CN107146222A - Medical Image Compression algorithm based on human anatomic structure similitude - Google Patents

Medical Image Compression algorithm based on human anatomic structure similitude Download PDF

Info

Publication number
CN107146222A
CN107146222A CN201710267790.3A CN201710267790A CN107146222A CN 107146222 A CN107146222 A CN 107146222A CN 201710267790 A CN201710267790 A CN 201710267790A CN 107146222 A CN107146222 A CN 107146222A
Authority
CN
China
Prior art keywords
region
image
density
segmentation
kidney
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
Application number
CN201710267790.3A
Other languages
Chinese (zh)
Other versions
CN107146222B (en
Inventor
闵秋莎
刘能
王志锋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong Normal University
Original Assignee
Huazhong Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huazhong Normal University filed Critical Huazhong Normal University
Priority to CN201710267790.3A priority Critical patent/CN107146222B/en
Publication of CN107146222A publication Critical patent/CN107146222A/en
Application granted granted Critical
Publication of CN107146222B publication Critical patent/CN107146222B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • 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

Abstract

The present invention relates to a kind of Medical Image Compression algorithm based on human anatomic structure similitude, traditional partitioning algorithm based on intensity is combined, the specific organ in abdominal CT data set is split by the present invention with anatomical knowledge.The priori anatomical knowledge for being primarily based on current data set obtains the candidate region of each organ, then accurately extracts the data of organ using the method based on density in the candidate region.Relative position of the present invention using organ in the body, can be applied to the image of different patient sizes.Secondly, this cutting techniques is performed in a progressive way, first candidate regions are substantially defined, then uses the dividing method based on density to refine target area, and this method causes segmentation precision more preferable.The present invention can be used for splitting single organ in medical image, and the anatomical variability being adapted between different patients, help to reduce segmentation error, ultimately help to improve follow-up squeeze operation.

