CN107578822A - A kind of pretreatment and feature extracting method for the multi-modal big data of medical treatment - Google Patents
A kind of pretreatment and feature extracting method for the multi-modal big data of medical treatment Download PDFInfo
- Publication number
- CN107578822A CN107578822A CN201710612240.0A CN201710612240A CN107578822A CN 107578822 A CN107578822 A CN 107578822A CN 201710612240 A CN201710612240 A CN 201710612240A CN 107578822 A CN107578822 A CN 107578822A
- Authority
- CN
- China
- Prior art keywords
- data
- modal
- big data
- feature
- pretreatment
- 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.)
- Granted
Links
Abstract
The present invention proposes a kind of analysis method that multi-modal big data is analyzed for medical institutions.Primarily directed to the analysis of the multi-modal big data of patient in hospital database.Multiple modal information data can be considered, be effectively prevented from during conventional data analysis, the generation of transmission network limited situation, and ensure Real-time Feedback user profile.The multidimensional partial least square model established, and convolutional neural networks method is combined, information loss can be reduced, the forecast model stablized, more detailed and accurate analysis report is provided for hospital.
Description
Technical field
The present invention relates to medical big data field, pretreatment and feature in particular to the multi-modal big data of hospital
Extraction.
Background technology
With the development of society, medical technology is also correspondingly continuously available raising.Domestic hospitals, which nearly all establish, to be belonged to
The data warehouse of oneself, and constantly area is tired out in the data and the big database of itself of historical record of various disease informations, in it
Appearance has all reached sizable scale.This is an important information resources for each hospital institution.For tradesman
The help of disease information is provided, observation disease development law and development trend over the years, is all served very important important
Property.But nowadays each medical institutions all suffer from such predicament, it is how to analyze the multi-modal big data of disease,
The utilization rate of disease information is improved, and finds out the information of needs exactly, makes brilliant decision-making.
The content of the invention
In order to solve the problem of pretreatment and the feature extraction of the multi-modal big data of medical treatment, the present invention proposes one kind and is directed to
Hospital analyzes the method for multi-modal big data, and proposes to design multi-density quantizer, using genetic algorithm and BP genetic algorithms etc.
Technology is predicted analysis.
A kind of pretreatment and feature extracting method for the multi-modal big data of medical treatment, as shown in figure 1, including following step
Suddenly:
Step 1. is pre-processed using S-G exponential smoothings to the multi-modal data of hospital.One is chosen before and after pending point
Segment data.Continuum method point forms single window and sorted, and takes median as smooth value.
Data after step 2. acquisition processing, using the method for the information quantization of the feature of multi-modal data, gather the doctor
The multi-modal big data of mechanism is treated, with reference to the load capacity of network transmission, designs multi-density quantizer
Local regression method of the step 3. based on correlation analysis, data mould is built using multidimensional partial least squares algorithm
Type, the method modeled using GA-BP, and the method for combining convolutional neural networks, extract valuable in patient history data's data
Information
Step 4 derives the novel information extraction algorithm of disease data, the dynamic evolution rule of patient disease is obtained, to disease
Disease makes Performance Evaluation index, and the scheme of rolling optimization is proposed for patient.
Brief description of the drawings
A kind of schematic diagrames for the pretreatment and feature extracting method for being directed to the multi-modal big data of medical treatment of Fig. 1.
Embodiment
Described S-G exponential smoothings, it is characterised in that suitable window is chosen first, then according to polynomial fitting method,
Data in each window are smoothed, by window data, Ran Houshi corresponding to the smooth value being calculated replacement
Between increased direction move a data point successively, new window is formed, untill traveling through all data point;
Its specific method is one matrix smooth window of selection in three-dimensional fluorescence spectrum so that window includes (2p+
1) × (2q+1) individual data point, the data point of its window can be expressed as:
(a-p,b-q,x(a-p,b-q,))…(a-p,b0,x(a-p,b0,)),…,(a-p,bq,x(a-p,bq,)) …..
(a0,b-q,x(a0,b-q,))…(a0,b0,x(a0,b0,)),…,(a0,bq,x(a0,bq,)) …..
(ap,b-q,x(ap,b-q,))…(ap,b0,x(ap,b0,)),…,(ap,bq,x(ap,bq,))
Wherein am(m=-p ..., p) is m-th of emission spectrum wavelength, bn(n=-q ..., q) it is n-th excitation spectrum ripple
It is long, x (am,bn) (m=-p ..., p, n=-q ..., q) be data point (am,bn) fluorescence intensity.
The smooth value calculation formula of each point is wherein in window:
Described multi-density quantizer, it is characterised in that can according to the situation of transmission network, dynamic regulation quantizer
Setting value.Because the situation of actual transmissions network is dynamic, multi-density quantizer ensures maximal efficiency quantized data, reaches pair
The high efficiency of transmission of multi-modal big data.By being write the data after quantification treatment shape of the output valve plus a Gaussian noise as
Formula, i.e.,:
Then the degree of load at corresponding moment, and the window value changed according to history big data statistics, binding number are obtained
According to the precision required for warehouse and load design multi-density quantizer.
Described multidimensional partial least squares algorithm structure data model, it is characterised in that multidimensional offset minimum binary is that one kind is more
D Data Model, during regression model foundation is carried out, the load vectors directly related with each dimension can be obtained, and to model
Each dimension do independent explanation, obtain regression model, be represented by:
Wherein, X is the matrix generated after multi-modal big data is handled, and F is number of components, and T is score matrix, and size is I rows F
Row, WJAnd WKIt is the weight matrix in J directions and K directions respectively, size is respectively J rows F row and K rows F row.
When the operation being predicted, by multi-modal data matrix Xw(I × J × K), prediction knot can be obtained by carrying out calculating
Fruit:By XwDimensionality reduction is to two-dimensional matrix Xw(I × JK), solve predictive variable YnewValue
The method of GA-BP modeling, it is characterised in that using genetic algorithm and BP algorithm (GA-BP) in turn to obtaining
Regression model is trained, and according to the index of correlation of disease, chooses wherein valuable packet, is substituted into Genetic Algorithm Model and is entered
Item modeling, until seeing network convergence.
Wherein BP e-learnings flow is selects the topological structure of 3 layers of BP networks, after the selection quantization of its input layer
Multi-modal data, then networking input normalization sample data, with reference to forecast sample simulated effect, when predicted value is square
Root error reaches certain index and just shifts to an earlier date deconditioning, directly exports the BP network models trained.
The convolutional neural networks method, it is characterised in that employ the output valve of front and rear transmission, backpropagation weight and
Biasing, it is adjacent in internal neutral net between neverous frontails unit connected using part, make the partial nerve member on upper strata, pass through god
Neuron through network internal is perceived, and contributes to from the multi-modal big data of medical treatment to extract the knowledge of profound level, so as to build
The vertical depth for multi-modal big data recognizes.
The first step establishes convolutional neural networks first, and its effect is exactly the local feature it can be found that data, is then utilized
Map in convolutional neural networks, share a convolutional neural networks core.Each of which map is by multiple neural unit groups
Into.
Then by realizing the full connection of characteristic and output layer, weight is adjusted using the mode of back-propagation nerve network
And biasing.Neutral net can be solved by gradient descent method.Because in actual applications, gradient descent method tends to
To gratifying result.
Convolutional neural networks core is exactly the implication of weight in fact, does not have to individually calculate in actual calculating process, but solid
The weight matrix for determining size goes on image to match.Weight sharing policy reduces the parameter of needs training so that trains what is come
The general Huaneng Group power of model is stronger.
The specific embodiment of the present invention is described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring the substantive content of the present invention.
Claims (6)
1. a kind of pretreatment and feature extracting method for the multi-modal big data of medical treatment, it is characterised in that comprise the following steps:
Step 1. is pre-processed using S-G exponential smoothings to the multi-modal data of hospital.A hop count is chosen before and after pending point
According to.Continuum method point forms single window and sorted, and takes median as smooth value;
Data after step 2. acquisition processing, using the method for the information quantization of the feature of multi-modal data, gather the therapeutic machine
The multi-modal big data of structure, with reference to the load capacity of network transmission, design multi-density quantizer;
Local regression method of the step 3. based on correlation analysis, data model is built using multidimensional partial least squares algorithm,
The method modeled using GA-BP, and the method for combining convolutional neural networks, extract valuable information in patient history data;
Step 4 derives the novel information extraction algorithm of disease data, obtains the dynamic evolution rule of patient disease, disease is done
Go out Performance Evaluation index, and the scheme of rolling optimization is proposed for patient.
2. according to claim 1 exist for the pretreatment of the multi-modal big data of medical treatment and feature extracting method, its feature
In:Described S-G exponential smoothings, choose suitable window first, then according to polynomial fitting method, in each window
Data are smoothed, by the smooth value being calculated replace corresponding to window data, then time increased direction is successively
A mobile data point, forms new window, untill traveling through all data points.
3. according to claim 1 exist for the pretreatment of the multi-modal big data of medical treatment and feature extracting method, its feature
In:Described multi-density quantizer, can be according to the situation of transmission network, the setting value of dynamic regulation quantizer.
4. according to claim 1 exist for the pretreatment of the multi-modal big data of medical treatment and feature extracting method, its feature
In:Described multidimensional partial least squares algorithm structure data model, multidimensional offset minimum binary is a kind of Multidimensional Data Model, is being entered
During row regression model is established, the load vectors directly related with each dimension are obtained, and independent explanation is done to each dimension of model, are obtained
To regression model, it is represented by:
X=T (WK⊙WJ)T+E
Wherein, F is number of components, and T is score matrix, and size arranges for I rows F, WJAnd WKIt is the weight square in J directions and K directions respectively
Battle array, size are respectively J rows F row and K rows F row.
5. according to claim 1 exist for the pretreatment of the multi-modal big data of medical treatment and feature extracting method, its feature
In:The method of the GA-BP modelings, is trained, foundation to obtained regression model in turn using genetic algorithm and BP algorithm
The requirement of disease index of correlation, chooses wherein valuable packet, substitutes into the modeling of Genetic Algorithm Model income, until network is received
Hold back.
6. according to claim 1 exist for the pretreatment of the multi-modal big data of medical treatment and feature extracting method, its feature
In:The convolutional neural networks method, employ the output valve of front and rear transmission, backpropagation weight and biasing, internal nerve
In network it is adjacent between neverous frontails unit using part connect, make upper strata partial nerve member, pass through the god inside neutral net
Perceived through member.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710612240.0A CN107578822B (en) | 2017-07-25 | 2017-07-25 | Pretreatment and feature extraction method for medical multi-modal big data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710612240.0A CN107578822B (en) | 2017-07-25 | 2017-07-25 | Pretreatment and feature extraction method for medical multi-modal big data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107578822A true CN107578822A (en) | 2018-01-12 |
CN107578822B CN107578822B (en) | 2020-12-15 |
Family
ID=61034174
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710612240.0A Active CN107578822B (en) | 2017-07-25 | 2017-07-25 | Pretreatment and feature extraction method for medical multi-modal big data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107578822B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109241041A (en) * | 2018-06-26 | 2019-01-18 | 广东工业大学 | A kind of preprocess method and device of injection molding equipment big data |
CN109448855A (en) * | 2018-09-17 | 2019-03-08 | 大连大学 | A kind of diabetes glucose prediction technique based on CNN and Model Fusion |
CN112001228A (en) * | 2020-07-08 | 2020-11-27 | 上海品览数据科技有限公司 | Video monitoring warehouse in-out counting system and method based on deep learning |
CN112712895A (en) * | 2021-02-04 | 2021-04-27 | 广州中医药大学第一附属医院 | Data analysis method of multi-modal big data for type 2 diabetes complications |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130117257A1 (en) * | 2011-11-03 | 2013-05-09 | Microsoft Corporation | Query result estimation |
CN105393252A (en) * | 2013-04-18 | 2016-03-09 | 数字标记公司 | Physiologic data acquisition and analysis |
CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
-
2017
- 2017-07-25 CN CN201710612240.0A patent/CN107578822B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130117257A1 (en) * | 2011-11-03 | 2013-05-09 | Microsoft Corporation | Query result estimation |
