CN106846317A - A kind of feature based extracts the method for retrieving medicine image with Similarity matching - Google Patents
A kind of feature based extracts the method for retrieving medicine image with Similarity matching Download PDFInfo
- Publication number
- CN106846317A CN106846317A CN201710106121.8A CN201710106121A CN106846317A CN 106846317 A CN106846317 A CN 106846317A CN 201710106121 A CN201710106121 A CN 201710106121A CN 106846317 A CN106846317 A CN 106846317A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- target
- images
- value
- 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.)
- Granted
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- 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
- G06T2207/10072—Tomographic images
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- 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
- G06T2207/10132—Ultrasound image
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Image Analysis (AREA)
Abstract
The method for retrieving medicine image with Similarity matching is extracted the invention discloses a kind of feature based, in the flow of radiotherapy planning, image registration is a critically important committed step.The application purpose of image registration is, in order to find one and the most suitable template image of target image, optimal registration result be can obtain after matched computing, to be used for clinical Target delineations, organ dose's simulation or treatment in radiotherapy planning.Therefore, it is considerable for how searching most suitable template image.The morphological feature of 10 and morphologic correlation is passed through in the method, with reference to history information, and give these features different differentiation weights, finally select 10 groups of most close images and relevant information, it is supplied to doctor to select, through considering the lower most suitable template image of selection.Ensure with the similarity of target image by template image after this method computing, can be provided, can so improve the applicability and accuracy of image registration computing in radiotherapy planning.
Description
Technical field
The present invention relates to the registering sample extraction method of the Target delineations of radiotherapy planning, and in particular to a kind of feature based is carried
Take the method for retrieving medicine image with Similarity matching.
Background technology
With continuing to develop for computer science and information technology, medical imaging technology is also developed rapidly, various
New imaging device is continued to bring out, such as computer tomography (CT), digital subtraction angiography (DSA), single photon hair tomography
Imaging (SPECT), magnetic resonance imaging (MRI) positron emission tomography (PET) etc..Various imaging techniques and inspection method are all
There are its advantage and deficiency, not a kind of imaging technique goes for the inspection and medical diagnosis on disease to all organs of human body
It is not that a kind of imaging technique can replace another imaging technique, but complements each other, is complementary to one another.It is correct in order to improve diagnosis
Rate is, it is necessary to comprehensively utilize the various image informations of patient.One obvious development trend of current medical image, is using letter
Breath integration technology, plurality of medical image is combined, the characteristics of make full use of different medical images, on piece image simultaneously
Multi-aspect information of the expression from human body, makes many situations such as structure, the function of inside of human body be reflected by image,
So as to more intuitively provide the information such as human dissection, physiology and pathology.Many image information fusions are realized, it is topmost to be exactly
Complete image registration, i.e. multiple image and reach corresponding to completely on geometric position and anatomical position in the spatial domain.
In radiotherapy field, image registration is also considerable problem.Traditional tumour radiation therapy process, is in treatment
Before beginning, the positioning CT based on patient by doctor carrys out Target delineations and jeopardizes organ to generate radiotherapy planning, Ran Hou manually
Keep radiotherapy planning constant in subsequent therapeutic process, some interval procedures are carried out to patient.Such Therapeutic mode is not examined
The anatomical structure for considering patient in therapeutic process changes, such as the change of gross tumor volume and position, the change of patient body profile,
The change of stomach and intestine expanded state and jeopardize change of organ site etc. around causing, cause the dosage of the actual receiving of patient inclined
From the prescribed dose of doctor, and then cause the decline of tumor control rate and the increase of Normal Tissue Complication probability.
Registration is applied in radiotherapy planning, seeks to find one and the most suitable template image of target image, matched
Optimal registration result can be obtained after computing, is used for clinical Target delineations, organ dose's simulation or treatment.Therefore, how to search
Rope is applied in the registration Algorithm of radiotherapy planning to most suitable template image, is considerable.Registering operation result is more accurate
Really, follow-up Target delineations or organ dose's simulation or clinical treatment can be made more accurate.
In recent years, with the development of science and technology image registration application has been gradually introduced in radiotherapy field, positive applies
In radiotherapy field.At present in the development of medical figure registration.Theory based on DEMONS is come the image registration mode for improveing
One of current main flow.No matter but using which kind of image registration mode, looking for most close sample to complete registering computing, absolutely
To being the method got twice the result with half the effort.
The content of the invention
Not enough present in conventional images registration technique in order to overcome, the present invention provides a kind of feature based and extracts and similar
The method for retrieving medicine image of matching, when making clinic in drawing target outline and jeopardizing organ, application image registration technique can be processed
In it is faster more accurate, meet clinical needs well.