Description

Medical Image Compression algorithm based on human anatomic structure similitude
Technical field
The present invention relates to Computer Applied Technology field, especially digital image processing field.
Background technology
Medical imaging technology plays indispensable effect in current clinical medicine, be the diagnosis state of an illness it is important according to According to.But, with the raising of medical imaging devices resolution ratio, acquired image produce huge data volume to image storage with Real-time Transmission brings huge pressure.Therefore seek a kind of effective compression algorithm to be necessary, the difficulty of the technology While degree is image being compressed, preferable image quality is kept.
Most of medical images are three-dimensional image sequences, and correlation in piece is not only existed, between time slice exist stronger phase Guan Xing, piece number is more, and correlation is stronger.This characteristic of medical image determines compression and the Ordinary image compression of medical image Difference, Medical Image Compression mainly reduces or removed the correlation of image using lossy compression method and Lossless Compression.
Transmission speed is improved using lossy compression and space is saved, under the conditions of given target bit rate, figure is rebuild As with should closely on square errors sense between original image, but lossy compression method is in order to produce higher pressure Contracting ratio, will inevitably to medical image bring it is a certain degree of degrade, the loss of key diagnostic information may be caused.
Lossless Compression is mainly using the method for predictive coding or transition coding.Predictive coding is made in field of medical images One of earliest compress technique, one of the key challenge of the compress technique is the accurate forecast model of generation, main at present Forecast model have the forecast model based on JPEG, based on context-adaptive forecast model and based on the adaptive of least square method Answer forecast model.
JPEG-LS obtains preferable effect in rest image Lossless Compression, its performance be even more than JPEG2000 without Compression is damaged, but it compresses just for single image, it is impossible to utilize the frame-to-frame correlation of image.Predicting Technique meter based on JPEG It is counted as relatively low, but is due to that JPEG predicts the outcome and can not be well adapted for specific image context, therefore it is used to compress multiple Poor-performing during miscellaneous image.
Lossless compression algorithm based on context-adaptive, the intensity gradient that can be put near current pixel has notable change When change, switch different sub- fallout predictors, phase is gone in front and rear 6 directions using optimal sub- fallout predictor execution up and down in pixel Close operation.The technology needs the longer coding and decoding time than the Predicting Technique based on JPEG, in addition, the parameter of forecast model (switching threshold and predictor coefficient) is that experiment is predefined, it is impossible to based on being adapted to by local data's characteristic of compression image The change of property, reduces decorrelation energy.
Adaptive approach based on least square method has been proven that to be had to the adaptive prediction scheme based on context It is significantly improved, the technology is by local optimum predictive coefficient come more new model, and to produce accurate predicted value, these optimizations are Number is generally calculated by lowest mean square principle.Although the adaptive approach based on least square method is during the coding and decoding stage Forecast model can be adaptively updated, but is still present in the another way of the adaptive approach of least square method, it Based on header, forecast model is directly generated by decoding, heavy calculating is eliminated in decoding stage, and there is provided quick to one The method of decoding.
Transition coding is another important class of compress technique in medical domain, and most of researchs concentrate on conversion stages Wavelet transformation in, it is relative to other kinds of conversion, the property in terms of decorrelation and positioning in both time domain and frequency domain Can comparative superiority.But wavelet transformation compression method will be expanded further, should and the combination of human-eye visual characteristic on Make an effort, improve picture quality, improve compression ratio, and combined with the Dominant Facies of other compression methods.
Medical image have different from general pattern feature, such as due to partial volume effect (PVE) phenomenon caused by Weak edge, borderline pixel causes borderline region " fuzzy " due to the average value with surrounding all pixels, therefore general Forecasting Methodology is not suitable for directly application and the compression of medical image, in addition, it is contemplated that the characteristic that medical image has, i.e., double Symmetry and the Anatomical Structure structural similarity across different patients are dissected in side, therefore propose the compression that one is medical image customization Algorithm is necessary.
The content of the invention
The purpose of the present invention is to propose to a kind of Medical Image Compression algorithm based on human anatomic structure similitude, this is one Plant the adaptive forecasting technique being oriented to dissection with local optimum.The compression scheme is using the anatomical features of patient come fixed in advance Different anatomic region in the medical data collection of position, then for the adaptive prediction model of each particular anatomical region, optimization Predictive factor, produces predicted value, and predicted value is then subtracted with actual value and obtains final predicated error, entropy code is eventually used for.
Lossless Compression is characterised by reversible process, wherein decompression data are numerically identical with initial data.It is this The technology of type is preferably as the loss of any diagnostic message in image may be led in the case of Medical Image Compression Cause serious consequence, such as mistaken diagnosis.Therefore what the compress technique of the present invention considered first is the scheme using Lossless Compression, is proposed Medical Image Compression algorithm based on human anatomic structure similitude, the method for " dividing and rule " is produced using cutting procedure to locate Medical images data sets are managed, multiple regions is divided the image into, then individually compresses different segmentations using the method for local optimum Region is to realize high compression ratio, and it comprises the following steps：
A kind of Medical Image Compression algorithm based on human anatomic structure similitude, including：
Step 1, CTC data sets are obtained.Data can be downloaded from free medical image data bank, and network address is：https://public.cancerimagingarchive.net/ncia/dataBasketDisplay.jsf；In webpage Selected in " Collection (s) " this plate in CT COLONOGRAPHY, " Image Modality (ies) " this plate Select after CT, it is possible to download CTC data sets.
Step 2, using specific anatomic region in density and dissection feature recognition CTC data sets, split, complete number According to the pretreatment stage of collection, specifically include：
Step 2.1, four regions outside identification scanning area, i.e., have -1024H constant density in four corners of image Region.Begin to use seed region growth algorithm from four angle points, then pass through these region parameters of simply record storage To represent the border between region and actual scanning data.
Step 2.2, different anatomic region is classified, according to different density values, CTC data sets is divided into 9 Primary categories, bony areas, soft tissue area, air section, PVE regions, adipose tissue regions desk region, undefined region With the air section outside patient's body.
Step 2.3, it is extracted after the part beyond whole body and scanning area, position and essence based on them are special Levy one residual pixel in image distributed in predefined classification.Histogram thresholding processing is carried out to image, passes through threshold Image is divided into roughly different regions by value.Then the segmentation based on density is applied to avoid erroneous segmentation in these regions.
Step 2.4, after each organ in CTC data sets has been compartmentalized, the density feature based on them is used Bone, bone PVE, colon, colon PVE, body PVE, clothes, desk and outer air zone are recognized with previous segmentation result, These cut zone are used to provide useful information to instruct subsequent cutting procedure, extract certain organs.Using based on density Segmentation with the combination of anatomical features is realized.After organ segmentation completes, the remaining voxel in body region is based on them Density feature be assigned to adipose tissue or lean tissue mass；
Step 3, based on the segmentation result in step 2, subsequently generating one optimized for each particular anatomical region is Row fallout predictor, is then applied to whole data set by the adaptive prediction model of the fallout predictor composition of these optimizations.
Step 3.1, linear prediction model, P=β are selected0+β1x1+β2x2+…+βNxN, wherein P is predicted value, β0、β1…βN It is the coefficient of forecast model, for soft tissue, air and adipose tissue regions, N are 58, and is 96 for bone region N.Each picture The optimized coefficients of element can be calculated by lowest mean square principle, and the process is repeated in each cut zone.
Step 3.2, it is that normal defines 8 discrete orientations in XZ planes：0°、45°、90°、135°、180°、225°、 270°、315°.If actual angle is not equal to one in this 8 angles, immediate angle is quantified as, in order to keep away Exempt from 45 °, 135 °, 225 °, and 315 ° of pixel is scaled, and is the normal exploitation added pattern on these directions.
Step 3.3, it is determined that after normal, the then predicted value based on each direction of cube template generation.Based on edge Fallout predictor in using 5 pixels template size, to reach that progress is maximized, complexity is minimized.Fringe region from Adaptive model is：M={ A, P { PT1, PT2, PT3, PT4, wherein A is special angle, and P is corresponding predictive factor.T1, T2, T3, T4 is original template position, and other templates are generated using rotation and translation.
Step 4, carried out using adaptive prediction model after decorrelation, by outline code, predictor parameter and residual error data hair It is sent to entropy coder.Final compressed file includes head and main body two parts.The head of this document includes outline code, fallout predictor ginseng The parameter on number, voxel size, slice numbers and border, the main body of file includes residual error data.
In a kind of above-mentioned Medical Image Compression algorithm based on human anatomic structure similitude, the step 2.3 is specific Including：
Step 2.3.1, histogram thresholding processing is carried out to CTC data sets, is analyzed the histogram of image, is found continuous main peak Between valley point, different regions are then divided the image into using the intensity level corresponding to valley point as threshold value, are specifically included：
Step 2.3.3.1, extracts skin and hypodermis, and the position in hypodermis region is determined using body contour is corroded Put, the summary of attack step point is calculated as：
Step-length=(skin thickness+hypodermis thickness)/Pixel Dimensions
Step 2.3.3.2, extracts lung areas, using lung in the position characteristics of body, positioned at the top of body, uses Simple threshold technology extracts the lung in top slice, based on this initial segmentation result, uses simple top-down point Segmentation method realizes the segmentation of lung.
Step 2.3.3.3, extracts liver area, first with liver in the relative position of body, removes around liver Muscle, the bone and lung's pixel in liver area is removed using threshold technology, applied anatomy and the rule based on density are gone Unless liver element, realizes the segmentation of liver area.
Step 2.3.3.4, extracts kidney portion region, kidney is two organellas positioned at backbone both sides, below thoracic cavity, According to the position feature of kidney, in top-down sequence section, the top of kidney is recognized by tracking rib, then position with Kidney in cutting into slices afterwards, Liang Geshen areas are expressed as：
Left kidney:(xcos60°-zsin60°)(xcos60°-zsin60°)/(0.13h*0.13h)+(zcos60°+
sin60°)(zcos60°+xsin60°)/(0.2v*0.2v)<=1
Right kidney:(xcos(-60)°-zsin(-60)°)(xcos(-60)°-zsin(-60)°)/(0.13h*
0.13h)+(zcos(-60)°+xsin(-60)°)(zcos(-60)°+xsin(-60)°)/(0.2v*0.
2v)<=1
Wherein (x, z) represents the coordinate of pixel in image, and h and v refer to the length and vertical axis of the trunnion axis of body region Length.Cos and sin represent the cosine value and sine value of the point and backbone formation angle in kidney region respectively.
Step 2.3.2, the density range of some organs is relatively, it is meant that histogram thresholding method can not correctly divide These objects are cut, now based on anatomical priori theoretical, using the relative position and shape of each organ in the body, automatically Detect the certain organs in CTC data sets.
The dividing method proposed in the present invention is to be combined traditional partitioning algorithm based on intensity with anatomical knowledge Get up, the specific organ in abdominal CT data set is split.It is primarily based on the priori anatomical knowledge of current data set The candidate region of each organ is obtained, then the number of organ is accurately extracted using the method based on density in the candidate region According to.This method has several advantages, first, using the relative position of organ in the body, can be applied to different patient sizes Image.Secondly, this cutting techniques is performed in a progressive way, is first substantially defined candidate regions, is then used based on density Dividing method refines target area, and this method causes segmentation precision more preferable.In a word, the technology that is proposed using dissection and Density feature instructs cutting procedure, can be used for splitting single organ in medical image, and be adapted to different trouble Anatomical variability between person, helps to reduce segmentation error, ultimately helps to improve follow-up squeeze operation.
A kind of forecast model for Medical Image Compression is proposed in the present invention, this model is based on two kinds of modeling methods Exploitation, it is respectively intended to processing inside and fringe region.The first method designed for interior zone is based on to optimal mould Plate identification is so as to generate effective fallout predictor.The fallout predictor can realize high precision of prediction with relatively low calculating cost.Place Manage fringe region and different edge directions are considered with second method, it can rotate forecast model to ensure in forecast model Consistent input pattern, this pattern can optimize edge prediction device.Forecast model proposed by the invention includes both types Fallout predictor and optimum prediction device is adaptively switched to according to the region characteristic compressed, solve adaptive based on context The problem of forecast model can not be based on the change of adaptability be carried out by local data's characteristic of compression image is answered, in addition, this prediction Model takes full advantage of inter-frame relation, compensate for that the defect of frame-to-frame correlation can not be utilized based on JPEG forecast models.
The final stage of the present invention is that residual error data is encoded using entropy coder, Prediction with Partial Matching (PPM) technologies distribute the probability of current sign based on previous context, can make whole compression performance Maximize, therefore PPM technologies are used as the entropy coder in compression scheme.
For the validity of the whole compression scheme that illustrates to include data decorrelation and entropy code stage, by present invention offer The compression result that is obtained with a variety of substitute technologies of compression result be compared, as a result show that compression method proposed by the present invention exists 12% is averagely improved on JPEG2000 and 3D-JPEG2000, and it is higher by 6% than standard 3D-JPEG4+PPMd methods, even if 3D- JPEG2000 and 3DJPEG4+PPM methods are using the correlation between section, and they can not effectively compressed edge region.By In the fact medical image has a large amount of edges, these technologies are performed poor in the context of medical images.
Compression algorithm proposed by the present invention combines a kind of novel Forecasting Methodology based on edge especially to handle medical science It is highly effective in terms of the PVE regions that view data is concentrated, the fallout predictor residual information amount associated with edge in reduction, it is remaining Region by the fallout predictor decorrelation of series of optimum, the as a result more preferable compression performance of Display Realization.
Brief description of the drawings
Fig. 1 is the block diagram of compression scheme proposed by the present invention.
Fig. 2 is the flow chart of cutting procedure.
Fig. 3 is the template for region recognition.
Fig. 4 is the explanation of four stereotypes.
Fig. 5 is the coding and decoding process in complete compression scheme technology
Embodiment
Below by embodiment, and with reference to accompanying drawing, technical scheme is described in further detail.
Embodiment：
The validity of the compression scheme of proposition is proved in the present invention using CTC data sets, Fig. 1 is to be used for CTC data Complete compression scheme schematic description.
Step 1, CTC data sets are obtained.
Step 2, using specific anatomic region in density and dissection feature recognition CTC data sets, split, complete number According to the pretreatment stage of collection.
Step 2.1, whole body is extracted from CTC data sets, and uses a series of chain codes (one of each section in data set Chain code) its profile is recorded, obtain body profile using Roberts edge detectors and use 4 connects chain representations.Work as coding body During profile, select borderline point and store its coordinate, encoder follows border and keeps tracking from one in a sequential manner Individual boundary pixel represents that the symbol of the direction of motion is stored using chain code, in order to further subtract to the direction of another boundary pixel The size of little profile file, Compressed Chain Code is carried out using lossless coding technique.
Step 2.2, four regions outside identification scanning area, i.e., have -1024HU in four corners of image The region of the constant density of (Hounsfield Units, medical image unit).The seed since four angle points can be used Algorithm of region growing realizes, then stores these region parameters in a very effective manner by simply recording and represents area Border and actual scanning data between domain.
Step 2.3, it is extracted after the part beyond whole body and scanning area, can be based on their position and this Residual pixel in image is distributed to one in predefined classification (for example by matter feature：Air, soft tissue, bone and PVE bodies Element).
Step 2.3.1,9 primary categories are subdivided into by CTC data sets, and these classifications are bone respectively, and soft tissue is internal empty Gas, PVE regions, adipose tissue, desk, clothes, the air of patient-external and undefined region.
Step 2.3.2, histogram thresholding processing is carried out to image, analyzes the histogram of image, is found between continuous main peak Valley point, the intensity level using valley point as threshold value divides the image into different zones.
Step 2.3.3, for that can not use the region of the correct Ground Split of histogram thresholding method, is applied in these candidate regions Dividing method based on density, with reference to anatomical priori, to avoid incorrect segmentation.
Step 2.3.3.1, extracts skin and hypodermis, and the position in hypodermis region is determined using body contour is corroded Put, the summary of attack step point may be calculated：
Step-length=(skin thickness+hypodermis thickness)/Pixel Dimensions
Step 2.3.3.2, extracts lung areas, using lung in the position characteristics of body, positioned at the top of body, uses Simple threshold technology extracts the lung in top slice, based on this initial segmentation result, uses simple top-down point Segmentation method realizes the segmentation of lung.
Step 2.3.3.3, extracts liver area, first with liver in the relative position of body, removes around liver Muscle, the bone and lung's pixel in liver area is removed using threshold technology, applied anatomy and the rule based on density are gone Unless liver element, realizes the segmentation of liver area.
Step 2.3.3.4, extracts kidney portion region, kidney is two organellas positioned at backbone both sides, below thoracic cavity, According to the position feature of kidney, in top-down sequence section, the top of kidney is recognized by tracking rib, then position with Kidney in cutting into slices afterwards, Liang Geshen areas can be expressed as：
Left kidney:(xcos60°-zsin60°)(xcos60°-zsin60°)/(0.13h*0.13h)+(zcos60°+
sin60°)(zcos60°+xsin60°)/(0.2v*0.2v)<=1
Right kidney:(xcos(-60)°-zsin(-60)°)(xcos(-60)°-zsin(-60)°)/(0.13h*
0.13h)+(zcos(-60)°+xsin(-60)°)(zcos(-60)°+xsin(-60)°)/(0.2v*0.
2v)<=1
Wherein (x, z) represents the coordinate of pixel in image, and h and v refer to the length and vertical axis of the trunnion axis of body region Length, cos and sin represent the cosine value and sine value of the point and backbone formation angle in kidney region respectively.
Step 2.4, cutting procedure is as shown in Fig. 2 after the region beyond body region and scanning area is extracted, be based on Their density feature and previous segmentation result recognize bone, bone PVE, colon, colon PVE, body PVE, clothes, desk and Outer air zone, these cut zone may be used to provide useful information to instruct subsequent cutting procedure, extract specific Organ.Realized using the segmentation based on density and the combination of anatomical features.After organ segmentation completes, in body region Density feature of the remaining voxel based on them is assigned to adipose tissue or lean tissue mass.
Step 2.4.1, automatic identification is completed by checking face adjacent known to the two of current pixel, automatic to know Other details is as follows：
Step 2.4.1.1, judges PN, PW value is whether in the range of bone density
If PN ∈ Gu Touquyu ＆PW ∈ bone regions
If PN, PW value belong to bone region, then X just belongs to bone region
X ∈ bone regions
Step 2.4.1.2, if being judged as NO for previous step, continues to judge PN, whether PW value is in atmospheric density In the range of, if being assumed to be very, according to medical science priori and the relative position of intracorporeal organ, it can be determined that X position is In lung areas, if true, then X belongs to lung areas, otherwise X belongs to colon regions.
If PN ∈ Kong Qiquyu ＆PW ∈ air sections
If LX ∈ lung areas X ∈ lung areas
Otherwise X ∈ colon regions
Step 2.4.1.3, if PN and PW be unsatisfactory for above it is assumed that continuing to judge PN and PW value whether in soft tissue Region, if in soft tissue area, its affiliated area is then judged further according to the position where X.
If PN ∈ Ruan Zuzhiquyu ＆PW ∈ soft tissue areas
If region X ∈ livers of LX ∈ livers region
If the left left kidney regions of kidney region X ∈ of LX ∈
If the right right kidney regions of kidney region X ∈ of LX ∈
If LX ∈ spleens region X ∈ spleens region
Otherwise X ∈ lean tissues
Step 2.4.1.4, if PN and PW be unsatisfactory for above it is assumed that continue judge whether PN, PW belong to bone The region of volume effect, if belonged to, then judges X affiliated areas according to the position of X position.
If PN ∈ bone volume effect Qu Yu ＆PW ∈ bone volume effects region
If LX is close to bony areas X ∈ bone volume effects region
If LX is close to colon regions X ∈ colon volume effects region
If LX is close to lung areas X ∈ lung volume effects region
Otherwise X ∈ adipose tissue regions
Step 2.4.1.5, if PN and PW are unsatisfactory for hypothesis above, then determined fringe region, according to X position Put and judge X affiliated areas.
If the volume effect region of the volume effect region X ∈ body regions of LX ∈ body regions
If LX ∈ hypodermis region X ∈ hypodermis region
Step 2.4.1.6, finally judges external scanning area, because external object density diversity ratio is larger, so not Need the position further according to X to judge the region belonging to it, it is only necessary to the region belonging to judging X according to PN and PW value.
If PN ∈ desk Qu Yu ＆PW ∈ desks region
X ∈ desks region
If PN ∈ Yi Fuquyu ＆PW ∈ Garment regions
X ∈ Garment regions
Otherwise air section external X ∈
Wherein PN represents the pixel value of a pixel above current pixel X, and PW represents one, current pixel X left sides pixel Pixel value, LX represents the position of current pixel, and X represents current pixel.PN, PW and X relative position are shown in Fig. 3.
Step 3, based on the segmentation result in step 2, subsequently generate for a series of of each particular anatomical region optimization Fallout predictor.Then the adaptive prediction model being made up of these fallout predictors optimized is applied to whole data set.
Step 3.1, based on internal forecast model, linear prediction model is selected, P=β are expressed as0+β1x1+β2x2+…+βNxN, wherein P is predicted value, β0、β1…βNIt is the coefficient of forecast model, for soft tissue, air and adipose tissue regions, N is 58, and be 96 for bone region N.The optimized coefficients of each pixel can be calculated by lowest mean square principle, and the process is each Repeated in cut zone.Adaptive model finally combines all these sub- fallout predictors, is expressed as M={ R, P { Pbone；Pliver； Pspleen；Plean_tissue；Pair_lung；Pair_colon；Psubcutaneous_tissue；Padipose_tissue；Ptable；Pclothing； Pexternal_air, wherein R is specific region, and P is corresponding fallout predictor.
Step 3.2, the forecast model based on edge, first has to estimate normal direction, can based on edge normal rotary template To ensure that template is alignd with fringe region all the time, in order to determine the normal direction and amplitude of current point, side is recognized using derivative Edge region, in the case of first derivative, local minimum and maximum represent the presence at edge, and it is density point to make y=f (x) Derivative at the function of cloth, point x can be expressed as
Corresponding both sides pixel edge detection mask is [- 1,1].
Step 3.3, it is that normal defines 8 discrete orientations in XZ planes using the discrete orientation of normal：0°、45°、90°、 135°、180°、225°、270°、315°.If actual angle is not equal to one in this 8 angles, it is quantified as most connecing Near angle, in order to avoid at 45 °, 135 °, 225 °, and 315 ° of pixel scaling, it is that the method line development on these directions is added Template.
Step 3.3.1, according to normal direction two stereotypes defined in XZ planes：One be along axis, it is another Individual is the angle at 45 ° with axis.Other templates are defined in a similar way, all stereotype details such as Fig. 4 institutes Show.
Step 3.4, it is determined that after normal, the then predicted value based on each direction of cube template generation.Based on edge Fallout predictor in using 5 pixels template size, to reach that progress is maximized, complexity is minimized.Fringe region from Adaptive model is：M={ A, P { PT1, PT2, PT3, PT4, wherein A is special angle, and P is corresponding predictive factor.T1, T2, T3, T4 is original template position, and other templates are generated using rotation and translation.
Step 4, carried out using adaptive prediction model after decorrelation, by outline code, predictor parameter and residual error data hair It is sent to entropy coder.Final compressed file includes head and main body two parts.The head of this document includes outline code, fallout predictor ginseng The parameter on number, voxel size, slice numbers and border, the main body of file includes residual error data.
New compression scheme proposed by the present invention, the compressibility of medical image is improved using the priori of anatomic information Can, the program is by based on anatomical cutting procedure, adaptive prediction model and entropy coder composition, in the coding and decoding stage Complete compression scheme it is as shown in Figure 5.
During coding stage, raw data set is initially divided into different anatomic regions, and is then each area The predictive factor of domain generation optimization, adaptive prediction (AAP) model carries out decorrelation using a serial Optimization Prediction device, so Sending residual error data afterwards is used for entropy code.Decoding process is with coded treatment just on the contrary, yet with the coefficient of each fallout predictor Have stored in the header in compressed data, it is possible to generate forecast model in the case of no significantly calculating cost. Lossless Compression is characterised by reversible process, wherein decompression data are numerically identical with initial data.
The type in the region being associated according to current pixel, decoder is switched to corresponding fallout predictor to generate predicted value, The value is added to the predicated error of storage to rebuild initial data.
Specific embodiment described herein is only to present invention explanation for example.The technical field of the invention Technical staff can be made various modifications or supplement to described specific embodiment or be substituted using similar mode, example Decorrelation operation such as is carried out using other forecast models, but without departing from spirit of the invention or surmounts appended letter of authorization Defined scope.

Claims (2)

1. a kind of Medical Image Compression algorithm based on human anatomic structure similitude, including：
Step 1, CTC data sets are obtained；
Step 2, using specific anatomic region in density and dissection feature recognition CTC data sets, split, complete data set Pretreatment stage, specifically include：
Step 2.1, four regions outside identification scanning area, i.e., in four corners of image the area of the constant density with -1024H Domain；Begin to use seed region growth algorithm from four angle points, then by simply record storage these region parameters come table Show the border between region and actual scanning data；
Step 2.2, different anatomic region is classified, according to different density values, CTC data sets is divided into 9 mainly Classification, bony areas, soft tissue area, air section, PVE regions, adipose tissue regions desk region, undefined region and trouble Air section outside person's body；
Step 2.3, it is extracted after the part beyond whole body and scanning area, position and substantive characteristics based on them will Residual pixel in image distributes to one in predefined classification；Histogram thresholding processing is carried out to image, will by threshold value Image is divided into roughly different regions；Then the segmentation based on density is applied to avoid erroneous segmentation in these regions；
Step 2.4, after each organ in CTC data sets has been compartmentalized, the density feature based on them and elder generation are used Preceding segmentation result recognizes bone, bone PVE, colon, colon PVE, body PVE, clothes, desk and outer air zone, these Cut zone is used to provide useful information to instruct subsequent cutting procedure, extracts certain organs；Conciliate using based on density The segmentation for cuing open the combination of feature is realized；After organ segmentation completes, the remaining voxel in body region is close based on them Degree feature is assigned to adipose tissue or lean tissue；
Step 3, based on the segmentation result in step 2, subsequently generate for a series of pre- of each particular anatomical region optimization Device is surveyed, the adaptive prediction model of the fallout predictor composition of these optimizations is then applied to whole data set；
Step 3.1, linear prediction model, P=β are selected0+β1x1+β2x2+…+βNxN, wherein P is predicted value, β0、β1…βNIt is pre- The coefficient of model is surveyed, for soft tissue, air and adipose tissue regions, N are 58, and be 96 for bone region N；Each pixel Optimized coefficients can be calculated by lowest mean square principle, and the process is repeated in each cut zone；
Step 3.2, it is that normal defines 8 discrete orientations in XZ planes：0°、45°、90°、135°、180°、225°、270°、 315°；If actual angle is not equal to one in this 8 angles, be quantified as immediate angle, in order to avoid 45 °, 135 °, 225 °, and 315 ° pixel scaling, be on these directions normal exploitation added pattern；
Step 3.3, it is determined that after normal, the then predicted value based on each direction of cube template generation；Based on the pre- of edge The template size using 5 pixels in device is surveyed, to reach that progress is maximized, complexity is minimized；Fringe region it is adaptive Model is：M={ A, P { PT1, PT2, PT3, PT4, wherein A is special angle, and P is corresponding predictive factor；T1, T2, T3, T4 are Original template position, other templates are generated using rotation and translation；
Step 4, carried out using adaptive prediction model after decorrelation, by outline code, predictor parameter and residual error data are sent to Entropy coder；Final compressed file includes head and main body two parts；The head of this document include outline code, predictor parameter, The parameter on voxel size, slice numbers and border, the main body of file includes residual error data.
2. a kind of Medical Image Compression algorithm based on human anatomic structure similitude according to claim 1, the step Rapid 2.3 specifically include：
Step 2.3.1, histogram thresholding processing is carried out to CTC data sets, analyzes the histogram of image, is found between continuous main peak Valley point, different regions are then divided the image into using the intensity level corresponding to valley point as threshold value, are specifically included：
Step 2.3.3.1, extracts skin and hypodermis, and the position in hypodermis region is determined using body contour is corroded, The summary of attack step point is calculated as：
Step-length=(skin thickness+hypodermis thickness)/Pixel Dimensions
Step 2.3.3.2, extract lung areas, using lung body position characteristics, positioned at the top of body, using simple Threshold technology extract the lung in top slice, based on this initial segmentation result, use simple top-down segmentation side Method realizes the segmentation of lung；
Step 2.3.3.3, extracts liver area, first with liver in the relative position of body, removes the muscle around liver, The bone and lung's pixel in liver area, the applied anatomy and rule based on density is gone unless liver are removed using threshold technology Dirty element, realizes the segmentation of liver area；
Step 2.3.3.4, extracts kidney portion region, kidney is two organellas positioned at backbone both sides, below thoracic cavity, according to The position feature of kidney, in top-down sequence section, the top of kidney is recognized by tracking rib, and then positioning is then cut Kidney in piece, Liang Geshen areas are expressed as：
Left kidney:(xcos60°-zsin60°)(xcos60°-zsin60°)/(0.13h*0.13h)+(zcos60°+sin60°) (zcos60°+xsin60°)/(0.2v*0.2v)<=1
Right kidney:(xcos(-60)°-zsin(-60)°)(xcos(-60)°-zsin(-60)°)/(0.13h*0.13h)+(zcos(- 60)°+xsin(-60)°)(zcos(-60)°+xsin(-60)°)/(0.2v*0.2v)<=1
Wherein (x, z) represents the coordinate of pixel in image, and h and v refer to the length and the length of vertical axis of the trunnion axis of body region Degree, cos and sin represent the cosine value and sine value of the point and backbone formation angle in kidney region respectively；
Step 2.3.2, the density range of some organs is relatively, it is meant that histogram thresholding method can not correctly Ground Split this A little objects, now based on anatomical priori theoretical, utilize the relative position and shape of each organ in the body, automatic detection Certain organs in CTC data sets.
CN201710267790.3A 2017-04-21 2017-04-21 Medical image compression method based on human anatomy structure similarity Active CN107146222B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710267790.3A CN107146222B (en) 2017-04-21 2017-04-21 Medical image compression method based on human anatomy structure similarity

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710267790.3A CN107146222B (en) 2017-04-21 2017-04-21 Medical image compression method based on human anatomy structure similarity

Publications (2)

Publication Number Publication Date
CN107146222A true CN107146222A (en) 2017-09-08
CN107146222B CN107146222B (en) 2020-03-10

Family

ID=59774988

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710267790.3A Active CN107146222B (en) 2017-04-21 2017-04-21 Medical image compression method based on human anatomy structure similarity

Country Status (1)

Country Link
CN (1) CN107146222B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109584233A (en) * 2018-11-29 2019-04-05 广西大学 Three-dimensional image segmentation method based on subjective threshold value and three-dimensional label technology
CN109949309A (en) * 2019-03-18 2019-06-28 安徽紫薇帝星数字科技有限公司 A kind of CT image for liver dividing method based on deep learning
CN110555853A (en) * 2019-08-07 2019-12-10 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN111373438A (en) * 2017-10-17 2020-07-03 透视诊断有限公司 Method and apparatus for imaging an organ
CN111694491A (en) * 2020-05-26 2020-09-22 珠海九松科技有限公司 Method and system for automatically selecting and zooming specific area of medical material by AI (artificial intelligence)
CN112419330A (en) * 2020-10-16 2021-02-26 北京工业大学 Temporal bone key anatomical structure automatic positioning method based on spatial relative position prior
US11182920B2 (en) * 2018-04-26 2021-11-23 Jerry NAM Automated determination of muscle mass from images

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1430185A (en) * 2001-12-29 2003-07-16 田捷 Ultralarge scale medical image surface reconstruction method based on single-layer surface tracking
WO2009109971A2 (en) * 2008-03-04 2009-09-11 Innovea Medical Ltd. Segmentation device and method
WO2012139205A1 (en) * 2011-04-13 2012-10-18 Hamid Reza Tizhoosh Method and system for binary and quasi-binary atlas-based auto-contouring of volume sets in medical images
CN104134210A (en) * 2014-07-22 2014-11-05 兰州交通大学 2D-3D medical image parallel registration method based on combination similarity measure
CN104270638A (en) * 2014-07-29 2015-01-07 武汉飞脉科技有限责任公司 Compression and quality evaluation method for region of interest (ROI) of CT (Computed Tomography) image
CN104933288A (en) * 2014-03-18 2015-09-23 三星电子株式会社 Apparatus and method for visualizing anatomical elements in a medical image

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1430185A (en) * 2001-12-29 2003-07-16 田捷 Ultralarge scale medical image surface reconstruction method based on single-layer surface tracking
WO2009109971A2 (en) * 2008-03-04 2009-09-11 Innovea Medical Ltd. Segmentation device and method
WO2012139205A1 (en) * 2011-04-13 2012-10-18 Hamid Reza Tizhoosh Method and system for binary and quasi-binary atlas-based auto-contouring of volume sets in medical images
CN104933288A (en) * 2014-03-18 2015-09-23 三星电子株式会社 Apparatus and method for visualizing anatomical elements in a medical image
CN104134210A (en) * 2014-07-22 2014-11-05 兰州交通大学 2D-3D medical image parallel registration method based on combination similarity measure
CN104270638A (en) * 2014-07-29 2015-01-07 武汉飞脉科技有限责任公司 Compression and quality evaluation method for region of interest (ROI) of CT (Computed Tomography) image

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
QIUSHA MIN ET AL；: "《An Edge-based Prediction Approach for Medical Image Compression》", 《2012 IEEE EMBS INTERNATIONAL CONFERENCE ON BIOMEDICAL ENGINEERING AND SCIENCES》 *
QIUSHA MIN ET AL；: "《Medical Image Compression Using Region-based》", 《2012 IEEE EMBS INTERNATIONAL CONFERENCE ON BIOMEDICAL ENGINEERING AND SCIENCES》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111373438A (en) * 2017-10-17 2020-07-03 透视诊断有限公司 Method and apparatus for imaging an organ
US11182920B2 (en) * 2018-04-26 2021-11-23 Jerry NAM Automated determination of muscle mass from images
CN109584233A (en) * 2018-11-29 2019-04-05 广西大学 Three-dimensional image segmentation method based on subjective threshold value and three-dimensional label technology
CN109949309A (en) * 2019-03-18 2019-06-28 安徽紫薇帝星数字科技有限公司 A kind of CT image for liver dividing method based on deep learning
CN109949309B (en) * 2019-03-18 2022-02-11 安徽紫薇帝星数字科技有限公司 Liver CT image segmentation method based on deep learning
CN110555853A (en) * 2019-08-07 2019-12-10 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN110555853B (en) * 2019-08-07 2022-07-19 杭州深睿博联科技有限公司 Method and device for segmentation algorithm evaluation based on anatomical priors
CN111694491A (en) * 2020-05-26 2020-09-22 珠海九松科技有限公司 Method and system for automatically selecting and zooming specific area of medical material by AI (artificial intelligence)
CN112419330A (en) * 2020-10-16 2021-02-26 北京工业大学 Temporal bone key anatomical structure automatic positioning method based on spatial relative position prior

Also Published As

Publication number Publication date
CN107146222B (en) 2020-03-10

Similar Documents

Publication Publication Date Title
CN107146222A (en) Medical Image Compression algorithm based on human anatomic structure similitude
Liu et al. A fast fractal based compression for MRI images
CN110648337A (en) Hip joint segmentation method, hip joint segmentation device, electronic apparatus, and storage medium
CN102984511A (en) Method and apparatus for creating a multi-resolution framework for improving medical imaging workflow
CN111242956A (en) U-Net-based ultrasonic fetal heart and fetal lung deep learning joint segmentation method
CN106203255A (en) A kind of pedestrian based on time unifying heavily recognition methods and system
CN114882051A (en) Automatic segmentation and three-dimensional reconstruction method for pelvic bone tumor based on multi-modal image
Karadimitriou Set redundancy, the enhanced compression model, and methods for compressing sets of similar images
Ratakonda et al. Lossless image compression with multiscale segmentation
US20140161329A1 (en) Method and system for binary and quasi-binary atlas-based auto-contouring of volume sets in medical images
Loganathan et al. Active contour based medical image segmentation and compression using biorthogonal wavelet and embedded zerotree
CN115719357A (en) Multi-structure segmentation method for brain medical image
Kalinin et al. A classification approach for anatomical regions segmentation
CN115018864A (en) Three-stage liver tumor image segmentation method based on adaptive preprocessing
Kassim et al. Hierarchical segmentation-based image coding using hybrid quad-binary trees
Moorthi et al. Region-based medical image compression in teleradiology
Dawod et al. Adaptive Slices in Brain Haemorrhage Segmentation Based on the SLIC Algorithm.
Eldib et al. Web image mining age estimation framework
Shi et al. A novel u-like network for the segmentation of thoracic organs
Abdellatif et al. Efficient ROI-based compression of mammography images
Soundarya et al. Comparison of hybrid codes for MRI brain image compression
Rupa et al. MRI brain image compression using spatial fuzzy clustering technique
Bhardwaj et al. Feature Extraction Based Domain Kickout Method For Fractal Image Compression
Sombutkaew et al. Adaptive quantization via fuzzy classified priority mapping for liver ultrasound compression
Dam et al. Efficient segmentation by sparse pixel classification

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