CN103778600B - Image processing system - Google Patents
Image processing system Download PDFInfo
- Publication number
- CN103778600B CN103778600B CN201210411592.7A CN201210411592A CN103778600B CN 103778600 B CN103778600 B CN 103778600B CN 201210411592 A CN201210411592 A CN 201210411592A CN 103778600 B CN103778600 B CN 103778600B
- Authority
- CN
- China
- Prior art keywords
- feature
- candidate tumor
- tumor profile
- profile
- image processing
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
A kind of image processing system for breast image includes: image collection device, for obtaining two-dimentional breast ultrasound wave image；Lesion detection device includes the interest region of tumor of breast for two-dimentional breast ultrasound wave Image detection M from acquisition, and detects score as it for the marking of each interest region, wherein M > 0；Multi-parameter divider for being partitioned into K candidate tumor profile from each of M interest region using K lesion segmentation algorithm respectively, and is each candidate tumor profile record detection score, wherein K > 0；Feature scoring device, for carrying out evaluation marking to each candidate tumor profile according at least one scheduled feature；Fusion device selects the tumor's profiles being finally partitioned into the M*K candidate tumor profile for the detection score and feature scores according to the M*K candidate tumor profile.
Description
Technical field
This application involves a kind of image processing system for breast ultrasound wave image, more particularly to it is a kind of from breast ultrasound
The interest region of the multiple tumors of breast of wave Image detection is simultaneously given a mark, using multiple lesion segmentation approach of multi-parameter to each
Interest region, which is split, to be handled and gives a mark, and the higher segmentation result of composite index is selected from multiple segmentation results
Image processing techniques.
Background technique
Breast cancer is the second largest killer of women, and early detection is the key that reduce the death rate (40% or more).It is super
Sound wave is used as the supplemental diagnostics test of Mammogram (X-ray) to be used for breast imaging more and more, when breast X-ray is examined
Sensibility may be occurred when reducing or when Mammogram is there are when unacceptable radiation risk by looking into, itself also conduct
First Line imaging technique is used.Therefore, computer-aided diagnosis (CAD) system can help the doctor to lack experience to avoid
Mistaken diagnosis, reduces the quantity of benign lesion biopsy under the premise of not mistaken diagnosis cancer, and reduces the variation of various detections.
In computer assisted breast ultrasound wave diagnostic system, core technology includes lesion detection and segmentation, and tumour
Segmentation is the key that determining tumour is benign or pernicious.The single testing result of existing lesion detection processing output, and tumour
Dividing processing carries out lesion segmentation processing for the single testing result.In the process, detection mistake or incorrect inspection
It surveys result and can all lead to the segmentation result of mistake using unsuitable parameter.Such area of computer aided breast ultrasound wave is examined
Disconnected not powerful enough, the accuracy of segmentation result is also unstable.
Summary of the invention
The purpose of the present invention is to provide a kind of image processing systems for breast ultrasound wave image, from breast ultrasound wave
The interest region of the multiple tumors of breast of Image detection is simultaneously given a mark, using multiple lesion segmentation approach of multi-parameter to each emerging
Interesting region is split processing, and the feature for carrying out multiple features to each tumor's profiles being partitioned into is given a mark, hereafter from more
The image processing techniques that the higher segmentation result of composite index is selected in a segmentation result makes segmentation result not vulnerable to partially not just
The influence of true testing result and/or partitioning parameters improper use, to improve tumor of breast detection and dividing processing result
Stability.
Another object of the present invention is to provide a kind of image processing system for breast ultrasound wave image, from having marked
The breast ultrasound wave image in multiple interest regions divides each interest region using multiple lesion segmentation approach of multi-parameter
Processing is cut, and the feature for carrying out multiple features to each tumor's profiles being partitioned into is given a mark, and selects higher point of characteristic index
The image processing techniques for cutting result makes influence of the segmentation result not vulnerable to partial segmentation parameter improper use, to improve mammary gland
The stability of lesion segmentation processing result.
According to an aspect of the present invention, a kind of image processing system for breast image is provided, comprising: image collection
Device, for obtaining two-dimentional breast ultrasound wave image；Lesion detection device, the two-dimentional breast ultrasound wave for being obtained from image collection device
Image detection M include the interest region (ROI) of tumor of breast, and detect score as it for each ROI marking, wherein M
> 0；Multi-parameter divider, for using the K lesion segmentation algorithm based on different parameter or parameter combination from the M respectively
Each of a ROI is partitioned into K candidate tumor profile, and is partitioned into the candidate for each candidate tumor profile record and swells
The detection score of the ROI of tumor profile, wherein K > 0；Feature scoring device, for according at least one scheduled feature to being partitioned into
Each of M*K candidate tumor profile carry out evaluation marking as its feature scores；Fusion device, for according to the M*K
The detection score and feature scores of a candidate tumor profile select a candidate swollen in the M*K candidate tumor profile
Tumor profile is as the tumor's profiles being finally partitioned into.
The image processing system can further include: combiner, for distinguishing M*K candidate tumor profile being partitioned into
Similarity between calculating, and be higher than in multiple candidate tumor profiles of predetermined value from every group of similarity and remove detection
Score is not highest candidate tumor profile, to obtain N number of candidate tumor profile, wherein N < < M*K；Wherein, feature is beaten
Device is divided to carry out evaluation marking as its feature to each of described N number of candidate tumor profile according at least one scheduled feature
Score, fusion device select a candidate tumor profile according to the detection score and feature scores of N number of candidate tumor profile.