CN105393252A (en) * | 2013-04-18 | 2016-03-09 | 数字标记公司 | Physiologic data acquisition and analysis |
CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
Non-Patent Citations (2)
Title |
---|
LIBAO YANG等: "MH-ARM: a Multi-mode and High-value Association Rule Mining Technique for Healthcare Data Analysis", 《2016 INTERNATIONAL CONFERENCE ON COMPUTATIONAL SCIENCE AND COMPUTATIONAL INTELLIGENCE》 * |
SHENG-CHENG LEE等: "A Multi-Channel Multi-Mode Physiological Signals Acquisition and Analysis Platform", 《 2013 IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS (ISCAS)》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109241041A (en) * | 2018-06-26 | 2019-01-18 | 广东工业大学 | A kind of preprocess method and device of injection molding equipment big data |
CN109241041B (en) * | 2018-06-26 | 2021-05-11 | 广东工业大学 | Preprocessing method and device for big data of injection molding equipment |
CN109448855A (en) * | 2018-09-17 | 2019-03-08 | 大连大学 | A kind of diabetes glucose prediction technique based on CNN and Model Fusion |
CN112001228A (en) * | 2020-07-08 | 2020-11-27 | 上海品览数据科技有限公司 | Video monitoring warehouse in-out counting system and method based on deep learning |
CN112712895A (en) * | 2021-02-04 | 2021-04-27 | 广州中医药大学第一附属医院 | Data analysis method of multi-modal big data for type 2 diabetes complications |
CN112712895B (en) * | 2021-02-04 | 2024-01-26 | 广州中医药大学第一附属医院 | Data analysis method of multi-modal big data aiming at type 2 diabetes complications |
Also Published As
Publication number | Publication date |
---|---|
CN107578822B (en) | 2020-12-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107578822A (en) | A kind of pretreatment and feature extracting method for the multi-modal big data of medical treatment | |
CN106779084B (en) | Machine learning system and method | |
CN106778014A (en) | A kind of risk Forecasting Methodology based on Recognition with Recurrent Neural Network | |
CN107169504B (en) | A kind of hand-written character recognition method based on extension Non-linear Kernel residual error network | |
CN107610123A (en) | A kind of image aesthetic quality evaluation method based on depth convolutional neural networks | |
CN108154430A (en) | A kind of credit scoring construction method based on machine learning and big data technology | |
CN106682385B (en) | Health information interaction system | |
CN105956150B (en) | A kind of method and device generating user's hair style and dressing collocation suggestion | |
CN106919951A (en) | A kind of Weakly supervised bilinearity deep learning method merged with vision based on click | |
CN108573303A (en) | It is a kind of that recovery policy is improved based on the complex network local failure for improving intensified learning certainly | |
CN103942571B (en) | Graphic image sorting method based on genetic programming algorithm | |
CN106650933A (en) | Deep neural network optimizing method based on coevolution and back propagation | |
CN110070116A (en) | Segmented based on the tree-shaped Training strategy of depth selects integrated image classification method | |
CN110991317B (en) | Crowd counting method based on multi-scale perspective sensing network | |
EP3991102A1 (en) | Pruning and/or quantizing machine learning predictors | |
CN114693064B (en) | Building group scheme generation performance evaluation method | |
CN107368707A (en) | Gene chip expression data analysis system and method based on US ELM | |
CN108710609A (en) | A kind of analysis method of social platform user information based on multi-feature fusion | |
CN108304925A (en) | A kind of pond computing device and method | |
CN109567313B (en) | Intelligent insole with biological characteristic recognition function | |
CN112907604A (en) | Self-adaptive super-pixel FCM (pixel-frequency modulation) method for fundus velveteen speckle image segmentation | |
CN110893100A (en) | Device and method for monitoring posture change based on plantar pressure sensor | |
CN116386899A (en) | Graph learning-based medicine disease association relation prediction method and related equipment | |
CN108962382A (en) | A kind of layering important feature selection method based on breast cancer clinic high dimensional data | |
CN106651001A (en) | Needle mushroom yield prediction method based on improved neural network and implementation system |
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 |