CN109887600A - A kind of analysis method of pair of non-small cell lung cancer prognosis Survival - Google Patents
A kind of analysis method of pair of non-small cell lung cancer prognosis Survival Download PDFInfo
- Publication number
- CN109887600A CN109887600A CN201910303300.XA CN201910303300A CN109887600A CN 109887600 A CN109887600 A CN 109887600A CN 201910303300 A CN201910303300 A CN 201910303300A CN 109887600 A CN109887600 A CN 109887600A
- Authority
- CN
- China
- Prior art keywords
- cell lung
- sample
- feature
- prognosis survival
- characteristic
- 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
- 238000004393 prognosis Methods 0.000 title claims abstract description 75
- 230000004083 survival effect Effects 0.000 title claims abstract description 61
- 238000004458 analytical method Methods 0.000 title claims abstract description 36
- 208000002154 non-small cell lung carcinoma Diseases 0.000 title claims abstract description 30
- 208000029729 tumor suppressor gene on chromosome 11 Diseases 0.000 title claims abstract description 29
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 40
- 210000004072 lung Anatomy 0.000 claims abstract description 29
- 208000000649 small cell carcinoma Diseases 0.000 claims abstract description 22
- 230000011218 segmentation Effects 0.000 claims abstract description 19
- 238000002474 experimental method Methods 0.000 claims abstract description 11
- 238000000034 method Methods 0.000 claims description 34
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 230000002685 pulmonary effect Effects 0.000 claims description 16
- 208000037841 lung tumor Diseases 0.000 claims description 11
- 239000005337 ground glass Substances 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 7
- 230000002708 enhancing effect Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000004069 differentiation Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000012352 Spearman correlation analysis Methods 0.000 claims description 3
- 238000012098 association analyses Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000002093 peripheral effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000005303 weighing Methods 0.000 claims description 3
- 238000011160 research Methods 0.000 abstract description 14
- 238000011161 development Methods 0.000 abstract description 5
- 230000018109 developmental process Effects 0.000 abstract description 5
- 238000010219 correlation analysis Methods 0.000 abstract description 4
- 238000000605 extraction Methods 0.000 abstract description 4
- 230000003902 lesion Effects 0.000 abstract description 4
- 239000000523 sample Substances 0.000 description 41
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 6
- 201000010099 disease Diseases 0.000 description 6
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 6
- 201000005202 lung cancer Diseases 0.000 description 6
- 208000020816 lung neoplasm Diseases 0.000 description 6
- 201000011510 cancer Diseases 0.000 description 4
- 208000024891 symptom Diseases 0.000 description 4
- 239000003814 drug Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 206010041067 Small cell lung cancer Diseases 0.000 description 2
- 238000004195 computer-aided diagnosis Methods 0.000 description 2
- 208000000587 small cell lung carcinoma Diseases 0.000 description 2
- 206010041823 squamous cell carcinoma Diseases 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 239000002671 adjuvant Substances 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 210000000621 bronchi Anatomy 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000007872 degassing Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000556 factor analysis Methods 0.000 description 1
- 210000004907 gland Anatomy 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000003849 large cell carcinoma Diseases 0.000 description 1
- 231100000518 lethal Toxicity 0.000 description 1
- 230000001665 lethal effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004899 motility Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000000391 smoking effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 239000000439 tumor marker Substances 0.000 description 1
Landscapes
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The present invention proposes the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival, contrived experiment of the present invention carries out the research of prognosis survival analysis to Patients with Non-small-cell Lung, non-small cell lung cancer prognostic analysis model is constructed based on CT images group feature, according to traditional image group research framework, the segmentation of tumour has been carried out to Patients with Non-small-cell Lung data, feature extraction, Feature Selection, the correlation analysis and prognosis survival analysis model modeling of image group feature and prognosis Survival, obtain image group Prognostic Factors relevant to Patients with Non-small-cell Lung prognosis existence conspicuousness and prognosis survival analysis model, to provide a series of data information that patient includes time-to-live and later period lesion developments for doctor, simultaneously, the performance of obtained prognosis survival model is evaluated, it ensure that prognosis is raw Deposit the accuracy of model.
Description
Technical field
The present invention relates to computer-aided medical science technical field more particularly to a kind of pair of non-small cell lung cancer prognosis existence feelings
The analysis method of condition.
Background technique
The World Health Organization (WHO) international cancer research institution (IARC) issues newest report recently and claims, and lung cancer is the whole world
The fastest-rising malignant tumour of morbidity and mortality in range, it is contemplated that will cause 1,800,000 people dead for 2018, and account for estimated cancer
The 18.4% of total toll.Classified according to histological type, lung cancer is divided into non-small cell lung cancer and Small Cell Lung Cancer, wherein
Non-small cell lung cancer (non-small cell lung cancer, NSCLC) accounts for the 80%~85% of lung cancer illness total number of persons, packet
Include squamous cell carcinoma (squamous carcinoma), gland cancer, large cell carcinoma.For non-small cell lung cancer compared to Small Cell Lung Cancer, growth division is slower, expands
Scattered transfer is later, and lethal is also relatively weak, but due to there are a large amount of individuals between the lesion of different Patients with Non-small-cell Lung
Between difference so that there is very big difference in the development speed that different patient suffers from the cancer prognosis state of an illness.Epidemiology statistics show largely
Patients with Non-small-cell Lung fail to receive suitable treatment in time due to not obtaining the prediction of accurate progression of the disease so that should
The death rate of class patients with lung cancer is up to 75%.Therefore, there is an urgent need to effective survival of patients time prediction model come to treatment and
The selection of check scheme is assisted, and to improve the therapeutic effect of non-small cell lung cancer, and then is improved the cure rate of patient and is deposited
Motility rate.
Image group is an emerging field in medicine, and the birth of the technology and image genomics are in disease research
Great potential in the diagnosing and treating of disease of superperformance and medical image it is inseparable.Radiation group passes through feature
It extracts, by the information MAP of tumor region to high-dimensional feature space, the prognosis of disease is then constructed by the method for machine learning
Model predicts the future development of disease, to treatment to disease and check the selection of scheme and instruct.CT shadow
It is easy to the features such as comparing as data possess the easy and result of acquisition, as one of the important mode in image group database,
It is widely used in the research of image group.
In recent years, it is directed to computer-aided diagnosis (computer aided diagnosis, CAD) technology and essence both at home and abroad
The research of quasi- medical treatment (Precision Medicine) is more and more burning hoter.Cad technique and accurate medical treatment are both needed to by image group
Means carry out quantitative analysis to tumour by extracting a large amount of image group features, to achieve the purpose that adjuvant clinical diagnoses.And
Correlation analysis is carried out to patient's prognosis Survival using the image group feature of extraction, building prognosis evaluation model prediction is suffered from
The prognosis Survival of person, so that doctor preferably be instructed to select the treatment of patient and check method.
From the point of view of current domestic and international present Research, the research of non-small cell carcinoma prognostic analysis generally goes out from clinical angle
Hair, firstly, (clinical stages, smoking history, whether there is or not brain metastes, tumor marker, doctors according to the intuitive Clinical symptoms of case sample
Learn sign etc.) quantizating index as case sample;Then, by traditional statistical method to Clinical symptoms and Prognostic significance
Single factor test survival analysis is carried out, Clinical symptoms relevant to patient's prognosis is obtained;Finally, by single factor analysis with patient's prognosis
Relevant Clinical symptoms substitutes into COX regression model and carries out multiplicity, obtains the Prognostic Factors of non-small cell lung cancer, helps to cure
It takes root and carries out more accurate prognosis evaluation to Patients with Non-small-cell Lung according to Prognostic Factors, design preferably treatment and check
Scheme, to extend the survival of patients time.And such methods, there is also limitation, the clinical information type that can be utilized is less,
And the feature of medicine sign class only shows the portion forms characteristic of tumor region, and from image group angle, it can obtain
It can effectively be solved to the more abundant image group feature of more and type to reflect the more implicit informations of tumour
Certainly Tumor Heterogeneity is difficult to the problem of being quantitatively evaluated.For current research Shortcomings, the present invention devises new lung cancer for prognosis
Research method carries out analysis to the prognosis survival state of non-small cell lung cancer and probes into, and obtains non-small thin based on CT images feature
Born of the same parents' lung cancer for prognosis analysis model predicts the prognosis life span of patient;Meanwhile contrived experiment tests experimental method
Card, so improve presently, there are deficiencies, obtain better prognostic analysis effect.
Summary of the invention
It is an object of the invention to propose the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival, to realize
The prognosis life span of patient is predicted.
In order to achieve the above objectives, the present invention proposes the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival, packet
Include following steps:
Step 1:CT image processing;
Step 1.1: pulmonary parenchyma coarse segmentation is carried out to the lung CT sequence of Patients with Non-small-cell Lung, and to segmentation after
It is repaired at pulmonary parenchyma edge;
Step 1.2: smart segmentation is carried out to the pulmonary parenchyma after rough cutting;
Step 1.3: the lung tumors in the pulmonary parenchyma after cutting to fine cut detect;
Step 1.4: smart segmentation and artificial correction are carried out to the lung tumors detected;
Step 1.5: in above-mentioned steps, while extracting the corresponding quantitative characteristic for describing tumour;
Step 2: characteristic processing;
Step 2.1: the characteristic of the quantitative characteristic is normalized;
Step 2.2: balancing the positive and negative sample size of the characteristic data set;
Step 2.3: filtering out highest preceding ten quantitative characteristics of weighted value;
Step 3: the association analysis of image group feature and prognosis Survival:
Step 3.1: filtering out and the significantly associated image group feature of prognosis Survival;
Step 3.2: by image group feature described in the quantitative characteristic filtered out in step 2.3 and step 3.1
Intersection is as image group Prognostic Factors;
Step 4: the foundation of prognosis survival model;
Step 4.1: using the cut-off life span of Patients with Non-small-cell Lung as dependent variable, with the image group prognosis because
Element is independent variable, establishes prognosis survival model;
Step 4.2: assessing the performance of the prognosis survival model.
Preferably, in step 1.1, it is real that lung is carried out using the lung CT sequence of " threshold method " to Patients with Non-small-cell Lung
Matter coarse segmentation, while lung edge is repaired using " Chain-Code-Method ";
In step 1.2, smart segmentation is carried out to the pulmonary parenchyma after rough cutting using " region-growing method ", goes degasification
Pipe, bronchial interference;
In step 1.3, the method that is combined with " Gaussian template matching method " with " detection of hessian matrix edge point "
The lung tumors in the pulmonary parenchyma after cutting to fine cut detect, and reject the interference of angiosomes;
In step 1.5, artificial correction is carried out to the lung tumors of over-segmentation and less divided.
Preferably, in step 2.1, the normalized is by the value of the characteristic of the quantitative characteristic
Section zooms to [0,1];
In step 2.2, " Smote algorithm " is utilized to balance the positive and negative sample size of characteristic data set;
In step 2.3, the characteristic is screened using " Relief feature weight algorithm ", filters out weight
It is worth highest preceding ten quantitative characteristics.
Preferably, in step 3.1, Spearman correlation analysis image group feature and prognosis Survival are utilized
Between correlation, with P < 0.05 be index screening go out statistically with prognosis Survival significantly associated image group spy
Sign.
Preferably, it in step 4.1, is established using Lasso regression fit generalized linear model and is based on image group feature
Prognosis survival model;
In step 4.2, contrived experiment is commented using average absolute value error, mean square error, R square value as evaluation index
Estimate the performance of the prognosis survival model.
Preferably, 124 Patients with Non-small-cell Lung are chosen to be detected.
Preferably, in step 1.5, corresponding 256 quantitative characteristics for describing tumour are extracted.
Preferably, the quantitative characteristic includes image group feature and patient clinical information.
Preferably, in step 1.4, different smart splitting schemes is taken for different types of tumour: (1) stand alone is swollen
Tumor: utilizing " method of fuzzy C-means clustering " to be split, obtain the binary mask of tumour, then by being multiplied with original image,
Tumour result after being divided;(2) ground glass type tumour: pass through the gray value of " filtering enhancing method " enhancing ground glass type tumour
Afterwards, the contrast for improving tumour and peripheral region, is then partitioned into ground glass type tumour using the method for cluster.
Preferably, the basic thought of described " Smote algorithm " is analysis to be carried out to minority class sample and according to minority class sample
This artificial synthesized new samples is added in data set, and experiment flow is as follows:
(1) is for each sample x in minority class, and using Euclidean distance as criterion calculation, it arrives minority class sample set SminIn
The distance of all samples obtains its k neighbour.
(2) samples multiplying power N according to one oversampling ratio of sample imbalance ratio setting to determine, for each minority
Class sample x randomly chooses several samples from its k neighbour, it is assumed that the neighbour selected is xn。
(3) the neighbour x that selects each at randomn, new sample is constructed according to following formula with original sample respectively
xnew;
xnew=x+rand (0,1) × (x-xn)
" the Relief feature weight algorithm " is a kind of characteristic weighing algorithm, according to the phase of each quantitative characteristic and classification
Closing property, which is characterized, distributes different weights, and the function that weight is less than some threshold value will be deleted, " Relief feature weight algorithm "
Ability of the correlation of middle quantitative characteristic and classification based on feature differentiation short distance sample, the algorithm select at random from training set D
Sample R is selected, nearest samples H, referred to as NearHit are then searched for from the sample of same type R, and from the sample of different R types
Nearest samples M, referred to as Near Miss are found in this, then according to the weight of each quantitative characteristic of following Policy Updates: if
R in feature and near point, which the distance between are hit, is less than R and the distance between closely, then it represents that the quantitative characteristic is conducive to area
Divide nearest same type and different classes of neighbours, then increases the weight of quantitative characteristic;On the contrary, if R and Near Hit
The distance between be greater than the distance between R and NearMiss, show the quantitative characteristic to distinguishing same type and different classes of
Nearest-neighbors have negative effect, then the weight of the quantitative characteristic reduces;It repeats the above process m times, finally obtains each quantitative
The weight of the average weight of feature, quantitative characteristic is bigger, and the classification capacity of quantitative characteristic is stronger, classifies to quantitative characteristic
Ability is weaker, and specific algorithm pseudocode is as follows:
If training dataset D, sample frequency in sampling m, feature weight divide threshold value be g, output be each feature weight T:
(1) sets 0 all feature weights, and T is empty set;
(2) .for i=1to m
1) a sample R is randomly choosed;
2) it is focused to find out the nearest samples H of R from similar sample, nearest samples M is found from inhomogeneity sample set;
3) for a=1to N
W (A)=W (A)-diff (A, R, H)/m+diff (A, R, M)/m
(3) .for A=1to N
if W(A)≥g
The A feature is added in T
End。
Compared with prior art, of the invention to be advantageous in that: contrived experiment of the present invention is to Patients with Non-small-cell Lung
The research of prognosis survival analysis is carried out, non-small cell lung cancer prognostic analysis model is constructed based on CT images group feature, is pressed
According to traditional image group research framework, to Patients with Non-small-cell Lung carried out the segmentation of tumour, feature extraction, Feature Selection,
The correlation analysis and prognosis survival analysis model modeling of image group feature and prognosis Survival, obtain and non-small cell lung
Cancer patient's prognosis is survived the relevant image group Prognostic Factors of conspicuousness, thus for doctor provide patient include the time-to-live and
A series of data information of later period lesion developments, meanwhile, the performance of obtained prognosis survival model is evaluated, is protected
The accuracy of prognosis survival model is demonstrate,proved.
Detailed description of the invention
Fig. 1 is to illustrate in one embodiment of the invention to the process of the analysis method of non-small cell lung cancer prognosis Survival
Figure.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, technical solution of the present invention will be made below into
Illustrate to one step.
As shown in Figure 1, the present invention proposes the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival, including following
Step:
Step 1:CT image processing;
Step 1.1: pulmonary parenchyma being carried out to the lung CT sequence for choosing 124 Patients with Non-small-cell Lung using " threshold method "
Coarse segmentation, while the pulmonary parenchyma edge after segmentation is repaired using " Chain-Code-Method ";
Step 1.2: smart segmentation being carried out to the pulmonary parenchyma after rough cutting using " region-growing method ", removes gas removing pipe, bronchus
Interference;
Step 1.3: with the method that " Gaussian template matching method " is combined with " detection of hessian matrix edge point " to essence
The lung tumors in pulmonary parenchyma after cutting are detected, and the interference of angiosomes is rejected;
Step 1.4: smart segmentation being carried out to the lung tumors detected, takes different essence point for different types of tumour
It cuts scheme: (1) stand alone tumour: utilizing " method of fuzzy C-means clustering " to be split, obtain the binary mask of tumour, so
Tumour result afterwards by being multiplied with original image, after being divided;(2) ground glass type tumour: pass through " filtering enhancing method " enhancing
After the gray value of ground glass type tumour, the contrast of tumour and peripheral region is improved, is then partitioned into mill using the method for cluster
Glass mould tumour;Artificial correction is carried out to the lung tumors of over-segmentation and less divided.
Step 1.5: in above-mentioned steps, while extracting corresponding 256 quantitative characteristics extracted for describing tumour, such as
Shown in table 1;
Step 2: characteristic processing;
Step 2.1: the characteristic of quantitative characteristic being normalized, normalized is by the spy of quantitative characteristic
The value interval of sign data zooms to [0,1];
Step 2.2: utilizing the positive and negative sample size of " Smote algorithm " balance characteristics data set;
The basic thought of " Smote algorithm " is that analysis and artificial synthesized new according to minority class sample is carried out to minority class sample
Sample is added in data set, and experiment flow is as follows:
(1) is for each sample x in minority class, and using Euclidean distance as criterion calculation, it arrives minority class sample set SminIn
The distance of all samples obtains its k neighbour.
(2) samples multiplying power N according to one oversampling ratio of sample imbalance ratio setting to determine, for each minority
Class sample x randomly chooses several samples from its k neighbour, it is assumed that the neighbour selected is xn。
(3) the neighbour x that selects each at randomn, new sample is constructed according to following formula with original sample respectively
xnew;
xnew=x+rand (0,1) × (x-xn)
Step 2.3: characteristic being screened using " Relief feature weight algorithm ", it is highest to filter out weighted value
Preceding ten quantitative characteristics;
" Relief feature weight algorithm " is a kind of characteristic weighing algorithm, according to the correlation of each quantitative characteristic and classification
It is characterized and distributes different weights, the function that weight is less than some threshold value will be deleted, fixed in " Relief feature weight algorithm "
Ability of the correlation of measure feature and classification based on feature differentiation short distance sample, the algorithm randomly choose sample from training set D
Then this R searches for nearest samples H, referred to as NearHit from the sample of same type R, and from the sample of different R types
Nearest samples M, referred to as NearMiss are found, then according to the weight of each quantitative characteristic of following Policy Updates: if feature
On R and near point the distance between hit and to be less than R and the distance between closely, then it represents that quantitative characteristic is conducive to distinguish nearest
Then same type and different classes of neighbours increase the weight of quantitative characteristic;On the contrary, if the distance between R and NearHit
Greater than the distance between R and NearMiss, show that the quantitative characteristic has differentiation same type and different classes of nearest-neighbors
There is negative effect, then the weight of the quantitative characteristic reduces;It repeats the above process m times, finally obtains being averaged for each quantitative characteristic
The weight of weight, quantitative characteristic is bigger, and the classification capacity of quantitative characteristic is stronger, and the ability classified to quantitative characteristic is weaker,
Specific algorithm pseudocode is as follows:
If training dataset D, sample frequency in sampling m, feature weight divide threshold value be g, output be each feature weight T:
(1) sets 0 all feature weights, and T is empty set;
(2) .for i=1to m
1) a sample R is randomly choosed;
2) it is focused to find out the nearest samples H of R from similar sample, nearest samples M is found from inhomogeneity sample set;
3) for a=1to N
W (A)=W (A)-diff (A, R, H)/m+diff (A, R, M)/m
(3) .for A=1to N
if W(A)≥g
The A feature is added in T
End。
Step 3: the association analysis of image group feature and prognosis Survival:
Step 3.1: using related between Spearman correlation analysis image group feature and prognosis Survival
Property, with P < 0.05 be index screening go out statistically with the significantly associated image group feature of prognosis Survival;Wherein, P value
Refer in a probabilistic model, statistical abstract (such as two groups of sample standard deviation value differences) is identical as actual observation data, or even more big
The probability that this event occurs;It is a possibility that null hypothesis null hypothesis is set up or performance is more serious.If p value and selected conspicuousness
Horizontal (0.05 or 0.01) compared to smaller, then null hypothesis can be denied and unacceptable, however this does not show null hypothesis directly
Correctly.P value is the stochastic variable of a Normal Distribution, is not known in actual use because the various factors such as sample exist
Property;The result of generation may bring dispute, and the p value of the model of general statistical calculations just has statistics meaning less than 0.05
Justice, otherwise model is invalid.
Step 3.2: using the intersection of image group feature in the quantitative characteristic filtered out in step 2.3 and step 3.1 as
Image group Prognostic Factors;
Step 4: the foundation of prognosis survival model;
Step 4.1: using the cut-off life span of Patients with Non-small-cell Lung as dependent variable, being with image group Prognostic Factors
Independent variable establishes the prognosis survival model based on image group feature using Lasso regression fit generalized linear model;
Step 4.2: contrived experiment is with average absolute value error (MeanAbsolute Error, MAE)), mean square error
(Mean Square Error, MSE), R square value (R-Squared) are used as evaluation index, assess the property of prognosis survival model
Energy.
In the present embodiment, contrived experiment of the present invention carries out the research of prognosis survival analysis, base to Patients with Non-small-cell Lung
Non-small cell lung cancer prognostic analysis model is constructed in CT images group feature, according to traditional image group research framework,
Segmentation, feature extraction, Feature Selection, image group feature and the prognosis existence feelings of tumour have been carried out to Patients with Non-small-cell Lung
The correlation analysis and prognosis survival analysis model modeling of condition obtain related to Patients with Non-small-cell Lung prognosis existence conspicuousness
Image group Prognostic Factors, so that providing patient for doctor includes time-to-live and a series of later period lesion developments
Data information, meanwhile, the performance of obtained prognosis survival model is evaluated, ensure that the accurate of prognosis survival model
Property.
Table 1
The above is only a preferred embodiment of the present invention, does not play the role of any restrictions to the present invention.Belonging to any
Those skilled in the art, in the range of not departing from technical solution of the present invention, to the invention discloses technical solution and
Technology contents make the variation such as any type of equivalent replacement or modification, belong to the content without departing from technical solution of the present invention, still
Within belonging to the scope of protection of the present invention.
Claims (10)
1. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival, which comprises the following steps:
Step 1:CT image processing;
Step 1.1: pulmonary parenchyma coarse segmentation being carried out to the lung CT sequence of Patients with Non-small-cell Lung, and real to the lung after segmentation
It is repaired at matter edge;
Step 1.2: smart segmentation is carried out to the pulmonary parenchyma after rough cutting;
Step 1.3: the lung tumors in the pulmonary parenchyma after cutting to fine cut detect;
Step 1.4: smart segmentation and artificial correction are carried out to the lung tumors detected;
Step 1.5: in above-mentioned steps, while extracting the corresponding quantitative characteristic for describing tumour;
Step 2: characteristic processing;
Step 2.1: the characteristic of the quantitative characteristic is normalized;
Step 2.2: balancing the positive and negative sample size of the characteristic data set;
Step 2.3: filtering out highest preceding ten quantitative characteristics of weighted value;
Step 3: the association analysis of image group feature and prognosis Survival:
Step 3.1: filtering out and the significantly associated image group feature of prognosis Survival;
Step 3.2: by the intersection of image group feature described in the quantitative characteristic filtered out in step 2.3 and step 3.1
As image group Prognostic Factors;
Step 4: the foundation of prognosis survival model;
Step 4.1: using the cut-off life span of Patients with Non-small-cell Lung as dependent variable, being with the image group Prognostic Factors
Independent variable establishes prognosis survival model;
Step 4.2: assessing the performance of the prognosis survival model.
2. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 1.1, pulmonary parenchyma coarse segmentation, while benefit are carried out using the lung CT sequence of " threshold method " to Patients with Non-small-cell Lung
Lung edge is repaired with " Chain-Code-Method ";
In step 1.2, smart segmentation is carried out to the pulmonary parenchyma after rough cutting using " region-growing method ", gas removing pipe is gone, props up
The interference of tracheae;
In step 1.3, with the method that " Gaussian template matching method " is combined with " detection of hessian matrix edge point " to essence
The lung tumors in the pulmonary parenchyma after cutting are detected, and the interference of angiosomes is rejected;
In step 1.5, artificial correction is carried out to the lung tumors of over-segmentation and less divided.
3. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 2.1, the normalized be the value interval of the characteristic of the quantitative characteristic is zoomed to [0,
1];
In step 2.2, " Smote algorithm " is utilized to balance the positive and negative sample size of characteristic data set;
In step 2.3, the characteristic is screened using " Relief feature weight algorithm ", filters out weighted value most
Preceding ten high quantitative characteristics.
4. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 3.1, using the correlation between Spearman correlation analysis image group feature and prognosis Survival, with
P < 0.05 be index screening go out statistically with the significantly associated image group feature of prognosis Survival.
5. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 4.1, the prognosis survival model based on image group feature is established using Lasso regression fit generalized linear model;
In step 4.2, contrived experiment assesses institute using average absolute value error, mean square error, R square value as evaluation index
State the performance of prognosis survival model.
6. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
124 Patients with Non-small-cell Lung are chosen to be detected.
7. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 1.5, corresponding 256 quantitative characteristics for describing tumour are extracted.
8. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
The quantitative characteristic includes image group feature and patient clinical information.
9. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 1, which is characterized in that
In step 1.4, different smart splitting scheme: (1) stand alone tumour is taken for different types of tumour: " Fuzzy C is equal for utilization
The method of value cluster " is split, and obtains the binary mask of tumour, swollen after being divided then by being multiplied with original image
Tumor result;(2) ground glass type tumour: after the gray value by " filtering enhancing method " enhancing ground glass type tumour, improve tumour with
Then the contrast of peripheral region is partitioned into ground glass type tumour using the method for cluster.
10. the analysis method of a kind of pair of non-small cell lung cancer prognosis Survival according to claim 3, feature exist
In, the basic thought of " the Smote algorithm " be minority class sample is carried out analysis and according to minority class sample it is artificial synthesized new
Sample is added in data set, and experiment flow is as follows:
(1) is for each sample x in minority class, and using Euclidean distance as criterion calculation, it arrives minority class sample set SminIn own
The distance of sample obtains its k neighbour.
(2) samples multiplying power N according to one oversampling ratio of sample imbalance ratio setting to determine, for each minority class sample
This x randomly chooses several samples from its k neighbour, it is assumed that the neighbour selected is xn。
(3) the neighbour x that selects each at randomn, new sample x is constructed according to following formula with original sample respectivelynew;
xnew=x+rand (0,1) × (x-xn)
" the Relief feature weight algorithm " is a kind of characteristic weighing algorithm, according to the correlation of each quantitative characteristic and classification
It is characterized and distributes different weights, the function that weight is less than some threshold value will be deleted, fixed in " Relief feature weight algorithm "
Ability of the correlation of measure feature and classification based on feature differentiation short distance sample, the algorithm randomly choose sample from training set D
Then this R searches for nearest samples H, referred to as Near Hit from the sample of same type R, and from the sample of different R types
Nearest samples M, referred to as Near Miss are found, then according to the weight of each quantitative characteristic of following Policy Updates: if feature
On R and near point the distance between hit and to be less than R and the distance between closely, then it represents that the quantitative characteristic is conducive to distinguish most
Then close same type and different classes of neighbours increase the weight of quantitative characteristic;On the contrary, if between R and Near Hit
Distance be greater than the distance between R and Near Miss, show the quantitative characteristic to distinguishing same type and different classes of nearest
Neighbours have negative effect, then the weight of the quantitative characteristic reduces;It repeats the above process m times, finally obtains each quantitative characteristic
Average weight, the weight of quantitative characteristic is bigger, and the classification capacity of quantitative characteristic is stronger, the ability classified to quantitative characteristic
Weaker, specific algorithm pseudocode is as follows:
If training dataset D, sample frequency in sampling m, feature weight divide threshold value be g, output be each feature weight T:
(1) sets 0 all feature weights, and T is empty set;
(2) .for i=1 to m
1) a sample R is randomly choosed;
2) it is focused to find out the nearest samples H of R from similar sample, nearest samples M is found from inhomogeneity sample set;
3) for a=1 to N
W (A)=W (A)-diff (A, R, H)/m+diff (A, R, M)/m
(3) .for A=1 to N
if W(A)≥g
The A feature is added in T
end。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910303300.XA CN109887600A (en) | 2019-04-16 | 2019-04-16 | A kind of analysis method of pair of non-small cell lung cancer prognosis Survival |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910303300.XA CN109887600A (en) | 2019-04-16 | 2019-04-16 | A kind of analysis method of pair of non-small cell lung cancer prognosis Survival |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109887600A true CN109887600A (en) | 2019-06-14 |
Family
ID=66937421
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910303300.XA Pending CN109887600A (en) | 2019-04-16 | 2019-04-16 | A kind of analysis method of pair of non-small cell lung cancer prognosis Survival |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109887600A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110415820A (en) * | 2019-08-08 | 2019-11-05 | 中南大学 | Non-small cell lung cancer appraisal procedure, device and electronic equipment based on big data |
CN110634571A (en) * | 2019-09-20 | 2019-12-31 | 四川省人民医院 | Prognosis prediction system after liver transplantation |
CN110942808A (en) * | 2019-12-10 | 2020-03-31 | 山东大学 | Prognosis prediction method and prediction system based on gene big data |
CN112037167A (en) * | 2020-07-21 | 2020-12-04 | 苏州动影信息科技有限公司 | Target area determination system based on image omics and genetic algorithm |
CN112349421A (en) * | 2020-10-26 | 2021-02-09 | 中国医学科学院肿瘤医院 | Treatment prognosis evaluation model for lung cancer patient |
CN112885409A (en) * | 2021-01-18 | 2021-06-01 | 吉林大学 | Colorectal cancer protein marker selection system based on feature selection |
CN113051525A (en) * | 2019-12-26 | 2021-06-29 | ***通信集团北京有限公司 | Key quality index KQI analysis method, device, equipment and medium |
CN113409949A (en) * | 2021-06-26 | 2021-09-17 | 南方医科大学南方医院 | Construction method and application of prediction model of non-small cell lung cancer immune check point inhibition treatment curative effect based on organ transfer spectrum |
CN113571194A (en) * | 2021-07-09 | 2021-10-29 | 清华大学 | Modeling method and device for hepatocellular carcinoma long-term prognosis prediction |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105005714A (en) * | 2015-06-18 | 2015-10-28 | 中国科学院自动化研究所 | Non-small cell lung cancer prognosis method based on tumor phenotypic characteristics |
CN105653858A (en) * | 2015-12-31 | 2016-06-08 | 中国科学院自动化研究所 | Image omics based lesion tissue auxiliary prognosis system and method |
CN107169497A (en) * | 2017-04-14 | 2017-09-15 | 中国科学院苏州生物医学工程技术研究所 | A kind of tumor imaging label extracting method based on gene iconography |
CN108647493A (en) * | 2018-05-09 | 2018-10-12 | 中国科学院昆明动物研究所 | A kind of clear cell carcinoma of kidney personalization prognostic evaluation methods based on multi-gene expression characteristic spectrum |
CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
-
2019
- 2019-04-16 CN CN201910303300.XA patent/CN109887600A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105005714A (en) * | 2015-06-18 | 2015-10-28 | 中国科学院自动化研究所 | Non-small cell lung cancer prognosis method based on tumor phenotypic characteristics |
CN105653858A (en) * | 2015-12-31 | 2016-06-08 | 中国科学院自动化研究所 | Image omics based lesion tissue auxiliary prognosis system and method |
CN107169497A (en) * | 2017-04-14 | 2017-09-15 | 中国科学院苏州生物医学工程技术研究所 | A kind of tumor imaging label extracting method based on gene iconography |
CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
CN108647493A (en) * | 2018-05-09 | 2018-10-12 | 中国科学院昆明动物研究所 | A kind of clear cell carcinoma of kidney personalization prognostic evaluation methods based on multi-gene expression characteristic spectrum |
Non-Patent Citations (3)
Title |
---|
JING GONG ET AL: "Computer-aided detection of pulmonary nodules using dynamic self-adaptive template matching and a FLDA classifier", 《PHYSICA MEDICA》 * |
WOOK-JIN CHOI ET AL: "Automated pulmonary nodule detection based on three-dimensional shape-based feature descriptor", 《COMPUTE METHODS AND PROGRAMS IN BIOMEDICINE》 * |
于博洋艺: "基于肺癌CT影像放射组学预后模型研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110415820A (en) * | 2019-08-08 | 2019-11-05 | 中南大学 | Non-small cell lung cancer appraisal procedure, device and electronic equipment based on big data |
CN110634571A (en) * | 2019-09-20 | 2019-12-31 | 四川省人民医院 | Prognosis prediction system after liver transplantation |
CN110942808A (en) * | 2019-12-10 | 2020-03-31 | 山东大学 | Prognosis prediction method and prediction system based on gene big data |
CN113051525A (en) * | 2019-12-26 | 2021-06-29 | ***通信集团北京有限公司 | Key quality index KQI analysis method, device, equipment and medium |
CN112037167A (en) * | 2020-07-21 | 2020-12-04 | 苏州动影信息科技有限公司 | Target area determination system based on image omics and genetic algorithm |
CN112037167B (en) * | 2020-07-21 | 2023-11-24 | 苏州动影信息科技有限公司 | Target area determining system based on image histology and genetic algorithm |
CN112349421A (en) * | 2020-10-26 | 2021-02-09 | 中国医学科学院肿瘤医院 | Treatment prognosis evaluation model for lung cancer patient |
CN112885409A (en) * | 2021-01-18 | 2021-06-01 | 吉林大学 | Colorectal cancer protein marker selection system based on feature selection |
CN113409949A (en) * | 2021-06-26 | 2021-09-17 | 南方医科大学南方医院 | Construction method and application of prediction model of non-small cell lung cancer immune check point inhibition treatment curative effect based on organ transfer spectrum |
CN113409949B (en) * | 2021-06-26 | 2022-09-02 | 南方医科大学南方医院 | Construction method and application of prediction model of non-small cell lung cancer immune check point inhibition treatment curative effect based on organ transfer spectrum |
CN113571194A (en) * | 2021-07-09 | 2021-10-29 | 清华大学 | Modeling method and device for hepatocellular carcinoma long-term prognosis prediction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109887600A (en) | A kind of analysis method of pair of non-small cell lung cancer prognosis Survival | |
CN109658411A (en) | A kind of correlation analysis based on CT images feature Yu Patients with Non-small-cell Lung prognosis situation | |
Butt et al. | Deep learning system to screen coronavirus disease 2019 pneumonia | |
CN106780460B (en) | A kind of Lung neoplasm automatic checkout system for chest CT images | |
CN104751178B (en) | Lung neoplasm detection means and method based on shape template matching combining classification device | |
CN107066829A (en) | Pathological analysis system and method based on cloud computing | |
CN109706065A (en) | Tumor neogenetic antigen load detection device and storage medium | |
CN108597601A (en) | Diagnosis of chronic obstructive pulmonary disease auxiliary system based on support vector machines and method | |
CN109978880A (en) | Lung tumors CT image is carried out sentencing method for distinguishing using high dimensional feature selection | |
CN110111895A (en) | A kind of method for building up of nasopharyngeal carcinoma far-end transfer prediction model | |
CN110781953B (en) | Lung cancer pathological section classification method based on multi-scale pyramid convolution neural network | |
CN108038352A (en) | Combination difference analysis and the method for association rule mining full-length genome key gene | |
CN113355421A (en) | Lung cancer early screening marker, model construction method, detection device and computer readable medium | |
CN101551854A (en) | A processing system of unbalanced medical image and processing method thereof | |
CN115100474B (en) | Thyroid gland puncture image classification method based on topological feature analysis | |
CN115537467A (en) | Establishment method and application of ovarian cancer survival prognosis prediction molecular model based on deep neural network | |
Jothi et al. | Soft set based feature selection approach for lung cancer images | |
Zhang et al. | Improved heterogeneous data fusion and multi‐scale feature selection method for lung cancer subtype classification | |
CN112927795B (en) | Breast cancer prediction system based on bagging algorithm | |
CN110010204A (en) | Prognosis biomarker recognition methods based on converged network and more marking strategies | |
Krawczyk et al. | Breast thermogram analysis using a cost-sensitive multiple classifier system | |
CN111554381A (en) | Artificial intelligent pathological diagnosis method and diagnosis model for renal clear cell carcinoma based on deep learning | |
Bhaskar et al. | Pulmonary lung nodule detection and classification through image enhancement and deep learning | |
CN116469570A (en) | Malignant tumor complication analysis method based on electronic medical record | |
Wang et al. | Applying feature selection for effective classification of microcalcification clusters in mammograms |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |