CN102289657A - Breast X ray image lump detecting system based on visual attention mechanism - Google Patents
Breast X ray image lump detecting system based on visual attention mechanism Download PDFInfo
- Publication number
- CN102289657A CN102289657A CN2011101227220A CN201110122722A CN102289657A CN 102289657 A CN102289657 A CN 102289657A CN 2011101227220 A CN2011101227220 A CN 2011101227220A CN 201110122722 A CN201110122722 A CN 201110122722A CN 102289657 A CN102289657 A CN 102289657A
- Authority
- CN
- China
- Prior art keywords
- entropy
- diagram
- pyramid
- amplitude
- image
- 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
Images
Abstract
The invention discloses a breast X ray image lump detecting system and a detecting method based on a visual attention mechanism, mainly solving the problems of low detection rate and high false positive rate of the prior art. The whole system comprises an image pre-processing module, a feature extraction module, a lump detecting module and a detection result outputting module. The image preprocessing module preprocesses an original image; the feature extraction module extracts features of the preprocessed image; the lump detecting module generates a Gauss pyramid for a feature image of the features, then processes the Gauss pyramid to obtain a total saliency map, and finally cuts the total saliency map to obtain candidate and doubtful lumps and filters false positive lumps; and the detection result outputting module outputs the detection result. The breast X ray image lump detecting system has the advantages of high detection rate and low false positive rate of the breast X ray image lump and can be used for the detection of a region of interest and the aided diagnosis of a medical image.
Description
Technical field
The invention belongs to technical field of image processing, particularly relate to medical image processing, the area-of-interest that can be used for medical image detects and auxiliary diagnosis.
Background technology
In recent decades, medical image has become one of field with fastest developing speed in the medical technology, and its result makes the clinician more direct, more clear to the observation of inside of human body diseased region, and diagnosis rate is also higher.Computer-aided diagnosis (Computer Aided Diagnosis abbreviates CAD as) technology is called as doctor's " the second eyes eyeball ", how research is effectively handled these medical image informations by image processing techniques, be used for auxiliary doctor's diagnosis even the planning etc. that undergos surgery, have important social benefit and application prospects.The Medical Image Processing technology develops so far as the key of computer-aided diagnosis, the intersection of each subject has been the inexorable trend of development, but wherein also have a lot of problems to need to be resolved hurrily, flourish along with tele-medicine particularly, to Medical Image Processing with to analyze requirement also more and more higher, so further study Medical Image Processing and analysis has crucial meaning.
Breast cancer is the modal cancer of harm whole world women's health, more and more obtains the concern of all circles in recent years.The mammary X-ray image is a kind of the most common and to the detection of breast cancer otherwise effective technique relatively, the lump detection method of mammary X-ray image generally has simple threshold method, obtain area-of-interest based on method, the pre-segmentation of wave filter after, extract the simple sorter of associating detects after the feature of area-of-interest method etc.But threshold method is not considered the characteristic of entire image only at the neighborhood of single pixel or pixel; General lump to definite shape of method based on wave filter has humidification, and lump edge variation complexity in the mammary X-ray image, this class methods scope of application is narrower; After the pre-segmentation area-of-interest is extracted feature, utilize sorter to judge whether area-of-interest is the method for lump again, because interpersonal personalized difference is big, the characteristic of lump varies, can not there be complete feature can express the difference of lump and normal region, also can't obtains very good result so the performance of sorter is good again.The recall rate that generally obtains when therefore adopting said method to carry out the lump detection is lower, and the false positive rate of lump is higher.
Summary of the invention
The objective of the invention is to overcome the deficiency of above-mentioned prior art,, lump characteristic big at the personalized difference of mammary X-ray image is difficult to expression, the low problem of recall rate, a kind of mammary X-ray image lump detection system based on vision noticing mechanism is proposed, with recall rate and its false positive rate of reduction that improves mammary X-ray image lump.
For achieving the above object, the invention provides mammary X-ray image lump detection system, comprising based on vision noticing mechanism:
The image pretreatment module is used to adopt the histogram equalization method, original mammary X-ray image is cut and enhancement process the image f (X) after being enhanced;
Characteristic extracting module is used for that the image f (X) after strengthening is carried out piece with the pixel of 2 * 2 sizes and divides, and obtains block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change;
The characteristic pattern generation module replaces block of pixels x with 4 eigenwerts that characteristic extracting module obtains respectively
iI=1,2,3..., N, form average figure M, variogram V, amplitude figure A and brightness rate of change figure I, and these 4 characteristic patterns are carried out normalized respectively, obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized;
The gaussian pyramid generation module, be used for each this pyramid of self-generating three floor heights to the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', this pyramid of the three floor heights M ' that obtains average figure M ' is u=1 (u), and 2,3, this pyramid of three floor heights V ' of variogram V ' is u=1 (u), and 2,3, (u) u=1 of this pyramid of three floor heights A ' of amplitude figure A ', 2,3 and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3;
The characteristic pattern conversion module comprises:
The preliminary making submodule is used for described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u), the every width of cloth image of I ' in (u) is made as X ' (u), X represents M, V, A and I, u=1,2,3;
Neighborhood entropy calculating sub module is used for each pixel x (u) to X '
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u);
The feature entropy diagram generates submodule, is used for the entropy H (x that will obtain
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3;
Difference block, be used to adopt central neighborhood difference method, respectively to described 4 three this pyramid of floor height M " (u), V " (u), A " (u) and I " (u), u=1,2,3 carry out difference processing, make each gaussian pyramid obtain two width of cloth difference diagrams, promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
Standardized module is used to adopt the standardization operator N () of Itti proposition to described difference diagram
With
V=1,2 carry out standardization respectively, obtain the average entropy difference diagram after the standardization
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2;
Characteristic remarkable picture generation module, two width of cloth difference diagrams of each gaussian pyramid that the standardization submodule is obtained respectively
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude that obtains 4 gaussian pyramid correspondences is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
Total significantly figure generation module is used for described remarkable figure
With
Carry out the linearity fusion and obtain a total significantly figure S;
Total significantly figure is cut apart module, is used for total significantly figure S is carried out two class cluster segmentation by k-means, obtains bianry image, and wherein highlighted part is the suspicious lump of candidate;
False positive lump filtering module is used for adopting the false positive lump of morphological feature and the suspicious lump of priori filtering candidate, obtains suspicious lump position;
The testing result output module is used for the image f (X) after position with suspicious lump corresponds to enhancing, and suspicious lump testing result is exported in the suspicious lump of mark position in f (X).
For achieving the above object, the invention provides mammary X-ray image lump detection method, comprise the steps: based on vision noticing mechanism
(1) adopts the histogram equalization method, original mammary X-ray image is cut and enhancement process the image f (X) after being enhanced;
(2) the image f (X) after will strengthening carries out piece with the pixel of 2 * 2 sizes and divides, and obtains block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change;
(3) 4 eigenwerts using step (2) to obtain respectively replace block of pixels x
iI=1,2,3, ..., N forms average figure M, variogram V, amplitude figure A and brightness rate of change figure I, and these 4 characteristic patterns are carried out normalized respectively, obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized;
(4) to each this pyramid of self-generating three floor heights of the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', this pyramid of the three floor heights M ' that obtains average figure M ' is u=1 (u), 2,3, this pyramid of three floor heights V ' of variogram V ' is u=1 (u), 2,3, (u) u=1 of this pyramid of three floor heights A ' of amplitude figure A ', 2,3 and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3;
(5) characteristic pattern in the pyramid is converted into the feature entropy diagram;
5a) establish described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u), the every width of cloth image of I ' in (u) be X ' (u), X represents M, V, A and I, u=1,2,3;
5b) to X ' each pixel x (u)
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u);
5c) with the entropy H (x that obtains
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3;
(6) adopt central neighborhood difference method, respectively to this pyramid of three floor heights M " (u), this pyramid of three floor heights V of variance entropy diagram " of average entropy diagram (u), this pyramid of three floor heights A of amplitude entropy diagram " (u) and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3 carry out difference processing, make each gaussian pyramid obtain two width of cloth difference diagrams, promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
(7) adopt the standardization operator N () of Itti proposition to average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2 carry out standardization respectively, obtain the average entropy difference diagram after the standardization
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2;
(8) two width of cloth difference diagrams of each gaussian pyramid that respectively step (7) is obtained
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude that obtains 4 gaussian pyramid correspondences is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
(9) with described remarkable figure
With
Carry out the linearity fusion and obtain a total significantly figure S;
(10) total significantly figure S is carried out two class cluster segmentation by k-means, obtain bianry image, wherein highlighted part is the suspicious lump of candidate;
(11) the false positive lump in employing morphological feature and the suspicious lump of priori filtering candidate obtains suspicious lump position;
(12) position of suspicious lump is corresponded to image f (X) after the enhancing, suspicious lump testing result is exported in the suspicious lump of mark position in f (X).
The present invention has the following advantages compared with prior art:
1) the present invention sets about from the vision noticing mechanism of human eye, and the imitation radiologist detects the mechanism of lump and carries out the lump detection, is a kind of new lump detection system, and obtains higher recall rate and lower false positive rate.
2) the present invention adopts average, variance, amplitude and 4 features of brightness rate of change according to characteristics such as the densification of lump center, high brightness in the mammary X-ray image, has better represented the characteristic of lump in the mammary X-ray image than the textural characteristics of general use.
3) the present invention compares with the existing method that generates remarkable figure based on vision noticing mechanism, behind generating feature figure gaussian pyramid, utilize the thought of information entropy that the characteristic pattern in the gaussian pyramid is converted into the feature entropy diagram, make among the remarkable figure that obtains the contrast of area-of-interest and background area higher.
The simulation experiment result shows, what the present invention proposed has improved the recall rate of lump based on the mammary X-ray image lump detection system of vision noticing mechanism, has reduced false positive rate.
Description of drawings
Fig. 1 is a virtual system synoptic diagram of the present invention;
Fig. 2 is a detection method process flow diagram of the present invention;
Fig. 3 carries out the synoptic diagram that piece is divided to image among the present invention;
Fig. 4 is the figure as a result of testing process of the present invention;
Fig. 5 is simulation result figure of the present invention.
Embodiment
With reference to Fig. 1, the virtual system of mammary X-ray image lump detection system based on vision noticing mechanism of the present invention for constituting by computer software, it comprises: image pretreatment module, characteristic extracting module, characteristic pattern generation module, gaussian pyramid generation module, characteristic pattern conversion module, difference block, standardized module, characteristic remarkable picture generation module, total significantly figure generation module, total significantly figure are cut apart module, false positive lump filtering module and testing result output module, these functional modules cooperatively interact, and finish the detection to mammary X-ray image lump jointly.Wherein:
The image pretreatment module adopts the histogram equalization method, original mammary X-ray image is cut and enhancement process the image f (X) after being enhanced;
Characteristic extracting module is carried out piece with the image f (X) after strengthening with the pixel of 2 * 2 sizes and is divided, and obtains block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change;
The characteristic pattern generation module replaces block of pixels x with 4 eigenwerts that characteristic extracting module obtains respectively
iI=1,2,3..., N, form average figure M, variogram V, amplitude figure A and brightness rate of change figure I, and these 4 characteristic patterns are carried out normalized respectively, obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized;
The gaussian pyramid generation module, to each this pyramid of self-generating three floor heights of the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', this pyramid of the three floor heights M ' that obtains average figure M ' is u=1 (u), and 2,3, this pyramid of three floor heights V ' of variogram V ' is u=1 (u), and 2,3, (u) u=1 of this pyramid of three floor heights A ' of amplitude figure A ', 2,3 and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3;
The characteristic pattern conversion module comprises that preliminary making submodule, neighborhood entropy calculating sub module and feature entropy diagram generate submodule.This preliminary making submodule, with described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u), the every width of cloth image of I ' in (u) is made as X ' (u), X represents M, V, A and I, u=1,2,3; This neighborhood entropy calculating sub module is to X ' each pixel x (u)
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u); This feature entropy diagram generates submodule, with the entropy H (x that obtains
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3;
Difference block, adopt central neighborhood difference method, respectively to described 4 three this pyramid of floor height M " (u), V " (u), A " (u) and I " (u), u=1,2,3 carry out difference processing, make each gaussian pyramid obtain two width of cloth difference diagrams, promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
Standardized module, the standardization operator N () that adopts the Itti proposition is to described difference diagram
With
V=1,2 carry out standardization respectively, obtain the average entropy difference diagram after the standardization
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2;
Characteristic remarkable picture generation module, two width of cloth difference diagrams of each gaussian pyramid that the standardization submodule is obtained respectively
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude that obtains 4 gaussian pyramid correspondences is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
Total significantly figure generation module is with described remarkable figure
With
Carry out linearity and merge, promptly
With
Obtain a total significantly figure S behind linear the fusion;
Total significantly figure is cut apart module, and total significantly figure S is carried out two class cluster segmentation by k-means, the bianry image after promptly obtaining cutting apart, and wherein highlighted part is the suspicious lump of candidate;
False positive lump filtering module adopts the false positive lump in morphological feature and the suspicious lump of priori filtering candidate, obtains suspicious lump position;
The testing result output module corresponds to image f (X) after the enhancing with the position of suspicious lump, and suspicious lump testing result is exported in the suspicious lump of mark position in f (X).
With reference to Fig. 2 and Fig. 4, the mammary X-ray image lump detection method based on vision noticing mechanism of the present invention comprises the steps:
The pre-service of the original mammary X-ray image of step 1..
1a) the original mammary gland X image shown in Fig. 4 (a) is adopted image level and the automatic cutting method of vertical computing machine, the artificial marking that exists in the background of excision image and the image, the mammary gland X image after obtaining cutting;
1b) adopt the histogram equalization enhancement process to remove noise, obtain the image f (X) after the enhancing shown in Fig. 4 (b) the mammary X-ray image after the cutting.
Image f (X) after the step 2. pair enhancing carries out piece and divides and extract eigenwert.
2a) with reference to Fig. 3, the image f (X) after strengthening is carried out piece with the pixel of 2 * 2 sizes divide, obtain block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, and establish x
iIn 4 pixels comprising be x
I1, x
I2, x
I3, x
I4
2b) to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change and is respectively:
Average:
Variance:
Amplitude: R (x
i)=max (l
Ik)-min (l
Ik)
The brightness rate of change:
L wherein
IkBe block of pixels x
iIn each pixel x
IkCorresponding pixel value, k=1,2,3,4, j=(k+1) mod 4.
Step 3. generating feature figure and each are carried out normalized respectively to characteristic pattern.
3a) 4 eigenwerts using step (2) to obtain respectively replace block of pixels x
iI=1,2,3 ..., N forms average figure M, variogram V, amplitude figure A and these 4 characteristic patterns of brightness rate of change figure I;
3b) respectively the pixel value of described 4 characteristic patterns is normalized to [0 ..., 1], obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized.
4a) respectively with average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' as gaussian pyramid ground floor M ' (1), V ' (1), A ' (1) and I ' (1);
4b) respectively described M ' (1), V ' (1), A ' (1) and I ' (1) are descended 2 samplings, image M ' (2), V ' (2), A ' (2) and the I ' (2) that obtains after 2 samplings under using is as the second layer of gaussian pyramid;
4c) respectively described M ' (2), V ' (2), A ' (2) and I ' (2) are descended 2 samplings, with image M ' (3), V ' (3), A ' (3) and the I ' (3) that obtain after following 2 samplings the 3rd layer as gaussian pyramid, this pyramid of the three floor heights M ' that promptly obtains average figure M ' (u), this pyramid of three floor heights V ' of variogram V ' (u), this pyramid of three floor heights A ' of amplitude figure A ' (u) and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3.
Step 5. is converted into the feature entropy diagram with the characteristic pattern in the gaussian pyramid.
5a) establish described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u) and the every width of cloth image of I ' in (u) be X ' (u), X represents M, V, A and I, u=1,2,3;
5b) to X ' each pixel x (u)
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u);
5c) with the entropy H (x that obtains
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3.
Step 6. adopts central neighborhood difference method, respectively to this pyramid of three floor heights M " (u), this pyramid of three floor heights V of variance entropy diagram " of average entropy diagram (u), this pyramid of three floor heights A of amplitude entropy diagram " (u) and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3 carry out difference processing.
6a) with this pyramid second layer of three floor heights M " this pyramid second layer of three floor heights of (2), variance entropy diagram V " (2) of average entropy diagram, this pyramid second layer of three floor heights A " this pyramid second layer of three floor heights of (2) and brightness rate of change entropy diagram I " (2) of amplitude entropy diagram, adopt cube interpolation method to amplify a twice, obtain amplifying the average entropy diagram after the twice
The variance entropy diagram
The amplitude entropy diagram
With brightness rate of change entropy diagram
6b) with the 3rd layer of M of this pyramid of three floor heights " the 3rd layer of V of this pyramid of three floor heights of (3), variance entropy diagram " (3) of average entropy diagram, the 3rd layer of A of this pyramid of three floor heights " the 3rd layer of I of this pyramid of three floor heights of (3) and brightness rate of change entropy diagram " (2) of amplitude entropy diagram, adopt cube interpolation method quadruplication, obtain the average entropy diagram after the quadruplication
The variance entropy diagram
The amplitude entropy diagram
With brightness rate of change entropy diagram
6c) use described M " (1), V " (1), A " (1) and I " (1) respectively with described
With
Subtract each other by pixel, obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
6d) use described M " (1), V " (1), A " (1) and I " (1) respectively with described
With
Subtract each other by pixel, obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
Promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
The standardization operator N () that step 7. adopts Itti to propose is handled each feature entropy difference diagram.
7a) with average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1, every width of cloth image is made as in 2
X represents M, V, A and I, v=1,2;
7c) find out
Maximal value W in all pixel values, and calculate all local peaked average w except that W;
7d) will
All pixel values multiply by (W-w)
2, obtain the feature entropy difference diagram after the standardization
, X represents M, V, A and I, v=1, and 2, promptly obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2.
Two width of cloth difference diagrams of each gaussian pyramid that step 8. obtains step (7)
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude of 4 gaussian pyramid correspondences that obtain is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
Be respectively:
Step 9. is significantly schemed average
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
Carry out linearity and merge, obtain the total significantly figure S behind linear fusion the shown in Fig. 4 (c),
Step 10. adopts the k-means clustering method to cut apart total significantly figure S and obtains the suspicious lump of candidate.
10a) significantly scheme all pixel values of S as the cluster sample with total;
10b) the class number of setting the k-means cluster is 2, and two value mu1 of cluster centre and the initialization formula of mu2 are respectively mu1=max (l
p)/3, mu2=2 * max (l
p)/3, wherein l
pRepresent the pixel value of S, p=1,2,3 ..., n, n are the pixel count that S comprises;
10c) distance of comparison each sample to two cluster centre mu1 and mu2 is selected wherein minimum value, and gives sample with the corresponding class label of this minimum value, obtains all target data sample label;
10d) calculate the mean value of each class target data, obtain new cluster centre;
10e) new cluster centre and former cluster centre are compared, if new cluster centre is different with former cluster centre, then turn back to step 10c), if new cluster centre is identical with former cluster centre, then two values with new cluster centre compare, and be that the pixel value of a class is with 255 replacements with the higher value of new cluster centre in respectively will total significantly figure S, to be that the pixel value of a class is with 0 replacement with the smaller value of new cluster centre among total significantly figure S, obtain bianry image L, L is the figure as a result that significantly schemes after S is cut apart total;
10f) the bianry image L after will cutting apart adopts cube interpolation method to amplify a twice, obtains amplifying shown in Fig. 4 (d) the bianry image L ' after the twice, and highlighted part is the suspicious lump of candidate among the figure.
Step 11. adopts morphological feature and priori, the false positive lump in the suspicious lump of filtering candidate.
11a) adopt eccentricity and dutycycle as morphological feature, gray average and number of pixels scope are as priori, and the threshold value of establishing eccentricity is that the threshold value of E, dutycycle is D, and the threshold value of gray average is G, and the threshold range of number of pixels is [Num1, Num2];
11b) adopt the position of the suspicious lump of each candidate among the bwlabel function mark bianry image L ' that matlab carries;
11c) position of the suspicious lump of each candidate of mark is corresponded to image f (X) after the enhancing shown in Fig. 4 (b), calculate the gray average G ' of the suspicious lump of each candidate
z, z=1,2,3 ..., n ', n ' they are the number of the suspicious lump of candidate;
11d) the regionprops function that adopts matlab to carry is obtained the eccentricity E ' of the suspicious lump of each candidate in bianry image L '
z, dutycycle D '
zWith number of pixels Num
zZ=1,2,3 ..., n ',
11e) if the gray average G ' of the suspicious lump of candidate
z, eccentricity E '
z, dutycycle D '
zWith number of pixels Num
zSatisfy G '
z<G, E '
z<E, D '
z<D or
In any one, think that then the suspicious lump of this candidate is the false positive lump, the suspicious lump of this candidate of filtering obtains the final suspicious lump position behind the filtering false positive lump shown in Fig. 4 (e).
Step 12. corresponds to image f (X) after the enhancing shown in Fig. 4 (b), the suspicious lump of mark position in f (X), the suspicious lump testing result of output shown in Fig. 4 (f) with the position of final suspicious lump.
Effect of the present invention can further specify mammary gland X image simulation data by following:
1. experiment condition
Emulation of the present invention is at windows 7, SPI, CPU Pentium (R) 4, basic frequency 2.4GHZ, software platform is the MatlabR2010a operation, the original mammary gland X image that emulation is selected for use derives from common data sets MIAS, obtains the mammary X-ray image of 50 width of cloth generation cancerations altogether, and this 50 width of cloth image comprises 54 lumps altogether.
2. emulation content and result
2.1) obtaining mammary gland X image simultaneously, the image information of acquisition also includes: the types of organization of breast, the position of pathological abnormalities, unusual size, the position of breast cancer, tumor type isostructuralism matter.According to the position of pathological abnormalities and unusual size information, in former figure,, be used to verify and adopt the inventive method to carry out the result that lump detects in the experiment with elliptic curve circle mark lump position.
2.2) experiment eccentricity threshold value E span is 0.88~0.95, duty cycle threshold D span is 0.32~0.42, the threshold value G of gray average is 170, number of pixels threshold value Num1=100, Num2=30000.
Adopt system of the present invention that the original mammary X-ray image of 50 width of cloth generation cancerations chosen is carried out lump and detect, testing result is as shown in table 1:
Table 1 testing result
The actual lump that comprises | Detect lump | Recall rate | False positive rate |
54 | 49 | 90.74% | 3.8 |
False positive rate is the false positive lump number that average every width of cloth image comprises in the table 1.
This experiment is to simulation result such as the Fig. 5 of 4 width of cloth mammary X-ray images in 50 width of cloth mammary X-ray images of choosing, wherein (a) and (b) among Fig. 5, (c), (d) have been for having marked the mammary X-ray image of lump position according to image information with the elliptic curve circle, (e), (f), (g), (h) be for detecting the result of lump with the present invention.From Fig. 5 as seen, mammary X-ray image the present invention of 4 width of cloth generation cancerations has correctly been detected the position of lump, and only had and respectively detect two false positive lumps in two width of cloth images.
Above result shows: the present invention can carry out lump to the mammary X-ray image effectively and detect, and has higher recall rate and lower false positive rate, is a kind of detection system that can effectively detect mammary X-ray image lump.
Claims (6)
1. mammary X-ray image lump detection system based on vision noticing mechanism comprises:
The image pretreatment module is used to adopt the histogram equalization method, original mammary X-ray image is cut and enhancement process the image f (X) after being enhanced;
Characteristic extracting module is used for that the image f (X) after strengthening is carried out piece with the pixel of 2 * 2 sizes and divides, and obtains block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change;
The characteristic pattern generation module replaces block of pixels x with 4 eigenwerts that characteristic extracting module obtains respectively
iI=1,2,3..., N, form average figure M, variogram V, amplitude figure A and brightness rate of change figure I, and these 4 characteristic patterns are carried out normalized respectively, obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized;
The gaussian pyramid generation module, be used for each this pyramid of self-generating three floor heights to the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', this pyramid of the three floor heights M ' that obtains average figure M ' is u=1 (u), and 2,3, this pyramid of three floor heights V ' of variogram V ' is u=1 (u), and 2,3, (u) u=1 of this pyramid of three floor heights A ' of amplitude figure A ', 2,3 and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3;
The characteristic pattern conversion module comprises:
The preliminary making submodule is used for described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u), the every width of cloth image of I ' in (u) is made as X ' (u), X represents M, V, A and I, u=1,2,3;
Neighborhood entropy calculating sub module is used for each pixel x (u) to X '
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u);
The feature entropy diagram generates submodule, is used for the entropy H (x that will obtain
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3;
Difference block, be used to adopt central neighborhood difference method, respectively to described 4 three this pyramid of floor height M " (u), V " (u), A " (u) and I " (u), u=1,2,3 carry out difference processing, make each gaussian pyramid obtain two width of cloth difference diagrams, promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
Standardized module is used to adopt the standardization operator N () of Itti proposition to described difference diagram
With
V=1,2 carry out standardization respectively, obtain the average entropy difference diagram after the standardization
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2;
Characteristic remarkable picture generation module, two width of cloth difference diagrams of each gaussian pyramid that the standardization submodule is obtained respectively
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude that obtains 4 gaussian pyramid correspondences is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
Total significantly figure generation module is used for described remarkable figure
With
Carry out the linearity fusion and obtain a total significantly figure S;
Total significantly figure is cut apart module, is used for total significantly figure S is carried out two class cluster segmentation by k-means, obtains bianry image, and wherein highlighted part is the suspicious lump of candidate;
False positive lump filtering module is used for adopting the false positive lump of morphological feature and the suspicious lump of priori filtering candidate, obtains suspicious lump position;
The testing result output module is used for the image f (X) after position with suspicious lump corresponds to enhancing, and suspicious lump testing result is exported in the suspicious lump of mark position in f (X).
2. the mammary X-ray image lump detection method based on vision noticing mechanism comprises the steps:
(1) adopts the histogram equalization method, original mammary X-ray image is cut and enhancement process the image f (X) after being enhanced;
(2) the image f (X) after will strengthening carries out piece with the pixel of 2 * 2 sizes and divides, and obtains block of pixels x
iI=1,2,3 ..., N, N are the number that f (X) piece is divided, to block of pixels x
iI=1,2,3 ..., N extracts average, variance, amplitude and 4 eigenwerts of brightness rate of change;
(3) 4 eigenwerts using step (2) to obtain respectively replace block of pixels x
iI=1,2,3, ..., N forms average figure M, variogram V, amplitude figure A and brightness rate of change figure I, and these 4 characteristic patterns are carried out normalized respectively, obtain average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' after the normalized;
(4) to each this pyramid of self-generating three floor heights of the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', this pyramid of the three floor heights M ' that obtains average figure M ' is u=1 (u), 2,3, this pyramid of three floor heights V ' of variogram V ' is u=1 (u), 2,3, (u) u=1 of this pyramid of three floor heights A ' of amplitude figure A ', 2,3 and (u) u=1 of this pyramid of three floor heights I ' of brightness rate of change figure I ', 2,3;
(5) characteristic pattern in the pyramid is converted into the feature entropy diagram;
5a) establish described 4 three this pyramid of floor height M ' (u), V ' (u), A ' (u), the every width of cloth image of I ' in (u) be X ' (u), X represents M, V, A and I, u=1,2,3;
5b) to X ' each pixel x (u)
tT=1,2,3 ..., n gets 7 * 7 windows, calculates the entropy H (x in 7 * 7 neighborhoods
t):
Wherein T is the number of grayscale levels that occurs in this neighborhood, p
TbBe the probability that gray level b occurs, p in this neighborhood
Tb=n
b'/49, n
b' the number of times that occurs in this neighborhood for gray level b, n are the number of pixels that X ' comprises in (u);
5c) with the entropy H (x that obtains
t) t=1,2,3 ..., n composition characteristic entropy diagram X " (u) X represents M, V, A and I, u=1,2; 3, with X " (u) replaces X ' (u), obtains this pyramid of three floor heights M of average entropy diagram " (u) u=1; 2,3, this pyramid of three floor heights V " (u) u=1,2 of variance entropy diagram, 3, this pyramid of three floor heights A of amplitude entropy diagram " (u) u=1,2,3 and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3;
(6) adopt central neighborhood difference method, respectively to this pyramid of three floor heights M " (u), this pyramid of three floor heights V of variance entropy diagram " of average entropy diagram (u), this pyramid of three floor heights A of amplitude entropy diagram " (u) and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3 carry out difference processing, make each gaussian pyramid obtain two width of cloth difference diagrams, promptly 4 gaussian pyramids separately correspondence obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2;
(7) adopt the standardization operator N () of Itti proposition to average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2 carry out standardization respectively, obtain the average entropy difference diagram after the standardization
Variance entropy difference diagram
Amplitude entropy difference diagram
With rate of change entropy difference diagram
V=1,2;
(8) two width of cloth difference diagrams of each gaussian pyramid that respectively step (7) is obtained
With
X represents M, V, A and I, the linearity significantly figure that permeates
The amplitude that obtains 4 gaussian pyramid correspondences is significantly schemed
Variance is significantly schemed
Amplitude is significantly schemed
Significantly scheme with rate of change
(9) with described remarkable figure
With
Carry out the linearity fusion and obtain a total significantly figure S;
(10) total significantly figure S is carried out two class cluster segmentation by k-means, obtain bianry image, wherein highlighted part is the suspicious lump of candidate;
(11) the false positive lump in employing morphological feature and the suspicious lump of priori filtering candidate obtains suspicious lump position;
(12) position of suspicious lump is corresponded to image f (X) after the enhancing, suspicious lump testing result is exported in the suspicious lump of mark position in f (X).
3. method according to claim 2, wherein the described employing histogram equalization of step 1 method is cut and enhancement process original mammary X-ray image, carries out as follows:
1a) the original mammary gland X image to input adopts image level and the automatic cutting method of vertical computing machine, the artificial marking that exists in the background of excision image and the image, the mammary gland X image after obtaining cutting;
1b) adopt the histogram equalization enhancement process to remove noise, the image f (X) after being enhanced to the mammary X-ray image after the cutting.
4. method according to claim 2, wherein step 4 is described to each this pyramid of self-generating three floor heights of the average figure M ' after the normalized, variogram V ', amplitude figure A ' and brightness rate of change figure I ', carries out as follows:
4a) respectively with average figure M ', variogram V ', amplitude figure A ' and brightness rate of change figure I ' as gaussian pyramid ground floor M ' (1), V ' (1), A ' (1) and I ' (1);
4b) respectively described M ' (1), V ' (1), A ' (1) and I ' (1) are descended 2 samplings, image M ' (2), V ' (2), A ' (2) and the I ' (2) that obtains after 2 samplings under using is as the second layer of gaussian pyramid;
4c) respectively described M ' (2), V ' (2), A ' (2) and I ' (2) are descended 2 samplings, image M ' (3), V ' (3), A ' (3) and the I ' (3) that obtains after 2 samplings under using is as the 3rd layer of gaussian pyramid.
5. method according to claim 2, the central neighborhood difference method of the described employing of step 6 wherein, respectively to this pyramid of three floor heights M " (u), this pyramid of three floor heights V of variance entropy diagram " of average entropy diagram (u), this pyramid of three floor heights A of amplitude entropy diagram " (u) and this pyramid of three floor heights I of brightness rate of change entropy diagram " (u), u=1,2,3 carry out difference processing, carry out as follows:
6a) described M " (2), V " (2), A " (2) and I " (2) are adopted cube interpolation method amplify a twice, obtain amplifying the average entropy diagram after the twice
The variance entropy diagram
The amplitude entropy diagram
With brightness rate of change entropy diagram
6b) described M " (3), V " (3), A " (3) and I " (3) are adopted cube interpolation method quadruplication, obtain the average entropy diagram after the quadruplication
The variance entropy diagram
The amplitude entropy diagram
With brightness rate of change entropy diagram
6c) use described M " (1), V " (1), A " (1) and I " (1) respectively with described
With
Subtract each other by pixel, obtain average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
6. method according to claim 2, wherein the standardization operator N () of the described employing of step 7 Itti proposition is to average entropy difference diagram
Variance entropy difference diagram
Amplitude entropy difference diagram
With brightness rate of change entropy difference diagram
V=1,2 carry out standardization respectively, carry out as follows:
7a) with described difference diagram
With
V=1, every width of cloth image is made as in 2
X represents M, V, A and I, v=1,2;
7c) find out
Maximal value W in all pixel values, and calculate all local peaked average w except that W;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011101227220A CN102289657A (en) | 2011-05-12 | 2011-05-12 | Breast X ray image lump detecting system based on visual attention mechanism |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011101227220A CN102289657A (en) | 2011-05-12 | 2011-05-12 | Breast X ray image lump detecting system based on visual attention mechanism |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102289657A true CN102289657A (en) | 2011-12-21 |
Family
ID=45336065
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2011101227220A Pending CN102289657A (en) | 2011-05-12 | 2011-05-12 | Breast X ray image lump detecting system based on visual attention mechanism |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102289657A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102945378A (en) * | 2012-10-23 | 2013-02-27 | 西北工业大学 | Method for detecting potential target regions of remote sensing image on basis of monitoring method |
CN102968786A (en) * | 2012-10-23 | 2013-03-13 | 西北工业大学 | Unsupervised potential target region detection method for remote sensing images |
CN103106671A (en) * | 2013-01-25 | 2013-05-15 | 西北工业大学 | Method for detecting interested region of image based on visual attention mechanism |
CN103455821A (en) * | 2012-05-29 | 2013-12-18 | 北京三星通信技术研究有限公司 | Device and method for analyzing images on basis of BI-RADS (breast imaging reporting and data system) |
CN104217437A (en) * | 2014-09-19 | 2014-12-17 | 西安电子科技大学 | Lesion area partitioning method for prostate KVCT image |
CN104657984A (en) * | 2015-01-28 | 2015-05-27 | 复旦大学 | Automatic extraction method of three-dimensional breast full-volume image regions of interest |
CN106997596A (en) * | 2017-04-01 | 2017-08-01 | 太原理工大学 | A kind of Lung neoplasm dividing method of the LBF movable contour models based on comentropy and joint vector |
CN107958453A (en) * | 2017-12-01 | 2018-04-24 | 深圳蓝韵医学影像有限公司 | Detection method, device and the computer-readable storage medium of galactophore image lesion region |
CN108320272A (en) * | 2018-02-05 | 2018-07-24 | 电子科技大学 | The method that image delusters |
CN115423806A (en) * | 2022-11-03 | 2022-12-02 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
-
2011
- 2011-05-12 CN CN2011101227220A patent/CN102289657A/en active Pending
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103455821A (en) * | 2012-05-29 | 2013-12-18 | 北京三星通信技术研究有限公司 | Device and method for analyzing images on basis of BI-RADS (breast imaging reporting and data system) |
CN103455821B (en) * | 2012-05-29 | 2019-05-28 | 北京三星通信技术研究有限公司 | Image analysis apparatus and method based on BI-RADS |
CN102968786B (en) * | 2012-10-23 | 2015-08-12 | 西北工业大学 | A kind of non-supervisory remote sensing images potential target method for detecting area |
CN102968786A (en) * | 2012-10-23 | 2013-03-13 | 西北工业大学 | Unsupervised potential target region detection method for remote sensing images |
CN102945378A (en) * | 2012-10-23 | 2013-02-27 | 西北工业大学 | Method for detecting potential target regions of remote sensing image on basis of monitoring method |
CN102945378B (en) * | 2012-10-23 | 2015-06-10 | 西北工业大学 | Method for detecting potential target regions of remote sensing image on basis of monitoring method |
CN103106671A (en) * | 2013-01-25 | 2013-05-15 | 西北工业大学 | Method for detecting interested region of image based on visual attention mechanism |
CN104217437A (en) * | 2014-09-19 | 2014-12-17 | 西安电子科技大学 | Lesion area partitioning method for prostate KVCT image |
CN104657984B (en) * | 2015-01-28 | 2018-10-16 | 复旦大学 | The extraction method of three-D ultrasonic mammary gland total volume interesting image regions |
CN104657984A (en) * | 2015-01-28 | 2015-05-27 | 复旦大学 | Automatic extraction method of three-dimensional breast full-volume image regions of interest |
CN106997596A (en) * | 2017-04-01 | 2017-08-01 | 太原理工大学 | A kind of Lung neoplasm dividing method of the LBF movable contour models based on comentropy and joint vector |
CN106997596B (en) * | 2017-04-01 | 2019-08-20 | 太原理工大学 | A kind of Lung neoplasm dividing method of the LBF movable contour model based on comentropy and joint vector |
CN107958453A (en) * | 2017-12-01 | 2018-04-24 | 深圳蓝韵医学影像有限公司 | Detection method, device and the computer-readable storage medium of galactophore image lesion region |
CN108320272A (en) * | 2018-02-05 | 2018-07-24 | 电子科技大学 | The method that image delusters |
CN115423806A (en) * | 2022-11-03 | 2022-12-02 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
CN115423806B (en) * | 2022-11-03 | 2023-03-24 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102289657A (en) | Breast X ray image lump detecting system based on visual attention mechanism | |
CN108053417B (en) | lung segmentation device of 3D U-Net network based on mixed rough segmentation characteristics | |
CN109363698B (en) | Method and device for identifying mammary gland image signs | |
CN107945179A (en) | A kind of good pernicious detection method of Lung neoplasm of the convolutional neural networks of feature based fusion | |
CN108805209A (en) | A kind of Lung neoplasm screening method based on deep learning | |
CN103337074B (en) | A kind of method based on active contour model segmentation mammary gland DCE-MRI focus | |
CN109447065A (en) | A kind of method and device of breast image identification | |
CN102999886B (en) | Image Edge Detector and scale grating grid precision detection system | |
CN110490892A (en) | A kind of Thyroid ultrasound image tubercle automatic positioning recognition methods based on USFaster R-CNN | |
CN108364294A (en) | Abdominal CT images multiple organ dividing method based on super-pixel | |
Palma et al. | Detection of masses and architectural distortions in digital breast tomosynthesis images using fuzzy and a contrario approaches | |
Cortina-Januchs et al. | Detection of pore space in CT soil images using artificial neural networks | |
CN104217213B (en) | A kind of medical image multistage sorting technique based on symmetric theory | |
CN111640128A (en) | Cell image segmentation method based on U-Net network | |
CN108564561A (en) | Pectoralis major region automatic testing method in a kind of molybdenum target image | |
CN104751160A (en) | Mammary gland image processing method based on sparse automatic coding depth network | |
Shao et al. | Application of U-Net and Optimized Clustering in Medical Image Segmentation: A Review. | |
CN107545561A (en) | method for analyzing breast image and electronic device thereof | |
CN104331864B (en) | Based on the processing of the breast image of non-down sampling contourlet and the significant model of vision | |
Soni et al. | CT scan based brain tumor recognition and extraction using Prewitt and morphological dilation | |
Pezeshki et al. | Mass classification of mammograms using fractal dimensions and statistical features | |
Rahmati et al. | A new preprocessing filter for digital mammograms | |
Elmorsy et al. | K3. A region growing liver segmentation method with advanced morphological enhancement | |
Arpana et al. | Feature extraction values for digital mammograms | |
Sreeja et al. | Detection of tumours in digital mammograms using wavelet based adaptive windowing method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111221 |