CN107273685A - A kind of data analysing method of multi-modal big data for clinical disease - Google Patents
A kind of data analysing method of multi-modal big data for clinical disease Download PDFInfo
- Publication number
- CN107273685A CN107273685A CN201710447993.0A CN201710447993A CN107273685A CN 107273685 A CN107273685 A CN 107273685A CN 201710447993 A CN201710447993 A CN 201710447993A CN 107273685 A CN107273685 A CN 107273685A
- Authority
- CN
- China
- Prior art keywords
- data
- disease
- msub
- big data
- mrow
- 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.)
- Pending
Links
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a kind of data analysing method of the multi-modal big data for clinical disease, including：The corresponding history big data of disease type is obtained from medical system；By history big data, corresponding rate of change designs multi-density quantizer under each mode；Multi-modal data method for digging is used to history big data, and combines convolutional neural networks method, the corresponding characteristic information of disease type is extracted；The individual dynamic evolution rule for the type that caught in history big data is derived according to characteristic information；The Performance Evaluation index of real time data is obtained with reference to the dynamic evolution rule of individual.As can be seen here, this method can efficiently against single mode data limitation, the hazards of disease can be considered, more detailed and accurate clinical criteria is provided for hospital, the dynamic evolution rule of individual can be provided simultaneously, decision-making foundation and technical support are provided for the early diagnosis early treatment of disease, diagnosis efficiency and diagnosis and treatment quality is improved.
Description
Technical field
The present invention relates to Analysis of Medical Treatment Data field, a kind of more particularly to multi-modal big data for clinical disease
Data analysing method.
Background technology
With the development of society, science and technology is also correspondingly improved constantly, national each hospital starts to store oneself
The mode of data becomes more diversified, therefore the concept of medical multi-modal big data has obtained the concern of numerous experts and scholars.
The multi-modal big data of clinical disease refers to adopt disease with vision (such as physiology, custom, psychology, environment) from different angles
Collect the set of obtained Various types of data.Multi-modal big data provides new Research Thinking for the diagnosis research of complex disease.Sea
Measure and correlation information between data and disease is imply in multi-modal medical data.
But clinically rely primarily on clinical manifestation at present and laboratory examination makes diagnosis, and with examining in recent years
The development of survey technology, it is multi-modal that the mankind are already available to the related a variety of inspection project results of more human body diseases etc.
Data, and often there is correlation between disease in these data.Medicinal aboundresources, case disease is complete, multi-modal sample
Amount is huge.But it is due to the effective utilization for the support and multi-modal data for lacking big data processing environment, current existing disease
Diagnosis research remains in traditional data research based on single mode.
Medical information is abundant in content various, and data structure is special, clinical disease big data may containing clear data, image,
Word, detection signal, audio or video information etc..The expression of many medical informations simultaneously, record have uncertain, fuzzy
Property, imperfection, the difficulty for increasing with the feature such as noise and redundancy medical data mining.Although nowadays scientific circles pair
The perception of data has much very important development with acquisition methods, but under the multi-modal big data environment of generalization, data
Amount is greatly and mode is various brings new challenge to the collection of data, transmission, cleaning and storage, causes existing method straight
Connect and be dissolved into the modeling of the information extraction based on the multi-modal big data of non-intrusion type generalization and disease Evolution.
In addition, the eating habit of patient, the climate characteristic of location, local conditions and customs, recent anxious state of mind etc.
There is substantial connection Deng the generation development all with disease.And this error promotes, with enlarge-effect, can more drill over time
It is stronger, cause Evolution to have the opposite effect actual therapeutic effect.Therefore, the non-intrusion type generalization multimode based on clinical disease
State medical treatment big data, substantially the error of disease Evolution and the actual state of an illness is one big on big data medical treatment way for development less
Obstacle.
In summary, it is desirable to realize that the research of the multi-modal big data of generalization is not easy to, and bigger problem be as
What tries to analyze inherence and the transient cause of disease incidence, and discloses the Evolution of disease.As can be seen here, how by many
The correlative study that mode big data obtains medical diagnosis on disease is those skilled in the art ground urgently to be resolved hurrily problem.
The content of the invention
It is an object of the invention to provide a kind of data analysing method of the multi-modal big data for clinical disease, for leading to
Cross the correlative study that multi-modal big data obtains medical diagnosis on disease.
In order to solve the above technical problems, the present invention provides a kind of data analysis of multi-modal big data for clinical disease
Method, including：
According to the disease type determined, the corresponding history big data of the disease type is obtained from medical system；
By the history big data, corresponding rate of change designs multi-density quantizer to be counted in real time under each mode
According to perception and acquisition methods；
Multi-modal data method for digging is used to the history big data, and combines convolutional neural networks method, is extracted
The corresponding characteristic information of the disease type；
Derive that the individual dynamic that the disease type is infected in the history big data is drilled according to the characteristic information
Law；
The Performance Evaluation index of the real time data is obtained with reference to the individual dynamic evolution rule.
Preferably, it is described by the history big data under each mode corresponding rate of change design multi-density quantizer with
The perception and acquisition methods for obtaining real time data are specifically included：
The history big data is segmented according to the rate of change, and solves the corresponding average value of each segmentation result；
The average value is divided into multiple set by the scope according to the deviation of each average value, and solves the set
Average value；
Calculate the deviation of the measurement data in the set, and the deviation according to the measurement data and the set is flat
The corresponding relation design anticipation function of average；
The measurement data is standardized according to the anticipation function；
Data cleansing is carried out to the measurement data after standardization, and sets described how close according to the load characteristic of transmission network
Metrization device.
Preferably, it is described that the history big data is specifically included using multi-modal data method for digging：
By the influence relation production Methods matrix of disease and hazards；
According to the relational matrix sets target function, and to the object function minimization；
Wherein, the object function is：
Wherein, MijFor the relational matrix, U is disease, and V is hazards, XuRepresent the feature of disease, XvFor it is dangerous because
The feature of element, R (U, V) is the constraint of the regularization to U and V, Ru(U,Xu) for U to XuRegularization constraint, Rv(V,Xv) for V to Xv
Regularization constraint, λ, λuAnd λvRespectively the weight of rule of correspondence bound term, is worth between 0~1, L (Ui,Vj,Mij) be
Ui、ViAnd MijRelation function.
Preferably, the combination convolutional neural networks method, extracts the corresponding characteristic information of the disease type specific
Including：
Type according to the affiliated mode of the measurement data sets up a variety of convolution kernels；
Mixing sampling is carried out to the convolution kernel and obtains the characteristic information；
The characteristic information is connected entirely with the output layer of the convolutional neural networks.
Preferably, the individual dynamic evolution rule is obtained especially by following steps：
Set up fuzzy dynamic between the characteristic information and diseased individuals Evolution using fuzzy logic system identification method
State treats model.
Preferably, the individual dynamic evolution rule with reference to described in obtains the Performance Evaluation index tool of the real time data
Body includes：
Object function in model is treated based on the fuzzy dynamic, disease Performance Evaluation is set up by iteration optimization algorithms
Index；
Calculate the minimum value of the disease performance evaluation index.
Preferably, in addition to：
Based on the disease performance evaluation index, optimal disease treatment sequence is obtained using rolling optimization algorithm.
Preferably, in addition to：
Feedback correction is carried out to the disease treatment sequence by the real time data.
Preferably, in addition to：
The dynamic evolution rule of the corresponding colony of the disease type is set up according to the individual dynamic evolution rule.
The data analysing method of multi-modal big data provided by the present invention for clinical disease, including：According to determination
Good disease type, obtains the corresponding history big data of disease type from medical system；By history big data in each mode
Under corresponding rate of change design multi-density quantizer to obtain perception and the acquisition methods of real time data；History big data is used
Multi-modal data method for digging, and convolutional neural networks method is combined, extract the corresponding characteristic information of disease type；According to special
Reference breath derives the individual dynamic evolution rule for the type that caught in history big data；Advised with reference to the dynamic evolution of individual
Rule obtains the Performance Evaluation index of real time data.As can be seen here, this method can be efficiently against conventional data analysis method only
Consider the limitation of Disease physiology single mode data, the hazards of disease can be considered, provided more for hospital
Detailed and accurate clinical criteria, while the individual dynamic evolution rule of multi-modal big data and disease can be provided, in advance
Survey the incidence trend of disease, decision-making foundation and technical support provided for the early diagnosis early treatment of disease, improve diagnosis efficiency with
Diagnosis and treatment quality.
Brief description of the drawings
In order to illustrate the embodiments of the present invention more clearly, the required accompanying drawing used in embodiment will be done simply below
Introduce, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for ordinary skill people
For member, on the premise of not paying creative work, other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is a kind of data analysing method of multi-modal big data for clinical disease provided in an embodiment of the present invention
Flow chart；
Fig. 2 is a kind of original that optimal disease treatment sequence is obtained based on rolling optimization algorithm provided in an embodiment of the present invention
Reason figure.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.Based on this
Embodiment in invention, those of ordinary skill in the art are not under the premise of creative work is made, and what is obtained is every other
Embodiment, belongs to the scope of the present invention.
The core of the present invention is to provide a kind of data analysing method of the multi-modal big data for clinical disease.
In order that those skilled in the art more fully understand the present invention program, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.
Fig. 1 is a kind of data analysing method of multi-modal big data for clinical disease provided in an embodiment of the present invention
Flow chart.As shown in figure 1, the data analysing method of the multi-modal big data for clinical disease, including：
S10：According to the disease type determined, the corresponding history big data of disease type is obtained from medical system.
It is appreciated that disease type has many kinds, for example, anorectal disease, particular type the present embodiment is repeated no more.
In addition, in medical system, the corresponding history big data of every kind of disease type that is all stored with, these data are True Datas, are had
Correlation between history big data is obtained individual corresponding to disease type by very high reference value, the present invention
Dynamic evolution rule, so that the Performance Evaluation index of real time data is obtained by this dynamic evolution rule, to complete to real-time
The analysis of data and obtain corresponding therapeutic scheme.
S11：By history big data, corresponding rate of change designs multi-density quantizer to be counted in real time under each mode
According to perception and acquisition methods.
In the present embodiment, in terms of data acquisition, data cleansing and quantifying compression three, go to build non-intrusion type generalization
The characteristics of perception of multi-modal data and acquisition methods, analysis of history data (rates of change of such as data) and with the clinical disease
The incidence relation of disease, for collecting method of the different design datas based on multi-sampling rate.Preferably embodiment,
Step S11 specifically includes S110-S114.
S110：History big data is segmented according to rate of change, and solves the corresponding average value of each segmentation result.
According to the characteristic of history big data, the rate of change of different modalities data is analyzed, data are entered based on this rate of change
Row segmentation.For different sections of data, pass through following equations average value.
Wherein, n is the number of s stage sampled datas, diFor all measured values in a stage.
S111：Average value is divided into multiple set by the scope according to the deviation of each average value, and solves being averaged for set
Value.
μsClose value is classified as a class, is defined as set omegas, and remember that this average value gathered is
S112：The deviation of measurement data in set of computations, and deviation and the average value of set according to measurement data
Corresponding relation designs anticipation function.
The deviation calculation formula of measurement data is：Wherein niFor set omegasAll surveys
Measure the number of data.
S113：Measurement data is standardized according to anticipation function.
For a measurement data with dynamic characteristic, its corresponding average value and deviation are predicted based on limited step, i.e.,
Design the average value and deviation of anticipation function f (d (k) ... d (k-N+1)) prediction d (k) moment measurement data.Based on d (k) moment
The average value and deviation of prediction, standardize measurement data, and standardization formula is as follows：
S114：Data cleansing is carried out to the measurement data after standardization, and sets many according to the load characteristic of transmission network
Intensity Quantizer.
Wherein, data cleansing is specially：A constant θ ＞ 1 is defined, by judging whether y (k) absolute value is more than θ, is come
Judge whether the measurement data at current d (k) moment belongs to outlier, realize that dynamic data is cleaned if it is not, then rejecting.
For the data after the standardization that is obtained as described above, the load characteristic based on transmission network need to be designed as follows
Multi-density quantizer, quantizer is that data are quantified, to facilitate data acquisition, cleaning and store.General quantizer is passed
The setting value of throughput rate is fixed, and is the drawbacks of such, when network transfer speeds are fast, easily causes network resources waste, network
When transmission speed is slow, valid data are easily lost.And multi-density quantizer can be according to the situation of transmission network, dynamic regulated quantity
Change the setting value of device, so as to obtain multi-density quantizer.Because multi-modal big data is high to performance index requirements, and real network
Situation be it is dynamic, when it is good when it is bad.For maximal efficiency quantized data, so will the state based on transmission network, real-time change
Setting value, steps below is exactly to explain how to go dynamic adjustment setting value.
The first step, writes quantization measured value as the form for quantifying output valve plus a Gaussian noise, i.e.,
Wherein, yi(k) it is actual measurement data,For the measurement data after quantization, qi(k) it is quantization error.In order to
It is easy to analysis below, it is assumed that qi(k) it is in interval [- 0.5ui,0.5ui] on meet average for 0, variance isIt is uniform
Distribution.Based on the analysis of history big data, the required precision of institute's gathered data, design pair are determined the need for providing different data sources
Answer the uniform quantizer of data source.
Second step, the bit number of each packet is analyzed according to transmission network, B1 is defined as.It is actual that the k moment is defined simultaneously
The data bit number of transmission is B2 (k).The present invention define the transmission channel k moment degree of load be：
3rd step, the window value of data variation is counted according to history big data, is defined as M, and design length is M storage
Device.This is a kind of common method of data processing, is, using some time span as span, subregion to be carried out to data herein,
The data of each time zone are referred to as window value.Defining average load degree in M length is：
The precision met with reference to required for data source and average degree of load design multi-density quantizer.
According to the average load degree for each period obtained, the size based on this value and control reference value, to adjust
The setting value of integral quantization device, reaches the effect of dynamic quantizer with this, that is, multi-density quantizer function mode.
By designing multi-density quantizer, transmission network handling capacity is limited in the case of can efficiently solving synchronous transfer
Problem, obtained the dynamic statistics outlier judgment criterion based on limited historical data, realized the dynamic of multi-modal big data
State is cleaned, and improves the utilization rate of Internet resources.
S12：Multi-modal data method for digging is used to history big data, and combines convolutional neural networks method, is extracted
The corresponding characteristic information of disease type.
Multi-modal big data method for digging, be first by the relation of disease and hazards it is abstract be a relational matrix
Mij, MijRepresent that disease i is influenceed relation by hazards j, can be 0 or 1, expression whether there is influence relation；Can also be real
Numerical value, represents impacted degree.
Mode is preferably carried out as one kind, history big data is specifically included using multi-modal data method for digging：
By the influence relation production Methods matrix of disease and hazards；
According to relational matrix sets target function, and to object function minimization；
Wherein, object function is：
Wherein, MijFor the relational matrix, U is disease, and V is hazards, XuRepresent the feature of disease, XvFor it is dangerous because
The feature of element, R (U, V) is the constraint of the regularization to U and V, Ru(U,Xu) for U to XuRegularization constraint, Rv(V,Xv) for V to Xv
Regularization constraint, λ, λuAnd λvRespectively the weight of rule of correspondence bound term, is worth between 0~1, L (Ui,Vj,Mij) be
Ui、ViAnd MijRelation function.
By minimization above object function, i.e., to object function minimization：
Influence relational matrix M between the disease and hazards that can be finally given by minimization object functionij, it is real
Now the excavation to disease risk factor is predicted.
By above-mentioned method for digging, the Sparse problem of multi-modal big data is solved, has obtained being based on convolutional Neural
The multi-modal big data relevance decision method of network, improves the analysis efficiency of multi-modal big data.
Above the hazards to the disease in history big data are excavated, and characteristic information is described below
Identification.Mode is preferably carried out as one kind, with reference to convolutional neural networks method, the corresponding characteristic information of disease type is extracted
Specifically include：
Type according to the affiliated mode of measurement data sets up a variety of convolution kernels.
Mixing sampling is carried out to convolution kernel and obtains characteristic information.
The output layer of characteristic information and convolutional neural networks is connected entirely.
The first step, sets up a variety of convolution kernels, and regional area is perceived it can be found that some local features of data, such as picture
On an angle, one section of arc, these essential characteristics be constitute animal vision basis；And in backpropagation neural network, own
Pixel be the chaotic point of a pile, relation each other is not mined.In convolutional neural networks each layer by multiple
Map is constituted, and each map is made up of multiple neural units, and same map all neural units share a convolution kernel and (weighed
Weight), convolution kernel often represents a feature, such as some convolution and represents one section of arc, then this convolution kernel is entirely being schemed
Rolled on piece, the larger region of convolution value is likely to be just one section of arc.It is exactly weight in fact to note convolution kernel, and we are not
Need individually to go to calculate a convolution, but the weight matrix of a fixed size is when going to match on image, this operation and volume
Product is similar, therefore we are referred to as convolutional neural networks.In fact, backpropagation neural network can also regard a kind of special volume as
Product neutral net, simply this convolution kernel is exactly certain layer of all weights, i.e. sensing region is whole image.Weight sharing policy
Reduce the parameter for needing to train so that the generalized ability for training the model come is stronger.
Second step, mixing sampling is carried out to different convolution kernels and obtains characteristic information.The purpose of sampling, which is mainly, obscures spy
The particular location levied, because after some feature is found out, its particular location is inessential, we only need to this feature
With other relative positions, such as one " 8 ", above we have obtained when one " o ", we require no knowledge about it in image
Particular location, it is only necessary to know below it but be one " o " we it is known that be one ' 8' because in picture " 8 "
In picture it is to the left or it is to the right do not affect us and recognize it, this strategy for obscuring particular location can be to deforming and distorting
Picture is identified, and realizes complementing one another for multi-modal data, and basis is provided accurately to portray the disease." characteristic information " is referred to
Be useful data, the data of disease information can be characterized.The patient body temperature of 30 days is for example recorded, if wherein there are four days bodies
Temperature abnormality, although this 30 days is all useful data, but that four days abnormal temperature datas are exactly characteristic information.
3rd step, realizes the full connection of the characteristic information and output layer of mixing sampling, and utilize similar back-propagating nerve
The mode of network, i.e., adjust weight and biasing by minimizing residual error.
Assuming that we have a fixed sample collection { (x, y) }, it includes m sample.We can be declined with batch gradient
Method solves neutral net.Specifically, for single sample (x, y), its cost function is：
J(w,b；X, y)=| | hw,b(x)-y||2
Target is to ask its function J (W, minimum value b) for parameter W and b.In order to solve neutral net, it would be desirable to
By each parameter W and b be initialized as very little, close to zero random value, batch gradient is used to object function afterwards
The optimization algorithm of descent method.(W, is b) non-convex function, and gradient descent method is likely to converge to local optimum because J
Solution.But in actual applications, gradient descent method usually leads to gratifying result.In above formula Section 1 J (W,
B) it is a mean square deviation.Section 2 is a regularization term (being also weight attenuation term), the purpose is to reduce the amplitude of weight,
Prevent overfitting.
S13：The individual dynamic evolution rule for the type that caught in history big data is derived according to characteristic information.
The step is that the characteristic information based on the multi-modal big data of generalization is studied the dynamic evolution rule of individual.
Mode is preferably carried out as one kind, individual dynamic evolution rule is obtained especially by following steps：Using fuzzy logic system
The fuzzy dynamic treatment model that system identification method is set up between characteristic information and diseased individuals Evolution.
Fuzzy dynamic treats model：
Wherein, θjFuzzy logic membership function is represented, u (t) represents treatment input, and y (t) represents the disease of diseased individuals
Disease output, state x (t)=[x1(t),...,xN(t)]TRepresent N number of characteristic information with disease rank correlation, such as body temperature, blood
Pressure, urea etc., the influence of nonlinear function f (x (t)) and g (x (t)) representative feature information to diseased individuals, dynamical system matrix
B and D represent influence of the control input to diseased individuals.
S14：The Performance Evaluation index of real time data is obtained with reference to the dynamic evolution rule of individual.
In specific implementation, disease performance evaluation index is to evaluate one of important parameter of disease therapeuticing effect, can be retouched
State the advancing of disease trend under therapeutic scheme effect.Mode is preferably carried out as one kind, with reference to the dynamic evolution of individual
The Performance Evaluation index that rule obtains real time data is specifically included：
Object function in model is treated based on fuzzy dynamic, disease Performance Evaluation is set up by iteration optimization algorithms and refers to
Mark；
Calculate the minimum value of disease performance evaluation index.
Wherein, disease performance evaluation index is specially：
Wherein, J (t) is that the disease performance evaluation index, matrix R are treatment input parameter to the disease Performance Evaluation
Affecting parameters, the u (t) of index are that treatment input parameter, Q are N number of characteristic information to the disease performance evaluation index
Affecting parameters, S be disease grade to the affecting parameters of the disease performance evaluation index,For the pre- of the characteristic information
Measured value,For the predicted value of the disease grade.
Disease Performance Evaluating Indexes above are related to the state of future time instanceAnd outputTherefore need
Budget is carried out to these variables using forecast model.Meanwhile, the J (t) includes treatment input parameter u (t) and disease grade forecast
ValueTherefore the treatment input of disease can be adjusted while disease grade is judged, that is, requires treatment input
Disease grade is kept to reach while minimum minimum, as described in above-mentioned formula.
The present invention has obtained the dynamic evolution rule of individual, describe accurately spy by fuzzy logic system identification method
Reference ceases the dynamic associations between disease, gives disease performance evaluation index, solves disease of clinical disease etc.
Level discrimination.
The data analysing method for the multi-modal big data for clinical disease that the present embodiment is provided, including：According to determination
Good disease type, obtains the corresponding history big data of disease type from medical system；By history big data in each mode
Under corresponding rate of change design multi-density quantizer to obtain perception and the acquisition methods of real time data；History big data is used
Multi-modal data method for digging, and convolutional neural networks method is combined, extract the corresponding characteristic information of disease type；According to special
Reference breath derives the individual dynamic evolution rule for the type that caught in history big data；Advised with reference to the dynamic evolution of individual
Rule obtains the Performance Evaluation index of real time data.As can be seen here, this method can be efficiently against conventional data analysis method only
Consider the limitation of Disease physiology single mode data, the hazards of disease can be considered, provided more for hospital
Detailed and accurate clinical criteria, while the individual dynamic evolution rule of multi-modal big data and disease can be provided, in advance
Survey the incidence trend of disease, decision-making foundation and technical support provided for the early diagnosis early treatment of disease, improve diagnosis efficiency with
Diagnosis and treatment quality.
On the basis of above-described embodiment, preferably embodiment, in addition to：
Based on disease performance evaluation index, optimal disease treatment sequence is obtained using rolling optimization algorithm.
Rolling optimization algorithm is the mechanism using closed loop feedback, and output result is fed back into input and optimizes adjustment
Export again afterwards, until obtaining optimal disease treatment sequence.Illustrated by taking anorectal disease as an example, the feature based on anorectal disease
The feedback correction regulation fuzzy dynamic of information treats the parameter of model, to obtain the prediction Active treatment model of anorectal disease.Institute
Obtained prediction Active treatment model can not only predict the development trend of anorectal disease, also use Active treatment scheme substitute with
Past static treatment scheme, repeatedly optimization calculates and rolls the corresponding treatment sequence of implementation, effectively overcomes in therapeutic process
Uncertainty, and can easily between processing feature information restriction relation, reach preferable therapeutic effect.
Based on designed disease performance evaluation index J (t), following optimal disease is found out using rolling optimization algorithm and controlled
Treat sequence.Resulting treatment sequence can be according to priori features information and following disease for the treatment of input prediction diseased individuals etc.
Level, is a kind of optimal control treatment sequence.Traditional therapeutic scheme is different from, rolling optimization algorithm has strict to treatment time
Limitation, be finite time section rolling optimization.Fig. 2 is that one kind provided in an embodiment of the present invention is obtained based on rolling optimization algorithm
To the schematic diagram of optimal disease treatment sequence.In the presence of gained optimum treatment sequence, minimum treatment input can make
Obtain disease performance evaluation index J (t) to minimize within the most short time, i.e., disease is fully recovered with most fast speed.
On the basis of above-described embodiment, preferably embodiment, in addition to：
Feedback correction is carried out to disease treatment sequence by real time data.
In clinical disease treatment optimization process, feedback correction is constantly carried out using real-time characteristic information, passes through disease
The mode of model ONLINE RECOGNITION and therapeutic scheme on-line amending, adjusting the parameter in the dynamic evolution rule of individual is used to compensate mould
The interference of type predicted treatment error and external environment to the diseased individuals.Define error vector：
Wherein, e is treatment error.
Using error, the method prediction future disease error of sampling weight is simultaneously compensated based on fuzzy dynamic treatment model correspondence
Forecast model treatment error, the predicted treatment vector after being corrected is：
The repetitive decision process on the basis of individual feedback information is built upon due to feedback correction, therefore the dynamic prediction is controlled
Treatment scheme can obtain preferable therapeutic effect, can be following to diseased individuals while disease performance evaluation index is optimized
Clinical disease grade makes prediction accurately.
On the basis of above-described embodiment, preferably embodiment, in addition to：
Dynamic evolution rule according to individual sets up the dynamic evolution rule of the corresponding colony of disease type.
In specific implementation, for the Evolution problem analysis between individual, the data to multiple fields are combined
Modeling, so that abundant data avoid over-fitting and more accurately described to individual disease traits.To the people in each field
Corresponding model between group's disease and medicine, it is intended that what they can be as far as possible is close, so as to constrain each other.Formally,
Intend minimizing following object function：
Wherein, UtWith VtThe model in t-th of field is represented, due to multiple UtPresence, human diseases' characteristic can be refined
Ground is modeled；And that last bound term, allow the information of different field to pass through world model U againaShared.
Except the content-data of human diseases and medicine in itself, also there is association between crowd.Due to crowd's social action
The part of properties of crowd is often reacted, as two crowds that are mutually related, its characteristic is often also more similar.By to people
The modeling of group's social action relation chain, we can set up more accurately people's group model U.Specifically, relation chain data can
For constraining people's group model of interrelated crowd.Form, intend minimizing following object function：
Wherein, DikStrength of association between expression crowd.
On the other hand, " new " crowd of no social action can also be modeled.That is, new crowd's disease
The model of disease can be indicated with associated human diseases' model so that model can to new human diseases and medicine it
Between relation accurately predicted, so as to realize the intervention to clinical disease.
The data analysing method of the multi-modal big data provided by the present invention for clinical disease has been carried out in detail above
It is thin to introduce.The embodiment of each in specification is described by the way of progressive, and what each embodiment was stressed is and other realities
Apply the difference of example, between each embodiment identical similar portion mutually referring to.For device disclosed in embodiment
Speech, because it is corresponded to the method disclosed in Example, so description is fairly simple, related part is referring to method part illustration
.It should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention, also
Some improvement and modification can be carried out to the present invention, these are improved and modification also falls into the protection domain of the claims in the present invention
It is interior.
It should also be noted that, in this manual, term " comprising ", "comprising" or its any other variant are intended to contain
Lid nonexcludability is included, so that process, method, article or equipment including a series of key elements not only will including those
Element, but also other key elements including being not expressly set out, or also include being this process, method, article or equipment
Intrinsic key element.In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that
Also there is other identical element in process, method, article or equipment including the key element.
Claims (9)
1. a kind of data analysing method of multi-modal big data for clinical disease, it is characterised in that including：
According to the disease type determined, the corresponding history big data of the disease type is obtained from medical system；
By the history big data, corresponding rate of change designs multi-density quantizer to obtain real time data under each mode
Perceive and acquisition methods；
Multi-modal data method for digging is used to the history big data, and combines convolutional neural networks method, is extracted described
The corresponding characteristic information of disease type；
The individual dynamic evolution rule that the disease type is infected in the history big data are derived according to the characteristic information
Rule；
The Performance Evaluation index of the real time data is obtained with reference to the individual dynamic evolution rule.
2. data analysing method according to claim 1, it is characterised in that it is described by the history big data in each mould
Corresponding rate of change design multi-density quantizer is specifically included with the perception for obtaining real time data with acquisition methods under state：
The history big data is segmented according to the rate of change, and solves the corresponding average value of each segmentation result；
The average value is divided into multiple set by the scope according to the deviation of each average value, and solves the flat of the set
Average；
Calculate the deviation of the measurement data in the set, and deviation and the average value of the set according to the measurement data
Corresponding relation design anticipation function；
The measurement data is standardized according to the anticipation function；
Data cleansing is carried out to the measurement data after standardization, and the multi-density amount is set according to the load characteristic of transmission network
Change device.
3. data analysing method according to claim 2, it is characterised in that described that multimode is used to the history big data
State data digging method is specifically included：
By the influence relation production Methods matrix of disease and hazards；
According to the relational matrix sets target function, and to the object function minimization；
Wherein, the object function is：
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</munder>
<mi>L</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>U</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>V</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&lambda;</mi>
<mi>R</mi>
<mrow>
<mo>(</mo>
<mi>U</mi>
<mo>,</mo>
<mi>V</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>&lambda;</mi>
<mi>u</mi>
</msub>
<msub>
<mi>R</mi>
<mi>u</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>U</mi>
<mo>,</mo>
<msub>
<mi>X</mi>
<mi>u</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>&lambda;</mi>
<mi>v</mi>
</msub>
<msub>
<mi>R</mi>
<mi>v</mi>
</msub>
<mo>(</mo>
<mrow>
<mi>V</mi>
<mo>,</mo>
<msub>
<mi>X</mi>
<mi>v</mi>
</msub>
</mrow>
<mo>)</mo>
<mo>;</mo>
</mrow>
Wherein, MijFor the relational matrix, U is disease, and V is hazards, XuRepresent the feature of disease, XvFor hazards
Feature, R (U, V) is the constraint of the regularization to U and V, Ru(U,Xu) for U to XuRegularization constraint, Rv(V,Xv) for V to XvRule
Then change constraint, λ, λuAnd λvRespectively the weight of rule of correspondence bound term, is worth between 0~1, L (Ui,Vj,Mij) it is Ui、ViWith
MijRelation function.
4. data analysing method according to claim 3, it is characterised in that the combination convolutional neural networks method, is carried
The corresponding characteristic information of the disease type is taken out to specifically include：
Type according to the affiliated mode of the measurement data sets up a variety of convolution kernels；
Mixing sampling is carried out to the convolution kernel and obtains the characteristic information；
The characteristic information is connected entirely with the output layer of the convolutional neural networks.
5. data analysing method according to claim 1, it is characterised in that the individual dynamic evolution rule is specifically led to
Following steps are crossed to obtain：
The fuzzy dynamic set up using fuzzy logic system identification method between the characteristic information and diseased individuals Evolution is controlled
Treat model.
6. data analysing method according to claim 5, it is characterised in that the dynamic evolution rule individual with reference to described in
The Performance Evaluation index that rule obtains the real time data is specifically included：
Object function in model is treated based on the fuzzy dynamic, disease Performance Evaluation is set up by iteration optimization algorithms and refers to
Mark；
Calculate the minimum value of the disease performance evaluation index.
7. data analysing method according to claim 6, it is characterised in that also include：
Based on the disease performance evaluation index, optimal disease treatment sequence is obtained using rolling optimization algorithm.
8. data analysing method according to claim 7, it is characterised in that also include：
Feedback correction is carried out to the disease treatment sequence by the real time data.
9. data analysing method according to claim 1, it is characterised in that also include：
The dynamic evolution rule of the corresponding colony of the disease type is set up according to the individual dynamic evolution rule.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447993.0A CN107273685A (en) | 2017-06-14 | 2017-06-14 | A kind of data analysing method of multi-modal big data for clinical disease |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447993.0A CN107273685A (en) | 2017-06-14 | 2017-06-14 | A kind of data analysing method of multi-modal big data for clinical disease |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107273685A true CN107273685A (en) | 2017-10-20 |
Family
ID=60067577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710447993.0A Pending CN107273685A (en) | 2017-06-14 | 2017-06-14 | A kind of data analysing method of multi-modal big data for clinical disease |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107273685A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198621A (en) * | 2018-01-18 | 2018-06-22 | 中山大学 | A kind of database data synthesis dicision of diagnosis and treatment method based on neural network |
| CN108538397A (en) * | 2017-12-23 | 2018-09-14 | 天津国科嘉业医疗科技发展有限公司 | A kind of influenza trend predicting system and method based on particle filter model |
| CN108846503A (en) * | 2018-05-17 | 2018-11-20 | 电子科技大学 | A kind of respiratory disease illness person-time dynamic prediction method neural network based |
| CN109754881A (en) * | 2017-11-03 | 2019-05-14 | 中国移动通信有限公司研究院 | A kind of appraisal procedure and device of community's screening scheme |
| CN110610747A (en) * | 2019-10-10 | 2019-12-24 | 桂林理工大学 | Deep learning-based micro chemical experiment system and method |
| CN110619011A (en) * | 2019-10-10 | 2019-12-27 | 桂林理工大学 | Big data analysis-based pollution treatment system and method for chemical laboratory in colleges and universities |
| CN111265317A (en) * | 2020-02-10 | 2020-06-12 | 上海牙典医疗器械有限公司 | Tooth orthodontic process prediction method |
| CN111462909A (en) * | 2020-03-30 | 2020-07-28 | 安徽科大讯飞医疗信息技术有限公司 | Disease evolution tracking and disease condition prompting method and device and electronic equipment |
| 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 |
| CN113707308A (en) * | 2021-08-31 | 2021-11-26 | 平安国际智慧城市科技股份有限公司 | Medical data analysis device and computer readable storage medium |
| CN115331116A (en) * | 2022-10-13 | 2022-11-11 | 潍坊绘圆地理信息有限公司 | Ground target accurate identification on-satellite fuzzy inference method based on multi-mode data |
| CN115376698A (en) * | 2022-10-25 | 2022-11-22 | 北京鹰瞳科技发展股份有限公司 | Apparatus, method, and storage medium for predicting progression of fundus disease |
-
2017
- 2017-06-14 CN CN201710447993.0A patent/CN107273685A/en active Pending
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109754881A (en) * | 2017-11-03 | 2019-05-14 | 中国移动通信有限公司研究院 | A kind of appraisal procedure and device of community's screening scheme |
| CN108538397A (en) * | 2017-12-23 | 2018-09-14 | 天津国科嘉业医疗科技发展有限公司 | A kind of influenza trend predicting system and method based on particle filter model |
| CN108198621A (en) * | 2018-01-18 | 2018-06-22 | 中山大学 | A kind of database data synthesis dicision of diagnosis and treatment method based on neural network |
| CN108198621B (en) * | 2018-01-18 | 2022-03-08 | 中山大学 | Database data comprehensive diagnosis and treatment decision method based on neural network |
| CN108846503A (en) * | 2018-05-17 | 2018-11-20 | 电子科技大学 | A kind of respiratory disease illness person-time dynamic prediction method neural network based |
| CN110610747B (en) * | 2019-10-10 | 2023-08-18 | 桂林理工大学 | Micro chemical experiment system and method based on deep learning |
| CN110610747A (en) * | 2019-10-10 | 2019-12-24 | 桂林理工大学 | Deep learning-based micro chemical experiment system and method |
| CN110619011A (en) * | 2019-10-10 | 2019-12-27 | 桂林理工大学 | Big data analysis-based pollution treatment system and method for chemical laboratory in colleges and universities |
| CN111265317A (en) * | 2020-02-10 | 2020-06-12 | 上海牙典医疗器械有限公司 | Tooth orthodontic process prediction method |
| CN111462909A (en) * | 2020-03-30 | 2020-07-28 | 安徽科大讯飞医疗信息技术有限公司 | Disease evolution tracking and disease condition prompting method and device and electronic equipment |
| CN111462909B (en) * | 2020-03-30 | 2024-04-05 | 讯飞医疗科技股份有限公司 | Disease evolution tracking and disease condition prompting method and device and electronic equipment |
| 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 |
| CN113707308A (en) * | 2021-08-31 | 2021-11-26 | 平安国际智慧城市科技股份有限公司 | Medical data analysis device and computer readable storage medium |
| CN115331116A (en) * | 2022-10-13 | 2022-11-11 | 潍坊绘圆地理信息有限公司 | Ground target accurate identification on-satellite fuzzy inference method based on multi-mode data |
| CN115376698A (en) * | 2022-10-25 | 2022-11-22 | 北京鹰瞳科技发展股份有限公司 | Apparatus, method, and storage medium for predicting progression of fundus disease |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107273685A (en) | A kind of data analysing method of multi-modal big data for clinical disease | |
| ȚĂRANU | Data mining in healthcare: decision making and precision. | |
| Xie et al. | Blockchain financial investment based on deep learning network algorithm | |
| CN112216396B (en) | Method for predicting drug-side effect relationship based on graph neural network | |
| Hayashi et al. | A comparison between two neural network rule extraction techniques for the diagnosis of hepatobiliary disorders | |
| CN107944559A (en) | A kind of entity relationship automatic identifying method and system | |
| CN103116707A (en) | Heart disease intelligent diagnostic method based on case reasoning | |
| Begum et al. | Classification of physiological signals for wheel loader operators using multi-scale entropy analysis and case-based reasoning | |
| Jiang et al. | A hybrid intelligent model for acute hypotensive episode prediction with large-scale data | |
| Mohammedqasim et al. | Real-time data of COVID-19 detection with IoT sensor tracking using artificial neural network | |
| Gupta et al. | Time-series imputation and prediction with bi-directional generative adversarial networks | |
| Hati | Convolutional neural network-long short term memory optimization for accurate prediction of airflow in a ventilation system | |
| Zhu et al. | Knowledge graph and deep learning based pest detection and identification system for fruit quality | |
| Abraham et al. | Convolutional neural network for biomedical applications | |
| Mehta et al. | Collaborative Intelligence in AgriTech: Federated Learning CNN for Bean Leaf Disease Classification | |
| Taherinezhad et al. | COVID-19 crisis management: Global appraisal using two-stage DEA and ensemble learning algorithms | |
| Wang et al. | Broad learning system with Takagi–Sugeno fuzzy subsystem for tobacco origin identification based on near infrared spectroscopy | |
| Mu et al. | Study on exercise muscle fatigue based on sEMG and ECG data fusion and temporal convolutional network | |
| Irshad et al. | Convolutional neural network enable optoelectronic system for predicting cardiac response by analyzing auction-based optimization algorithms | |
| Guo et al. | Classification of domestic refuse in medical institutions based on transfer learning and convolutional neural network | |
| DE202023101305U1 (en) | An intelligent health and fitness data management system using artificial intelligence with IoT devices | |
| Sinha et al. | Automated detection of coronary artery disease using machine learning algorithm | |
| Abdulkareem et al. | A manifold intelligent decision system for fusion and benchmarking of deep waste-sorting models | |
| Tang et al. | Explainable and efficient deep early warning system for cardiac arrest prediction from electronic health records | |
| Alharbi et al. | Spatiotemporal wavelet-domain neuroimaging of chaotic EEG seizure signals in epilepsy diagnosis and prognosis with the use of graph convolutional LSTM networks |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171020 |
|
| WD01 | Invention patent application deemed withdrawn after publication |