The image processing system can further include: preprocessor, for being held to the two-dimentional breast ultrasound wave image of acquisition
Row pretreatment, the pretreatment include executing Denoising disposal and/or image to the two-dimentional breast ultrasound wave image of acquisition
Enhancing processing；Wherein, multi-parameter divider executes at the lesion segmentation to by pretreated two-dimentional breast ultrasound wave image
Reason.
Deformation site model method, template matching method or Adaboost method can be used to execute the M for lesion detection device
The detection and marking of a ROI comprising tumor of breast.
The different parameter can be different the number of iterations, different ratios or different methods, and described swollen
Tumor dividing method is Level Set Method, figure cutting method, region growing methods or watershed algorithm.
Feature scoring device can be used support vector regression method each according at least one feature calculation in following characteristics
The Jaccard index of candidate tumor profile is as its feature scores: textural characteristics, space characteristics, strength characteristic, contour feature or
Tumoral character.
The contour feature can be grey-scale contrast, intensity contrast or Gestar feature.
The tumoral character can be rear acoustic signature or echo mode feature.
The detection score of each candidate tumor profile and feature scores can be normalized respectively for fusion device, according to following
Formula calculates the composite score of each candidate tumor profile, and selects the highest candidate tumor profile of composite score:
Scorecombined=wds×NDS+wrs×NRS
Wherein, ScorecombinedIt is the composite score of candidate tumor profile, wdsAnd wrsIt is to confer to candidate tumor profile respectively
Detection score and feature scores weight, NDS and NRS are the detection score and feature scores of candidate tumor profile respectively.
Fusion device can be used s-score normalization algorithm, min-max normalization algorithm, Tanh estimator or
Double sigmoid algorithm executes the normalized.
Support vector regression method can be used in feature scoring device, according in following characteristics at least one feature and candidate
The detection score of tumor's profiles calculates the Jaccard index of each candidate tumor profile as its feature scores: textural characteristics, sky
Between feature, strength characteristic, contour feature or tumoral character, wherein fusion device selects in the M*K candidate tumor profile
The highest candidate tumor profile of feature scores is as the tumor's profiles being finally partitioned into.
The different parameter can be different the number of iterations, different ratios, different steps, different denoisings
Processing or different image enchancing methods, and the lesion segmentation approach is Level Set Method, figure cutting method, region growth
Method or watershed algorithm.
According to another aspect of the present invention, a kind of image processing system for breast image is provided, comprising: image collection
Device, for obtaining the two-dimentional breast ultrasound wave image for being labeled with M tumour ROI, wherein M > 0；Multi-parameter divider, for dividing
The K lesion segmentation algorithm based on different parameter or parameter combination is not used to be partitioned into K from each of described M ROI
Candidate tumor profile, wherein K > 0；Feature scoring device, for being waited according at least one scheduled feature to M*K be partitioned into
It selects each of tumor's profiles to carry out evaluation marking as its feature scores, and is selected from the M*K candidate tumor profile
The highest candidate tumor profile of feature scores is selected as the tumor's profiles being finally partitioned into.
The different parameter can be different the number of iterations, different ratios or different steps, and described swollen
Tumor dividing method is Level Set Method, figure cutting method, region growing methods or watershed algorithm.
Feature scoring device can be used support vector regression method each according at least one feature calculation in following characteristics
The Jaccard index of candidate tumor profile is as its feature scores: textural characteristics, space characteristics, strength characteristic, contour feature or
Tumoral character.
The contour feature can be grey-scale contrast, intensity contrast or Gestar feature.The tumoral character can be with
It is rear acoustic signature or echo mode feature.
According to another aspect of the present invention, a kind of image processing method for breast image is provided, comprising: a) obtain two
Tie up breast ultrasound wave image；B) from the two-dimentional breast ultrasound wave Image detection M of the acquisition ROI comprising tumor of breast, and it is
Each ROI marking detects score as it, wherein M > 0；C) K based on different parameter or parameter combination are used to swell respectively
Tumor partitioning algorithm is partitioned into K candidate tumor profile from each of described M ROI, and remembers for each candidate tumor profile
Record is partitioned into the detection score of the ROI of the candidate tumor profile, wherein K > 0；D) according at least one scheduled feature pair
Each of M*K candidate tumor profile being partitioned into carries out evaluation marking as its feature scores；E) it is waited according to described M*K
The detection score and feature scores for selecting tumor's profiles select a candidate tumor wheel in the M*K candidate tumor profile
Exterior feature is as the tumor's profiles being finally partitioned into.
The method can further include: g) to M*K candidate tumor profile being partitioned into calculate separately between phase
Like degree, and being higher than removal detection score in multiple candidate tumor profiles of predetermined value from every group of similarity is not highest time
Tumor's profiles are selected, to obtain N number of candidate tumor profile, wherein N < < M*K；Wherein, in step d), according to it is scheduled extremely
A few feature carries out evaluation marking as its feature scores to each of described N number of candidate tumor profile, and in step
E) in, according to the detection score and feature scores of N number of candidate tumor profile, a candidate tumor profile is selected.
The method can further include: before executing step b), executes pre- place to the two-dimentional breast ultrasound wave image of acquisition
Reason, the pretreatment include executing at Denoising disposal and/or image enhancement to the two-dimentional breast ultrasound wave image of acquisition
Reason；Wherein, it in step c), handles the lesion segmentation is executed by pretreated two-dimentional breast ultrasound wave image.
In step b), deformation site model method, template matching method or Adaboost method can be used to execute the M
The detection and marking of a ROI comprising tumor of breast.
In step c), the different parameter can be different the number of iterations, different ratios or different steps,
And the lesion segmentation approach is Level Set Method, figure cutting method, robs segmentation method or watershed algorithm.
In step d), it can be used support vector regression method each according at least one feature calculation in following characteristics
The Jaccard index of candidate tumor profile is as its feature scores: textural characteristics, space characteristics, strength characteristic, contour feature or
Tumoral character.
The contour feature can be grey-scale contrast, intensity contrast or Gestar feature.The tumoral character can be with
It is rear acoustic signature or echo mode feature.
In step e), the detection score of each candidate tumor profile and feature scores can be normalized respectively, root
The composite score of each candidate tumor profile is calculated according to following formula, and selects the highest candidate tumor profile of composite score:
Scorecombined=wds×NDS+wrs×NRS
Wherein, ScorecombinedIt is the composite score of candidate tumor profile, wdsAnd wrsIt is to confer to candidate tumor profile respectively
Detection score and feature scores weight, NDS and NRS are the detection score and feature scores of candidate tumor profile respectively.
In step e), can be used s-score normalization algorithm, min-max normalization algorithm, Tanh estimator or
Double sigmoid algorithm executes the normalized.
In step d), support vector regression method can be used, according at least one feature and time in following characteristics
The detection score of tumor's profiles is selected to calculate the Jaccard index of each candidate tumor profile as its feature scores: textural characteristics,
Space characteristics, strength characteristic, contour feature or tumoral character.It wherein, can be from the M*K candidate tumor wheel in step e)
Select the highest candidate tumor profile of feature scores as the tumor's profiles being finally partitioned into wide.
The different parameter can be different the number of iterations, different ratios, different steps, different denoisings
Processing or different image enchancing methods, and the lesion segmentation approach is Level Set Method, figure cutting method, robs segmentation method
Or watershed algorithm.
According to another aspect of the present invention, a kind of image processing method for breast image is provided, comprising: a) obtain mark
It is marked with the two-dimentional breast ultrasound wave image of M tumour ROI, wherein M > 0；B) it uses respectively based on different parameter or parameter group
K lesion segmentation algorithm of conjunction is partitioned into K candidate tumor profile from each of described M ROI, wherein K > 0；C) basis
At least one scheduled feature carries out evaluation marking as its feature point to each of the M*K candidate tumor profile being partitioned into
Number；D) select feature scores highest candidate tumor profiles as being finally partitioned into from the M*K candidate tumor profile
Tumor's profiles.
In step b), the different parameter can be different the number of iterations, different ratios or different steps,
And the lesion segmentation approach is Level Set Method, figure cutting method, robs segmentation method or watershed algorithm.
In step c), it can be used support vector regression method each according at least one feature calculation in following characteristics
The Jaccard index of candidate tumor profile is as its feature scores: textural characteristics, space characteristics, strength characteristic, contour feature or
Tumoral character.
The contour feature can be grey-scale contrast, intensity contrast or Gestar feature.The tumoral character can be with
It is rear acoustic signature or echo mode feature.
Detailed description of the invention
By the description carried out with reference to the accompanying drawing, above and other purpose of the invention and feature will become more clear
Chu, in which:
Fig. 1 is the logic diagram and its execution figure for showing the image processing system of an exemplary embodiment of the present invention
As the schematic diagram of processing；
Fig. 2 is the flow chart for showing the image processing method of an exemplary embodiment of the present invention；
The candidate tumor profile that the image processing method that Fig. 3 shows an exemplary embodiment of the present invention is partitioned into；
Fig. 4 shows the flow chart of image processing method in accordance with an alternative illustrative embodiment of the present invention.
Specific embodiment
Hereinafter, with reference to the accompanying drawings to the embodiment that the present invention will be described in detail.
Fig. 1 is the logic diagram and its execution figure for showing the image processing system of an exemplary embodiment of the present invention
As the schematic diagram of processing.
Referring to Fig.1, image processing system include image collection device 100, lesion detection device 110, multi-parameter divider 120,
Feature scoring device 140 and fusion device 150.
Image collection device 100 obtains two-dimentional breast ultrasound wave image (source figure shown in image collection device downside as shown in figure 1
Picture).Image collection device 100 can obtain the two-dimentional breast ultrasound wave image from supersonic imaging device connected to it, can also
To read the two-dimentional breast ultrasound wave image from information storage medium.Preferred embodiment in accordance with the present invention, image procossing system
System further includes preprocessor (not shown), and preprocessor is for executing pretreatment to the two-dimentional breast ultrasound wave image of acquisition, institute
Stating pretreatment includes executing Denoising disposal and/or image enhancement processing to the two-dimentional breast ultrasound wave image of acquisition.It is swollen
110 pairs of tumor detector execute lesion detection by pretreated two-dimentional breast ultrasound wave image.
(or by pretreated) two-dimentional breast ultrasound wave image that lesion detection device 110 is obtained from image collection device 100
It detects M and includes the interest region (ROI) of tumor of breast, and detect score as it for the marking of each interest region, wherein
M > 0.Each ROI is the rectangular region for including practical tumour, these regions can be overlapped.Lesion detection device 110 in Fig. 1
Downside shows the ROI of multiple light collimation mark notes.Deformation site model (DPM) method, template matching can be used in lesion detection device 110
Method or Adaboost method execute the detection and marking of the M ROI comprising tumor of breast, but of the present invention swollen
Tumor detection method is not limited to the above method.
For this M ROI, multi-parameter divider 120 uses the K tumour based on different parameter or parameter combination respectively
Partitioning algorithm is partitioned into K candidate tumor profile from each of described M ROI, and records for each candidate tumor profile
It is partitioned into the detection score of the ROI of the candidate tumor profile, wherein K > 0.The different parameter be (but being not limited to) no
Same the number of iterations, different ratios or different steps, and the lesion segmentation approach can be (but are not limited to) level
Set method, figure cutting method, region growing methods or watershed algorithm.In the embodiment comprising preprocessor, the difference
Parameter further include different Denoising disposals or different image enchancing methods.By this step, image processing system is obtained
The M*K candidate tumor profiles being partitioned into.Processing by the step is shown to obtain on the downside of multi-parameter divider 120 in Fig. 1
Part candidate tumor profile.Wherein, two candidate tumor profiles of the top are closely similar.
An exemplary embodiment of the present invention, in order to save computing resource and operation time, image processing system is also wrapped
Include combiner 130.Combiner 130 to the M*K candidate tumor profile that multi-parameter divider 120 is partitioned into calculate separately each other it
Between similarity, and removal detection score is not most in multiple candidate tumor profiles that every group of similarity is higher than predetermined value
High candidate tumor profile, to obtain N number of candidate tumor profile, wherein N < < M*K.If image processing system does not consider
The factor of operation time and calculation process amount may not include combiner 130.
Feature scoring device 140 be used for according at least one scheduled feature to M*K candidate tumor profile being partitioned into or
Each of N number of candidate tumor profile that person retains after 130 merging treatment of combiner carries out evaluation marking as its feature
Score.An exemplary embodiment of the present invention, feature scoring device 140 uses support vector regression (SVR) method, according to texture
The each candidate tumor profile of at least one feature calculation in feature, space characteristics, strength characteristic, contour feature or tumoral character
Jaccard index as its feature scores.The contour feature is grey-scale contrast, intensity contrast or Gestar feature.
The tumoral character is rear acoustic signature or echo mode feature.
Fusion device 150 is used for detection score and feature scores according to the M*K or N number of candidate tumor profile, from institute
State selects a candidate tumor profile as the tumor's profiles being finally partitioned into M*K or N number of candidate tumor profile.Specifically
Ground, fusion device use s-score normalization algorithm, min-max normalization algorithm, Tanh estimator or double first
The detection score of each candidate tumor profile and feature scores are normalized sigmoid algorithm respectively, according to the following formula
The composite score of each candidate tumor profile is calculated, and selects the highest candidate tumor profile of composite score:
Scorecombined=wds×NDS+wrs×NRS
Wherein, ScorecombinedIt is the composite score of candidate tumor profile, wdsAnd wrsIt is to confer to candidate tumor profile respectively
Detection score and feature scores weight, NDS and NRS are the detection score and feature scores of candidate tumor profile respectively.
Optional exemplary embodiment according to the present invention, feature scoring device 140 uses SVR method, according to textural characteristics, sky
Between at least one feature and candidate tumor profile in feature, strength characteristic, contour feature or tumoral character detection score
The Jaccard index of each candidate tumor profile is calculated as its feature scores；In this case, fusion device 150 is from described
Select the highest candidate tumor profile of feature scores as the tumor's profiles being finally partitioned into M*K candidate tumor profile.
Image processing system according to the present invention detects multiple ROI from two-dimentional breast ultrasound wave image, the multiple
The basis that ROI is divided as mammary gland；The present invention also uses multiple lesion segmentation approach based on multi-parameter to the multiple ROI
Lesion segmentation processing is carried out, obtains greater number of candidate tumor profile, and multiple spies are carried out to each candidate tumor profile
The evaluation of sign is given a mark；As a result, image processing system can comprehensive detection processing in marking and evaluation processing feature marking, from
And the comprehensive highest candidate tumor profile lesion segmentation result final as its of giving a mark is selected according to certain standard.This place
Reason mode it is more traditional only detect a tumor region, carry out a kind of lesion segmentation processing only for this tumor region
Mode is compared, and can more ensure to export the Stability and veracity of result, without vulnerable to the single error occurred in any link
Influence.
In some cases, before by computer aided system execution processing, doctor may be super in the mammary gland of shooting in advance
Sound wave image is marking region existing for one or more tumours.At this point, image processing system can be by two-dimentional breast ultrasound wave shadow
As and one or more tumours for marking in advance existing for region as input, wherein respectively by mark in advance one or
Region existing for multiple tumours is handled as ROI.In accordance with an alternative illustrative embodiment of the present invention, at image of the invention
Reason system includes image collection device 200, multi-parameter divider 120 and feature scoring device 140.Image collection device 100 can obtain
It is labeled with the two-dimentional breast ultrasound wave image of M tumour ROI, wherein M > 0.Image collection device 100 can be as needed, to mark
Tumour ROI carry out processing (ROI is such as adjusted to rectangle) appropriate.Multi-parameter divider 120 is for respectively using being based on
K lesion segmentation algorithm of different parameters or parameter combination is partitioned into K candidate tumor wheel from each of described M ROI
It is wide, wherein K > 0, to be partitioned into M*K candidate tumor profile altogether.Feature scoring device 140 is as previously mentioned, for according to predetermined
At least one feature evaluation marking is carried out as its feature scores to each of the M*K candidate tumor profile being partitioned into；
In addition, selected from M*K candidate tumor profile described in feature scoring device 140 the highest candidate tumor profile of feature scores as
The tumor's profiles being finally partitioned into.
The image processing method of an exemplary embodiment of the present invention is described in detail hereinafter with reference to Fig. 2-Fig. 4.
Fig. 2 is the flow chart for showing the image processing method of an exemplary embodiment of the present invention.
Referring to Fig. 2, in step S100, image processing system obtains two-dimentional breast ultrasound wave image.Image processing system can
The two-dimentional breast ultrasound wave image is obtained from supersonic imaging device connected to it, can also be read from information storage medium
The two dimension breast ultrasound wave image.
The ultrasonograph of supersonic imaging device shooting often contains such as spot " noise ".It is according to the present invention
Preferred embodiment, in order to obtain preferable image processing effect, in step S105, two-dimentional mammary gland of the image processing system to acquisition
Ultrasonograph executes pretreatment, the pretreatment include the two-dimentional breast ultrasound wave image of acquisition is executed Denoising disposal with
And/or person's image enhancement processing.But step S105 is optional step, rather than the step of having to carry out.
Hereafter, in step S110, from (or by pretreated) two-dimentional breast ultrasound wave shadow of image processing system acquisition
Score is detected as it as M ROI comprising tumor of breast of detection, and for each ROI marking, wherein M > 0.Shape can be used
Become site model (DPM) method, template matching method or AdaBoost method and executes the M ROI's comprising tumor of breast
Detection and marking.
In step S120, image processing system uses the K lesion segmentation based on different parameter or parameter combination respectively
Algorithm is partitioned into K candidate tumor profile from each of described M ROI, and is the record segmentation of each candidate tumor profile
The detection score of the ROI of the candidate tumor profile out, wherein K > 0.By the processing of step S120, M*K time will be obtained
Select tumor's profiles.Fig. 3 shows the ROI detected from the two-dimentional breast ultrasound wave image in left side and carries out 3 lesion segmentations
The 3 candidate tumor profiles (right side) that algorithm is split.
Here, the K lesion segmentation algorithm is the kinds of tumors partitioning algorithm using different parameter or parameter combination.
The different parameter is different the number of iterations, different ratios or different steps.In the pretreatment for executing step S105
In the case where, the different parameter can further include different Denoising disposals or different image enchancing methods.The tumour
Dividing method is Level Set Method, figure cutting method, region growing methods or watershed algorithm.
Hereafter, in step S130, image processing system to M*K candidate tumor profile being partitioned into calculate separately each other it
Between similarity, and removal detection score is not most in multiple candidate tumor profiles that every group of similarity is higher than predetermined value
High candidate tumor profile, to obtain N number of candidate tumor profile, wherein N < < M*K.
If image processing system does not consider that operation time is too long and the excessive influence of calculation process amount, can not execute
S130。
In step S140, image processing system is according at least one scheduled feature in N number of candidate tumor profile
Each of carry out evaluation marking as its feature scores.It, will be to the M*K being partitioned into the case where being not carried out step S130
Each of candidate tumor profile carries out evaluation marking as its feature scores.
An exemplary embodiment of the present invention, image processing system using SVR method according in following characteristics at least
The Jaccard index of each candidate tumor profile of one feature calculation is as its feature scores: textural characteristics, space characteristics, strong
Spend feature, contour feature or tumoral character.Wherein, the contour feature is grey-scale contrast, intensity contrast or Gestar special
Sign, the tumoral character is rear acoustic signature or echo mode feature.But the present invention is not limited to use support vector regression side
Other characteristics of image method for evaluating similarity also can be used in method, and present invention is also not necessarily limited to feature listed here
Carry out evaluation marking.
Hereafter, in step S150, image processing system is according to the detection score and feature of N number of candidate tumor profile point
Number selects a candidate tumor profile as the final tumor's profiles being partitioned into from N number of candidate tumor profile.
Using multiple candidate tumor profiles and its detection score and feature scores as input, according to preference and stress, it can
The overall merit of candidate tumor profile is executed to use a variety of fusion methods, to select the highest candidate tumor of overall merit
Profile is as the tumor's profiles being finally partitioned into.
An exemplary embodiment of the present invention, image processing system is by the detection score of each candidate tumor profile and spy
Sign score is normalized respectively, calculates the composite score of each candidate tumor profile according to the following formula, and select to integrate
The highest candidate tumor profile of score:
Scorecombined=wds×NDS+wrs×NRS
Wherein, ScorecombinedIt is the composite score of candidate tumor profile, wdsAnd wrsIt is to confer to candidate tumor profile respectively
Detection score and feature scores weight, NDS and NRS are the detection score and feature scores of candidate tumor profile respectively.
Here it is possible to using s-score normalization algorithm, min-max normalization algorithm, Tanh estimator or
Double sigmoid algorithm executes the normalized of the score.
According to an alternative embodiment of the invention, in step S140, image processing system uses support vector regression method, root
Each candidate tumor profile is calculated according to the detection score of at least one feature and candidate tumor profile in following characteristics
Jaccard index is as its feature scores: textural characteristics, space characteristics, strength characteristic, contour feature or tumoral character.In step
Rapid S150, image processing system select the highest candidate tumor profile of feature scores in the M*K candidate tumor profile
As the tumor's profiles being finally partitioned into.
Another alternative embodiment according to the present invention, the final tumor's profiles being partitioned into of image processing system output.
Fig. 4 shows the flow chart of image processing method in accordance with an alternative illustrative embodiment of the present invention.
Referring to Fig. 4, in step S200, image processing system obtain one of two-dimentional breast ultrasound wave image and mark or
Multiple regions.Image processing system is using each region of mark as ROI.Here, suppose that there is M ROI.
In step S220, image processing system uses the K lesion segmentation based on different parameter or parameter combination respectively
Algorithm is partitioned into K candidate tumor profile from each of described M ROI, wherein K > 0.It can be such as step S120 execution in Fig. 2
Dividing processing described here, unlike, there is no detection scores in the situation of input tab area.
In step S240, image processing system is according at least one scheduled feature to M*K candidate tumor being partitioned into
Each of profile carries out evaluation marking as its feature scores, and selects feature from the M*K candidate tumor profile
The highest candidate tumor profile of score is as the tumor's profiles being finally partitioned into.
According to the description to exemplary embodiment of the present invention as can be seen that image processing system and method for the invention can
Multi-parameter segmentation is executed to each ROI detected to obtain multiple candidate tumor profiles, and to the multiple candidate swollen
Tumor profile carries out feature marking, so that the score of comprehensive detection and feature scores choose optimal candidate tumor profile as final
Segmentation result, this processing mode it is more traditional only detect a tumor region, only for this tumor region carry out
A kind of mode of lesion segmentation processing is compared, and can more ensure to export the Stability and veracity of result, without vulnerable to any ring
The influence of the single error occurred in section.
Further, it is also possible to which the breast image for being labelled with ROI to doctor executes multi-parameter lesion segmentation, and according to each
The feature marking for the candidate tumor profile being partitioned into chooses optimal candidate tumor profile as final segmentation result, improves
The stability of the result of lesion segmentation makes segmentation result not vulnerable to one or partial parameters improper use or the shadow of single error
It rings.
Although show and describing the present invention with reference to preferred embodiment, it will be understood by those skilled in the art that not
In the case where being detached from the spirit and scope of the present invention that are defined by the claims, these embodiments can be carry out various modifications and
Transformation.
Claims (17)
1. a kind of image processing system for breast image, comprising:
Image collection device, for obtaining two-dimentional breast ultrasound wave image；
Lesion detection device, two-dimentional breast ultrasound wave Image detection M for obtaining from image collection device include tumor of breast
Interest region, and score is detected as it for the marking of each interest region, wherein M > 0；
Multi-parameter divider, for using the K lesion segmentation algorithm based on different parameter or parameter combination from the M respectively
Each of a interest region is partitioned into K candidate tumor profile, and is described in each candidate tumor profile record is partitioned into
The detection score in the interest region of candidate tumor profile, wherein K > 0；
Feature scoring device, for according at least one scheduled feature to each of the M*K candidate tumor profile being partitioned into
Evaluation marking is carried out as its feature scores；
Fusion device, for being waited according to the detection score in the interest region where the M*K candidate tumor profile and described M*K
The feature scores for selecting tumor's profiles select a candidate tumor profile as final in the M*K candidate tumor profile
The tumor's profiles being partitioned into.
2. image processing system as described in claim 1, further includes:
Combiner, for the similarity between being calculated separately to M*K candidate tumor profile being partitioned into, and from every group
Similarity is higher than the detection point that the interest region where candidate tumor profile is removed in multiple candidate tumor profiles of predetermined value
Number is not highest candidate tumor profile, to obtain N number of candidate tumor profile, wherein N < < M*K；
Wherein, feature scoring device comments each of described N number of candidate tumor profile according at least one scheduled feature
Valence marking is used as its feature scores, fusion device according to the detection score in the interest region where N number of candidate tumor profile and
The feature scores of N number of candidate tumor profile, select a candidate tumor profile.
3. image processing system as described in claim 1, further includes:
Preprocessor, for executing pretreatment to the two-dimentional breast ultrasound wave image of acquisition, the pretreatment includes to acquisition
Two-dimentional breast ultrasound wave image executes Denoising disposal and/or image enhancement processing；
Wherein, multi-parameter divider executes the lesion segmentation processing to by pretreated two-dimentional breast ultrasound wave image.
4. image processing system as described in claim 1, wherein lesion detection device uses deformation site model method, template
Matching process or Adaboost method execute the detection and marking in the M interest regions comprising tumor of breast.
5. image processing system as described in claim 1, wherein the different parameter is different the number of iterations, difference
Ratio or different methods, and the lesion segmentation approach be Level Set Method, figure cutting method, region growing methods or
Watershed algorithm.
6. image processing system as claimed in claim 4, wherein feature scoring device using support vector regression method according to
The Jaccard index of each candidate tumor profile of at least one feature calculation in lower feature is as its feature scores: texture is special
Sign, space characteristics, strength characteristic, contour feature or tumoral character.
7. image processing system as claimed in claim 6, wherein the contour feature is grey-scale contrast, intensity contrast
Or Gestar feature.
8. image processing system as claimed in claim 6, wherein the tumoral character is that rear acoustic signature or echo mode are special
Sign.
9. image processing system as described in claim 1, wherein fusion device is by the region of interest where each candidate tumor profile
The detection score in domain and the feature scores of each candidate tumor profile are normalized respectively, calculate each time according to the following formula
The composite score of tumor's profiles is selected, and selects the highest candidate tumor profile of composite score:
Scorecombined=wds×NDS+wrs×NRS
Wherein, ScorecombinedIt is the composite score of candidate tumor profile, wdsAnd wrsIt is to confer to where candidate tumor profile respectively
Interest region detection score and candidate tumor profile feature scores weight, NDS and NRS are candidate tumor profile respectively
The feature scores of the detection score and candidate tumor profile in the interest region at place.
10. image processing system as claimed in claim 9, wherein fusion device uses s-score normalization algorithm, min-max
Normalization algorithm, Tanh estimator or double sigmoid algorithm execute the normalized.
11. image processing system as claimed in claim 4,
Wherein, feature scoring device use support vector regression method, according in following characteristics at least one feature and candidate
The detection score in the interest region where tumor's profiles calculates the Jaccard index of each candidate tumor profile as its feature point
Number: textural characteristics, space characteristics, strength characteristic, contour feature or tumoral character,
Wherein, fusion device selected in the M*K candidate tumor profile the highest candidate tumor profile of feature scores as
The tumor's profiles being finally partitioned into.
12. image processing system as claimed in claim 3, wherein the different parameter is different the number of iterations, difference
Ratio, different steps, different Denoising disposals or different image enchancing methods, and the lesion segmentation approach is
Level Set Method, figure cutting method, region growing methods or watershed algorithm.
13. a kind of image processing system for breast image, comprising:
Image collection device, for obtaining the two-dimentional breast ultrasound wave image for being labeled with M tumour interest region, wherein M > 0；
Multi-parameter divider, for using the K lesion segmentation algorithm based on different parameter or parameter combination from the M respectively
Each of a interest region is partitioned into K candidate tumor profile, wherein K > 0；
Feature scoring device, for according at least one scheduled feature to each of the M*K candidate tumor profile being partitioned into
Evaluation marking is carried out as its feature scores, and selects the highest time of feature scores from the M*K candidate tumor profile
Select tumor's profiles as the tumor's profiles being finally partitioned into.
14. image processing system as claimed in claim 13, wherein the different parameter is different the number of iterations, no
Same ratio or different steps, and the lesion segmentation approach is Level Set Method, figure cutting method, region growing methods
Or watershed algorithm.
15. image processing system as claimed in claim 13, wherein feature scoring device using support vector regression method according to
The Jaccard index of each candidate tumor profile of at least one feature calculation in following characteristics is as its feature scores: texture
Feature, space characteristics, strength characteristic, contour feature or tumoral character.
16. image processing system as claimed in claim 15, wherein the contour feature is grey-scale contrast, intensity contrast
Degree or Gestar feature.
17. image processing system as claimed in claim 15, wherein the tumoral character is that rear acoustic signature or echo mode are special
Sign.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210411592.7A CN103778600B (en) | 2012-10-25 | 2012-10-25 | Image processing system |
| KR1020130021930A KR102128325B1 (en) | 2012-10-25 | 2013-02-28 | Image Processing System |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210411592.7A CN103778600B (en) | 2012-10-25 | 2012-10-25 | Image processing system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103778600A CN103778600A (en) | 2014-05-07 |
| CN103778600B true CN103778600B (en) | 2019-02-19 |
Family
ID=50570799
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210411592.7A Active CN103778600B (en) | 2012-10-25 | 2012-10-25 | Image processing system |
Country Status (2)
| Country | Link |
|---|---|
| KR (1) | KR102128325B1 (en) |
| CN (1) | CN103778600B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102315574B1 (en) * | 2014-12-03 | 2021-10-20 | 삼성전자주식회사 | Apparatus and method for classification of data, apparatus and method for segmentation of region of interest |
| KR102333479B1 (en) | 2015-03-31 | 2021-12-03 | 한국전자통신연구원 | Method and apparatus for breast lesion diagnosis |
| JP2018517207A (en) * | 2015-05-18 | 2018-06-28 | コーニンクレッカ フィリップス エヌ ヴェＫｏｎｉｎｋｌｉｊｋｅ Ｐｈｉｌｉｐｓ Ｎ．Ｖ． | Self-recognizing image segmentation method and system |
| WO2017092615A1 (en) | 2015-11-30 | 2017-06-08 | 上海联影医疗科技有限公司 | Computer aided diagnosis system and method |
| CN105488800A (en) * | 2015-11-30 | 2016-04-13 | 上海联影医疗科技有限公司 | Feature extracting method and computer-aided diagnosis method and system |
| CN106683137B (en) * | 2017-01-11 | 2019-12-31 | 中国矿业大学 | Artificial mark based monocular and multiobjective identification and positioning method |
| CN110569837B (en) * | 2018-08-31 | 2021-06-04 | 创新先进技术有限公司 | Method and device for optimizing damage detection result |
| WO2021087687A1 (en) * | 2019-11-04 | 2021-05-14 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic image analyzing method, ultrasonic imaging system and computer storage medium |
| CN111554383B (en) * | 2020-04-24 | 2023-09-05 | 浙江杜比医疗科技有限公司 | Neural network for breast tumor detection and detection system thereof |
| KR102667231B1 (en) * | 2020-12-21 | 2024-05-20 | 주식회사 인피니트헬스케어 | System and method for contouring a set of medical images based on deep learning algorighm and anatomical characteristics |
| CN115089292A (en) * | 2022-06-09 | 2022-09-23 | 北京三春晖医疗器械有限公司 | Needle distribution control system and method and tumor therapeutic apparatus |
| CN115641430B (en) * | 2022-10-12 | 2024-01-30 | 阿里巴巴（中国）有限公司 | Method, device, medium and computer equipment for determining interest surface |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887514A (en) * | 2009-05-13 | 2010-11-17 | 联咏科技股份有限公司 | Human face detecting device and human face detecting method thereof |
| CN102113335A (en) * | 2009-06-09 | 2011-06-29 | 索尼公司 | Image processing apparatus, method and program |
| CN102467741A (en) * | 2010-11-18 | 2012-05-23 | 索尼公司 | Method and device for detecting spot area in image |
| CN102479327A (en) * | 2010-11-23 | 2012-05-30 | 现代摩比斯株式会社 | Object recognition system, and obstacle recognition system and method for automobile |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100609249B1 (en) * | 1997-02-06 | 2006-11-30 | 코닌클리케 필립스 일렉트로닉스 엔.브이. | Image object tracking method and corresponding system |
| US20050129297A1 (en) * | 2003-12-15 | 2005-06-16 | Kamath Vidya P. | Classification of breast lesion method and system |
| KR100863745B1 (en) * | 2005-08-18 | 2008-10-16 | 주식회사 메디슨 | Apparatus and method for processing an ultrasound image in an ultrasound image diagnostic system |
| GB0602739D0 (en) * | 2006-02-10 | 2006-03-22 | Ccbr As | Breast tissue density measure |
| US8218850B2 (en) * | 2006-02-10 | 2012-07-10 | Synarc Inc. | Breast tissue density measure |
| IL182332A (en) * | 2006-03-31 | 2013-04-30 | Given Imaging Ltd | System and method for assessing a patient condition |
| US20100158332A1 (en) * | 2008-12-22 | 2010-06-24 | Dan Rico | Method and system of automated detection of lesions in medical images |
| KR101439412B1 (en) * | 2010-11-22 | 2014-09-11 | 한국전자통신연구원 | Method of segmenting lesions in images |
-
2012
- 2012-10-25 CN CN201210411592.7A patent/CN103778600B/en active Active
-
2013
- 2013-02-28 KR KR1020130021930A patent/KR102128325B1/en active IP Right Grant
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887514A (en) * | 2009-05-13 | 2010-11-17 | 联咏科技股份有限公司 | Human face detecting device and human face detecting method thereof |
| CN102113335A (en) * | 2009-06-09 | 2011-06-29 | 索尼公司 | Image processing apparatus, method and program |
| CN102467741A (en) * | 2010-11-18 | 2012-05-23 | 索尼公司 | Method and device for detecting spot area in image |
| CN102479327A (en) * | 2010-11-23 | 2012-05-30 | 现代摩比斯株式会社 | Object recognition system, and obstacle recognition system and method for automobile |
Also Published As
| Publication number | Publication date |
|---|---|
| KR20140057130A (en) | 2014-05-12 |
| KR102128325B1 (en) | 2020-07-01 |
| CN103778600A (en) | 2014-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103778600B (en) | Image processing system | |
| Ghatwary et al. | Early esophageal adenocarcinoma detection using deep learning methods | |
| Kavur et al. | Comparison of semi-automatic and deep learning-based automatic methods for liver segmentation in living liver transplant donors | |
| Kim et al. | Machine-learning-based automatic identification of fetal abdominal circumference from ultrasound images | |
| US9996918B2 (en) | Method for distinguishing pulmonary artery and pulmonary vein, and method for quantifying blood vessels using same | |
| Zheng et al. | Machine learning based vesselness measurement for coronary artery segmentation in cardiac CT volumes | |
| Wang et al. | Vessel segmentation using implicit model-guided level sets | |
| CN104657984A (en) | Automatic extraction method of three-dimensional breast full-volume image regions of interest | |
| CN109478327B (en) | Method for automatic detection of systemic arteries in Computed Tomography Angiography (CTA) of arbitrary field of view | |
| US9109391B2 (en) | Method and branching determination device for determining a branching point within a hollow organ | |
| Rueda et al. | Feature-based fuzzy connectedness segmentation of ultrasound images with an object completion step | |
| KR20150027637A (en) | Method and Apparatus for registering medical images | |
| US10521906B2 (en) | Method and image processor for evaluating a contrast agent-enhanced magnetic resonance slice image of a heart | |
| CN107274399A (en) | A kind of Lung neoplasm dividing method based on Hession matrixes and 3D shape index | |
| Abbas et al. | Combined spline and B-spline for an improved automatic skin lesion segmentation in dermoscopic images using optimal color channel | |
| Sakellarios et al. | Novel methodology for 3D reconstruction of carotid arteries and plaque characterization based upon magnetic resonance imaging carotid angiography data | |
| JP6862147B2 (en) | Image processing device, operation method of image processing device, image processing system | |
| CN111899850A (en) | Medical image information processing method, display method and readable storage medium | |
| Divyashree et al. | Breast cancer mass detection in mammograms using gray difference weight and mser detector | |
| CN102855483B (en) | Handle the method and apparatus and breast cancer diagnosis apparatus of ultrasonoscopy | |
| CN104021391B (en) | Handle the method and apparatus and breast cancer diagnosis apparatus of ultrasound image | |
| WO2023156290A1 (en) | Method and system for computer aided diagnosis based on morphological characteristics extracted from 3-dimensional medical images | |
| Affane et al. | Robust deep 3-D architectures based on vascular patterns for liver vessel segmentation | |
| US20070121787A1 (en) | System and method for airway detection | |
| US20070086637A1 (en) | Distance Transform Based Vessel Detection for Nodule Segmentation and Analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |