CN104732520A - Cardio-thoracic ratio measuring algorithm and system for chest digital image - Google Patents

Abstract

The invention discloses a cardio-thoracic ratio measuring algorithm and system for a chest digital image. The algorithm concretely includes the following steps that the loaded chest image is corrected; down-sampling and transverse mean filtering are performed on the corrected image to obtain a preprocessed image; each pixel point in the preprocessed image is detected to obtain row data candidate peak value points, wherein binarization processing is performed on the row data candidate peak value points; edge lines on the left and right sides of each lung are grown; columns, to be searched for, of the lung apexes are scanned to obtain column data candidate peak value points, so that the positions of the lung apexes are obtained; a cardiopulmonary interest area is divided to obtain a heart area and a lung area; the heart area and the lung area in the divided image are used for calculating the cardio-thoracic ratio and the cardio-thoracic area ratio. The calculation operation is high, robustness is good, the cardio-thoracic ratio and the cardio-thoracic area ratio of a patient can be accurately measured, auxiliary diagnosis is performed on cardiomegaly and other diseases, and the working efficiency of a doctor is improved.

Description

A kind of ambition of chest digitized video is than Measurement Algorithm and system

Technical field

The invention belongs to digital medical images process field, be specifically related to a kind of ambition of chest digitized video than Measurement Algorithm and system.The present invention is for splitting cardiopulmonary area measure patient ambition ratio and ambition area ratio, and measurement result can be used for carrying out auxiliary diagnosis to diseases such as cardiomegalys.

Background introduction

Imaging of medical is the technology of a kind of advanced person of decipher human body Different Organs health status, and digitized X-ray film has become a kind of method of popular diagnosis different syndromes.Because cost is lower, the first diagnosis basis of health examination and different syndromes can be widely used in.

In clinical, Chest Image is usually used in the diseases such as the such as cardiomegaly of evaluate cardiac and lung areas.Evaluate cardiac hypertrophy has a variety of diagnosis index, wherein ambition ratio (cardiothoracic ratio, CTR) be radiologist's usual indagation standard diagnostics index, it is defined as the ratio of the horizontal maximum gauge being the horizontal maximum gauge of heart and the left and right lobe of the lung; Another is diagnosis index more intuitively---and ambition area ratio (cardiothoracic area ratio, CTAR) is defined as the ratio of the area of heart area and lung's area-of-interest.

Ambition is the earliest rely on doctor naked eyes to the observation of film and judgement than index evaluation, and basis for estimation is the experience of doctor, is easily subject to the impact of other various subjective factors, inconvenient operation and work efficiency is lower.Since digitizing film comes into operation, domestic and international researchist develops different computer based ambition and compares measuring technique.Current ambition is roughly divided into three classes than measuring technique: (1) first kind method is by the mark heart scope manual on digitizing film of doctor and lung's scope, and computing machine carries out computing and obtains a result.The diagnostic method based on film of this method in essence with early stage is similar, and work efficiency is not improved; (2) Equations of The Second Kind method utilizes the pixel grey scale information of image to carry out the segmentation in cardiopulmonary region, and fixing is divided into two classes by pixel in image.Because cardiopulmonary region can be subject to the interference of its hetero-organization, cause the segmentation precision based on gray scale lower.(3) the 3rd class methods are based on the multi-mode matching method of cardiopulmonary.Rabat image has the feature of smooth excessiveness at each organ intersection of health, based on multimode matching dividing method this be fade-in the feature of image gradually gone out under be difficult to reach accurate matching effect, and make such method cannot reach good balance in precision and time complexity in the variation of patient and image-forming condition.To sum up, these methods all cannot meet the requirement of high precision, real-time and robotization simultaneously.

Summary of the invention

For the defect existed in above-mentioned technology or deficiency, Automatic measuring algorithm is compared in the ambition that the invention provides a kind of chest digitized video.

To achieve these goals, the present invention adopts following technical scheme to be solved:

The ambition of chest digitized video, than a Measurement Algorithm, specifically comprises the steps:

Step 1: the Chest Image loaded is calibrated, obtains calibrating rear image;

Step 2: carry out down-sampling and horizontal mean filter process to calibration chart picture, obtains image after pre-service;

Step 3: image X after the pre-service that step 2 is obtained _fin each pixel carry out detection and obtain row data candidate peak point, according to row data candidate peak point, binary conversion treatment is carried out to image after pre-service, obtains the bianry image of lung's edge point-of-interest;

Step 4: the bianry image X utilizing lung's edge point-of-interest _iin all lungs edge point-of-interest, grow the edge lines of the lung left and right sides;

Step 5: image chooses apex pulmonis row to be searched after the pre-service that step 2 obtains, scans apex pulmonis row to be searched, obtains column data candidate peak point, obtains apex pulmonis position by column data candidate peak point;

Step 6: cardiopulmonary area-of-interest is determined in the apex pulmonis position that the lung left and right sides edge lines obtained by step 4 and step 5 obtain, splits the rough segmentation obtaining heart area and lung areas to cardiopulmonary area-of-interest; Rim detection is carried out to cardiopulmonary region of interest area image and obtains edge image, utilize edge image to be split image accurately;

Step 7: the heart in the accurate segmentation image utilizing step 6 to obtain and lung areas calculate ambition ratio and ambition area ratio.

Further, the concrete steps of described step 1 are as follows:

1.1) laterally calibration: loading size is the Chest Image of M × N, to described chest image from top to bottom the individual element scanning is carried out respectively to centre in the left and right two ends of row, and obtain the pixel that left and right two gray scales are greater than 50 respectively, the horizontal ordinate of two described pixels is respectively B _l, B _r; If the pixel count on these two pixel distance Chest Image borders is all more than 16, then by B in Chest Image _lall row on-16 row left sides and B _rthe pixel of all row on+16 row the right is deleted, and has laterally calibrated;

1.2) longitudinal alignment: respectively along the B of the Chest Image after laterally calibrating _l+ 16 and B _r-16 row carry out picture element scan, run into the ordinate S that in two row, first gray scale is greater than 50 respectively _land S _rstop, both calculating average if S row distance upper end boundary pixel number is more than 16, then delete S-16 capable more than row pixel, longitudinal alignment completes, and obtains calibrating rear image.

Further, the concrete steps of described step 2 are as follows:

2.1) after the calibration obtained step 1, image adopts bilinear interpolation to carry out filtering process, obtains the down-sampled images X that size is 256 × 256 _d;

2.2) to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process, obtaining size is image X after the pre-service of 256 × 256 _f; Wherein, to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process and refer to and utilize following formula to process:

$y_{i,j}=\frac{x_{i,j-1}+x_{i,j}+x_{i,j+1}}{3}$

In formula, x _i,jfor current location is at down-sampled images X _din gray scale, y _i,jfor the filtering result of current location.

Further, the concrete steps of described step 3 are as follows:

3.1) image X after pre-service step 2 obtained _fin the individual element scanning from left to right of often row, for each pixel in every row with its be expert in the pixel value of last pixel do first order difference computing and obtain difference result corresponding to current pixel; If the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is row data candidate peak point p _j, wherein, p is gray scale, and subscript j represents the position in current pixel row residing for it;

3.2) for each row data candidate peak point p _j, get left and right apart from it be 5 two pixel p _j-5and p _j+5, according to following formulae discovery image line crest impact degree:

${\mathrm{\θ}}_1\text{=arctan}\left(\frac{p_j-p_{j-5}}{t}\right),$

${\mathrm{\θ}}_2\text{=arctan}\left(\frac{p_j-p_{j+5}}{t}\right),$

θ＝π-θ ₁-θ ₂，

0 < tan (θ)≤1.1, then p if satisfy condition _jfor lung's edge point-of-interest;

By 3.1 and step 3.2) traversal pre-service after image X _fall pixels after, obtain multiple lungs edge point-of-interest; Thus obtain the bianry image that size is lung's edge point-of-interest of 256 × 256, be designated as X _i.

Further, the concrete steps of described step 4 are as follows:

4.1) the bianry image X of the lung's edge point-of-interest obtained with step 3 _ibe often classified as unit, individual element scans from bottom to top, if a certain Lie Zhong continuous print lung edge point-of-interest number is no less than 10, then records the vertical line segment be made up of these continuous print point-of-interests; To the bianry image X of lung's edge point-of-interest _iin after all row all scan, obtain the vertical line segment aggregate L={l comprising the vertical line segment of k bar ₁, l ₂l _k;

4.2) by the bianry image X of lung's edge point-of-interest _ibe divided into left and right two parts, the position l that the connected region area formed for the lung's edge point-of-interest on the vertical line segment in these two parts is maximum _land l _r, by two vertical line segment l _land l _rthe average of pixel ordinate is bottom designated as I _b;

4.3) image X after the pre-treatment _fin, by vertical line segment l _lwith vertical line segment l _rall as the lower edge lines of the initial two panels lobe of the lung; Two described edge lines are handled as follows respectively: current edge lines pixel is topmost designated as y _i,j, compare y _{i-1, j-1}, y _{i-1, j}and y _{i-1, j+1}the gray-scale value of position, includes the pixel of gray scale maximum value position in current edge lines, more repeatedly carries out above-mentioned process, until current edge lines grow into image X after pre-service _fupper end till, obtain the edge lines of the lung left and right sides respectively; Horizontal ordinate minimum value I on the lobe of the lung edge lines on the record left side _land the horizontal ordinate maximal value I on the lobe of the lung edge lines on the right _r.

Further, the concrete steps of described step 5 are as follows:

5.1) image X after the pre-service obtained from step 2 _fin choose apex pulmonis row to be searched; Get I _l+ 16 (I _r-I _l) row, row, row and row are as apex pulmonis row to be searched;

5.2) each row in above-mentioned 6 row row to be searched are handled as follows: individual element scanning from top to bottom, each pixel is done first order difference computing with the pixel value of last pixel in its column and obtains difference result corresponding to current pixel, if the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is column data candidate peak point; After 6 row row to be searched are all processed, obtain multiple column data candidate peak point, using the ordinate mean value of nearest for ordinate in them two column data candidate peak points as apex pulmonis position ordinate the coboundary of cardiopulmonary area-of-interest is defined as

Further, the concrete steps of described step 6 are as follows:

6.1) 4 the parameter I obtained by step 4 and step 5 _l, I _r, I _t, I _bwith edge, the lung left and right sides lines l that step 4 obtains _l, l _r, obtain cardiopulmonary part region of interest area image ROI; Ostu threshold segmentation algorithm is performed to ROI and obtains bianry image, 11 × 11 medium filterings are performed to this bianry image and obtains bianry image ROI after filtering _binary; To bianry image ROI after filtering _binaryfill l _l, l _rlung space with after segmentation, generates rough lung areas and rough left and right ventricles margo border of the lung;

6.2) bianry image ROI after scan-filtering from top to bottom _binaryevery a line, for every a line, searched for the horizontal ordinate at this row middle left and right cardiopulmonary edge to its left and right sides by this row central point with and check the horizontal ordinate at cardiopulmonary edge of 5 row of being separated by respectively with them with if both horizontal ordinate spans | i _j+5-i _j| >8 is (such as the horizontal ordinate at left cardiopulmonary edge its with the absolute value of difference of horizontal ordinate be greater than 8), then coordinate (i, j) is designated as heart lower boundary summit; Connect the heart lower boundary summit of arranged on left and right sides, with bianry image ROI after rough left and right ventricles margo border of the lung and filtering _binaryupper end be expert at and surround region and be rough heart area;

6.3) it is 3 × 7 that the cardiopulmonary part region of interest area image ROI obtained step 6.1 carries out yardstick, and standard deviation is do Canny rim detection after the Gaussian smoothing of 5, obtains edge image ROI _canny; Bianry image ROI after the filtering that step 6.1 obtains _binaryin rough left and right ventricles margo border of the lung pixel position correspond to edge image ROI _cannyin location of pixels Horizon Search edge, if search edge within the specific limits, then rough heart area is filled to edge, thus is split image ROI accurately _seg.

Another object of the present invention is, provides a kind of ambition of chest digitized video than automatic measurement system, specifically comprises as lower module:

Calibration module: the Chest Image loaded is calibrated, obtains calibrating rear image;

Pretreatment module: carry out down-sampling and horizontal mean filter process to calibration chart picture, obtains image after pre-service;

Binarization block: image X after the pre-service that pretreatment module is obtained _fin each pixel carry out detection and obtain row data candidate peak point, according to row data candidate peak point, binary conversion treatment is carried out to image after pre-service, obtains the bianry image of lung's edge point-of-interest;

Lung's marginal growth module: the bianry image X utilizing lung's edge point-of-interest _iin all lungs edge point-of-interest, grow the edge lines of the lung left and right sides;

Apex pulmonis position detecting module: image chooses apex pulmonis row to be searched after the pre-service that pretreatment module obtains, scans apex pulmonis row to be searched, obtains column data candidate peak point, obtains apex pulmonis position by column data candidate peak point;

Cardiopulmonary region segmentation module: cardiopulmonary area-of-interest is determined in the apex pulmonis position that the lung left and right sides edge lines obtained by lung's marginal growth module and apex pulmonis position detecting module obtain, splits the rough segmentation obtaining heart area and lung areas to cardiopulmonary area-of-interest; Rim detection is carried out to cardiopulmonary region of interest area image and obtains edge image, utilize edge image to be split image accurately;

Ambition is than computing module: the heart in the accurate segmentation image utilizing cardiopulmonary region segmentation module to obtain and lung areas calculate ambition ratio and ambition area ratio.

Further, described lung marginal growth module has been used for following flow process:

4.1) the bianry image X of the lung's edge point-of-interest obtained with binarization block _ibe often classified as unit, individual element scans from bottom to top, if a certain Lie Zhong continuous print lung edge point-of-interest number is no less than 10, then records the vertical line segment be made up of these continuous print point-of-interests; To the bianry image X of lung's edge point-of-interest _iin after all row all scan, obtain the vertical line segment aggregate L={l comprising the vertical line segment of k bar ₁, l ₂l _k;

4.2) by the bianry image X of lung's edge point-of-interest _ibe divided into left and right two parts, the position l that the connected region area formed for the lung's edge point-of-interest on the vertical line segment in these two parts is maximum _land l _r, by two vertical line segment l _land l _rthe average of pixel ordinate is bottom designated as I _b;

4.3) image X after the pre-treatment _fin, by vertical line segment l _lwith vertical line segment l _rall as the lower edge lines of the initial two panels lobe of the lung; Two described edge lines are handled as follows respectively: current edge lines pixel is topmost designated as y _i,j, compare y _{i-1, j-1}, y _{i-1, j}and y _{i-1, j+1}the gray-scale value of position, includes the pixel of gray scale maximum value position in current edge lines, more repeatedly carries out above-mentioned process, until current edge lines grow into image X after pre-service _fupper end till, obtain the edge lines of the lung left and right sides respectively; Horizontal ordinate minimum value I on the lobe of the lung edge lines on the record left side _land the horizontal ordinate maximal value I on the lobe of the lung edge lines on the right _r

Described apex pulmonis position detecting module has been used for following flow process:

5.1) image X after the pre-service obtained from pretreatment module _fin choose apex pulmonis row to be searched; Get row, row, row and row are as apex pulmonis row to be searched;

5.2) each row in above-mentioned 6 row row to be searched are handled as follows: individual element scanning from top to bottom, each pixel is done first order difference computing with the pixel value of last pixel in its column and obtains difference result corresponding to current pixel, if the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is column data candidate peak point; After 6 row row to be searched are all processed, obtain multiple column data candidate peak point, using the ordinate mean value of nearest for ordinate in them two column data candidate peak points as apex pulmonis position ordinate the coboundary of cardiopulmonary area-of-interest is defined as

Further, described cardiopulmonary region segmentation module has been used for following flow process:

6.1) by 4 parameter I _l, I _r, I _t, I _bwith edge, the lung left and right sides lines l that step 4 obtains _l, l _r, obtain cardiopulmonary part region of interest area image ROI; Ostu threshold segmentation algorithm is performed to ROI and obtains bianry image, 11 × 11 medium filterings are performed to this bianry image and obtains bianry image ROI after filtering _binary; To bianry image ROI after filtering _binaryfill l _l, l _rlung space with after segmentation, generates rough lung areas and rough left and right ventricles margo border of the lung;

6.2) bianry image ROI after scan-filtering from top to bottom _binaryevery a line, for every a line, searched for the horizontal ordinate at this row middle left and right cardiopulmonary edge to its left and right sides by this row central point with and check the horizontal ordinate at cardiopulmonary edge of 5 row of being separated by respectively with them with if both horizontal ordinate spans | i _j+5-i _j| >8 is (such as the horizontal ordinate at left cardiopulmonary edge its with the absolute value of difference of horizontal ordinate be greater than 8), then coordinate (i, j) is designated as heart lower boundary summit; Connect the heart lower boundary summit of arranged on left and right sides, with bianry image ROI after rough left and right ventricles margo border of the lung and filtering _binaryupper end be expert at and surround region and be rough heart area;

6.3) it is 3 × 7 that the cardiopulmonary part region of interest area image ROI obtained step 6.1 carries out yardstick, and standard deviation is do Canny rim detection after the Gaussian smoothing of 5, obtains edge image ROI _canny; Bianry image ROI after the filtering _binaryin rough left and right ventricles margo border of the lung pixel position correspond to edge image ROI _cannyin location of pixels Horizon Search edge, if search edge within the specific limits, then rough heart area is filled to edge, thus is split image ROI accurately _seg.

Advantage of the present invention and innovative point as follows:

1, automatically and fast and effectively can split Chest Image cardiac and lung areas, measure ambition than and ambition area ratio index.

2, first carried out simple image calibration to the Chest Image loaded, effectively fixing health position in the picture, gets rid of the interference of image space, reduces the hunting zone of subsequent searches step.

3, have employed when splitting cardiopulmonary region by the thick segmentation step to essence, coarse segmentation is carried out under the filtering condition of large scale, and near the result of coarse segmentation, meticulous searching edge lines, complete segmentation; Can effectively reduce by the thick segmentation to essence the impact that noise and its hetero-organization cause segmentation result, obtain segmentation result comparatively accurately.

To sum up, algorithm of the present invention and system effectiveness high, accuracy rate is high, and has good robustness, can be good at taking into account precision and real-time.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of ambition than Measurement Algorithm of chest digitized video of the present invention.

Fig. 2 is the original chest digitized video loaded in embodiment.

Fig. 3 is the image after pre-service in embodiment.

Fig. 4 is the lung's edge point-of-interest image in embodiment.

Fig. 5 is lung's edge schematic diagram of the growth in embodiment.

Fig. 6 is the cardiopulmonary area-of-interest be partitioned in embodiment, i.e. cardiopulmonary ROI region.

Fig. 7 is the schematic diagram in the cardiopulmonary area-of-interest in embodiment after cardiopulmonary segmentation, represents the cardiopulmonary region after segmentation.In figure, black region is interested lung areas, and dark gray areas is heart area, and light grey and white portion is the region beyond cardiopulmonary.

Cardiopulmonary regional processing result schematic diagram in Fig. 8 embodiment, the white portion be positioned in the middle part of image is heart area, and two pieces of regions of heart area both sides are interested lung areas.CTR:0.458，CTAR:0.400。

Fig. 9 is a width Chest Image result schematic diagram.

Figure 10 is a width Chest Image result schematic diagram.

Figure 11 is a width Chest Image result schematic diagram.

Figure 12 is a width Chest Image result schematic diagram.

Figure 13 is a width Chest Image result schematic diagram.

Figure 14 is the schematic diagram of ambition than measuring system of chest digitized video of the present invention.

See the drawings and specific embodiments, further explanation is explained to the present invention below.

Embodiment

With reference to Fig. 1, the concrete steps of algorithm of the present invention are as follows:

Step 1: the Chest Image loaded is calibrated, obtains calibrating rear image.The object of this step can make patient body roughly be in the middle part of Chest Image; Concrete steps are as follows:

1.1) laterally calibration: loading size is the Chest Image of M × N (M > 256, N > 256), to described chest image from top to bottom the individual element scanning is carried out respectively to centre in the left and right two ends of row, and obtain the pixel that left and right two gray scales are greater than 50 respectively, the horizontal ordinate of two described pixels is respectively B _l, B _r; If the pixel count on these two pixel distance Chest Image borders is all more than 16, then by B in Chest Image _lall row on-16 row left sides and B _rthe pixel of all row on+16 row the right is deleted, and has laterally calibrated;

1.2) longitudinal alignment: respectively along the B of the Chest Image after laterally calibrating _l+ 16 and B _r-16 row carry out picture element scan, run into the ordinate S that in two row, first gray scale is greater than 50 respectively _land S _rstop, both calculating average if S row distance upper end boundary pixel number is more than 16, then delete S-16 capable more than row pixel, longitudinal alignment completes, and obtains calibrating rear image;

Step 2: carry out down-sampling and horizontal mean filter process to calibration chart picture, obtains image after pre-service.This step makes graphical rule reduce by down-sampling, and treatment effeciency is improved, and effectively reduces picture noise by filtering.Concrete steps are as follows:

2.1) after the calibration obtained step 1, image adopts bilinear interpolation to carry out filtering process, obtains the down-sampled images X that size is 256 × 256 _d;

2.2) to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process, obtaining size is image X after the pre-service of 256 × 256 _f.Wherein, to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process and refer to and utilize following formula to process:

$y_{i,j}=\frac{x_{i,j-1}+x_{i,j}+x_{i,j+1}}{3}$

In formula, x _i,jfor current location is at down-sampled images X _din gray scale, y _i,jfor the filtering result of current location;

Step 3: image X after the pre-service that step 2 is obtained _fin each pixel carry out detection and obtain row data candidate peak point, according to row data candidate peak point, binary conversion treatment is carried out to image after pre-service, obtains the bianry image of lung's edge point-of-interest; Concrete steps are as follows:

3.1) image X after pre-service step 2 obtained _fin the individual element scanning from left to right of often row, for each pixel in every row with its be expert in the pixel value of last pixel do first order difference computing and obtain difference result corresponding to current pixel; If the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is row data candidate peak point p _j, wherein, p is gray scale, and subscript j represents the position in current pixel row residing for it;

3.2) for each row data candidate peak point p _j, get left and right apart from it be 5 two pixel p _j-5and p _j+5, according to following formulae discovery image line crest impact degree:

${\mathrm{\&theta;}}_1\text{=arctan}\left(\frac{p_j-p_{j-5}}{t}\right),$

${\mathrm{\&theta;}}_2\text{=arctan}\left(\frac{p_j-p_{j+5}}{t}\right),$

θ＝π-θ ₁-θ ₂，

0 < tan (θ)≤1.1, then p if satisfy condition _jfor lung's edge point-of-interest;

By 3.1 and step 3.2) traversal pre-service after image X _fall pixels after, obtain multiple lungs edge point-of-interest; Thus obtain the bianry image that size is lung's edge point-of-interest of 256 × 256, be designated as X _i;

Step 4: the bianry image X utilizing lung's edge point-of-interest _iin all lungs edge point-of-interest, grow the edge lines of the lung left and right sides; Concrete steps are as follows:

4.1) the bianry image X of the lung's edge point-of-interest obtained with step 3 _ibe often classified as unit, individual element scans from bottom to top, if a certain Lie Zhong continuous print lung edge point-of-interest number is no less than 10, then records the vertical line segment be made up of these continuous print point-of-interests; To the bianry image X of lung's edge point-of-interest _iin after all row all scan, obtain the vertical line segment aggregate L={l comprising the vertical line segment of k bar ₁, l ₂l _k;

4.2) by the bianry image X of lung's edge point-of-interest _ibe divided into left and right two parts, the position l that the connected region area formed for the lung's edge point-of-interest on the vertical line segment in these two parts is maximum _land l _r, by two vertical line segment l _land l _rthe average of pixel ordinate is bottom designated as I _b;

4.3) image X after the pre-treatment _fin, by vertical line segment l _lwith vertical line segment l _rall as the lower edge lines of the initial two panels lobe of the lung; Two described edge lines are handled as follows respectively: current edge lines pixel is topmost designated as y _i,j, compare y _{i-1, j-1}, y _{i-1, j}and y _{i-1, j+1}the gray-scale value of position, includes the pixel of gray scale maximum value position in current edge lines, more repeatedly carries out above-mentioned process, until current edge lines grow into image X after pre-service _fupper end till, obtain the edge lines of the lung left and right sides respectively; Horizontal ordinate minimum value I on the lobe of the lung edge lines on the record left side _land the horizontal ordinate maximal value I on the lobe of the lung edge lines on the right _r;

Step 5: image chooses apex pulmonis row to be searched after the pre-service that step 2 obtains, scans apex pulmonis row to be searched, obtains column data candidate peak point, obtains apex pulmonis position by column data candidate peak point; Concrete steps are as follows:

5.1) image X after the pre-service obtained from step 2 _fin choose apex pulmonis row to be searched; Get row, row, row and row (namely respectively the inside certain distance of edge lines of the distance lung left and right sides two row and lay respectively at the left and right row of these two row, totally six row; In the present invention, this distance to be taken as lung's width ) as apex pulmonis row to be searched;

5.2) each row in above-mentioned 6 row row to be searched are handled as follows: individual element scanning from top to bottom, each pixel is done first order difference computing with the pixel value of last pixel in its column and obtains difference result corresponding to current pixel, if the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is column data candidate peak point; After 6 row row to be searched are all processed, obtain multiple column data candidate peak point, using the ordinate mean value of nearest for ordinate in them two column data candidate peak points as apex pulmonis position ordinate the coboundary of cardiopulmonary area-of-interest is defined as

Step 6: cardiopulmonary area-of-interest is determined in the apex pulmonis position that the lung left and right sides edge lines obtained by step 4 and step 5 obtain, splits the rough segmentation obtaining heart area and lung areas to cardiopulmonary area-of-interest; Rim detection is carried out to cardiopulmonary region of interest area image and obtains edge image, utilize edge image to be split image accurately; Concrete steps are as follows:

6.1) 4 the parameter I obtained by step 4 and step 5 _l, I _r, I _t, I _bwith edge, the lung left and right sides lines l that step 4 obtains _l, l _r, obtain cardiopulmonary part region of interest area image ROI as shown in Figure 6; Ostu threshold segmentation algorithm is performed to ROI and obtains bianry image (being divided into two classes by gray scale by cardiopulmonary region), 11 × 11 medium filterings are performed to this bianry image and obtains bianry image ROI after filtering _binary; To bianry image ROI after filtering _binaryfill l _l, l _rlung space with after segmentation, generates rough lung areas and rough left and right ventricles margo border of the lung (cardiopulmonary edge refers to the separatrix of heart and lung);

6.2) bianry image ROI after scan-filtering from top to bottom _binaryevery a line, for every a line, searched for the horizontal ordinate at this row middle left and right cardiopulmonary edge to its left and right sides by this row central point with and check the horizontal ordinate at cardiopulmonary edge of 5 row of being separated by respectively with them with if both horizontal ordinate spans | i _j+5-i _j| >8 is (such as the horizontal ordinate at left cardiopulmonary edge its with the absolute value of difference of horizontal ordinate be greater than 8), then coordinate (i, j) is designated as heart lower boundary summit; Connect the heart lower boundary summit of arranged on left and right sides, with bianry image ROI after rough left and right ventricles margo border of the lung and filtering _binaryupper end be expert at and surround region and be rough heart area;

6.3) it is 3 × 7 that the cardiopulmonary part region of interest area image ROI obtained step 6.1 carries out yardstick, and standard deviation is do Canny rim detection after the Gaussian smoothing of 5, obtains edge image ROI _canny(high-low threshold value wherein, in Canny edge detection algorithm is respectively 30 and 100); Bianry image ROI after the filtering that step 6.1 obtains _binaryin rough left and right ventricles margo border of the lung pixel position correspond to edge image ROI _cannyin location of pixels Horizon Search edge, if within the specific limits (in experiment for left and right each 8 pixels scope in) search edge, then rough heart area is filled to edge, thus is split image ROI accurately _seg(as shown in Figure 7).The purpose of design of this step: due to bianry image ROI after filtering _binarymiddle edge lines are single but inaccurate (may differ from several pixel <8), edge image ROI _cannymiddle existence is edge lines accurately, but also have other unwanted lines pixel interference, so in this step, at ROI _binarylaterally find lines pixel near shown lines, acquired results is and compares ROI _binarymore accurate, eliminate ROI again simultaneously _cannyin the interference of other lines pixels.

Step 7: utilize heart and lung areas to calculate ambition ratio and ambition area ratio:

Ambition is than being obtained by following formulae discovery:

$\mathrm{CTR}=\frac{{\mathrm{IN}}_r-{\mathrm{IN}}_l}{I_r-I_l};$

In formula, I _lrefer to accurate segmentation image ROI _segthe horizontal ordinate minimum value at middle left lung edge, I _lrefer to accurate segmentation image ROI _segthe horizontal ordinate maximal value at middle right side lung edge; IN _lrefer to the horizontal ordinate minimum value of the left side edge of heart area, IN _rrefer to the horizontal ordinate maximal value of the right side edge of heart area;

The accurate segmentation image ROI that obtains of traversal step 6 line by line _segmiddle pixel, adds up the number of pixels n of lung areas respectively _lwith the number of pixels n of heart area _c; Then ambition area ratio is obtained by following formulae discovery:

$\mathrm{CTAR}=\frac{n_L}{n_C}.$

Effect of the present invention is further illustrated by following emulation experiment:

This experiment PC used installs the operating system of 64 Window 7, and its CPU is Intel (R) Pentium (R) Dual CPU [email protected], internal memory 4.00GB.Development environment is Visual studio 2010.

Experiment use 50 width Chest Image digital film carries out experiment test.Following statistics is carried out to validity of the present invention and working time:

Experimental image number	Effective number of results	Null result number	Basis discrimination	Processing time (ms)
					50	48	2	96％	11544

Experiment display, the technology in the present invention achieves higher basic discrimination in test pattern, and average every width image only needs 230ms.

Embodiment

The present embodiment carries out cardiopulmonary segmentation to the original Chest Image digital film as shown in Figure 2 loaded and carries out ambition than measuring.First the Chest Image loaded is calibrated, down-sampling, the pretreatment operation such as filtering, image (as shown in Figure 3) after the pre-service obtained; Then row peak point is calculated to image after pre-service, obtain the binary image (as shown in Figure 4) of lung's edge point-of-interest; Utilize the binary image of lung's edge point-of-interest to grow lung's left and right edges, obtain the left and right edge image of lung (as shown in Figure 5).Lung's width is reached inside image distance lung edge to after pre-service position calculation row peak point, obtains apex pulmonis position, thus obtains cardiopulmonary area-of-interest (as shown in Figure 6).Use by being slightly partitioned into heart area and lung areas (as shown in Figure 7) to the segmentation strategy of essence.The trans D of heart area, lung areas and cardiopulmonary is carried out indicating (as shown in Figure 8) and carries out the calculating of ambition ratio, ambition area ratio respectively.Finally illustrate other five width Chest Image results (Fig. 9-13).

Claims (10)

1. the ambition of chest digitized video is than a Measurement Algorithm, it is characterized in that, specifically comprises the steps:

Step 1: the Chest Image loaded is calibrated, obtains calibrating rear image;

Step 2: carry out down-sampling and horizontal mean filter process to calibration chart picture, obtains image after pre-service;

Step 3: image X after the pre-service that step 2 is obtained _fin each pixel carry out detection and obtain row data candidate peak point, according to row data candidate peak point, binary conversion treatment is carried out to image after pre-service, obtains the bianry image of lung's edge point-of-interest;

Step 4: the bianry image X utilizing lung's edge point-of-interest _iin all lungs edge point-of-interest, grow the edge lines of the lung left and right sides;

Step 5: image chooses apex pulmonis row to be searched after the pre-service that step 2 obtains, scans apex pulmonis row to be searched, obtains column data candidate peak point, obtains apex pulmonis position by column data candidate peak point;

Step 6: cardiopulmonary area-of-interest is determined in the apex pulmonis position that the lung left and right sides edge lines obtained by step 4 and step 5 obtain, splits the rough segmentation obtaining heart area and lung areas to cardiopulmonary area-of-interest; Rim detection is carried out to cardiopulmonary region of interest area image and obtains edge image, utilize edge image to be split image accurately;

Step 7: the heart in the accurate segmentation image utilizing step 6 to obtain and lung areas calculate ambition ratio and ambition area ratio.

2. the ambition of chest digitized video as claimed in claim 1 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 1 are as follows:

1.1) laterally calibration: loading size is the Chest Image of M × N, to described chest image from top to bottom the individual element scanning is carried out respectively to centre in the left and right two ends of row, and obtain the pixel that left and right two gray scales are greater than 50 respectively, the horizontal ordinate of two described pixels is respectively B _l, B _r; If the pixel count on these two pixel distance Chest Image borders is all more than 16, then by B in Chest Image _lall row on-16 row left sides and B _rthe pixel of all row on+16 row the right is deleted, and has laterally calibrated;

1.2) longitudinal alignment: respectively along the B of the Chest Image after laterally calibrating _l+ 16 and B _r-16 row carry out picture element scan, run into the ordinate S that in two row, first gray scale is greater than 50 respectively _land S _rstop, both calculating average if S row distance upper end boundary pixel number is more than 16, then delete S-16 capable more than row pixel, longitudinal alignment completes, and obtains calibrating rear image.

3. the ambition of chest digitized video as claimed in claim 1 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 2 are as follows:

2.1) after the calibration obtained step 1, image adopts bilinear interpolation to carry out filtering process, obtains the down-sampled images X that size is 256 × 256 _d;

2.2) to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process, obtaining size is image X after the pre-service of 256 × 256 _f; Wherein, to down-sampled images X _din the pixel of each coordinate position (i, j) carry out 1 × 3 horizontal mean filter process and refer to and utilize following formula to process:

$y_{i,j}=\frac{x_{i,j-1}+x_{i,j}+x_{i,j+1}}{3}$

In formula, x _i,jfor current location is at down-sampled images X _din gray scale, y _i,jfor the filtering result of current location.

4. the ambition of chest digitized video as claimed in claim 1 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 3 are as follows:

3.1) image X after pre-service step 2 obtained _fin the individual element scanning from left to right of often row, for each pixel in every row with its be expert in the pixel value of last pixel do first order difference computing and obtain difference result corresponding to current pixel; If the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is row data candidate peak point p _j, wherein, p is gray scale, and subscript j represents the position in current pixel row residing for it;

3.2) for each row data candidate peak point p _j, get left and right apart from it be 5 two pixel p _j-5and p _j+5, according to following formulae discovery image line crest impact degree:

${\mathrm{\&theta;}}_1=\arctan \left(\frac{p_j-p_{j-5}}{t}\right),$

${\mathrm{\&theta;}}_2=\arctan \left(\frac{p_j-p_{j+5}}{t}\right),$

θ＝π-θ ₁-θ ₂，

0 < tan (θ)≤1.1, then p if satisfy condition _jfor lung's edge point-of-interest;

By 3.1 and step 3.2) traversal pre-service after image X _fall pixels after, obtain multiple lungs edge point-of-interest; Thus obtain the bianry image that size is lung's edge point-of-interest of 256 × 256, be designated as X _i.

5. the ambition of chest digitized video as claimed in claim 1 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 4 are as follows:

4.1) the bianry image X of the lung's edge point-of-interest obtained with step 3 _ibe often classified as unit, individual element scans from bottom to top, if a certain Lie Zhong continuous print lung edge point-of-interest number is no less than 10, then records the vertical line segment be made up of these continuous print point-of-interests; To the bianry image X of lung's edge point-of-interest _iin after all row all scan, obtain the vertical line segment aggregate L={l comprising the vertical line segment of k bar ₁, l ₂l _k;

4.2) by the bianry image X of lung's edge point-of-interest _ibe divided into left and right two parts, the position l that the connected region area formed for the lung's edge point-of-interest on the vertical line segment in these two parts is maximum _land l _r, by two vertical line segment l _land l _rthe average of pixel ordinate is bottom designated as I _b;

4.3) image X after the pre-treatment _fin, by vertical line segment l _lwith vertical line segment l _rall as the lower edge lines of the initial two panels lobe of the lung; Two described edge lines are handled as follows respectively: current edge lines pixel is topmost designated as y _i,j, compare y _{i-1, j-1}, y _{i-1, j}and y _{i-1, j+1}the gray-scale value of position, includes the pixel of gray scale maximum value position in current edge lines, more repeatedly carries out above-mentioned process, until current edge lines grow into image X after pre-service _fupper end till, obtain the edge lines of the lung left and right sides respectively; Horizontal ordinate minimum value I on the lobe of the lung edge lines on the record left side _land the horizontal ordinate maximal value I on the lobe of the lung edge lines on the right _r.

6. the ambition of chest digitized video as claimed in claim 5 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 5 are as follows:

5.1) image X after the pre-service obtained from step 2 _fin choose apex pulmonis row to be searched; Get row, row, row and row are as apex pulmonis row to be searched;

5.2) each row in above-mentioned 6 row row to be searched are handled as follows: individual element scanning from top to bottom, each pixel is done first order difference computing with the pixel value of last pixel in its column and obtains difference result corresponding to current pixel, if the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is column data candidate peak point; After 6 row row to be searched are all processed, obtain multiple column data candidate peak point, using the ordinate mean value of nearest for ordinate in them two column data candidate peak points as apex pulmonis position ordinate the coboundary of cardiopulmonary area-of-interest is defined as

7. the ambition of chest digitized video as claimed in claim 6 is than Measurement Algorithm, and it is characterized in that, the concrete steps of described step 6 are as follows:

6.1) 4 the parameter I obtained by step 4 and step 5 _l, I _r, I _t, I _bwith edge, the lung left and right sides lines l that step 4 obtains _l, l _r, obtain cardiopulmonary part region of interest area image ROI; Ostu threshold segmentation algorithm is performed to ROI and obtains bianry image, 11 × 11 medium filterings are performed to this bianry image and obtains bianry image ROI after filtering _binary; To bianry image ROI after filtering _binaryfill l _l, l _rlung space with after segmentation, generates rough lung areas and rough left and right ventricles margo border of the lung;

6.2) bianry image ROI after scan-filtering from top to bottom _binaryevery a line, for every a line, searched for the horizontal ordinate at this row middle left and right cardiopulmonary edge to its left and right sides by this row central point with and check the horizontal ordinate at cardiopulmonary edge of 5 row of being separated by respectively with them with if both horizontal ordinate spans | i _j+5-i _j| >8 is (such as the horizontal ordinate at left cardiopulmonary edge its with the absolute value of difference of horizontal ordinate be greater than 8), then coordinate (i, j) is designated as heart lower boundary summit; Connect the heart lower boundary summit of arranged on left and right sides, with bianry image ROI after rough left and right ventricles margo border of the lung and filtering _binaryupper end be expert at and surround region and be rough heart area;

6.3) it is 3 × 7 that the cardiopulmonary part region of interest area image ROI obtained step 6.1 carries out yardstick, and standard deviation is do Canny rim detection after the Gaussian smoothing of 5, obtains edge image ROI _canny; Bianry image ROI after the filtering that step 6.1 obtains _binaryin rough left and right ventricles margo border of the lung pixel position correspond to edge image ROI _cannyin location of pixels Horizon Search edge, if search edge within the specific limits, then rough heart area is filled to edge, thus is split image ROI accurately _seg.

8. the ambition of chest digitized video is than a measuring system, it is characterized in that, specifically comprises the module be connected successively as follows:

Calibration module: the Chest Image loaded is calibrated, obtains calibrating rear image;

Pretreatment module: carry out down-sampling and horizontal mean filter process to calibration chart picture, obtains image after pre-service;

Binarization block: image X after the pre-service that pretreatment module is obtained _fin each pixel carry out detection and obtain row data candidate peak point, according to row data candidate peak point, binary conversion treatment is carried out to image after pre-service, obtains the bianry image of lung's edge point-of-interest;

Lung's marginal growth module: the bianry image X utilizing lung's edge point-of-interest _iin all lungs edge point-of-interest, grow the edge lines of the lung left and right sides;

Apex pulmonis position detecting module: image chooses apex pulmonis row to be searched after the pre-service that pretreatment module obtains, scans apex pulmonis row to be searched, obtains column data candidate peak point, obtains apex pulmonis position by column data candidate peak point;

Cardiopulmonary region segmentation module: cardiopulmonary area-of-interest is determined in the apex pulmonis position that the lung left and right sides edge lines obtained by lung's marginal growth module and apex pulmonis position detecting module obtain, splits the rough segmentation obtaining heart area and lung areas to cardiopulmonary area-of-interest; Rim detection is carried out to cardiopulmonary region of interest area image and obtains edge image, utilize edge image to be split image accurately;

Ambition is than computing module: the heart in the accurate segmentation image utilizing cardiopulmonary region segmentation module to obtain and lung areas calculate ambition ratio and ambition area ratio.

9. the ambition of chest digitized video as claimed in claim 8 is than Measurement Algorithm, and it is characterized in that, described lung marginal growth module has been used for following flow process:

4.1) the bianry image X of the lung's edge point-of-interest obtained with binarization block _ibe often classified as unit, individual element scans from bottom to top, if a certain Lie Zhong continuous print lung edge point-of-interest number is no less than 10, then records the vertical line segment be made up of these continuous print point-of-interests; To the bianry image X of lung's edge point-of-interest _iin after all row all scan, obtain the vertical line segment aggregate L={l comprising the vertical line segment of k bar ₁, l ₂l _k;

4.2) by the bianry image X of lung's edge point-of-interest _ibe divided into left and right two parts, the position l that the connected region area formed for the lung's edge point-of-interest on the vertical line segment in these two parts is maximum _land l _r, by two vertical line segment l _land l _rthe average of pixel ordinate is bottom designated as I _b;

4.3) image X after the pre-treatment _fin, by vertical line segment l _lwith vertical line segment l _rall as the lower edge lines of the initial two panels lobe of the lung; Two described edge lines are handled as follows respectively: current edge lines pixel is topmost designated as y _i,j, compare y _{i-1, j-1}, y _{i-1, j}and y _{i-1, j+1}the gray-scale value of position, includes the pixel of gray scale maximum value position in current edge lines, more repeatedly carries out above-mentioned process, until current edge lines grow into image X after pre-service _fupper end till, obtain the edge lines of the lung left and right sides respectively; Horizontal ordinate minimum value I on the lobe of the lung edge lines on the record left side _land the horizontal ordinate maximal value I on the lobe of the lung edge lines on the right _r

Described apex pulmonis position detecting module has been used for following flow process:

5.1) image X after the pre-service obtained from pretreatment module _fin choose apex pulmonis row to be searched; Get row, row, row and row are as apex pulmonis row to be searched;

5.2) each row in above-mentioned 6 row row to be searched are handled as follows: individual element scanning from top to bottom, each pixel is done first order difference computing with the pixel value of last pixel in its column and obtains difference result corresponding to current pixel, if the sign symbol of the difference result that the difference result that current pixel is corresponding is corresponding from its last pixel is different, then think that current pixel is column data candidate peak point; After 6 row row to be searched are all processed, obtain multiple column data candidate peak point, using the ordinate mean value of nearest for ordinate in them two column data candidate peak points as apex pulmonis position ordinate the coboundary of cardiopulmonary area-of-interest is defined as

10. the ambition of chest digitized video as claimed in claim 9 is than Measurement Algorithm, and it is characterized in that, described cardiopulmonary region segmentation module has been used for following flow process:

6.1) by 4 parameter I _l, I _r, I _t, I _bwith edge, the lung left and right sides lines l that step 4 obtains _l, l _r, obtain cardiopulmonary part region of interest area image ROI; Ostu threshold segmentation algorithm is performed to ROI and obtains bianry image, 11 × 11 medium filterings are performed to this bianry image and obtains bianry image ROI after filtering _binary; To bianry image ROI after filtering _binaryfill l _l, l _rlung space with after segmentation, generates rough lung areas and rough left and right ventricles margo border of the lung;

6.2) bianry image ROI after scan-filtering from top to bottom _binaryevery a line, for every a line, searched for the horizontal ordinate at this row middle left and right cardiopulmonary edge to its left and right sides by this row central point with and check the horizontal ordinate at cardiopulmonary edge of 5 row of being separated by respectively with them with if both horizontal ordinate spans | i _j+5-i _j| >8 is (such as the horizontal ordinate at left cardiopulmonary edge its with the absolute value of difference of horizontal ordinate be greater than 8), then coordinate (i, j) is designated as heart lower boundary summit; Connect the heart lower boundary summit of arranged on left and right sides, with bianry image ROI after rough left and right ventricles margo border of the lung and filtering _binaryupper end be expert at and surround region and be rough heart area;

6.3) it is 3 × 7 that the cardiopulmonary part region of interest area image ROI obtained step 6.1 carries out yardstick, and standard deviation is do Canny rim detection after the Gaussian smoothing of 5, obtains edge image ROI _canny; Bianry image ROI after the filtering _binaryin rough left and right ventricles margo border of the lung pixel position correspond to edge image ROI _cannyin location of pixels Horizon Search edge, if search edge within the specific limits, then rough heart area is filled to edge, thus is split image ROI accurately _seg.