不同的观测值得到多粒度层次的用药粒度决策模型Sk＝(P，Pk∪{d})，其中k表示层数，包含了所能构造的所有粒度层；多粒度层次的用药粒度决策模型所构造的粒度层次数是I，由细粒度向粗粒度逐层构造，对象O＝{x1，x2，…，xn}，每一层的属性集记为

Figure GDA0003697453700000051
1≤k≤I，
Figure GDA0003697453700000052
分别表示第k层用药粒度决策模型中的第1个条件属性、第2个条件属性、…、第m个条件属性。In the single-granularity-level medication granularity decision-making model S=(O, P∪{d}), the multi-granularity-level medication granularity decision-making model Sk = (P , P k ∪{d}), where k represents the number of layers, including all the granularity layers that can be constructed; the number of granularity layers constructed by the multi-granularity level of medication granularity decision model is I, from fine-grained to coarse-grained layer by layer Construct, object O={x 1 , x 2 ,..., x n }, the attribute set of each layer is denoted as
Figure GDA0003697453700000051
1≤k≤I,
Figure GDA0003697453700000052
respectively represent the first condition attribute, the second condition attribute, ... and the mth condition attribute in the k-th layer of medication granularity decision-making model.

本实施例中，步骤A中所构建的多粒度层次的用药粒度决策模型中，条件属性P包括患者的基本情况和用药量，P＝{a1，a2，…，a10}，依次为病重程度、性别、年龄、药物A、药物B、药物C、药物D、药物E、药物F和药物G；多粒度层次的用药粒度决策模型的层数k为3，；将患者的药量用最小单位即片或粒表示时，可以得到第一层用药粒度决策模型S1＝(O，P1∪{d})，如图1所示。当患者的药物用量用盒或者瓶表示时，我们可以得到第二层用药粒度决策模型S2＝(O，P2∪{d})，如图2所示。当患者的药物用量用疗程表示时，我们可以得到第三层用药粒度决策模型S3＝(O，P3∪{d})，如图3所示。In this embodiment, in the multi-granularity-level medication granularity decision-making model constructed in step A, the condition attribute P includes the patient's basic condition and medication dosage, P={a 1 , a 2 , . . . , a 10 }, in the order of Severity, gender, age, drug A, drug B, drug C, drug D, drug E, drug F, and drug G; the number of layers k of the multi-granularity-level medication granularity decision model is 3; When represented by the smallest unit, namely tablet or grain, the first-level drug granularity decision model S 1 =(O, P 1 ∪{d}) can be obtained, as shown in Fig. 1 . When the patient's drug dosage is represented by a box or a bottle, we can obtain the second-level drug granularity decision model S 2 =(O, P 2 ∪{d}), as shown in Figure 2. When the patient's drug dosage is represented by the course of treatment, we can obtain the third-level drug granularity decision model S 3 =(O, P 3 ∪{d}), as shown in Figure 3.

设多粒度层次的用药粒度决策模型所构造的粒度层次数是I，由细粒度向粗粒度逐层构造，对象O＝{x1，x2，…，xn}，每一层的属性集记为

Figure GDA0003697453700000053
1≤k≤I，
Figure GDA0003697453700000054
分别表示第k层用药粒度决策模型中的第1个条件属性、第2个条件属性、…、第m个条件属性，即得到了多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})。Assume that the number of granularity levels constructed by the multi-granularity level of medication granularity decision model is I, and it is constructed layer by layer from fine-grained to coarse-grained, and the object O={x 1 , x 2 , . marked as
Figure GDA0003697453700000053
1≤k≤I,
Figure GDA0003697453700000054
respectively represent the first condition attribute, the second condition attribute, ..., the mth condition attribute in the drug granularity decision-making model of the k-th layer, that is, the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}).

B：判断癌症患者病重程度，若癌症患者为Ⅰ期和Ⅱ期患者则选取全局最优用药粒度，进入步骤C；若癌症患者为Ⅲ期和Ⅳ期患者则选取局部最优用药粒度，进入步骤F；B: Determine the severity of the cancer patient. If the cancer patient is in stage I and II, select the global optimal medication granularity and go to step C; if the cancer patient is in stage III and IV, select the local optimal medication granularity and enter step F;

C：根据患者数据的条数进行判断，若患者数据条数小于等于5000条，进入步骤D；若新增患者数据大于5000条，则进入步骤E；其中，患者数据包括对象O、条件属性P和决策属性d；C: Judging according to the number of patient data, if the number of patient data is less than or equal to 5000, go to step D; if the new patient data is greater than 5000, go to step E; wherein, the patient data includes object O, condition attribute P and decision attribute d;

D：使用协调性方法计算全局最优用药粒度；D: Use the coordination method to calculate the global optimal medication granularity;

D1：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按条件属性进行划分，假设对象x1的每个条件属性

Figure GDA0003697453700000061
和对象x2的每个条件属性
Figure GDA0003697453700000062
一一对应相同，则把x1和x2划分为一类，以此类推，记录划分结果并得到每一层划分结果的集合，记为RO。D1: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to conditional attributes, assuming that each conditional attribute of object x 1
Figure GDA0003697453700000061
and each condition property of object x 2
Figure GDA0003697453700000062
The one-to-one correspondence is the same, then divide x 1 and x 2 into one class, and so on, record the division results and obtain a set of division results for each layer, denoted as R O .

D2：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按决策属性进行划分，假设对象x1、x2和x3的药物治疗有效，而对象x4和x5的药物治疗无效，则按照决策属性进行划分，将x1、x2和x3划分为一类，将x4和x5划分为一类，依次类推，记录划分结果，得到所有对象在每一层上划分的集合，记为Rr。D2: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to decision attributes, assuming that the objects x 1 , x 2 and x 3 The drug treatment is effective, but the drug treatment of the objects x 4 and x 5 is invalid, then it is divided according to the decision attribute, and x 1 , x 2 and x 3 are divided into one class, x 4 and x 5 are divided into one class, and then By analogy, record the division result, and obtain the set of all objects divided on each layer, denoted as R r .

D3：比较划分结果RO和划分结果Rr，如果

Figure GDA0003697453700000063
则判断此层决策模型是协调的；如果
Figure GDA0003697453700000064
则判断此层决策模型是不协调；首次出现不协调的粒度层数的上一层即为全局最优用药粒度；RO表示对象按照属性划分所得到的集合，Rr表示对象按照药物治疗效果划分所得到的集合。D3: Compare the division result R O and the division result R r , if
Figure GDA0003697453700000063
Then it is judged that the decision model of this layer is coordinated; if
Figure GDA0003697453700000064
Then it is judged that the decision-making model at this layer is incongruous; the upper layer of the granularity layer where inconsistency occurs for the first time is the global optimal drug granularity; RO represents the set obtained by dividing the object according to the attribute, and R r represents the object according to the drug treatment effect. Divide the resulting set.

综上，步骤D1至步骤D3中，在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，如果

Figure GDA0003697453700000065
则称多粒度层次的用药粒度决策模型的此层决策模型是协调的，否则称多粒度层次的用药粒度决策模型的此层决策模型是不协调的，并以此确定全局最优用药粒度。To sum up, in steps D1 to D3, on each layer of the multi-granularity-level medication granularity decision-making model S k =(O, P k ∪{d}), if
Figure GDA0003697453700000065
The decision model at this layer of the multi-granularity level of medication granularity decision-making model is said to be coordinated, otherwise the decision model at this level of the multi-granularity level of medication granularity decision-making model is said to be uncoordinated, and the global optimal medication granularity is determined accordingly.

E：使用树形结构方法计算全局最优用药粒度。E: Use the tree structure method to calculate the global optimal medication granularity.

用树形结构求解全局最优粒度，将步骤A得到的多粒度层次的用药粒度决策模型转换成树形存储，转换过程如图4至图5所示，在图4所示的多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中，O中存在5个对象，分别为x1、x2、x3、x4和x5；条件属性是a1，a2和a3，决策属性用d表示。Use the tree structure to solve the global optimal granularity, and convert the multi-granularity-level drug granularity decision model obtained in step A into tree-shaped storage. The conversion process is shown in Figure 4 to Figure 5. In the medication granularity decision model Sk = (O, P k ∪{d}), there are 5 objects in O, namely x 1 , x 2 , x 3 , x 4 and x 5 ; the condition attributes are a 1 , a 2 and a 3 , the decision attribute is denoted by d.

观察图5树形结构可得出：构造完成的多粒度层次的用药粒度决策模型的每一层都是由多个树组成的森林，树形选取方法和协调性方法的不同之处就在于划分结果和协调性可以同时获得，节省了大量时间。树的每一层表示每一个条件属性，由上至下，依次为条件属性a1，a2和a3，叶子结点的下方存储的是对象O和决策属性d。根据同一叶子结点下方决策属性d的一致性判断此层决策模型是否协调，首次出现不协调的粒度层数的上一层即为全局最优用药粒度。Observing the tree structure in Figure 5, it can be concluded that each layer of the constructed multi-granularity-level medication granularity decision model is a forest composed of multiple trees. The difference between the tree selection method and the coordination method lies in the division Results and coordination can be obtained at the same time, saving a lot of time. Each layer of the tree represents each condition attribute, from top to bottom, the condition attributes a 1 , a 2 and a 3 in sequence, and the object O and the decision attribute d are stored below the leaf nodes. According to the consistency of the decision attribute d under the same leaf node, it is judged whether the decision model of this layer is coordinated.

鉴于患者数据属于持续增长类型，本发明主要讨论的是条件属性不变，对象批量增长的情况下选取全局最优粒度。在现有患者数据的基础上，批量增加数据，分别使用步骤D中协调性的计算全局最优用药粒度的方法和步骤E中使用树形结构方法计算全局最优用药粒度的方法，时间性能上的结果对比如图6所示。Considering that the patient data belongs to the type of continuous growth, the present invention mainly discusses the selection of the global optimal granularity under the condition that the conditional attributes remain unchanged and the objects grow in batches. On the basis of the existing patient data, data is added in batches, and the method of calculating the global optimal medication granularity in a coordinated manner in step D and the method of calculating the global optimal medication granularity by using a tree structure method in step E are respectively used. The results are compared as shown in Figure 6.

结果表明：两种算法计算结果一致的同时，随着对象的增加，两种求解全局最优粒度的方法所用时间均呈增长趋势。当增加对象个数为5000时，两种方法所用时间相差不多，仅为1秒左右，但是当增加对象个数大于5000，且越来越大时，在时间性能上，树形结构方法的优势就显示出来了。这为癌症患者选择用药粒度节省了大量宝贵时间。The results show that while the calculation results of the two algorithms are consistent, with the increase of the objects, the time used by the two methods to solve the global optimal granularity is increasing. When the number of added objects is 5000, the time used by the two methods is similar, only about 1 second, but when the number of added objects is greater than 5000 and becomes larger and larger, in terms of time performance, the tree structure method has the advantage is displayed. This saves a lot of valuable time for cancer patients in choosing medication granularity.

F：根据患者数据的条数进行判断，如果患者数据的条数小于等于1000条，则进入步骤G；如果患者数据的条数大于1000条，则进入步骤H；F: Judging according to the number of patient data, if the number of patient data is less than or equal to 1000, go to step G; if the number of patient data is greater than 1000, go to step H;

G：利用串行方法计算局部最优用药粒度：G: Use the serial method to calculate the local optimal medication granularity:

G1：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按条件属性进行划分，假设对象x1的每个属性

Figure GDA0003697453700000071
和对象x2的每个属性
Figure GDA0003697453700000072
一一对应相同，则把x1和x2划分为一类，以此类推，记录划分结果并得到每个对象划分结果的集合，记为
Figure GDA0003697453700000073
即对象x在第k层用药粒度决策模型上按条件属性P划分的结果。G1: On each layer of the multi-granularity-level medication granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to conditional attributes, assuming that each attribute of the object x 1
Figure GDA0003697453700000071
and each property of object x 2
Figure GDA0003697453700000072
The one-to-one correspondence is the same, then divide x 1 and x 2 into one class, and so on, record the division result and get the set of division results for each object, denoted as
Figure GDA0003697453700000073
That is, the result of object x being divided by conditional attribute P on the k-th layer of medication granularity decision model.

G2：在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中的每一层上，把各个对象按决策属性进行划分，假设对象x1和x2的药物治疗有效，而对象x3，x4和x5的药物治疗无效，则按照决策属性进行划分，将x1和x2划分为一类，将x3，x4和x5划分为一类，依次类推，记录划分结果并得到各个对象所在的集合，记为[x]d，即对象x按决策属性d划分的结果。G2: On each layer of the multi-granularity-level drug granularity decision-making model Sk = (O, P k ∪{d}), each object is divided according to decision attributes, assuming the drug treatment of objects x 1 and x 2 Effective, and the drug treatment of objects x 3 , x 4 and x 5 is ineffective, then divide according to the decision attribute, divide x 1 and x 2 into one class, divide x 3 , x 4 and x 5 into one class, in turn By analogy, record the division result and obtain the set where each object is located, denoted as [x] d , that is, the result of the division of object x according to the decision attribute d.

G3：比较步骤2中各个对象x所在集合[x]d是否包含步骤1中各个对象x所在集合

Figure GDA0003697453700000074
若
Figure GDA0003697453700000075
则判断该对象x所在层的决策模型是协调的；若
Figure GDA0003697453700000076
则判断该对象x所在层的决策模型是不协调；首次出现不协调的粒度层数的上一层即为对象x的局部最优用药粒度，即
Figure GDA0003697453700000077
且
Figure GDA0003697453700000078
时，第k层粒度是关于对象x的局部最优用药粒度；并由此得到所有对象的局部最优用药粒度的集合。G3: Compare whether the set [x] d of each object x in step 2 contains the set of each object x in step 1
Figure GDA0003697453700000074
like
Figure GDA0003697453700000075
Then it is judged that the decision model of the layer where the object x is located is coordinated; if
Figure GDA0003697453700000076
Then it is judged that the decision model of the layer where the object x is located is incongruous; the upper layer of the granularity level where inconsistency first appears is the local optimal drug granularity of the object x, that is,
Figure GDA0003697453700000077
and
Figure GDA0003697453700000078
When , the granularity of the kth layer is the local optimal medication granularity of object x; and thus the set of local optimal medication granularity of all objects is obtained.

综上，步骤G1至步骤G3中，根据各个对象的协调性选择局部最优用药粒度，即在多粒度层次的用药粒度决策模型Sk＝(O，Pk∪{d})中，对于x∈O，给定k，1≤k≤I，若

Figure GDA0003697453700000081
且
Figure GDA0003697453700000082
即对象x所在的第k层的决策模型是协调的且第k+1层的决策模型是不协调的，则判断第k层粒度是关于对象x的局部最优用药粒度，并由此得到所有对象的局部最优用药粒度的集合。To sum up, in steps G1 to G3, the local optimal medication granularity is selected according to the coordination of each object, that is, in the multi-granularity level medication granularity decision model Sk = (O, P k ∪{d}), for x ∈O, given k, 1≤k≤I, if
Figure GDA0003697453700000081
and
Figure GDA0003697453700000082
That is, the decision model of the kth layer where the object x is located is coordinated and the decision model of the k+1th layer is not coordinated, then it is judged that the kth layer granularity is the local optimal medication granularity for the object x, and thus all A collection of locally optimal medication granularities for an object.

H：利用并行方法计算局部最优用药粒度；H: Use parallel method to calculate the local optimal drug granularity;

根据所使用的计算机的处理器的内核数量N，将若干条患者数据平均划分为N组，然后所使用的计算机的处理器的N个内核分别按照步骤G所述的方法，分别计算对应的一组患者数据，最终得到所有对象的局部最优用药粒度的集合。例如所使用的计算机的处理器的内核数量为4，则将若干条患者数据平均划分为4组，然后所使用的计算机的处理器的4个内核分别按照步骤G所述的方法，分别计算对应的一组患者数据，即计算机的处理器的每一个内核分别计算对应的一组患者数据。According to the number of cores N of the processor of the computer used, several pieces of patient data are evenly divided into N groups, and then the N cores of the processor of the computer used respectively calculate the corresponding one according to the method described in step G. Group patient data, and finally obtain a set of local optimal medication granularity for all subjects. For example, the number of cores of the processor of the computer used is 4, then several pieces of patient data are equally divided into 4 groups, and then the 4 cores of the processor of the computer used are calculated according to the method described in step G respectively. A set of patient data, that is, each core of the processor of the computer calculates a corresponding set of patient data respectively.

通过依次增加对象的个数，使用步骤G和步骤H分别求取新增对象的局部最优粒度，结果表明，随着对象个数的增加，步骤G和步骤H两种求解局部最优粒度的方式所耗时间均呈增长趋势。但是总体而言，步骤H计算局部最优粒度所花费的时间比步骤G计算大约节省了一半的时间，增加的对象个数越大，节省的时间越多，如图7所示。By increasing the number of objects in turn, step G and step H are used to obtain the local optimal granularity of the newly added objects respectively. The time-consuming methods are increasing. But in general, the time spent in calculating the local optimal granularity in step H is about half the time spent in calculating in step G, and the greater the number of objects added, the greater the time savings, as shown in Figure 7.

如图8所示：批量增加对象的个数为1000时，步骤G求解局部最优粒度的时间本身就不多，所以步骤H效果并不明显，但是批量增加的对象越多，步骤G求解局部最优粒度所用的时间越多，步骤H的效果就越明显，由图8可得，步骤H求解局部最优粒度所用时间大约为步骤G求解局部最优粒度所用时间的一半，能够得出，当需要求解每个患者的最优用药粒度时，使用步骤H求解方式能节省大约一半的时间，早日为患者的治疗提供自动化的辅助工具。As shown in Figure 8: When the number of objects added in batches is 1000, the time for step G to solve the local optimal granularity itself is not much, so the effect of step H is not obvious, but the more objects are added in batches, step G solves the local optimal granularity. The more time it takes for the optimal particle size, the more obvious the effect of step H. It can be seen from Figure 8 that the time used to solve the local optimal particle size in step H is about half of the time used in step G to solve the local optimal particle size. It can be concluded that, When it is necessary to solve the optimal medication granularity of each patient, using the solution method of step H can save about half of the time, and provide an automated auxiliary tool for the treatment of patients as soon as possible.

本发明通过利用多粒度决策模型来解决癌症患者的用药选取问题；针对不同情况，给出了计算全局最优用药粒度的两种方法，通过分析癌症患者病重程度，在为Ⅰ期和Ⅱ期患者所建立的多粒度决策模型中，选用全局最优用药粒度；在为Ⅲ期和Ⅳ期患者所建立的多粒度决策模型中，选用局部最优用药粒度；同时，本发明还在新增患者数据条数大于5000条的情况下，选取树形结构求解全局最优用药粒度的方法，提升了时间效率；对选取局部最优用药粒度的过程做了并行，解决了时间过长的问题，节省了将近一半的时间。The invention solves the problem of drug selection of cancer patients by using a multi-granularity decision-making model; according to different situations, two methods for calculating the global optimal drug granularity are provided. In the multi-granularity decision-making model established by the patient, the global optimal medication granularity is selected; in the multi-granularity decision-making model established for stage III and IV patients, the local optimal medication granularity is selected; at the same time, the present invention also adds new patients. When the number of data items is greater than 5000, the tree structure is selected to solve the global optimal medication granularity method, which improves the time efficiency; the process of selecting the local optimal medication granularity is parallelized, which solves the problem of excessive time and saves money. nearly half the time.

对于抗癌药物的选取，医生一般基于现有的经验，存在一定的误诊率，不仅造成病人时间和金钱的浪费，还会延误病情。本文基于多粒度决策模型，提出了两种选取全局最优用药粒度的方法，并给出了具体算法，并行了选取局部最优粒度的过程。结果表明：新增患者数据量大于5000的情况下，树形结构选择全局最优粒度的方法在时间性能上较协调性的方法有很大的优势，并行选取局部最优粒度所用时间大约为串行所用时间的一半，为医生选取抗癌药物提供了自动化的辅助工具。For the selection of anti-cancer drugs, doctors generally based on the existing experience, there is a certain rate of misdiagnosis, which not only causes a waste of patients' time and money, but also delays the treatment. Based on the multi-granularity decision-making model, this paper proposes two methods for selecting the global optimal drug granularity, and gives a specific algorithm to parallelize the process of selecting the local optimal granularity. The results show that when the amount of newly added patient data is greater than 5000, the method of selecting the global optimal granularity in the tree structure has great advantages over the coordinated method in terms of time performance, and the time used to select the local optimal granularity in parallel is approximately It takes half of the time spent on the line, providing an automated aid for doctors to choose anti-cancer drugs.

Claims (4)

1. An optimal medication granularity calculation device based on a multi-granularity decision model is characterized in that: comprises a processor and a memory; the memory has a computer program stored therein, and the computer program, when executed by the processor, performs the following steps:
a: on the basis of the single-granularity-level medication granularity decision model, a multi-granularity-level medication granularity decision model is constructed in a medication system of a cancer patient by taking different granularity levels for the medication dosage of the medicament; then entering the step B;
in the step a, the single-granularity-level medication granularity decision model is a tuple S ═ (O, P ═ d }); wherein O is a group of cancer patients, and O ═ x is used 1 ，x 2 ，...，x n Denotes wherein x 1 Is patient 1, x 2 Patient 2, and so on, x n Is patient n; p is a condition attribute and comprises the basic condition and the dosage of the patient, and P is { a ═ a 1 ，a 2 ，...，a n The decision attribute d represents the treatment effect of the medicine and is represented by Y or N, Y represents the treatment effectiveness, and N represents the treatment ineffectiveness;
in the step a, in the constructed multi-granularity-level medication granularity decision model, the condition attribute P includes the basic condition and the dosage of the patient, and P ═ a 1 ，a 2 ，...，a 10 The disease severity, sex, age, drug A, drug B, drug C, drug D, drug E, drug F and drug G in sequence; the number k of layers of the multi-granularity-layer medicine-use granularity decision model is 3, and when the medicine amount of a patient is expressed by tablets or granules, a first-layer medicine-use granularity decision model S is obtained 1 ＝(O，P 1 U { d }); when the medicine dosage of the patient is represented by a box or a bottle, a second layer medicine dosage decision model S is obtained 2 ＝(O，P 2 U { d }); when the drug dosage of the patient is expressed by the course of treatment, a third layer of drug dosage granularity decision model S is obtained 3 ＝(O，P 3 ∪{d})；
In the single-granularity-level medication granularity decision model S ═ (O, Pu { d }), a multi-granularity-level medication granularity decision model S is obtained by taking different observed values for the drug dosage in the condition attribute set P k ＝(O，P k U { d }), wherein k represents the number of layers and contains all the granularity layers which can be constructed; the number of granularity layers constructed by the multi-granularity-layer medicine granularity decision model is I, the multi-granularity-layer medicine granularity decision model is constructed layer by layer from fine granularity to coarse granularity, and an object O is { x ═ x 1 ，x 2 ，...，x n H, attribute set of each layer is recorded as
Figure FDA0003697453690000011
Figure FDA0003697453690000012
Respectively representing the 1 st condition attribute, the 2 nd condition attribute, … and the mth condition attribute in the k-th layer medication granularity decision model;
b: judging the severity of the cancer patient, selecting the globally optimal drug granularity if the cancer patient is a stage I patient and a stage II patient, and entering the step C; if the cancer patient is a patient in stage III and stage IV, selecting local optimal drug granularity, and entering step F;
c: judging according to the number of the patient data, and entering the step D if the number of the patient data is less than or equal to 5000; if the number of newly added patient data is larger than 5000, entering the step E; wherein the patient data comprises an object O, a condition attribute P and a decision attribute d;
d: calculating the global optimal drug granularity by using a coordination method;
e: calculating the global optimal medication granularity by using a tree structure method;
the step E comprises the following specific steps:
solving the global optimum granularity by using a tree structure, converting the multi-granularity-level medication granularity decision model obtained in the step A into a tree storage, and using the multi-granularity-level medication granularity decision model S k ＝(O，P k In U { d }), each layer of a multi-granularity-level drug granularity decision model is a forest consisting of a plurality of trees, each layer of the tree represents each condition attribute, an object O and a decision attribute d are stored below leaf nodes, whether the decision model in the layer is coordinated or not is judged according to the consistency of the decision attribute d below the same leaf node, and the upper layer of the first-appearing uncoordinated granularity layer is the globally optimal drug granularity;
f: judging according to the number of the patient data, and entering the step G if the number of the patient data is less than or equal to 1000; if the number of pieces of patient data is more than 1000, entering step H;
g: calculating local optimal drug granularity by using a serial method;
h: and calculating the local optimal medication granularity by using a parallel method.
2. The device for calculating optimal medication granularity based on the multi-granularity decision model as claimed in claim 1, wherein the step D comprises the following specific steps:
d1: medication granularity decision model S at multiple granularity levels k ＝(O，P k Each layer of U { d }) divides each object according to condition attribute, and sets object x 1 Each condition attribute of
Figure FDA0003697453690000021
And object x 2 Each condition attribute of
Figure FDA0003697453690000022
The one-to-one correspondence is the same, then x is 1 And x 2 Dividing into one class by analogy, recording the division result and obtaining a set of the division result of each layer, and marking as R O ；
D2: medication granularity decision model S at multiple granularity levels k ＝(O，P k Each layer of U { d }) divides each object according to decision attributes, and sets object x 1 、x 2 And x 3 Is effective, and subject x 4 And x 5 If the drug treatment is not effective, the decision attribute is divided, and x is divided 1 、x 2 And x 3 Is divided into one class, x 4 And x 5 Dividing into one class, repeating the above steps, recording the division result to obtain the set of all objects divided on each layer, and recording as R r ；
D3: comparing the division results R O And a division result R r If, if
Figure FDA0003697453690000023
Judging that the decision model of the layer is harmonious; if it is not
Figure FDA0003697453690000024
Judging that the decision model of the layer is uncoordinated; the upper layer of the first uncoordinated granularity layer is the global optimal medication granularity; r O Representing a set of objects divided by attributes, R r Representing the set of subjects divided by the effect of the medication.
3. The device for calculating optimal medication granularity based on the multi-granularity decision model as claimed in claim 1, wherein the step G comprises the following specific steps:
g1: medication granularity decision model S at multiple granularity levels k ＝(O，P k At each level of U { d }), each object is divided according to conditional attributes, and an object x is assumed to be 1 Each attribute of (2)
Figure FDA0003697453690000025
And object x 2 Each attribute of (2)
Figure FDA0003697453690000026
Figure FDA0003697453690000027
The one-to-one correspondence is the same, then x is 1 And x 2 Dividing into one class, repeating the above steps, recording the division results and obtaining the set of the division results of each object, and recording the set as
Figure FDA0003697453690000028
Namely, the result of dividing the object x on the k-th layer medication granularity decision model according to the condition attribute P;
g2: medication granularity decision model S at multiple granularity levels k ＝(O，P k At each layer of U { d }), each object is divided according to decision attributes, and an object x is assumed to be 1 And x 2 Is effective, and subject x 3 ，x 4 And x 5 If the drug treatment is not effective, dividing the disease according to decision attributes and dividing x 1 And x 2 Is divided into one class, x 3 ，x 4 And x 5 Dividing into one class, repeating the above steps, recording the division result and obtaining the set of each object, and marking as [ x ]] d I.e. the result of the partitioning of the object x by the decision attribute d;
g3: comparing the set [ x ] of each object x in the step 2] d Whether the set of the objects x in the step 1 is included
Figure FDA0003697453690000031
If it is
Figure FDA0003697453690000032
Judging that the decision model of the layer where the object x is located is coordinated; if it is
Figure FDA0003697453690000033
Judging that the decision model of the layer where the object x is located is uncoordinated; the top layer of the first-occurring uncoordinated granularity layer is the locally optimal medication granularity of the object x, i.e. the
Figure FDA0003697453690000034
Eyes of a user
Figure FDA0003697453690000035
When, the k-th layer granularity is the locally optimal medication granularity for subject x; and thus a set of locally optimal medication granularities for all subjects.
4. The device for calculating optimal medication granularity based on a multi-granularity decision model according to claim 3, wherein in the step H, the plurality of pieces of patient data are averagely divided into N groups according to the number N of cores of the processor of the computer, then the N cores of the processor of the computer calculate a corresponding group of patient data respectively according to the method in the step G, and finally the set of local optimal medication granularity of all the objects is obtained.
CN201910625179.2A 2019-07-11 2019-07-11 A computing device for optimal medication granularity based on multi-granularity decision-making model Active CN110322946B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910625179.2A CN110322946B (en) 2019-07-11 2019-07-11 A computing device for optimal medication granularity based on multi-granularity decision-making model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910625179.2A CN110322946B (en) 2019-07-11 2019-07-11 A computing device for optimal medication granularity based on multi-granularity decision-making model

Publications (2)

Publication Number Publication Date
CN110322946A CN110322946A (en) 2019-10-11
CN110322946B true CN110322946B (en) 2022-09-16

Family

ID=68121971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910625179.2A Active CN110322946B (en) 2019-07-11 2019-07-11 A computing device for optimal medication granularity based on multi-granularity decision-making model

Country Status (1)

Country Link
CN (1) CN110322946B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119673352A (en) * 2024-12-05 2025-03-21 中国医学科学院阜外医院 Method, device and electronic device for determining blood sugar management strategy

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103336791A (en) * 2013-06-06 2013-10-02 湖州师范学院 Hadoop-based fast rough set attribute reduction method
CN107679368A (en) * 2017-09-11 2018-02-09 宁夏医科大学 PET/CT high dimensional feature level systems of selection based on genetic algorithm and varied precision rough set
CN109542949A (en) * 2018-11-07 2019-03-29 太原理工大学 A kind of decision information system knowledge acquisition method based on type vector
CN109859815A (en) * 2019-01-24 2019-06-07 西安交通大学 A kind of auxiliary medicine decision-making method and intelligent auxiliary medicine system

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2003269395A1 (en) * 2002-09-16 2004-04-30 Optimata, Ltd. An interactive technique for optimizing drug development from the pre-clinical phases through phase-iv
US20140372158A1 (en) * 2013-06-12 2014-12-18 Fair Isaac Corporation Determining Optimal Decision Trees

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103336791A (en) * 2013-06-06 2013-10-02 湖州师范学院 Hadoop-based fast rough set attribute reduction method
CN107679368A (en) * 2017-09-11 2018-02-09 宁夏医科大学 PET/CT high dimensional feature level systems of selection based on genetic algorithm and varied precision rough set
CN109542949A (en) * 2018-11-07 2019-03-29 太原理工大学 A kind of decision information system knowledge acquisition method based on type vector
CN109859815A (en) * 2019-01-24 2019-06-07 西安交通大学 A kind of auxiliary medicine decision-making method and intelligent auxiliary medicine system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Monotonic uncertainty measures for attribute reduction in probabilistic rough set model;Guoyin Wang et al;《International Journal of Approximate Reasoning》;20151231;全文 *
多粒度决策系统属性约简的最优粒度选择;史进玲等;《计算机科学》;20180228;全文 *
广义多尺度决策系统的局部最优粒度选择;顾沈明等;《山东大学学报(理学版)》;20180831;全文 *

Also Published As

Publication number Publication date
CN110322946A (en) 2019-10-11

Similar Documents

Publication Publication Date Title
CN112150209B (en) A Construction Method of CNN-LSTM Time Series Prediction Model Based on Cluster Center
Banu et al. Disease forecasting system using data mining methods
WO2017152802A1 (en) Intelligent system and method for converting textual medical report into structured data
Khaing Data mining based fragmentation and prediction of medical data
CN112199425A (en) Medical big data center based on mixed database structure and construction method thereof
CN113808693B (en) A drug recommendation method based on graph neural network and attention mechanism
CN108304887A (en) Naive Bayesian data processing system and method based on minority class sample synthesis
CN105335804A (en) Community health service system
CN112992370B (en) An unsupervised electronic medical record-based medical behavior compliance assessment method
Byeon et al. Deep neural network model for enhancing disease prediction using auto encoder based broad learning
CN108573758A (en) An intelligent medical big data service system and application method
WO2022166158A1 (en) System for performing long-term hazard prediction on hemodialysis complications on basis of convolutional survival network
CN114999574B (en) Parallel identification and analysis method and system for intestinal flora big data
CN108962382B (en) A hierarchical important feature selection method based on high-dimensional clinical data of breast cancer
Alturki et al. Predictors of readmissions and length of stay for diabetes related patients
CN110277172A (en) A clinical medication behavior analysis system based on efficient negative sequence mining mode and its working method
CN106326642A (en) Method for establishing medical consultation fee lattice model based on big data analysis
CN110322946B (en) A computing device for optimal medication granularity based on multi-granularity decision-making model
CN118173198A (en) Multi-relation contrast learning drug interaction prediction method, system, medium and equipment
Karthiga et al. Heart disease analysis system using data mining techniques
Solanki et al. Analysis of application of data mining techniques in healthcare
US20240285231A1 (en) System and Method for Determining a Stool Condition
CN112784886B (en) Brain image classification method based on multi-layer maximum spanning tree graph core
CN116646012A (en) LncRNA-disease association prediction method based on unbalanced neighbor constrained random walk
CN109994211A (en) A kind of modeling method of the chronic kidney disease progression risk based on EHR data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant