CN110223770A - Gastroenterology electronic data analysis method - Google Patents
Gastroenterology electronic data analysis method Download PDFInfo
- Publication number
- CN110223770A CN110223770A CN201910454851.6A CN201910454851A CN110223770A CN 110223770 A CN110223770 A CN 110223770A CN 201910454851 A CN201910454851 A CN 201910454851A CN 110223770 A CN110223770 A CN 110223770A
- Authority
- CN
- China
- Prior art keywords
- function
- analysis
- mistake
- result
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- 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
- G16H70/60—ICT specially adapted for the handling or processing of medical references relating to pathologies
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Theoretical Computer Science (AREA)
- Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The embodiment of the present invention discloses a kind of gastroenterology electronic data analysis method, it include: to obtain enteroscopy result and pathologic finding conclusion, and enteroscopy result and pathologic finding conclusion are compared, show that intestinal cancer analyzes result, result statistical nature is analyzed for intestinal cancer, construct nonhomogeneous Poisson process NHPP class diseases analysis reliability growth model, Training Support Vector Machines SVM model, using model accuracy rate as evaluation index, using diseases analysis reliability growth model each time reliability result as penalty factor, associated losses function optimizes SVM model together and carries out intestinal cancer judgement.The embodiment of the present invention can analyze gastroenterology electronic data, obtain more accurate judging result.
Description
Technical field
The present invention relates to data analysis field more particularly to a kind of gastroenterology electronic data analysis methods.
Background technique
The medical datas such as intestinal inspection result can reflect enteron aisle situation, if to the medical datas such as intestinal inspection result into
Row analysis, it will be appreciated that intestinal cancer situation.How the current needs of analysis are carried out to medical datas such as intestinal inspection results to solve
Technical problem.
Summary of the invention
The embodiment of the present invention provides a kind of gastroenterology electronic data analysis method, can to gastroenterology electronic data into
Row analysis obtains analysis result.
The embodiment of the present invention adopts the following technical scheme that
A kind of gastroenterology electronic data analysis method, comprising:
Obtain enteroscopy result and pathologic finding conclusion, and by the enteroscopy result and the pathologic finding conclusion
It is compared, show that intestinal cancer analyzes result;
Result statistical nature is analyzed for the intestinal cancer, building nonhomogeneous Poisson process NHPP class diseases analysis reliability increases
Long model;Training Support Vector Machines SVM model, using model accuracy rate as evaluation index, by the diseases analysis reliability growth
Model each time reliability result as penalty factor, associated losses function optimizes SVM model together, carries out intestinal cancer and sentences
It is disconnected.
Optionally, the acquisition enteroscopy result and pathologic finding conclusion, and by the enteroscopy result and described
Pathologic finding conclusion is compared, and show that intestinal cancer analysis result includes:
Characteristic information is extracted in enteroscopy report and pathologic finding report from the same sufferer, is believed according to individual features
Breath carries out report splicing integration;During integrating to data, it is subject to pathological examination results;Wherein, the extraction
Characteristic information includes: the extraction of the extraction of text-type feature, the extraction of temporal characteristics and patient's essential information feature;
The text-type feature is integrated into construction feature space, proposes corresponding disease category from pathologic finding report
Descriptor as label (Label), construction output space (Label Space), the attribute for being characterized each feature in space takes
Value and output space setting numeric coding rule, by the numeric coding that sets rule, to the data reporting after integration into
Line number value indicates, becomes computer and algorithm model can recognize and the numeric type data (Numerical of study
Data);
After carrying out numeralization expression to data, using pathologic finding conclusion as final conclusion, by enteroscopy conclusion and disease
Reason checks that conclusion is compared, and using the moon as time partition dimension, counts the analysis result of intestinal cancer every month.
Optionally, the extraction of the text-type feature includes: that the text-processing done to enteroscopy text data is known as
Medical Language processing, it is main during this to complete the participle to enteroscopy text data, noise elimination, and extract specific disease
The positive description of disease is used as characteristic information;
The extraction of the temporal characteristics includes: the character string type number for being Interval Coding division check data in audit report
According to, including the date, the moon is therefrom extracted as temporal characteristics;
The extraction of patient's essential information feature includes: that the feature of extraction includes gender, age, occupation.
Optionally, described to analyze result statistical nature for the intestinal cancer, construct nonhomogeneous Poisson process NHPP class disease
Analyzing reliability growth model includes:
Nonhomogeneous Poisson process NHPP is set;
Construct NHPP class calculated result reliability growth model frame;
Construct the NHPP class diseases analysis reliability growth model to tally with the actual situation;
To the parameter Estimation in the NHPP class diseases analysis reliability growth model to tally with the actual situation.
Optionally, the setting nonhomogeneous Poisson process NHPP includes:
One random counting process { N (t), t >=0 } is set meets A1 to A4:N and represent a counting process, can indicate
Quantity number, t indicate the time;
A1:N (0)=0;
A2:{ N (t), t >=0 } it is an independent incremental process;
A3:P [N (t+ Δ t)-N (t)]=1=λ (t) Δ t+o (Δ t);
λ (t) indicates that the intensity function of nonhomogeneous Poisson process, Δ t indicate a time interval, and (Δ t) indicates Δ t's to o
Higher-order shear deformation function;
A4:P {-N (s) >=2 N (t) }=o (Δ t);
Then claiming { N (t), t >=0 } is the nonhomogeneous Poisson process with intensity λ (t), as λ (t)=λ, nonhomogeneous Poisson
Process is exactly common homogeneous Poisson processes;
The probability distribution formula of nonhomogeneous Poisson process is as follows:
S indicates next time, identical as t meaning;
N (t): the cumulative error analysis times found within [0, the t] period;
M (t): the desired value of cumulative error diagnosis number, m (t)=E [N (t)] in [0, the t] period;
X (t): until moment t, it is detected and belongs to the error analysis number that repetition mistake is examined;
A (t): disease mistake examines total function, indicates until moment t, the diseases analysis errors number counted in case
Summation;
a0: errors number is analyzed present in case when statistics starts;
B: analysis error rate indicates that each mistake examines the probability counted in case;
P (t): mistake examines repetitive rate function, indicate in moment t, and the error analysis being each detected, which belongs to, to be repeated
Probability;
R (x | t): disease reliability function is indicated since moment t in the t+x period, the reliability of diseases analysis.
Optionally, the building NHPP class calculated result reliability growth model frame includes:
Based on the probability distribution formula of the nonhomogeneous Poisson process, set:
B1: the cumulative error analysis times N (t) to time t obeys the Poisson process that mean function is m (t).Any time
Interval t to t+ Δ t in desired mistake examine generation number examined to the remaining mistake of t moment it is several proportional.
B2: disease mistake examine quantity in varying environment, different moments be it is different, disease error analysis sum is at any time
Variation.
B3: identical mistake examines situation and is likely to occur in different time sections, and mistake examines the function that repetitive rate is the time.
B4: each error analysis is independent from each other in case, and consequence seriousness caused by each analysis mistake is different;
By assuming B1: having B5, B5:m (t+ Δ t)=b (a (t)-x (t)) Δ t+o (Δ t)
(a (t)-x (t)) is indicated until moment t, is detected and is not belonging to repeat the wrong analysis errors number examined
So as to obtain the differential equation
By assuming that B3 has B7, B7:
Indicate x (t) to t carry out derivation, behind be also same meaning;
B8, B8 can be obtained by equation B6, equation B7:
Indicate x (t) to t carry out derivation, behind be also same meaning;
Equation B6, equation B7 primary condition be B9, B10, B9:m (0)=0, B10:x (0)=0;
B11 can be obtained by formula B8, formula B10,
B11:
Exp is the meaning of index, and t, u are the distinct symbols for indicating the time, and dt, du quadrature;
It is B12 to which the cumulative analysis mistake mean function of model can be solved by formula B6, formula B9:
Since the cumulative error analysis times N (t) until moment t obeys the nonhomogeneous Poisson distribution of mean value m (t), institute
With B13:
According to nonhomogeneous Poisson be distributed property, Reliability Function B14:
R (x | t)=1-P {-N (t)=0 N (t+x) }=1-exp [- (m (t+x)-m (t))].
Optionally, the NHPP class diseases analysis reliability growth model to tally with the actual situation that constructs includes:
The total function of analysis mistake describes B15:a (t)=a with following function0(1+αt)
Wherein, the size of α < 0, α determine the wrong size for examining total function decrease speed;
Mistake, which examines repetitive rate function p (t), should meet following condition: p (t) ∈ [0,1] and p (t) is decreasing function, when t →
When ∞, p (t) → 0.Therefore, following function can be chosen and examine repetitive rate function B16 to define mistake:
The size of wherein k > 0, k determine the wrong speed for examining repetitive rate variation;
It brings formula B15, formula B16 into formula B11, formula B12, obtains:
B17:
B18:
After obtaining the mean function m (t) of cumulative analysis errors number, so that it may be obtained in m (t) using Parameter Estimation Method
Parameter.
Optionally, the parameter Estimation packet in the described pair of NHPP class diseases analysis reliability growth model to tally with the actual situation
It includes:
The parameter in formula B18 is estimated using Maximum Likelihood Estimation Method, examining mean function m (t) by mistake can obtain seemingly
Right function such as B19:
Wherein, (ti, ni) occur in pairs, niIndicate tiCounted in moment case institute it is wrong examine number only and;tiIt indicates
Counting on mistake and examining number summation is niAt the time of, L (parameters | (ti, ni)) indicate maximum likelihood function, subsequent is to ask
Product code, exp are to indicate exponent arithmetic;
Take natural logrithm that can obtain B20 formula B19:
Ln indicates that logarithm operation, ∑ indicate summation operation,!Indicate factorial operation;
Each estimates of parameters can be acquired to above-mentioned formula B20 differential.
Optionally, the Training Support Vector Machines SVM model, using model accuracy rate as evaluation index, by the disease point
Reliability growth model is analysed in the reliability result of each time as penalty factor, associated losses function optimizes SVM mould together
Type, carrying out intestinal cancer judgement includes:
Using linear separability support vector machines learning algorithm, i.e., maximal margin method come Training Support Vector Machines SVM model with
Optimal separating hyperplane is found, algorithm description is as follows:
Input: linear separability training dataset T={ (x1, y1), (x2, y2) ..., (xn, yn), wherein xi∈ χ=Rn,
yi∈ γ={ -1 ,+1 } is ith feature vector, also referred to as example, yiFor xiClass label, work as yiWhen=+ 1, claim xiFor positive example;
yi=-1 at that time, claims xiBe negative example, (xi, yi) it is known as sample point;
Output: largest interval separating hyperplance and categorised decision function;
(1) it constructs and solves constrained optimization problem:
s.t yi(w·xi+ b) -1 >=0, i=1,2 ..., N B22
Acquire optimal solution w*, b*;Min expression is minimized, and w, b indicate to constitute two ginsengs of largest interval separating hyperplance
Number, Xi indicate the example in training dataset;
(2) separating hyperplance is thus obtained are as follows:
w*·x+b*=0 B23
Categorised decision function are as follows:
F (x)=sign (w*·x+b*) B24
Wherein: | | w | | it is the L of w2Norm, (w, b) are given hyperplane;
By using above-mentioned algorithm Training Support Vector Machines model, optimal separating hyperplane is found, feature space is drawn
It is divided into two parts, a part is positive class, and a part is negative class, so that data set be classified.
Optionally, after obtaining hyperplane further include:
To each sample point (xi, yi) introduce a slack variable ξi>=0, so that function interval is greater than plus slack variable
Equal to 1, at this point, constraint condition becomes: yi(w·xi+b)≥1-ξi.Meanwhile to each slack variable ξi, pay a cost
Function ξi, objective function is also by originalBecome:N value range is all nonnegative integers
Set;
Wherein, C > 0 is known as punishment parameter, its value is different under different Question backgrounds, to misclassification when C value is big
Punishment increases, and C value hour reduces the punishment of misclassification, and the objective function after change includes keeping interval big as far as possible and misclassified gene
Quantity two layers of meaning small as far as possible, C belong to reconcile the relationship of the two variable;
The problems of value of C is combined with disease reliability issues, value, that is, disease Reliability Function of C is in each time
The calculated result of section, objective function become:To disease misclassification when data calculating reliability is big
Punishment increase, reliability hour, which punishes the misclassification of disease, to be reduced, at this point, by NHPP class diseases analysis reliability growth mould
Type is applied in the optimization method of basic SVM model, trains satisfactory more granularity enteroscopy report analysis models.
Gastroenterology electronic data analysis method based on the above-mentioned technical proposal obtains enteroscopy result and pathologic finding
Conclusion, and enteroscopy result and pathologic finding conclusion are compared, show that intestinal cancer is analyzed as a result, analyzing result for intestinal cancer
Statistical nature constructs nonhomogeneous Poisson process NHPP class diseases analysis reliability growth model, Training Support Vector Machines SVM mould
Type, using model accuracy rate as evaluation index, using diseases analysis reliability growth model each time reliability result as
Penalty factor, associated losses function optimize together SVM model carry out intestinal cancer judgement, so as to gastroenterology electronic data into
Row analysis, obtains more accurate intestinal cancer judging result.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows and meets implementation of the invention
Example, and be used to explain the principle of the present invention together with specification.
Fig. 1 is the flow chart of the gastroenterology electronic data analysis method shown in the embodiment of the present invention.
Fig. 2 is the feature extraction schematic diagram shown in the embodiment of the present invention.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistented with the present invention.On the contrary, they be only with it is such as appended
The example of device and method being described in detail in claims, some aspects of the invention are consistent.
As shown in Figure 1, the embodiment of the present invention provides a kind of gastroenterology electronic data analysis method, comprising:
11, enteroscopy result and pathologic finding conclusion are obtained, and by the enteroscopy result and the pathologic finding
Conclusion is compared, and show that intestinal cancer analyzes result.
For example, obtaining the analysis error rate of intestinal cancer every month.
Example is applied in the invention of this reality, is integrated enteroscopy report and pathologic finding report, sample data set is formed, after integration
Audit report data carry out the processing such as the numeralization expression of feature extraction and report;
12, result statistical nature is analyzed for the intestinal cancer, constructs nonhomogeneous Poisson process NHPP (non-
Homogeneous Poisson process, referred to as: NHPP) class diseases analysis reliability growth model.
Example is applied in the invention of this reality, which, which had not only considered identical mistake during diseases analysis, which examines situation, to repeat, but also
It considers that mistake examines recurrence probability and mistake examines sum may all change at any time, model is made more to tally with the actual situation.
13, Training Support Vector Machines SVM model, with model accuracy rate (accuracy) for evaluation index, by the disease
Reliability growth model is analyzed in the reliability result of each time as penalty factor, associated losses function optimizes SVM together
Model carries out intestinal cancer judgement.
The gastroenterology electronic data analysis method of the embodiment of the present invention obtains enteroscopy result and pathologic finding knot
By, and enteroscopy result and pathologic finding conclusion are compared, show that intestinal cancer is analyzed as a result, for intestinal cancer analysis result system
Meter feature, building nonhomogeneous Poisson process NHPP class diseases analysis reliability growth model, Training Support Vector Machines SVM model,
Using model accuracy rate as evaluation index, using diseases analysis reliability growth model each time reliability result as punishment
The factor, associated losses function optimizes SVM model together and carries out intestinal cancer judgement, so as to divide gastroenterology electronic data
Analysis, obtains more accurate intestinal cancer judging result.
In one embodiment, the acquisition enteroscopy result and pathologic finding conclusion, and by the enteroscopy knot
Fruit and the pathologic finding conclusion are compared, and show that intestinal cancer analysis result includes:
By enteroscopy report (Colonoscopy report) and pathologic finding report (Pathological report)
It is integrated, i.e., characteristic information is extracted in the enteroscopy report from the same sufferer and pathologic finding report, according to corresponding special
Sign information field (such as admission number) carries out report splicing integration;Enteroscopy report includes personal patient information, checks finding, intestines
Mirror analyzes a plurality of types of fields such as result, check data;And pathologic finding report includes pathologic finding position, pathologic finding institute
See and the information such as pathologic finding record.In medical domain, pathological examination results are considered as " goldstandard ", i.e., the generation of canceration whether
It can be obtained by carrying out pathological examination as a result, " ill " or " disease-free " can correctly be distinguished.Therefore, we to data into
During row integration, be subject to pathological examination results, enteroscopy result and pathological examination results it is inconsistent will be considered to intestines
Spectroscopy the result is that mistake;Wherein, described to extract the extraction that characteristic information includes: the extraction of text-type feature, temporal characteristics
With the extraction of patient's essential information feature;Wherein, the extraction of the text-type feature: the text that enteroscopy text data is done
Present treatment is known as Medical Language processing, and participle, the noise of enteroscopy text data are eliminated in main completion during this, and
The positive description for extracting particular condition is used as characteristic information;Wherein, the extraction of the temporal characteristics: in view of the season of disease incidence
Saving sexual factor, i.e., seasonal, periodic changing rule would generally be presented in the disease incidence of certain diseases, thus it is considered that disease
Breaking-out and inherent implicit contacted between season there are certain.Check data is the word that Interval Coding divides in audit report
Serial type data, including date are accorded with, therefrom extract the moon as temporal characteristics;Wherein, the extraction of patient's essential information feature:
The processing that we are done is mainly unified to lead to the problem of that expression is inconsistent, and the feature of extraction includes since data source is different
Gender, age, occupation etc..
To as shown in Figure 2 before and after audit report feature extraction.
Construction feature space is integrated by the text-type feature to extraction.Meanwhile it being mentioned from pathologic finding report
It is each to be characterized space as label (Label), construction output space (Label Space) for the descriptor of corresponding disease category out
The attribute value and output space setting numeric coding rule of a feature, by the numeric coding rule set, to integration
Data reporting afterwards carries out numeralization expression, becomes computer and algorithm model can recognize and the numeric type data of study
(Numerical Data);
After carrying out numeralization expression to data, using pathologic finding conclusion as final conclusion, by enteroscopy conclusion and disease
Reason checks that conclusion is compared, and using the moon as time partition dimension, counts the analysis analysis of intestinal cancer every month as a result, being convenient for subsequent
The building of intestinal cancer reliability growth model.
In one embodiment, described to analyze result statistical nature for the intestinal cancer, construct nonhomogeneous Poisson process
NHPP class diseases analysis reliability growth model includes:
Nonhomogeneous Poisson process NHPP is set;
Construct NHPP class calculated result reliability growth model frame;
Construct the NHPP class diseases analysis reliability growth model to tally with the actual situation;
To the parameter Estimation in the NHPP class diseases analysis reliability growth model to tally with the actual situation.
The embodiment of the present invention proposes a kind of new NHPP alanysis reliability growth model.The model had both considered identical
Analysis error situation may repeat, it is contemplated that the wrong recurrence probability of analysis and mistake are examined sum and may all be become at any time
Change, improves such prediction and evaluation capacity for calculating reliability growth model, model is made more to tally with the actual situation.
Nonhomogeneous Poisson process (non-homogeneous Poisson process, abbreviation NHPP) is Poisson process
One popularization, below we will provide the definition of nonhomogeneous Poisson process.In one embodiment, the setting nonhomogeneous Poisson
Process NHPP includes: that one random counting process { N (t), t >=0 } of setting meets A1 to A4:N and represents a counting process, can be with
Indicate quantity number, t indicate the time;
A1:N (0)=0;
A2:{ N (t), t >=0 } it is an independent incremental process;
A3:P [N (t+ Δ t)-N (t)]=1=λ (t) Δ t+o (Δ t);
λ (t) indicates that the intensity function of nonhomogeneous Poisson process, Δ t indicate a time interval, and (Δ t) indicates Δ t's to o
Higher-order shear deformation function;
A4:P {-N (s) >=2 N (t) }=o (Δ t);
Then claiming { N (t), t >=0 } is the nonhomogeneous Poisson process with intensity λ (t), as λ (t)=λ, nonhomogeneous Poisson
Process is exactly common homogeneous Poisson processes;
The probability distribution formula of nonhomogeneous Poisson process is as follows:
S indicates next time, identical as t meaning;
N (t): the cumulative error analysis times found within [0, the t] period;
M (t): the desired value of cumulative error analysis times in [0, the t] period, m (t)=E [N (t)];
X (t): until moment t, it is detected and belongs to the error analysis number that repetition mistake is examined;
A (t): disease mistake examines total function, indicates until moment t, the diseases analysis errors number counted in case
Summation:
a0: errors number is analyzed present in case when statistics starts;
B: analysis error rate indicates that each mistake examines the probability counted in case;
P (t): mistake examines repetitive rate function, indicate in moment t, and the error analysis being each detected, which belongs to, to be repeated
Probability;
R (x | t): disease reliability function is indicated since moment t in the t+x period, the reliability of diseases analysis.
The reliability growth model that the embodiment of the present invention is established mainly considers following two problem:
(1) diseases analysis process can continually introduce new mistake and examine case, i.e. diseases analysis mistake sum function a (t) be with
Time is changed;
(2) each mistake examines case and cannot be guaranteed to repeat, that is, there are problems that mistake examines repetitive rate size.While with
The migration of time, department's Medical Devices are increasingly advanced and doctors experience gradually increases, mistake examine repetitive rate function also with
Time changes.
It is needed to construct NHPP class diseases analysis reliability growth model frame to warp based on original reliability model
The NHPP class model of allusion quotation is modified and is supplemented, and assumed condition below is formd:
In one embodiment, the building NHPP class calculated result reliability growth model frame includes:
Based on the probability distribution formula of the nonhomogeneous Poisson process, set:
B1: the cumulative error analysis times N (t) to time t obeys the Poisson process that mean function is m (t).Any time
Interval t to t+ Δ t in desired mistake examine generation number examined to the remaining mistake of t moment it is several proportional.
B2: disease mistake examine quantity in varying environment, different moments be it is different, disease error analysis sum is at any time
Variation.
B3: identical mistake examines situation and is likely to occur in different time sections, and mistake examines the function that repetitive rate is the time.
B4: each error analysis is independent from each other in case, and consequence seriousness caused by each analysis mistake is different;
By assuming B1: having B5, B5:m (t+ Δ t)=b (a (t)-x (t)) Δ t+o (Δ t)
(a (t)-x (t)) is indicated until moment t, is detected and is not belonging to repeat the wrong analysis errors number examined
So as to obtain differential equation B6:
By assuming that B3 has B7, B7:
Indicate x (t) to t carry out derivation, behind be also same meaning;
B8, B8 can be obtained by equation B6, equation B7:
Equation B6, equation B7 primary condition be B9, B10, B9:m (0)=0, B10:x (0)=0;
B11 can be obtained by formula B8, formula B10,
B11:
Exp is the meaning of index, and t, u are the distinct symbols for indicating the time, and dt, du quadrature;
It is B12 to which the cumulative analysis mistake mean function of model can be solved by formula B6, formula B9:
Since the cumulative error analysis times N (t) until moment t obeys the nonhomogeneous Poisson distribution of mean value m (t), institute
With B13:
According to nonhomogeneous Poisson be distributed property, Reliability Function B14:
R (x | t)=1-P {-N (t)=0 N (t+x) }=1-exp [- (m (t+x)-m (t))].
In one embodiment, described to construct the NHPP class diseases analysis reliability growth model packet to tally with the actual situation
It includes:
The total function of analysis mistake describes B15:a (t)=a with following function0(1+αt)
Wherein, the size of α < 0, α determine the wrong size for examining total function decrease speed;
Mistake, which examines repetitive rate function p (t), should meet following condition: p (t) ∈ [0,1] and p (t) is decreasing function, when t →
When ∞, p (t) → 0.Therefore, following function can be chosen and examine repetitive rate function B16 to define mistake:
The size of wherein k > 0, k determine the wrong speed for examining repetitive rate variation;
It brings formula B15, formula B16 into formula B11, formula B12, obtains:
B17:
B18:
After obtaining the mean function m (t) of cumulative analysis errors number, so that it may be obtained in m (t) using Parameter Estimation Method
Parameter.
In one embodiment, the ginseng in the described pair of NHPP class diseases analysis reliability growth model to tally with the actual situation
Number is estimated
The parameter in formula B18 is estimated using Maximum Likelihood Estimation Method, examining mean function m (t) by mistake can obtain seemingly
Right function such as B19:
Wherein, (ti, ni) occur in pairs, niIndicate tiCounted in moment case institute it is wrong examine number only and;tiIt indicates
Counting on mistake and examining number summation is niAt the time of, L (parameters | (ti, ni)) indicate maximum likelihood function, subsequent is to ask
Product code, exp are to indicate exponent arithmetic;
Take natural logrithm that can obtain B20 formula B19:
Ln indicates that logarithm operation, ∑ indicate summation operation,!Indicate factorial operation;
Each estimates of parameters can be acquired to above-mentioned formula B20 differential.
Support vector machines (Support Vector Machine, abbreviation SVM) model is a kind of two disaggregated models, is being solved
It needs specifically to be promoted when more classification problems.The model shows brilliance in text categorization task and high dimensional data
Can, and become the mainstream technology for leading machine learning trend.Its basic model is the line in particular space with largest interval
Property classifier, core is that original training set is mapped to high-dimensional feature space, wherein non-linear separation characteristic is sentenced by High-dimensional Linear
Replaced other function, it is a kind of supervised learning method that can be widely applied to statistical classification and regression analysis.Meanwhile it
Many unique advantages are shown in terms of solving small sample, non-linear and high dimensional pattern identification, and can be applied to other
In Machine Learning Problems, such as Function Fitting.Supporting vector machine model is VC dimension theory and structure based on Statistical Learning Theory
Least risk principle, according to limited sample information model complexity (i.e. to the study precision of specific training sample,
Accuracy) seek optimal trade-off between learning ability (identifying the ability of arbitrary sample without error), to obtain
Best Generalization Ability (or generalization ability).In SVM model, kernel function directly determines support vector machines and kernel method
Final performance, but the selection of kernel function depends on particular problem, and Multiple Kernel Learning also can be used and pass through the multiple kernel functions of study
The optimal combination of acquisition improves the performance of model as final kernel function.
In one embodiment, the Training Support Vector Machines SVM model, using model accuracy rate as evaluation index, by institute
Diseases analysis reliability growth model is stated in the reliability result of each time as penalty factor, associated losses function is excellent together
Change SVM model, carrying out intestinal cancer judgement includes:
Using linear separability support vector machines learning algorithm, i.e., maximal margin method come Training Support Vector Machines SVM model with
Optimal separating hyperplane is found, algorithm description is as follows:
Input: linear separability training dataset T={ (x1, y1), (x2, y2) ..., (xn, yn), wherein xi∈ χ=Rn,
yi∈ γ={ -1 ,+1 } is ith feature vector, also referred to as example, yiFor xiClass label, work as yiWhen=+ 1, claim xiFor positive example;
Work as yiWhen=- 1, claim xiBe negative example, (xi, yi) it is known as sample point;
Output: largest interval separating hyperplance and categorised decision function;
(1) it constructs and solves constrained optimization problem:
s.t yi(w·xi+ b) -1 >=0, i=1,2 ..., N B22
Acquire optimal solution w*, b*;Min expression is minimized, and w, b indicate to constitute two ginsengs of largest interval separating hyperplance
Number, XiIndicate the example in training dataset;
(2) separating hyperplance is thus obtained are as follows:
w*·x+b*=0 B23
Categorised decision function are as follows:
F (x)=sign (w*·x+b*) B24
Wherein: | | w | | it is the L of w2Norm, (w, b) are given hyperplane;
By using above-mentioned algorithm Training Support Vector Machines model, optimal separating hyperplane is found, feature space is drawn
It is divided into two parts, a part is positive class, and a part is negative class, so that data set be classified.
Currently, support vector machines (SVM) model is applied to clinic by existing many researchs, the classification that disease may be implemented is known
Not, the probability of happening of predictive disease.The embodiment of the present invention is by constructing a kind of more granularity enteroscopy report analysis models to this
The text of mode input patient's enteroscopy describes data, can automate generation analysis result.
In one embodiment, after obtaining hyperplane further include:
To each sample point (xi, yi) introduce a slack variable ξi>=0, so that function interval is greater than plus slack variable
Equal to 1, at this point, constraint condition becomes: yi(w·xi+b)≥1-ξi.Meanwhile to each slack variable ξi, pay a cost
Function ξi, objective function is also by originalBecome:
Wherein, C > 0 is known as punishment parameter, its value is different under different Question backgrounds, to misclassification when C value is big
Punishment increases, and C value hour reduces the punishment of misclassification, and the objective function after change includes keeping interval big as far as possible and misclassified gene
Quantity two layers of meaning small as far as possible, C belong to reconcile the relationship of the two variable;
The problems of value of C is combined with disease reliability issues, value, that is, disease Reliability Function of C is in each time
The calculated result of section, objective function become:N value range is the set of all nonnegative integers;
The punishment of disease misclassification is increased when data calculating reliability is big, reliability hour, which punishes the misclassification of disease, to be reduced, this
When, NHPP class diseases analysis reliability growth model is applied in the optimization method of basic SVM model, trains and conform to
The more granularity enteroscopy report analysis models asked.
The gastroenterology electronic data analysis method of the embodiment of the present invention obtains enteroscopy result and pathologic finding knot
By, and enteroscopy result and pathologic finding conclusion are compared, show that intestinal cancer is analyzed as a result, for intestinal cancer analysis result system
Meter feature, building nonhomogeneous Poisson process NHPP class diseases analysis reliability growth model, Training Support Vector Machines SVM model,
Using model accuracy rate as evaluation index, using diseases analysis reliability growth model each time reliability result as punishment
The factor, associated losses function optimizes SVM model together and carries out intestinal cancer judgement, so as to divide gastroenterology electronic data
Analysis, obtains more accurate intestinal cancer judging result.
Various embodiments of the present invention are described above, above description is exemplary, and non-exclusive, and
It is not limited to disclosed each embodiment.Without departing from the scope and spirit of illustrated each embodiment, for this skill
Many modifications and changes are obvious for the those of ordinary skill in art field.The selection of term used herein, purport
In the principle, practical application or improvement to the technology in market for best explaining each embodiment, or make the art
Other those of ordinary skill can understand each embodiment disclosed herein.
Those skilled in the art will readily occur to its of the disclosure after considering specification and practicing disclosure disclosed herein
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.
Claims (10)
1. a kind of gastroenterology electronic data analysis method characterized by comprising
Enteroscopy result and pathologic finding conclusion are obtained, and the enteroscopy result and the pathologic finding conclusion are carried out
Compare, show that intestinal cancer analyzes result;
Result statistical nature is analyzed for the intestinal cancer, constructs nonhomogeneous Poisson process NHPP class diseases analysis reliability growth mould
Type;
Training Support Vector Machines SVM model increases the NHPP class diseases analysis reliability using model accuracy rate as evaluation index
Long model each time reliability result as penalty factor, associated losses function optimizes SVM model together, carries out intestinal cancer
Judgement.
2. the method according to claim 1, wherein the acquisition enteroscopy result and pathologic finding conclusion,
And be compared the enteroscopy result and the pathologic finding conclusion, show that intestinal cancer analysis result includes:
Extract characteristic information in enteroscopy report from the same sufferer and pathologic finding report, according to individual features information into
Row report splicing integration;During integrating to data, it is subject to pathological examination results;Wherein, the extraction feature
Information includes: the extraction of the extraction of text-type feature, the extraction of temporal characteristics and patient's essential information feature;
The text-type feature is integrated into construction feature space, proposes retouching for corresponding disease category from pathologic finding report
Predicate is compiled as label, construction output space, the attribute value and output space setting numerical value for being characterized each feature in space
Code rule carries out numeralization expression to the data reporting after integration, becomes calculating by the numeric coding rule set
Machine and algorithm model can recognize and the numeric type data of study;
After carrying out numeralization expression to data, using pathologic finding conclusion as final conclusion, enteroscopy conclusion and pathology are examined
It comes to an end by being compared, using the moon as time partition dimension, counts the analysis result of intestinal cancer every month.
3. according to the method described in claim 2, it is characterized in that,
The extraction of the text-type feature includes: that the text-processing done to enteroscopy text data is known as at Medical Language
Reason, it is main during this to complete the participle to enteroscopy text data, noise elimination, and the positive for extracting particular condition is retouched
It states as characteristic information;
The extraction of the temporal characteristics includes: the character string data for being Interval Coding division check data in audit report,
Including the date, the moon is therefrom extracted as temporal characteristics;
The extraction of patient's essential information feature includes: that the feature of extraction includes gender, age, occupation.
4. the method according to claim 1, wherein described analyze result statistical nature, structure for the intestinal cancer
Building nonhomogeneous Poisson process NHPP class diseases analysis reliability growth model includes:
Nonhomogeneous Poisson process NHPP is set;
Construct NHPP class calculated result reliability growth model frame;
Construct the NHPP class disease reliability growth model to tally with the actual situation;
To the parameter Estimation in the NHPP class disease reliability growth model to tally with the actual situation.
5. according to the method described in claim 4, it is characterized in that, the setting nonhomogeneous Poisson process NHPP includes:
One random counting process { N (t), t >=0 } is set meets A1 to A4:N and represent a counting process, can indicate quantity
Number, t indicate the time;
A1:N (0)=0;
A2:{ N (t), t >=0 } it is an independent incremental process;
A3:P [N (t+ Δ t)-N (t)]=1=λ (t) Δ t+o (Δ t);
λ (t) indicates that the intensity function of nonhomogeneous Poisson process, Δ t indicate a time interval, the o (high-order of Δ t) expression Δ t
Infinitely small function;
A4:P {-N (s) >=2 N (t) }=o (Δ t);
Then claiming { N (t), t >=0 } is the nonhomogeneous Poisson process with intensity λ (t), as λ (t)=λ, nonhomogeneous Poisson process
It is exactly common homogeneous Poisson processes;
The probability distribution formula of nonhomogeneous Poisson process is as follows:
S indicates next time, identical as t meaning;
N (t): the cumulative analysis errors number found within [0, the t] period;
M (t): the desired value of cumulative analysis errors number in [0, the t] period, m (t)=E [N (t)];
X (t): until moment t, it is detected and belongs to the analysis errors number that repetition mistake is examined;
A (t): disease mistake examines total function, indicates until moment t, the analysis errors number summation counted in case;
a0: errors number is calculated present in case when statistics starts;
B: calculating error rate, indicates the probability that each mistake is counted in case;
P (t): mistake repetitive rate function indicates that the mistake being each detected belongs to the probability repeated in moment t;
R (x | t): disease reliability function is indicated since moment t in the t+x period, the reliability of diseases analysis.
6. method according to claim 4 or 5, which is characterized in that the building NHPP class calculated result reliability growth
Model framework includes:
Based on the probability distribution formula of the nonhomogeneous Poisson process, set:
B1: the cumulative analysis errors number N (t) to time t obeys the Poisson process that mean function is m (t), arbitrary time span
Number occurs for desired analysis mistake proportional to the remaining analysis error number of t moment in t to t+ Δ t.
B2: disease mistake examine quantity in varying environment, different moments be it is different, diseases analysis mistake sum is to change over time
's.
B3: identical mistake examines situation and is likely to occur in different time sections, and mistake examines the function that repetitive rate is the time.
B4: each analysis mistake is independent from each other in case, and consequence seriousness caused by each analysis mistake is different;
By assuming B1: there is a B5, B5:m (t+ Δ t)=b (a (t)-x (t)) Δ t+o (Δ t),
(a (t)-x (t)) is indicated until moment t, be detected and be not belonging to repeat the wrong analysis errors number examined so as to
Obtain differential equation B6:
By assuming that B3 has B7, B7:
B8, B8 can be obtained by equation B6, equation B7:
Indicate that x (t) carries out derivation to t;
Equation B6, equation B7 primary condition be B9, B10, B9:m (0)=0, B10:x (0)=0;
B11 can be obtained by formula B8, formula B10,
B11:
Exp is the meaning of index, and t, u are the distinct symbols for indicating the time, and dt, du quadrature;
It is B12 to which the cumulative analysis mistake mean function of model can be solved by formula B6, formula B9:
Since the cumulative analysis errors number N (t) until moment t obeys the nonhomogeneous Poisson distribution of mean value m (t), so
B13:
According to nonhomogeneous Poisson be distributed property, Reliability Function B14:
R (x | t)=1-P {-N (t)=0 N (t+x) }=1-exp [- (m (t+x)-m (t))].
7. according to the method described in claim 6, it is characterized in that, described construct the NHPP class diseases analysis to tally with the actual situation
Reliability growth model includes:
The total function of analysis mistake describes B15:a (t)=a with following function0(1+αt)
Wherein, the size of α < 0, α determine the wrong size for examining total function decrease speed;
Mistake, which examines repetitive rate function p (t), should meet following condition: p (t) ∈ [0,1] and p (t) is decreasing function, as t → ∞,
p(t)→0.Therefore, following function can be chosen and examine repetitive rate function B16 to define mistake:
The size of wherein k > 0, k determine the wrong speed for examining repetitive rate variation;
It brings formula B15, formula B16 into formula B11, formula B12, obtains:
B17:
B18:
After obtaining the mean function m (t) of cumulative analysis errors number, so that it may obtain the ginseng in m (t) using Parameter Estimation Method
Number.
8. the method according to the description of claim 7 is characterized in that the described pair of NHPP class diseases analysis to tally with the actual situation can
Parameter Estimation in property model of growth includes:
The parameter in formula B18 is estimated using Maximum Likelihood Estimation Method, likelihood letter can be obtained by examining mean function m (t) by mistake
Number such as B19:
Wherein, (ti, ni) occur in pairs, niIndicate tiCounted in moment case institute it is wrong examine number only and;tiIndicate statistics
Examining number summation to mistake is niAt the time of, L (parameters | (ti, ni)) indicate maximum likelihood function, subsequent is quadrature symbol
Number, exp is to indicate exponent arithmetic;
Take natural logrithm that can obtain B20 formula B19:
Ln indicates that logarithm operation, ∑ indicate summation operation,!Indicate factorial operation;
Each estimates of parameters can be acquired to above-mentioned formula B20 differential.
9. the method according to claim 1, wherein the Training Support Vector Machines SVM model, accurate with model
Rate is evaluation index, using the diseases analysis reliability growth model each time reliability result as penalty factor,
Associated losses function optimizes SVM model together, carries out intestinal cancer judgement and includes:
Using linear separability support vector machines learning algorithm, i.e. maximal margin method carrys out Training Support Vector Machines SVM model to find
Optimal separating hyperplane, algorithm description are as follows:
Input: linear separability training dataset T={ (x1, y1), (x2, y2) ..., (xn, yn), wherein xi∈ χ=Rn, yi∈
γ={ -1 ,+1 } is ith feature vector, also referred to as example, yiFor xiClass label, work as yiWhen=+ 1, claim xiFor positive example;Work as yi
When=- 1, claim xiBe negative example, (xi, yi) it is known as sample point;
Output: largest interval separating hyperplance and categorised decision function;
(1) it constructs and solves constrained optimization problem:
s.t yi(w·xi+ b) -1 >=0, i=1,2 ..., N B22
Acquire optimal solution w*, b*;Min expression is minimized, and w, b indicate to constitute two parameters of largest interval separating hyperplance, Xi
Indicate the example in training dataset;
(2) separating hyperplance is thus obtained are as follows:
w*·x+b*=0 B23
Categorised decision function are as follows:
F (x)=sign (w*·x+b*) B24
Wherein: | | w | | it is the L of w2Norm, (w, b) are given hyperplane;
By using above-mentioned algorithm Training Support Vector Machines model, optimal separating hyperplane is found, feature space is divided into
Two parts, a part are positive classes, and a part is negative class, so that data set be classified.
10. according to the method described in claim 9, it is characterized in that, after obtaining hyperplane further include:
To each sample point (xi, yi) introduce a slack variable ξi>=0, so that function interval is more than or equal to plus slack variable
1, at this point, constraint condition becomes: yi(w·xi+b)≥1-ξi.Meanwhile to each slack variable ξi, pay a cost function
ξi, objective function is also by originalBecome:N value range is the collection of all nonnegative integers
It closes;
Wherein, C > 0 is known as punishment parameter, its value is different under different Question backgrounds, to the punishment of misclassification when C value is big
Increase, C value hour reduces the punishment of misclassification, and the objective function after change includes making interval number big and misclassified gene as far as possible
Two layers of meaning small as far as possible is measured, C belongs to the variable of reconciliation the relationship of the two;
The problems of value of C is combined with disease reliability issues, value, that is, disease Reliability Function of C is in each period
Calculated result, objective function become:Data calculating reliability punishes disease misclassification when big
Increase is penalized, reliability hour, which punishes the misclassification of disease, to be reduced, at this point, NHPP class diseases analysis reliability growth model is answered
In optimization method for basic SVM model, satisfactory more granularity enteroscopy report analysis models are trained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454851.6A CN110223770A (en) | 2019-05-29 | 2019-05-29 | Gastroenterology electronic data analysis method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454851.6A CN110223770A (en) | 2019-05-29 | 2019-05-29 | Gastroenterology electronic data analysis method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110223770A true CN110223770A (en) | 2019-09-10 |
Family
ID=67818706
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910454851.6A Withdrawn CN110223770A (en) | 2019-05-29 | 2019-05-29 | Gastroenterology electronic data analysis method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110223770A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112215835A (en) * | 2020-10-22 | 2021-01-12 | 刘茗露 | Information processing method and device for template report in image-text system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006313523A (en) * | 2005-04-05 | 2006-11-16 | Yoshiro Mizuno | Prescription, in-hospital order, pharmacy and task processing support system in receipt |
CN105303041A (en) * | 2015-10-20 | 2016-02-03 | 沈阳国际旅行卫生保健中心 | Human digestive system health risk early-warning system |
CN107633156A (en) * | 2017-10-13 | 2018-01-26 | 合肥工业大学 | Endoscopy intelligent decision support system for minimally-invasive treatment |
CN108346466A (en) * | 2018-01-22 | 2018-07-31 | 杭州米天基因科技有限公司 | A method of building lung cancer computer aided detection model using support vector machines |
CN108766559A (en) * | 2018-05-22 | 2018-11-06 | 合肥工业大学 | Clinical decision support method and system for intelligent disorder in screening |
-
2019
- 2019-05-29 CN CN201910454851.6A patent/CN110223770A/en not_active Withdrawn
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006313523A (en) * | 2005-04-05 | 2006-11-16 | Yoshiro Mizuno | Prescription, in-hospital order, pharmacy and task processing support system in receipt |
CN105303041A (en) * | 2015-10-20 | 2016-02-03 | 沈阳国际旅行卫生保健中心 | Human digestive system health risk early-warning system |
CN107633156A (en) * | 2017-10-13 | 2018-01-26 | 合肥工业大学 | Endoscopy intelligent decision support system for minimally-invasive treatment |
CN108346466A (en) * | 2018-01-22 | 2018-07-31 | 杭州米天基因科技有限公司 | A method of building lung cancer computer aided detection model using support vector machines |
CN108766559A (en) * | 2018-05-22 | 2018-11-06 | 合肥工业大学 | Clinical decision support method and system for intelligent disorder in screening |
Non-Patent Citations (2)
Title |
---|
YAN ZHANG 等: "Studies on Application of Support Vector Machine in Diagnose of Coronary Heart Disease", 《2012 SIXTH INTERNATIONAL CONFERENCE ON ELECTROMAGNETIC FIELD PROBLEMS AND APPLICATIONS》 * |
王江源 等: "支持向量机在急腹症智能诊断中的应用研究", 《中国卫生统计》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112215835A (en) * | 2020-10-22 | 2021-01-12 | 刘茗露 | Information processing method and device for template report in image-text system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11922348B2 (en) | Generating final abnormality data for medical scans based on utilizing a set of sub-models | |
US11281978B2 (en) | Distributed rule-based probabilistic time-series classifier | |
US20120271612A1 (en) | Predictive modeling | |
US11379685B2 (en) | Machine learning classification system | |
Ollech et al. | A random forest-based approach to identifying the most informative seasonality tests | |
Daumé III | Non-parametric Bayesian areal linguistics | |
CN111477329B (en) | Method for evaluating psychological state based on image-text combination | |
CN111477328B (en) | Non-contact psychological state prediction method | |
CN111243753A (en) | Medical data-oriented multi-factor correlation interactive analysis method | |
US20140244293A1 (en) | Method and system for propagating labels to patient encounter data | |
Rupp et al. | Exbehrt: Extended transformer for electronic health records | |
Zhang et al. | Exploring unsupervised multivariate time series representation learning for chronic disease diagnosis | |
CN110223770A (en) | Gastroenterology electronic data analysis method | |
Özkan et al. | Effect of data preprocessing on ensemble learning for classification in disease diagnosis | |
CN110223771A (en) | Gastroenterology electronic data analysis method based on NHPP | |
Li et al. | Bayesian nested latent class models for cause-of-death assignment using verbal autopsies across multiple domains | |
CN110633368A (en) | Deep learning classification method for early colorectal cancer unstructured data | |
CN110211697A (en) | Gastroenterology electronic data analysis method based on SVM model | |
CN115660871A (en) | Medical clinical process unsupervised modeling method, computer device, and storage medium | |
CN112489751A (en) | Hybrid enhanced physical examination report automatic generation method and device and storage medium | |
US20230409926A1 (en) | Index for risk of non-adherence in geographic region with patient-level projection | |
Ryyppö | Residual neural network in the identification of rare diseases | |
Havtorn | Uncertainty and the Medical Interview: Towards Self-Assessment in Machine Learning Models | |
Sood | Interpreting Black-Box Models Using Hierarchical and Temporal Feature Abstractions | |
Hasan | Machine learning methods for the analysis of clinical conversation |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190910 |