CN109166108A

CN109166108A - A kind of automatic identifying method of CT images pulmonary abnormalities tissue

Info

Publication number: CN109166108A
Application number: CN201810922391.0A
Authority: CN
Inventors: 张新财
Original assignee: Shanghai Rongda Information Technology Co Ltd
Current assignee: Shanghai Rongda Information Technology Co Ltd
Priority date: 2018-08-14
Filing date: 2018-08-14
Publication date: 2019-01-08
Anticipated expiration: 2038-08-14
Also published as: CN109166108B

Abstract

The present invention relates in medically computer implemented automatic diagosis technology, a kind of automatic identifying method for the CT images pulmonary abnormalities tissue realized using template difference comparison technology.The parameter vector mainly extracted with the CT lung image of one group of arm's length standard is compared, the method for being quickly found out abnormal structure and automatic marking as template group with real-time dynamic CT lung image.

Description

A kind of automatic identifying method of CT images pulmonary abnormalities tissue

Technical field

The present invention relates in medically computer implemented automatic diagosis technology, more specifically, being related to a kind of using mould Plate difference comparison technology and the automatic identifying method of CT images pulmonary abnormalities tissue realized.

Background technique

The a large amount of extensive use of CT images in medicine, has had resulted in two problems: on the one hand, to each patient one The secondary information content for checking image generated is increasing, and the image information that doctor's daily requirement is read is too many, and working strength is got over Come bigger；On the other hand, diagnostic level is highly dependent on the experience and ability of diagnostician, and mistaken diagnosis happens occasionally.By medicine The development of image format DICOM standard and computer-aided diagnosis technology realizes that the automation of the abnormal structure of CT images is known Not, the accuracy rate of diagosis diagnosis is provided, it is more and more important.

It realizes that the main application target of the automatic identifying method of CT images pulmonary abnormalities tissue shows two aspects: 1. existing On the basis of handling a large amount of patient CT lung image data, suspicious lesion is found out as precisely as possible, emphasis mark, It is supplied to the reference that doctor's behaviours further diagnose, the workload that doctor reads piece is reduced, the accuracy rate of diagnosis is provided.2. big Size medical image processing method and modeling is realized on the basis of largely analyzing certain special diseases, easy in iconography The normal tissue and lesion information obscured are differentiated by the selection of model parameter, and real lesion is screened out to come.

Summary of the invention

Defect in view of the prior art, the present invention relates to adopt in medically computer implemented automatic diagosis technology, one kind The automatic identifying method for the CT images pulmonary abnormalities tissue realized with template difference comparison technology.It is mainly normal with one group The parameter vector that the CT lung image of standard extracts is compared, quickly as template group with real-time dynamic CT lung image The method for finding abnormal structure and automatic marking.

We filter out the lung CT image of 50 Healthy Peoples according to gender, age bracket, fat rank respectively first As standard form, amount to 50, template number, male numbers from M1, M2 ... ... M25, and women is from F1, F2 ... ... F25 Then number generates template data by step.

CT standard video template characteristic the present invention relates to building be by the two-dimensional polar coordinates of multiple groups, node type, curvature, The data such as distance and contraction-expansion factor describe each minutiae feature between directionality, efficient association point, and these compare Template saves in the database in strict accordance with packet numbering.

In actual CT lung images comparison process, we will determine age and the gender of patient first, then carry out with The identical analysis processing step of template image, until finally obtaining its characteristic.Then corresponding grouping feature mould is found again Plate data carry out the processing of comparison one by one respectively.

During template matching, three main processing steps are needed:

(1) we obtain and the difference block S with 5 lung image template datas of group respectively_n(may also not have) and area grade Other β_n(dividing 0 ~ 10 grade by block area)；

(2) in S_nIn, overlapping region s ' all existing for more than 2 template above_(n)Respectively by the validity of block closure Carry out image removing.If image closure is invalid, from s '_(n)It is deleted in list；

(3) each s ' is calculated_(n)Central point, using maximum Edge Distance as radius r, finally in original CT images Circle remembers effective abnormal block out.

Detailed description of the invention

Attached drawing 1: being that CT lung of the present invention template characteristic generates schematic diagram；

Attached drawing 2: being the schematic diagram of CT lung image comparison calculation process of the present invention.

Specific embodiment

The present invention will be further described with reference to the accompanying drawings and examples:

We filter out the lung CT image conduct of 50 Healthy Peoples according to gender, age bracket, fat rank respectively first Standard form amounts to 50, and template number, male numbers from M1, M2 ... ... M25, and women is numbered from F1, F2 ... ... F25, Then template data is generated by step, as shown in Figure 1.

CT lung template characteristic in attached drawing 1 generates, and is classified according to gender, by young (0 years old -6 years old), teenager (7 years old-17 years old), young (18 years old-40 years old), middle aged (41-65 years old) and old (after 66 years old) grouping is established.Each template There is the ID number of oneself, it is gender (1: male, 2: female) that ID, which be 3: first,；Second represents group number (1-5)；Third position It represents with the serial number in group.

CT standard video template characteristic minutiae point is by two-dimensional polar coordinates, node type, curvature, directionality, effectively pass The data such as distance and contraction-expansion factor describe each minutiae feature between connection point, and these comparison templates in strict accordance with point Group # saves in the database.

During template matching, firstly for the CT lung images of any input, it is intended to carry out and template processing step It is identical to pre-process to obtain the characteristic of minutiae point.Then corresponding templates group is selected, is compared one by one.Finally respectively Out with the difference block S of same group of 5 lung image template datas_n(may also not have) and area rank β_nIt (is divided by block area 0 ~ 10 grade), it then finds out comprising S_nThe minimum rectangle of block, and four coordinate points are saved, for β_nValue also to save.Allow There are multiple difference blocks, S_n∈ Diff_Area, data structure are as follows:

struct Diff_Area {

Short Block_Number;// difference block serial number

Struct BLOCK_INDEX [Block_Number];// difference block coordinate

Short Scale[Block_Number];// difference block area rank

};

Difference block coordinate data structure is as follows:

Struct BLOCK_INDEX {

Int Top_X;X-coordinate on // block

Int Top_Y;Y coordinate on // block

Int Down_X;X-coordinate under // block

Int Down_Y;Y coordinate under // block

Int Left_X;The left X-coordinate of // block

Int Left_Y;The left Y coordinate of // block

Int Right_X;The right X-coordinate of // block

Int Right_Y;The right Y coordinate of // block

}；

In this alignment algorithm, the second step of progress is that method is: respectively by five template matchings go out difference block between into Capable operation of seeking common ground.In S_nIn, overlapping region s ' all existing for more than 2 template above_(n)It is accordingly to be regarded as effectively.

For effective block, the operation of image closed is carried out respectively.If superthreshold, it is invalid to be considered as closure, then from s’_(n)It is deleted in list.

Final step is to calculate each s '_(n)Central point, using this maximum Edge Distance as radius r, finally in original Circle remembers that effective abnormal block, alignment algorithm process are as shown in Figure 2 out in the CT images of beginning.

The foundation of CT lung image standard feature template is to need expert doctor to screen from existing CT file store to dispatch troops Complete modeling work.For the CT lung image to be checked, is then automatically read by computer and complete the institutes such as to compare and mark There is working link, without manually being intervened.Example below is then with CT lung image standard feature template and comparison process Objective for implementation.

1) data structure of CT lung image standard feature template is established

Template data be by distance between two-dimensional polar coordinates, node type, curvature, directionality, efficient association point and it is flexible because The data such as son describe each minutiae feature, and these comparison templates are stored in database in strict accordance with packet numbering In.Therefore its key data structure is defined as follows

struct CT_ BellowsForm {

char ID[4];// template number

int GroupID;// packet numbering

char Sex;// gender: M: male, F: female

date Birthday;// the date of birth

int Block_ToT_Number;// block total quantity

int Block_Serial_No;// block serial number

struct Block_Area *PArea；// block pointer

struct Block_ Direct *PDir;// block directional diagram pointer

struct Block_ Curvature *PCur;// block directional curvature

struct Block_ KeyPoint *PKeyPoint;// key node coordinate

struct Block_KeyPoint_ Distance *PDist;// key point vector table

struct Block_ StretchFactor *PFact；// key point contraction-expansion factor

int *PScale;// block amount of area rank (1 ~ 10)

}

struct Block_Area {

Short Block_Number;// block serial number

Struct BLOCK_INDEX *PNo;// block coordinate

Short Scale[Block_Number];// block area rank

} *PArea;

Difference block coordinate data structure is as follows:

Struct BLOCK_INDEX {

Int Top_X;X-coordinate on // block

Int Top_Y;Y coordinate on // block

Int Down_X;X-coordinate under // block

Int Down_Y;Y coordinate under // block

Int Left_X;The left X-coordinate of // block

Int Left_Y;The left Y coordinate of // block

Int Right_X;The right X-coordinate of // block

Int Right_Y;The right Y coordinate of // block

} *PNo;

2) CT lung image pre-processes

L baseline drift processing

L image

L image segmentation

L block divides

l … ...

3) realization of alignment algorithm

We are it needs to be determined that refuse to recognize rate (FAR), misclassification rate (FRR) and similarity (Similarity) three matched default first Amount is three groups:

0.1%、 0.02%、0.75%;

0.05%、0.06%、0.85%;

0.03%、0.08%、0.93%;

It is also required to simultaneously to every important indicator in image, block, minutiae point, sets up reasonable comparison threshold by largely calculating Value, does not illustrate herein.

The ECGWaveVerify (WaveData, WaveTemplate, VerifyParameter) of alignment algorithm;

4) CT images template reconstructs

Template reconstruction of function name are as follows:

ECGWaveTemplateRefactoring(OldWaveTemplate, NewWaveTemplate ,Parameter);

5) CT images are template optimized

Template optimized function name are as follows:

ECGWaveTemplateOptimize (WaveTemplate, Parameter);

Effect of the invention is designed from the angle of CT lung image template, can be improved CT images template to a certain extent Precision and reliability promote the accuracy rate diagnosed in advance automatically.

Claims

1. a kind of automatic identifying method for the CT images pulmonary abnormalities tissue realized using template difference comparison technology, feature It is, the CT lung image standard for establishing one group of arm's length standard form extracts its vector as template characteristic data.

2. the method according to claim 1, wherein template group and the comparison of the CT lung image acquired in real time are Using difference block cutting techniques, the method for finding out abnormal block.

3. the method according to claim 1, wherein the spy that the vector with CT lung images size is adaptive Sign.

4. the method according to claim 1, wherein the two-dimensional pole of the multiple groups obtained after CT images are pre-processed is sat Mark, node type, curvature, directional diagram, distance and contraction-expansion factor between efficient association point, as lung's surrounding tissue template characteristic Parameter vector.