To achieve the above object, the present invention provides the medical image retrieval side of a kind of feature based extraction and Similarity matching
Method, comprises the following steps：
Step 1, the image and medical history data that read patient, the data include image information and text information；
Step 2, the data obtained by step 1 is pre-processed, be divided into image preprocessing and pre-processed with word；
Step 2.1, image preprocessing：Original image is made to turn shelves, normalization, segmentation, target information is extracted；The target
Information includes target image and target image profile；Wherein, if original image is 2DDICOM images, directly carry out turning shelves；If
Original image is 3D DICOM images, then 3DDICOM images first are converted into 2D DICOM image sets, then carries out turning shelves；
Step 2.2, word pretreatment：In by the text information of patient, patient gender, age, disease information, disease are obtained
Cure the disease position, pathology and image report, previously whether done radiotherapy and whether had related complication；
Step 3, from the pretreated image of step 2.1 or image sets, extract required for 10 features, set up the mesh
The image ID of logo image；
10 features are as follows respectively：
Feature 1：2D image sets quantity or frame number；
Feature 2：Image outline longitudinal direction most major axis；
Feature 3：The horizontal most major axis of image outline；
Feature 4：Take at the bounding box longitudinal directions 1/4 of image outline, the most major axis of image outline；
Feature 5：Take at the bounding box longitudinal directions 1/2 of image outline, the most major axis of image outline；
Feature 6：Take at the bounding box longitudinal directions 3/4 of image outline, the most major axis of image outline；
Feature 7：Take at the bounding box horizontal 1/4 of image outline, the most major axis of image outline；
Feature 8：Take at the bounding box horizontal 1/2 of image outline, the most major axis of image outline；
Feature 9：Take at the bounding box horizontal 3/4 of image outline, the most major axis of image outline；
Feature 10：Image volume or area；
Step 4, from the pretreated text information of step 2.2, first pass through preliminary screening, find out identical disease, phase
With the image sets of therapentic part；
All images in step 5, the image sets for extracting the image ID and step 4 of target image obtained by step 3
ID is compared, and the similarity of feature 1 is compared first, looks for similar image sets, incongruent, is excluded；
Step 6, will be screened via step 5 after image sets, the similarity of feature 2-10 is compared with target image ID again,
Similar image sets are looked for, it is incongruent, exclude；
Step 7, the image sets screened via step 6 calculate image similarity index with target image group respectively
(SSIM), compare numerical value more convergence 1, then it represents that more close, extract most close preceding 10 groups of images, it is incongruent, exclude；
Step 8,10 groups of most close images are shown in hospital system end, for clinician's selection, terminated.
As a further improvement on the present invention, described image information includes CT images, pyramidal CT image, ultrasonoscopy, MRI
Image, PET image and X-Ray images；The text information includes patient's basic document, related history, related complication, disease
Type, disease treatment position, pathology and diagnostic imaging report information.
As a further improvement on the present invention, the step 2.1 includes：
Step 2.1.1, differentiate that image be 2D DICOM images or 3D DICOM images, it is of the invention at least wherein it
A kind of image；
If step 2.1.2,3D DICOM images, then transferring files are 2D DICOM sequence image groups；
Step 2.1.3,2D DICOM images or image sets are converted into .bmp or .jpeg forms；
Step 2.1.4, image is done histogram equalization, calculation step is as follows：
A, pending image statisticses its histogram to giving, obtain：
Pr(rj)=nj/ N, j=0,1 ..., L-1 (1)
The histogram that b, basis are counted is converted using cumulative distribution function；
C, replace old gray scale with new gray scale, obtain Sk；
Wherein N is the sum of pixel in piece image；njIt is the pixel of j-th stage gray scale；Pr(rj) represent original image gray scale
The probability distribution that level occurs；rkIt is k-th gray level；T(rk) it is to set up corresponding between input picture and output image gray level
Relation, i.e., the probability that new gray level occurs, repositions Cumulative Distribution Function Sk；
Step 2.1.5, self-adaption binaryzation image is done to the operation result of step 2.1.4 using Otsu methods, obtain base
This image outline edge；
After step 2.1.6, step 2.1.5 are completed, it will tentatively separate target image elementary contour；
Step 2.1.7, the image elementary contour according to step 2.1.6, finding out can encase the minimum square of image
Bounding box, then do an expansion algorithm and erosion algorithm to image, can obtain target image integrity profile edge.
As a further improvement on the present invention, in step 2.1.5：
Otsu methods are called Da-Jin algorithm, and though its histogram of image in calculating process whether there is it is obvious bimodal, can
Obtain remembering that f (i, j) is the gray value at MxN images (i, j) points；
Assuming that f (i, j) value [0, m-1], note p (k) is the frequency of gray value k, then have：
Assuming that being that the target that Threshold segmentation goes out is respectively with background with gray value t：F (i, j)≤t } and { f (i, j)>T },
Then target part ratio：ω0(t)=∑0≤i≤tp(i) (4)
Target part is counted：N0(t)=MN ∑s0≤i≤tp(i) (5)
Background parts ratio：ω1(t)=∑t≤i≤m-1p(i) (6)
Background parts are counted：N1(t)=MN ∑st≤i≤m-1p(i) (7)
Target mean：μ0(t)=∑0≤i≤tip(i)/ω0(t) (8)
Background mean value：μ1(t)=∑t≤i≤m-1ip(i)/ω1(t) (9)
Grand mean：μ=ω0(t)μ0(t)+ω1(t)μ1(t) (10)
Da-Jin algorithm points out to ask the formula of image optimal threshold g be：
It is actually inter-class variance value in the bracket of formula the right, target and background two parts that threshold value g is partitioned into are constituted
Entire image, and desired value μ0T (), probability is ω1(t), background value μ1T (), probability is ω0T (), grand mean is μ, root
The formula is obtained final product according to the definition of variance.
As a further improvement on the present invention, the screening threshold value in the step 5 is 95% or 90% or 85% or 80%
Or 75%.
As a further improvement on the present invention, the screening threshold value in the step 6 is 95% or 90% or 85% or 80%
Or 75%.
As a further improvement on the present invention, in step 7：
Give two images and be respectively defined as x and y, two structural similarities of image can be obtained in such a way：
Wherein μxIt is the average value of x, μyIt is the average value of y,It is the difference of x,It is the variance of y, σxyIt is the association side of x and y
Difference；C1=(K1L)2, C2=(K2L)2It is the constant for maintaining stabilization, L is the dynamic range of pixel value, K1=0.01, K2=
0.03, the scope of structural similarity is -1 to 1；When two images are the same, the value of SSIM is equal to 1.
As a further improvement on the present invention, the screening threshold value in the step 7 is the numerical value of most convergence 1.
As a further improvement on the present invention, in the step 2 characteristic ID of image is carried in pretreatment, the step 3
Take, the quick screening ratio in the step 4 in screening, the step 5, the step 6, the step 7 of pictograph information
To algorithm, be to be realized by based on GPU, CPU or distributed cloud computing platform.
Compared with prior art, beneficial effects of the present invention are：
The method for retrieving medicine image with Similarity matching is extracted the invention provides a kind of feature based, radiotherapy hook is applied to
The method for drawing registering field；The application of registration seeks to find one with the most suitable Prototype drawing of target image in radiotherapy planning
Picture, after matched computing obtains optimal registration result, uses for clinical Target delineations, organ dose's simulation or treatment.Therefore, such as
It is considerable that what searches most suitable template image.Pass through the shape facility of 10 and morphologic correlation in the method,
With reference to history information, and give these features different differentiation weights, finally select most close 10 groups of images and its related letter
Breath (will be comprising synthesis options such as male, women, child, old men in 10 groups of images), there is provided doctor's selection is given, under considering
Select most suitable template image.Image after this method computing, the similarity that can provide image ensures, so can be with
The applicability and accuracy of registering computing in radiotherapy planning are improved, is met clinical needs well.
The present invention looks for most like registering sample image by the algorithm accelerated based on GPU, is to realize in clinic
In radiotherapy, patient lie down on one's sick bed after a few minutes within complete registration with dosage simplation verification.Efficiency high of the present invention, section
Make an appointment and human cost, clinical needs are met well, can clinically be applicable, with significant social effect.
Brief description of the drawings
Fig. 1 is extracted and the method for retrieving medicine image of Similarity matching for feature based disclosed in an embodiment of the present invention
Flow chart；
Fig. 2 is the feature extraction explanatory diagram of step 3 in Fig. 1.
Specific embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained on the premise of creative work is not made, belongs to the scope of protection of the invention.
The method for retrieving medicine image with Similarity matching is extracted the invention discloses a kind of feature based, radiotherapy target is applied to
Registering field is delineated in area.In the flow of radiotherapy planning, image registration is a critically important committed step.Scheme in radiotherapy planning
As the application purpose of registration is, in order to find one and the most suitable template image of target image, to be can obtain most after matched computing
Good registration result, uses for clinical Target delineations, organ dose's simulation or treatment.Therefore, most suitable template how is searched
Image is considerable.In the method through 10 with the morphological feature of morphologic correlation, with reference to history information, and give this
The different differentiation weight of a little features, finally selects 10 groups of most close images and relevant information, there is provided doctor's selection is given, through synthesis
Consider the most suitable template image of lower selection.By after this method computing, template image can be provided similar to target image
Degree guarantee, can so improve the applicability and accuracy of image registration computing in radiotherapy planning.
The present invention is described in further detail below in conjunction with the accompanying drawings：
As shown in figure 1, the present invention provides the method for retrieving medicine image of a kind of feature based extraction and Similarity matching, its bag
Include following steps：
Step 1, the image and medical history data that read patient, data include image information and text information；Image information bag
Include CT images, pyramidal CT image, ultrasonoscopy, MRI image, PET image and X-Ray images；Text information includes that patient is basic
Data, related history, related complication, disease type, disease treatment position, pathology and diagnostic imaging report information.
Step 2, the data obtained by step 1 is pre-processed, be specifically divided into image preprocessing and pre-processed with word；
Step 2.1, image preprocessing：Original image (generally DICOM picture formats) is done and turns shelves, Image normalization,
Image segmentation, finally extracts target image image information associated therewith, for the nonstandard image of image taking (group) or
There is the image (group) of image quality issues, automatic decision is retained or deleted；Wherein, if original image is 2D DICOM images,
Then directly carry out turning shelves；If original image is 3D DICOM images, 3D DICOM images are first converted into 2D DICOM images
Group, then carries out turning shelves；Wherein：
Step 2.1.1, differentiate that original image be 2D DICOM images or 3D DICOM images, it is of the invention at least its
One of plant image；
If step 2.1.2,3D DICOM images, then transferring files are 2D DICOM sequence image groups；
Step 2.1.3,2D DICOM images or image sets are converted into .bmp or .jpeg forms, are easy to successive image
Treatment, is at least converted into one kind of in the present invention；
Step 2.1.4, image is done histogram equalization, calculation step is as follows：
A, pending image statisticses its histogram to giving, obtain：
Pr(rj)=nj/ N, j=0,1 ..., L-1 (1)
The histogram that b, basis are counted is converted using cumulative distribution function；
C, replace old gray scale with new gray scale, obtain Sk；
Wherein N is the sum of pixel in piece image；njIt is the pixel of j-th stage gray scale；Pr(rj) represent original image gray scale
The probability distribution that level occurs；rkIt is k-th gray level；T(rk) it is to set up corresponding between input picture and output image gray level
Relation, i.e., the probability that new gray level occurs, repositions Cumulative Distribution Function Sk；
Step 2.1.5, self-adaption binaryzation image is done to the operation result of step 2.1.4 using Otsu methods, obtain base
This image outline edge, comprises the following steps：
Otsu methods are a kind of Dynamic Binarization methods of overall situationization, are called Da-Jin algorithm, difference hair also known as between maximum kind, are
Statistics based on entire image is levied, and realizes the automatic selection of threshold value.Its principle is that image histogram is split with a certain gray value
Into two classes, the pixel number and average gray of this two class are calculated respectively, then calculate their inter-class variance.When being divided into
Two inter-class variances it is maximum when, this gray value is just as the threshold value of image binaryzation treatment.The use scope of Da-Jin algorithm is wider,
No matter it is obvious bimodal that the histogram of image whether there is, can obtain remembering that f (i, j) is the gray value at MxN images (i, j) points；
Assuming that f (i, j) value [0, m-1], note p (k) is the frequency of gray value k, then have：
Assuming that being that the target that Threshold segmentation goes out is respectively with background with gray value t：F (i, j)≤t } and { f (i, j)>T },
Then target part ratio：ω0(t)=∑0≤i≤tp(i) (4)
Target part is counted：N0(t)=MN ∑s0≤i≤tp(i) (5)
Background parts ratio：ω1(t)=∑t≤i≤m-1p(i) (6)
Background parts are counted：N1(t)=MN ∑st≤i≤m-1p(i) (7)
Target mean：μ0(t)=∑0≤i≤tip(i)/ω0(t) (8)
Background mean value：μ1(t)=∑t≤i≤m-1ip(i)/ω1(t) (9)
Grand mean：μ=ω0(t)μ0(t)+ω1(t)μ1(t) (10)
Da-Jin algorithm points out to ask the formula of image optimal threshold g be：
It is actually inter-class variance value in the bracket of formula the right, target and background two parts that threshold value g is partitioned into are constituted
Entire image, and desired value μ0T (), probability is ω1(t), background value μ1T (), probability is ω0T (), grand mean is μ, root
The formula is obtained final product according to the definition of variance；Because variance is a kind of measurement of intensity profile uniformity, variance yields is bigger, illustrates pie graph picture
Two parts difference it is bigger, when partial target mistake is divided into background or part background mistake is divided into target and can all cause two parts difference to become
It is small, therefore the segmentation for making inter-class variance maximum means that misclassification probability is minimum.
After step 2.1.6, step 2.1.5 are completed, it will tentatively separate target image elementary contour；
Step 2.1.7, the image elementary contour according to step 2.1.6, find out Bounding box, that is, minimum can be wrapped
Firmly image is square, and an expansion algorithm (dilation) and corrosion (erosion) algorithm are then done to image, can obtain mesh
Logo image integrity profile edge；The image of original position need to be paid special attention to Bounding box are extracted, if image Bounding
The area of box is less than certain size, then judge that this figure has the abnormal problem of positioning, is deleted.
Step 2.2, word pretreatment：In by the written historical materials information of patient, patient gender is obtained, the age, tumor type,
Whether therapentic part, pathology and image report previously did radiotherapy, if the relevant information such as have related complication ...；Wherein mesh
Mark tumor information will apply the image ID in data bank to screen with the information of tumor locus；And patient gender, the age, pathology with
The relevant informations such as image report will be applied in step 8；
Step 3, from the pretreated image of step 2.1 or image sets, extract required for 10 features, set up the mesh
The image ID of logo image；
As shown in Fig. 2 10 features are as follows respectively：
Feature 1：2D image sets quantity or frame number；
Feature 2：Image outline longitudinal direction most major axis；
Feature 3：The horizontal most major axis of image outline；
Feature 4：Take at the bounding box longitudinal directions 1/4 of image outline, the most major axis of image outline；
Feature 5：Take at the bounding box longitudinal directions 1/2 of image outline, the most major axis of image outline；
Feature 6：Take at the bounding box longitudinal directions 3/4 of image outline, the most major axis of image outline；
Feature 7：Take at the bounding box horizontal 1/4 of image outline, the most major axis of image outline；
Feature 8：Take at the bounding box horizontal 1/2 of image outline, the most major axis of image outline；
Feature 9：Take at the bounding box horizontal 3/4 of image outline, the most major axis of image outline；
Feature 10：Image volume or area；
Step 4, from the pretreated text information of step 2.2, first find out identical disease, the figure at identical treatment position
As group；First will this information send into data bank in do preliminary screening, find out identical disease, the image sets at identical treatment position, most
The image sets for meeting are proposed afterwards；
All images in step 5, the image sets for extracting the image ID and step 4 of target image obtained by step 3
ID compares, the main similarity for comparing feature 1, during due to clinically image taking, has rule for each position or organ
Fixed shooting image separation criteria, therefore, this parameter can be used as tentatively judging the characteristic standard of object construction size；Look for
Image sets characteristics of image ID similar to target image characteristics ID, incongruent in data bank, excludes；Wherein, screening threshold value is
95% or 90% or 85% or 80% or 75%, it is 95% preferably to screen threshold value, that is, look for characteristics of image ID and mesh in data bank
Logo image characteristic ID similarity>More than 95% image sets；
Step 6, will be screened via step 5 after image sets, the similarity of feature 2-10 is compared with target image ID again,
Similar image sets are looked for, it is incongruent, exclude；Wherein, screening threshold value is 95% or 90% or 85% or 80% or 75%,
It is preferred that screening threshold value is 95%, that is, look for similarity>More than 95% image sets；
Step 7, will be screened via step 6 after qualified image sets, respectively to target image group calculate image it is similar
Degree index (SSIM), compares numerical value, and its screening threshold value is 10 numerical value of most convergence 1, and more convergence 1 then represents more close, extracts
Most close preceding 10 groups of images, it is incongruent, exclude；
Give two images and be respectively defined as x and y, two structural similarities of image can be obtained in such a way：
Wherein μxIt is the average value of x, μyIt is the average value of y,It is the difference of x,It is the variance of y, σxyIt is the association side of x and y
Difference；C1=(K1L)2, C2=(K2L)2It is the constant for maintaining stabilization, L is the dynamic range of pixel value, K1=0.01, K2=
0.03, the scope of structural similarity is -1 to 1；When two images are the same, the value of SSIM is equal to 1.
Step 8,10 groups of most close images are shown in hospital system end, and show its related essential information, such as:Year
The relevant informations such as age, sex, medical history, there is provided the selection reference of registering template image is done to clinician, is terminated.
Preferably, in pretreatment, the step 3 in step 2 of the present invention the characteristic ID of image extract, pictograph in step 4
Quick screening alignment algorithm in screening, step 5, step 6, the step 7 of information, it is by based on GPU, CPU or distributed cloud
What calculating platform was realized.
The method for retrieving medicine image with Similarity matching is extracted the invention provides a kind of feature based, radiotherapy hook is applied to
The method for drawing registering field；The application of registration seeks to find one with the most suitable Prototype drawing of target image in radiotherapy planning
Picture, after matched computing obtains optimal registration result, uses for clinical Target delineations, organ dose's simulation or treatment.Therefore, such as
It is considerable that what searches most suitable template image.Pass through the shape facility of 10 and morphologic correlation in the method,
With reference to history information, and give these features different differentiation weights, finally select most close 10 groups of images and its related letter
Breath (will be comprising synthesis options such as male, women, child, old men in 10 groups of images), there is provided doctor's selection is given, under considering
Select most suitable template image.Image after this method computing, the similarity that can provide image ensures, so can be with
The applicability and accuracy of registering computing in radiotherapy planning are improved, is met clinical needs well.
The present invention looks for most like registering sample image by the algorithm accelerated based on GPU, is to realize in clinic
In radiotherapy, patient lie down on one's sick bed after a few minutes within complete registration with dosage simplation verification.Efficiency high of the present invention, section
Make an appointment and human cost, clinical needs are met well, can clinically be applicable, with significant social effect.
The preferred embodiments of the present invention are these are only, is not intended to limit the invention, for those skilled in the art
For member, the present invention can have various modifications and variations.All any modifications within the spirit and principles in the present invention, made,
Equivalent, improvement etc., should be included within the scope of the present invention.
Claims (9)
1. a kind of feature based extracts the method for retrieving medicine image with Similarity matching, it is characterised in that comprise the following steps：
Step 1, the image and medical history data that read patient, the data include image information and text information；
Step 2, the data obtained by step 1 is pre-processed, be divided into image preprocessing and pre-processed with word；
Step 2.1, image preprocessing：Original image is made to turn shelves, normalization, segmentation, target information is extracted；The target information
Including target image and target image profile；Wherein, if original image is 2D DICOM images, directly carry out turning shelves；If former
Beginning image is 3D DICOM images, then 3D DICOM images first are converted into 2D DICOM image sets, then carries out turning shelves；
Step 2.2, word pretreatment：In by the text information of patient, obtain patient gender, age, disease information, disease and cure the disease
Whether whether position, pathology and image are reported, had previously done radiotherapy and had related complication；
Step 3, from the pretreated image of step 2.1 or image sets, extract required for 10 features, set up the target figure
The image ID of picture；
10 features are as follows respectively：
Feature 1：2D image sets quantity or frame number；
Feature 2：Image outline longitudinal direction most major axis；
Feature 3：The horizontal most major axis of image outline；
Feature 4：Take at the bounding box longitudinal directions 1/4 of image outline, the most major axis of image outline；
Feature 5：Take at the bounding box longitudinal directions 1/2 of image outline, the most major axis of image outline；
Feature 6：Take at the bounding box longitudinal directions 3/4 of image outline, the most major axis of image outline；
Feature 7：Take at the bounding box horizontal 1/4 of image outline, the most major axis of image outline；
Feature 8：Take at the bounding box horizontal 1/2 of image outline, the most major axis of image outline；
Feature 9：Take at the bounding box horizontal 3/4 of image outline, the most major axis of image outline；
Feature 10：Image volume or area；
Step 4, from the pretreated text information of step 2.2, first pass through preliminary screening, find out identical disease, it is identical to control
Treat the image sets at position；
All image ID in step 5, the image sets for extracting the image ID and step 4 of target image obtained by step 3 do
Compare, the similarity of feature 1 is compared first, look for similar image sets, it is incongruent, exclude；
Step 6, will be screened via step 5 after image sets, the similarity of feature 2-10 is compared with target image ID again, look for
Similar image sets, it is incongruent, exclude；
Step 7, the image sets screened via step 6 calculate image similarity index (SSIM) with target image group respectively,
Compare numerical value more convergence 1, then it represents that more close, extract most close preceding 10 groups of images, it is incongruent, exclude；
Step 8,10 groups of most close images are shown in hospital system end, for clinician's selection, terminated.
2. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
Stating image information includes CT images, pyramidal CT image, ultrasonoscopy, MRI image, PET image and X-Ray images；The word
Information includes patient's basic document, related history, related complication, disease type, disease treatment position, pathology and diagnostic imaging
Report information.
3. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
Stating step 2.1 includes：
Step 2.1.1, differentiate that image be 2D DICOM images or 3D DICOM images, it is of the invention in it is at least one kind of
Image；
If step 2.1.2,3D DICOM images, then transferring files are 2D DICOM sequence image groups；
Step 2.1.3,2D DICOM images or image sets are converted into .bmp or .jpeg forms；
Step 2.1.4, image is done histogram equalization, calculation step is as follows：
A, pending image statisticses its histogram to giving, obtain：
Pr(rj)=nj/ N, j=0,1 ..., L-1 (1)
The histogram that b, basis are counted is converted using cumulative distribution function；
C, replace old gray scale with new gray scale, obtain Sk；
Wherein N is the sum of pixel in piece image；njIt is the pixel of j-th stage gray scale；Pr(rj) represent that original image gray level goes out
Existing probability distribution；rkIt is k-th gray level；T(rk) it is the corresponding pass set up between input picture and output image gray level
System, i.e., the probability that new gray level occurs, repositions Cumulative Distribution Function Sk；
Step 2.1.5, self-adaption binaryzation image is done to the operation result of step 2.1.4 using Otsu methods, obtain parent map
As contour edge；
After step 2.1.6, step 2.1.5 are completed, it will tentatively separate target image elementary contour；
Step 2.1.7, the image elementary contour according to step 2.1.6, find out the minimum square Bounding that can encase image
Box, then does an expansion algorithm and erosion algorithm to image, can obtain target image integrity profile edge.
4. feature based as claimed in claim 3 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that
In step 2.1.5：
Otsu methods are called Da-Jin algorithm, no matter its histogram of image in calculating process whether there is obvious bimodal, can obtain
Note f (i, j) is the gray value at MxN images (i, j) points；
Assuming that f (i, j) value [0, m-1], note p (k) is the frequency of gray value k, then have：
Assuming that being that the target that Threshold segmentation goes out is respectively with background with gray value t：F (i, j)≤t } and { f (i, j)>T },
Then target part ratio：ω0(t)=∑0≤i≤tp(i) (4)
Target part is counted：N0(t)=MN ∑s0≤i≤tp(i) (5)
Background parts ratio：ω1(t)=∑t≤i≤m-1p(i) (6)
Background parts are counted：N1(t)=MN ∑st≤i≤m-1p(i) (7)
Target mean：μ0(t)=∑0≤i≤tip(i)/ω0(t) (8)
Background mean value：μ1(t)=∑t≤i≤m-1ip(i)/ω1(t) (9)
Grand mean：μ=ω0(t)μ0(t)+ω1(t)μ1(t) (10)
Da-Jin algorithm points out to ask the formula of image optimal threshold g be：
It is actually inter-class variance value in the bracket of formula the right, target and background two parts that threshold value g is partitioned into constitute whole
Width image, and desired value μ0T (), probability is ω1(t), background value μ1T (), probability is ω0T (), grand mean is μ, according to side
Poor definition obtains final product the formula.
5. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
It is 95% or 90% or 85% or 80% or 75% to state the screening threshold value in step 5.
6. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
It is 95% or 90% or 85% or 80% or 75% to state the screening threshold value in step 6.
7. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that
In step 7：
Give two images and be respectively defined as x and y, two structural similarities of image can be obtained in such a way：
Wherein μxIt is the average value of x, μyIt is the average value of y,It is the difference of x,It is the variance of y, σxyIt is the covariance of x and y；C1
=(K1L)2, C2=(K2L)2It is the constant for maintaining stabilization, L is the dynamic range of pixel value, K1=0.01, K2=0.03,
The scope of structural similarity is -1 to 1；When two images are the same, the value of SSIM is equal to 1.
8. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
State the numerical value that the screening threshold value in step 7 is most convergence 1.
9. feature based as claimed in claim 1 extracts the method for retrieving medicine image with Similarity matching, it is characterised in that institute
The characteristic ID for stating image in pretreatment, the step 3 in step 2 extracts, the screening of pictograph information in the step 4,
Quick screening alignment algorithm in the step 5, the step 6, the step 7, it is by based on GPU, CPU or distributed cloud
What calculating platform was realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710106121.8A CN106846317B (en) | 2017-02-27 | 2017-02-27 | Medical image retrieval method based on feature extraction and similarity matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710106121.8A CN106846317B (en) | 2017-02-27 | 2017-02-27 | Medical image retrieval method based on feature extraction and similarity matching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106846317A true CN106846317A (en) | 2017-06-13 |
| CN106846317B CN106846317B (en) | 2021-09-17 |
Family
ID=59134147
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710106121.8A Active CN106846317B (en) | 2017-02-27 | 2017-02-27 | Medical image retrieval method based on feature extraction and similarity matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106846317B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679219A (en) * | 2017-10-19 | 2018-02-09 | 广州视睿电子科技有限公司 | Matching process and device, interactive intelligent tablet computer and storage medium |
| CN107688787A (en) * | 2017-09-01 | 2018-02-13 | 宜宾学院 | Proximal interphalangeal joint lines recognition methods based on Gabor wavelet |
| CN108921828A (en) * | 2018-06-15 | 2018-11-30 | 湖南科技大学 | Not significant weld joint recognition method under a kind of complex scene |
| CN109101525A (en) * | 2018-06-19 | 2018-12-28 | 黑龙江拓盟科技有限公司 | A kind of medical image comparison method based on image comparison identification |
| CN109166108A (en) * | 2018-08-14 | 2019-01-08 | 上海融达信息科技有限公司 | A kind of automatic identifying method of CT images pulmonary abnormalities tissue |
| CN109934934A (en) * | 2019-03-15 | 2019-06-25 | 广州九三致新科技有限公司 | A kind of medical image display methods and device based on augmented reality |
| US10510145B2 (en) | 2017-12-27 | 2019-12-17 | Industrial Technology Research Institute | Medical image comparison method and system thereof |
| WO2020233254A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Medical data analysis system integrating structured image data |
| CN112206063A (en) * | 2020-09-01 | 2021-01-12 | 广东工业大学 | Multi-mode multi-angle dental implant registration method |
| CN112395441A (en) * | 2019-08-14 | 2021-02-23 | 杭州海康威视数字技术股份有限公司 | Object retrieval method and device |
| CN112508773A (en) * | 2020-11-20 | 2021-03-16 | 小米科技(武汉)有限公司 | Image processing method and device, electronic device and storage medium |
| CN112950623A (en) * | 2021-03-29 | 2021-06-11 | 云印技术(深圳)有限公司 | Mark identification method and system |
| CN115938591A (en) * | 2023-02-23 | 2023-04-07 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy-based dose distribution interval determination device and electronic equipment |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050201516A1 (en) * | 2002-03-06 | 2005-09-15 | Ruchala Kenneth J. | Method for modification of radiotherapy treatment delivery |
| CN101373479A (en) * | 2008-09-27 | 2009-02-25 | 华中科技大学 | Method and system for searching computer picture of mammary gland x-ray radiography |
| CN102306239A (en) * | 2011-07-22 | 2012-01-04 | 李宝生 | Method for evaluating and optimizing radiotherapy dose based on cone beam CT (Computer Tomography) image CT value correction technology |
| CN102509286A (en) * | 2011-09-28 | 2012-06-20 | 清华大学深圳研究生院 | Target region sketching method for medical image |
| CN103247046A (en) * | 2013-04-19 | 2013-08-14 | 深圳先进技术研究院 | Automatic target volume sketching method and device in radiotherapy treatment planning |
| CN103345746A (en) * | 2013-06-25 | 2013-10-09 | 上海交通大学 | Medical image diagnostic method based on CT-PET |
| CN104036109A (en) * | 2014-03-14 | 2014-09-10 | 上海大图医疗科技有限公司 | Image based system and method for case retrieving, sketching and treatment planning |
| CN104117151A (en) * | 2014-08-12 | 2014-10-29 | 章桦 | Optimization method of online self-adaption radiotherapy plan |
| CN104338240A (en) * | 2014-10-31 | 2015-02-11 | 章桦 | Automatic optimization method for on-line self-adaption radiotherapy plan and device |
| CN105956198A (en) * | 2016-06-20 | 2016-09-21 | 东北大学 | Nidus position and content-based mammary image retrieval system and method |
-
2017
- 2017-02-27 CN CN201710106121.8A patent/CN106846317B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050201516A1 (en) * | 2002-03-06 | 2005-09-15 | Ruchala Kenneth J. | Method for modification of radiotherapy treatment delivery |
| CN101373479A (en) * | 2008-09-27 | 2009-02-25 | 华中科技大学 | Method and system for searching computer picture of mammary gland x-ray radiography |
| CN102306239A (en) * | 2011-07-22 | 2012-01-04 | 李宝生 | Method for evaluating and optimizing radiotherapy dose based on cone beam CT (Computer Tomography) image CT value correction technology |
| CN102509286A (en) * | 2011-09-28 | 2012-06-20 | 清华大学深圳研究生院 | Target region sketching method for medical image |
| CN103247046A (en) * | 2013-04-19 | 2013-08-14 | 深圳先进技术研究院 | Automatic target volume sketching method and device in radiotherapy treatment planning |
| CN103345746A (en) * | 2013-06-25 | 2013-10-09 | 上海交通大学 | Medical image diagnostic method based on CT-PET |
| CN104036109A (en) * | 2014-03-14 | 2014-09-10 | 上海大图医疗科技有限公司 | Image based system and method for case retrieving, sketching and treatment planning |
| CN104117151A (en) * | 2014-08-12 | 2014-10-29 | 章桦 | Optimization method of online self-adaption radiotherapy plan |
| CN104338240A (en) * | 2014-10-31 | 2015-02-11 | 章桦 | Automatic optimization method for on-line self-adaption radiotherapy plan and device |
| CN105956198A (en) * | 2016-06-20 | 2016-09-21 | 东北大学 | Nidus position and content-based mammary image retrieval system and method |
Non-Patent Citations (8)
| Title |
|---|
| ELIANA M.V´ASQUEZ OSORIO B.SC.等: "A Novel Flexible Framework with Automatic Feature Corresp ondence Optimization for Non-Rigid Registration in Radiotherapy", 《MEDICAL PHYSICS》 * |
| GIOVANNI MAURO CATTANEO等: "Target delineation in post-operative radiotherapy of brain gliomas:Interobserver variability and impact of image registration of MR(pre-operative) images on treatment planning CT scans", 《RADIOTHERAPY AND ONCOLOGY》 * |
| HUA ZHANG等: "Effect of compressed sensing reconstruction on target and organ delineation in cone-beam CT of head-and-neck and breast cancer patients", 《RADIOTHERAPY AND ONCOLOGY》 * |
| YUHAN YANG等: "Contour Propagation Using Feature-Based Deformable Registration for Lung Cancer", 《HINDAWI》 * |
| 张瑞: "基于SIFT的三维特征提取及其在医学影像配准中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王远瑞: "基于形变配准的交互式靶区勾画系统的设计与实现", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
| 胡玉兰: "医学图像融合在前列腺癌调强放疗靶区勾画中的应用研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
| 袁峥玺等: "图像配准技术在图像引导放疗中的应用", 《中国医学物理学杂志》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107688787A (en) * | 2017-09-01 | 2018-02-13 | 宜宾学院 | Proximal interphalangeal joint lines recognition methods based on Gabor wavelet |
| CN107688787B (en) * | 2017-09-01 | 2020-09-29 | 宜宾学院 | Near-end interphalangeal joint line identification method based on Gabor wavelet |
| CN107679219A (en) * | 2017-10-19 | 2018-02-09 | 广州视睿电子科技有限公司 | Matching process and device, interactive intelligent tablet computer and storage medium |
| US10510145B2 (en) | 2017-12-27 | 2019-12-17 | Industrial Technology Research Institute | Medical image comparison method and system thereof |
| CN108921828A (en) * | 2018-06-15 | 2018-11-30 | 湖南科技大学 | Not significant weld joint recognition method under a kind of complex scene |
| CN108921828B (en) * | 2018-06-15 | 2022-04-22 | 湖南科技大学 | Method for identifying insignificant weld joint in complex scene |
| CN109101525A (en) * | 2018-06-19 | 2018-12-28 | 黑龙江拓盟科技有限公司 | A kind of medical image comparison method based on image comparison identification |
| CN109166108B (en) * | 2018-08-14 | 2022-04-08 | 上海融达信息科技有限公司 | Automatic identification method for abnormal lung tissue of CT (computed tomography) image |
| CN109166108A (en) * | 2018-08-14 | 2019-01-08 | 上海融达信息科技有限公司 | A kind of automatic identifying method of CT images pulmonary abnormalities tissue |
| CN109934934A (en) * | 2019-03-15 | 2019-06-25 | 广州九三致新科技有限公司 | A kind of medical image display methods and device based on augmented reality |
| CN109934934B (en) * | 2019-03-15 | 2023-09-29 | 广州九三致新科技有限公司 | Medical image display method and device based on augmented reality |
| WO2020233254A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Medical data analysis system integrating structured image data |
| CN112395441A (en) * | 2019-08-14 | 2021-02-23 | 杭州海康威视数字技术股份有限公司 | Object retrieval method and device |
| CN112206063A (en) * | 2020-09-01 | 2021-01-12 | 广东工业大学 | Multi-mode multi-angle dental implant registration method |
| CN112508773A (en) * | 2020-11-20 | 2021-03-16 | 小米科技(武汉)有限公司 | Image processing method and device, electronic device and storage medium |
| CN112508773B (en) * | 2020-11-20 | 2024-02-09 | 小米科技(武汉)有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112950623A (en) * | 2021-03-29 | 2021-06-11 | 云印技术(深圳)有限公司 | Mark identification method and system |
| CN115938591A (en) * | 2023-02-23 | 2023-04-07 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy-based dose distribution interval determination device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106846317B (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106846317A (en) | A kind of feature based extracts the method for retrieving medicine image with Similarity matching | |
| Wang et al. | A deep learning‐based autosegmentation of rectal tumors in MR images | |
| Lee et al. | Pixel-level deep segmentation: artificial intelligence quantifies muscle on computed tomography for body morphometric analysis | |
| Huang et al. | MSFCN-multiple supervised fully convolutional networks for the osteosarcoma segmentation of CT images | |
| US11455732B2 (en) | Knowledge-based automatic image segmentation | |
| CN111008984B (en) | Automatic contour line drawing method for normal organ in medical image | |
| Liu et al. | Automatic whole heart segmentation using a two-stage u-net framework and an adaptive threshold window | |
| Peng et al. | Segmentation of lung in chest radiographs using hull and closed polygonal line method | |
| US20210225000A1 (en) | Method and device for stratified image segmentation | |
| Jun Guo et al. | Automated left ventricular myocardium segmentation using 3D deeply supervised attention U‐net for coronary computed tomography angiography; CT myocardium segmentation | |
| Dai et al. | Automated delineation of head and neck organs at risk using synthetic MRI‐aided mask scoring regional convolutional neural network | |
| US9727975B2 (en) | Knowledge-based automatic image segmentation | |
| Yang et al. | DCU-Net: Multi-scale U-Net for brain tumor segmentation | |
| Ashok et al. | A systematic review of the techniques for the automatic segmentation of organs-at-risk in thoracic computed tomography images | |
| Peng et al. | A-LugSeg: Automatic and explainability-guided multi-site lung detection in chest X-ray images | |
| Javaid et al. | Multi-organ segmentation of chest CT images in radiation oncology: comparison of standard and dilated UNet | |
| Sallemi et al. | Towards a computer aided prognosis for brain glioblastomas tumor growth estimation | |
| Wu et al. | Coarse-to-fine lung nodule segmentation in CT images with image enhancement and dual-branch network | |
| JPWO2020110774A1 (en) | Image processing equipment, image processing methods, and programs | |
| Zhou et al. | Detection and semiquantitative analysis of cardiomegaly, pneumothorax, and pleural effusion on chest radiographs | |
| Artzi et al. | Automatic segmentation, classification, and follow‐up of optic pathway gliomas using deep learning and fuzzy c‐means clustering based on MRI | |
| Liu et al. | LLRHNet: multiple lesions segmentation using local-long range features | |
| Wang et al. | Multi-view fusion segmentation for brain glioma on CT images | |
| Rosas González et al. | 3D automatic brain tumor segmentation using a multiscale input U-Net network | |
| Kieselmann et al. | Auto-segmentation of the parotid glands on MR images of head and neck cancer patients with deep learning strategies |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |