CN102749601A - Image processing apparatus, image processing method and magnetic resonance imaging apparatus - Google Patents

Abstract

The present invention provides an image processing apparatus, an image processing method and a magnetic resonance imaging apparatus, which can increase image quality. In one embodiment, the image processing apparatus is provided with a lowpass filter application unit, a bandpass filter application unit, a highpass filter application unit, an edge mask generating unit and a composing unit, wherein the lowpass filter application unit is corresponding to a data application lowpass filter corresponding to an original image, the bandpass filter application unit is corresponding to a data application bandpass filter, the highpass filter application unit is corresponding to a data application highpass filter, the edge mask generating unit generates an edge mask according to data obtained through using the bandpass filter; and the composing unit performs mask processing through using the data through applying the highpass filter, and performs composing processing through using the data undergone the mask processing and the data obtained through using the lowpass filter.

Description

Image processing apparatus, image processing method and MR imaging apparatus

The application advocates the right of priority of Patent Application No. 13/092,382 that on April 22nd, 2011 applied for and the japanese patent application No. 2012-74390 that applied on March 28th, 2012, and quotes the full content of above-mentioned Japanese patent application in this application.

Technical field

Embodiment relates to image processing apparatus, image processing method and magnetic resonance imaging (imaging) device.

Background technology

At present, in the image processing field of magnetic resonance imaging, known gibbs the is arranged ring artifact (ringing artifacts) of (Gibbs).Take place because ring artifact is block (truncation) owing to data (data), therefore, also be called as gibbs artifact (truncation artifacts) etc.

The prior art document

Patent documentation

Patent documentation 1: No. 5001429 instructions of United States Patent (USP)

Patent documentation 2: No. 5285157 instructions of United States Patent (USP)

Patent documentation 3: No. 5345173 instructions of United States Patent (USP)

Patent documentation 4: No. 5602934 instructions of United States Patent (USP)

Summary of the invention

The problem that the present invention will solve is, a kind of image processing apparatus, image processing method and MR imaging apparatus that can improve image quality is provided.

The image processing apparatus that embodiment relates to possesses: low-pass filter (filter) application portion, BPF. application portion, Hi-pass filter application portion, edge mask (edge mask) generation portion, synthetic portion.Above-mentioned application of low pass filters portion is for that obtain through magnetic resonance imaging, corresponding with original (original) image k space or the data of image space, application of low-pass filters.Above-mentioned BPF. application portion uses BPF. for above-mentioned data.Above-mentioned Hi-pass filter application portion uses Hi-pass filter for above-mentioned data.Above-mentioned edge mask generation portion generates the edge mask at the edge that extracts the object in the above-mentioned original image according to having used the data that above-mentioned BPF. gets.Above-mentioned synthetic portion use used above-mentioned Hi-pass filter and data carry out mask process based on above-mentioned edge mask, and use carried out above-mentioned mask process data with having used above-mentioned low-pass filter data synthesize processing.

Play the effect that can improve image quality.

Description of drawings

Fig. 1 is senior (high-level) frame (block) figure of MRI (the Magnetic Resonance Imaging) system (system) that relates to of embodiment.The MRI system that embodiment relates to is through obtaining the MRI data, and carries out Filtering Processing for the MRI data that obtain, thereby stresses edge (edge) and details in the reconstructed image, reduces the ring artifact of gibbs simultaneously.

Fig. 2 is the skeleton diagram of the Filtering Processing that relates to of embodiment.

Fig. 3 is the skeleton diagram that is fit to computing machine (computer) program (program) code (code) structure of embodiment.In embodiment, at least one data processor (processor) handle be stored in storage part, at least one data array (dataarray).

Symbol description

8 electrodes

9 subjects

10 stands

11 subject platforms

12 static magnetic field B ₀Magnet

14 G _x, G _y, G _zThe gradient magnetic field coil group

16 RF coil assemblies

18 imaging bodies

20 system's inscapes

22 MRI systems control divisions

24 display parts

26 keyboard/mouse

28 printers

30 MRI sequence control parts

32 G _x, G _y, G _zThe gradient magnetic field coil driver

34 RF sending parts

36 transmission/receiving keys

38 MRI data obtain the program code structure

40 RF acceptance divisions

42 MRI data processing divisions (microprocessor, I/O, storage part)

44 image reconstruction program code structures

46 MR image storage parts

50 programs/data store (for both sides or a side who cuts down ring artifact and noise artifacts, edge of image after stress rebuilding simultaneously and other details, storage is used in the k space MRI data that obtain are carried out the program code structure of Filtering Processing)

Embodiment

Below, image processing apparatus, image processing method and the MR imaging apparatus that embodiment relates to (below, suitably be called " MRI system ") is described.

Fig. 1 is the high level block diagram of the MRI system that relates to of expression embodiment.The various interconnected system inscape 20 that MRI system shown in Figure 1 comprises stand 10 (being illustrated by the summary cross section) and interconnects and play a role.At least stand 10 is configured in shielding (shield) chamber (room) usually.A MRI system shape shown in Figure 1 comprises static magnetic field B ₀Magnet 12, G _x, G _y, and G _zThe configuration of the roughly coaxial circles tubular of gradient magnetic field coil (coil) group (set) 14 and RF coil assembly (assembly) 16.Along this horizontal axis of wanting pixel array cylindraceous, exist with the imaging body (volume) 18 shown in the mode of the head that surrounds the subject 9 that supports through subject platform (table) 11 in fact.

MRI systems control division 22 possesses the input/output end port (port) that is connected with display part 24, keyboard (keyboard)/mouse (mouse) 26 and printer (printer) 28.Certainly, display part 24 also can be as also possessing control input, to have a multifarious touch-screen (touch screen).

MRI systems control division 22 is connected with MRI sequence (sequence) control part 30 interfaces (interface).MRI sequence control part 30 control G _x, G _y, and G _zGradient magnetic field coil driver (driver) 32 and RF sending part 34 and transmission/receiving key (swich) 36 (when same RF coil is used to send and receives both sides).Self-evident for those skilled in the art; Through on the health of subject, pasting more than one suitable electrode 8, can be to a side or the both sides of MRI sequence control part 30 output cardiogram (ECG (electrocardiogram)) signals and peripheral pulse ripple synchronizing signal.For use specific data are obtained operator's input that sequential parameter (parameter) defines and system's input a side perhaps both sides generate the MR image, MRI sequence control part 30 is also visited (access) and is used to carry out the program code structure 38 that the MRI data obtain the optimum of sequence.

System's inscape 20 of MRI system comprises the RF acceptance division 40 of data handling part 42 being supplied with input in order to process the view data of finishing dealing with that is used for to display part 24 transmissions.MRI data processing division 42 also constitutes can visit (for example, in order to store a side or the both sides by MR view data that obtains according to the process (process) of embodiment and image reconstruction program code structure 44 and intermediate result data) image reconstruction program code structure 44 and MR image storage part 46.

In addition, in Fig. 1, the general figure of MRI system program/data store 50 is shown.In this MRI system program/data store 50; Institute's stored program code structure (for example; For emphasical edge of image of being rebuild and other details; Cut down a side or the both sides of ring artifact and other pseudo-shadow simultaneously, in order the MRI data that obtained to be carried out Filtering Processing in the k space), be stored in the computer-readable storage medium of the many data processing inscapes that can visit the MRI system.Self-evident for a person skilled in the art; Also can MRI system program/data store 50 be divided into the various computing machines in the process computer of the system 20 that needs such stored program code structure when running well at once; And at least a portion is directly linked (promptly; Replace being stored in MRI systems control division 22 commonly, or directly link).

In fact, self-evident for a person skilled in the art, Fig. 1 is the figure that the general MRI system that increased some changes for the embodiment of stating after can carrying out in this instructions has been simplified very to heavens.The inscape of system can be divided into " square frame (box) " of the set of various logic; Usually; Comprise a plurality of digital signal processors (Digital Signal Processor:DSP), little (micro) processor, towards the treatment circuit of specific use (for example, high-speed a/d conversion, high-speed Fourier (Fourier) conversion, ARRAY PROCESSING use etc.).Usually; If (the perhaps clock of stated number (clock) cycle (cycle)) takes place each clock period, then these treating apparatus are respectively the physical data treatment circuit gets into the clock action type of another physical state from certain physical state " state machine (state machine) ".

In action; Not only treatment circuit (for example; CPU (Central Processing Unit), register (register), impact damper (buffer), computing unit (unit) etc.) physical state change to another from certain clock period progressively clock period; The physical state of the data storage medium that is bonded (for example, magnetic-based storage media position storage part) also in the system acting of that kind from certain state to another state variation.For example; When MR imaging reconstruction process finishes; The memory location of the computer-readable addressable data value of the storage medium of physical property (for example; The performance of the multidigit scale-of-two of pixel value) array from several in advance state (for example, all without exception for " 0 " value or all be " 1 " value) become new state.Under this new state, the physical state of the physical location of such array (for example, pixel (pixel) value) changes between minimum value and maximal value, the phenomenon of the physics of performance real world and situation (for example, the tissue of the subject in the shooting body space).Self-evident for a person skilled in the art, the array as the data value of being stored is represented and constitutes physical property to construct.That is to say; Above-mentioned array constitutes; When being read into successively in the command register and carrying out, produce the particular sequence of operating state, and be formed in the particular configuration of computer-controlled program code mobile the MRI system in by the more than one CPU of system's inscape 20.

The MRI system that following embodiment relates to provides a kind of perhaps method after both sides' the improvement of a side of generation and demonstration of a side who obtains and handle or both sides and MR image of the MRI of carrying out data.

Fig. 2 is the skeleton diagram of the Filtering Processing that relates to of embodiment.The MRI system that following embodiment relates to is in zone, k space (space of the fourier transform of the view data in the real space); For MRI data array (array) filter application; The generation LPF (below; Suitably be called " low pass (low-pass) filtering ") data array of finishing dealing with, bandpass filtering (below; Suitably be called " band logical (band-pass) filtering ") data array of finishing dealing with and high-pass filtering (below, suitably be called " high pass (high-pass) the filtering ") data array of finishing dealing with.In addition, the k spatial data array that the MRI system accomplishes bandpass filtering treatment carries out fourier transform, in image-region, gets its absolute value, and carries out threshold process and emergence (feathering) processing.In addition, the MRI system generates (" gray scale (gray-scale) "), edge mask (edge mask) data array of the successive value with fuzzy (fuzzy) according to the data array of having implemented threshold process and having sprouted wings and handle.On the other hand, the k spatial data array that the MRI system finishes dealing with high-pass filtering carries out fourier transform, and in image-region, carries out threshold process as required, generates sharpening mask data array.And the MRI system multiply by sharpening mask data array with edge mask data array, and its multiplied result and the k spatial data array that low-pass filtering treatment is accomplished are carried out fourier transform and the data array addition of getting its absolute value.Like this, the MRI system that embodiment relates to is in order to represent potential anatomy better, and the ring artifact of gibbs has been cut down in generation and the Filtering Processing of noise artifacts is accomplished image.

Like this, the MRI system that relates to of embodiment for example possesses: application of low pass filters portion, BPF. application portion, Hi-pass filter application portion, edge mask generation portion and synthetic portion.Application of low pass filters portion is for the data that obtain through magnetic resonance imaging, promptly corresponding with the original image k space or the data of image space, application of low-pass filters.BPF. application portion uses BPF. for these data.Hi-pass filter application portion uses Hi-pass filter for these data.Edge mask generation portion generates the edge mask at the edge that has extracted the object in the original image according to the data of having used BPF..Synthetic portion uses the data of having used Hi-pass filter to carry out the mask process based on the edge mask, and uses data of having carried out mask process and the data of having used low-pass filter to synthesize processing.

In addition; In following embodiment; The data in the k space of edge mask generation portion through will having used BPF. are carried out fourier transform, and carry out threshold process and sprout wings and handle for the data of the image space after the fourier transform, thereby generate the edge mask.In addition, synthetic portion will use the data in the k space of Hi-pass filter and carry out fourier transform, and after having carried out threshold process for the data of the image space after the fourier transform, carry out the mask process based on the edge mask.

In addition, in following embodiment, application of low pass filters portion and BPF. application portion use the low-pass filter and the BPF. of the size that is adjusted into the data that are fit to obtain through magnetic resonance imaging.On the other hand, Hi-pass filter application portion uses and is adjusted into the Hi-pass filter that is fit to than the size of the data that obtain through magnetic resonance imaging data array container big, the k space.

In addition, the MRI system that relates to of following embodiment can also possess contrary fourier transform portion.Contrary fourier transform portion is through being collected the data in k space by a plurality of key element coils; And respectively the data in the k space of each key element coil are carried out fourier transform; Generate the data of the image space of each key element coil; And the data of the image space of synthetic each the key element coil that is generated, thereby when generating original image, the data of this original image are carried out the data that contrary fourier transform generates the k space.At this moment, application of low pass filters portion, BPF. application portion and Hi-pass filter application portion are for the data in the k space that generates through contrary fourier transform portion, application of low-pass filters, BPF. and Hi-pass filter respectively.In addition, these each ones for example are installed in MRI data processing division 42, and in MRI data processing division 42, carry out.In addition, these each ones for example also can be installed on the image processing apparatus of other different with the MRI system, in image processing apparatus, carry out.At this moment, required data also can obtain in the MRI system, through the operator, perhaps via network (network) etc., input image processing unit.

The ring artifact that the blocking of k spatial data causes gibbs (below, ring artifact).Especially, when the matrix (matrix) of the view data that is obtained hour, ring artifact is in the execution of MR imaging, should overcome the most ancient and one of problem of difficulty.

In the MR imaging, under the leaning magnetic field, the value that determines to spatiality is converted into the measured value of frequency.When in the MRI data that obtained, when not comprising all high spatial frequency information shown in the anatomy part of image conversion object, in the MR image ring artifact takes place.The width decision of sample window and the shape of Singh's function of image convolution.The convolution of this Singh's function causes uncontinuity or the high-contrast edges of picture characteristics of approximate intensity of the ring artifact of gibbs.When coding (encoding) step (step) for example (is counted after a little while; ≤196); Receive the influence that the time of obtaining or dynamic contrast (contrast) or MRI obtain the restriction on the sequence; Especially, ring artifact makes us uncomfortable along perhaps both sides and occurring of the side of phase encoding (PE (Phase Encoding)) direction and section coding (SE (Slice Encoding)) direction.Ring artifact is along reading (RO (Read Out)) direction, according to the general long window (for example, more than 256) of obtaining, becomes more not noticeable, perhaps, in fact can produce bad influence to obtaining the time, and becoming is not problem.

Pseudo-shadow is observed and reports by the situation of the algorithm process of the ring artifact of flexible Application reduction all the year round in the MRI product according to the direct comparison of the mirage phantom (phantom) and the image in the subject of identical resolution.But,,, also have room for improvement for the picture quality that as a result of produces and the both sides or the side that implement efficient (for example, processing time, complexity etc.) for the trial in the past of that kind.

The LPF in general employed k space and the weighted mean of image space make trickle image detail unintelligible, reduce whole marginal sharpness.On the other hand; For example; Use that very exquisite Gegenbauer rebuilds, spread similar Sigma (Sigma) filtering with the anisotropy of popularizing through " TV (Total Variation) " qualification data extrapolation method and other extrapolation (CT (Computed Tomography))) be complicated; In most cases be repetition, have slack-off tendency, and have the tendency that the piecewise produces the product of the caricature that kind with certain characteristic.In MRI, usually, because the contrast of soft tissue is expected trickle tone, such product can't be seen as MRI naturally.

In following embodiment, proposed to reduce ring artifact and noise artifacts simultaneously, stress that simultaneously the edge is purpose, simple, rapid and suitable easy method.In the method; To original (raw) data in k space (below; Suitably be called " untreatment data "), window function (windowing function), the band that multiply by low pass (below, suitably be called " LP (low-pass) ") respectively logical (below; Suitably be called " BP (band-pass) ") the window function of window function and high pass (below, suitably be called " HP (high-pass) ").In addition, in the untreatment data in k space, comprise the k spatial data that generates from image-region through the contrary fourier transform of two dimension.Then, as shown in Figure 2 through multiply by the product that window function obtains, carry out two-dimentional fourier transform, and in image-region, combine.

Employed window function preferably carries out optimal treatment in the k space.Particularly, preferred window function optionally collects the edge with lower resolution, in threshold process and after sprouting wings, for can process have diffusible diffusion edge mask and middle territory～in control carefully in the scope of spatial frequency in high territory.Purpose at this is, stresses to realize suitable and correct unanimity between scope and the edge mask at the effective edge of HP Laplce sharpening mask.The slow motion of other that image-region convolution action slowly and the Tuscany edge detector institute general edge thinning that uses etc. are a large amount of is that (a) is replaced as k space action of equal value, still (b) avoidance.This is because purpose is not the refinement of pixel edge, but the fuzzy region around this edge.The sharpening mask carries out anti-phase to the LP smoothed image and adds via the edge mask of deriving according to (in the sharpening mask, multiply by each pixel) BP wave filter.

In Fig. 2, illustrate based on have suitably consistent low pass, band is logical and the algorithm of the embodiment of the k spatial window function of high pass.Use the logical frequency field wave filter of band to be considered to important new function based on the method for this embodiment in order to carry out the edge to extract.It is reported that this embodiment is being removed ring artifact and noise from low resolution, intermediate-resolution and high-resolution MR image, stress that simultaneously the edge describes, effective when improving trickle visual conspicuousness.The image quality that the whole picture quality beguine of finishing dealing with obtains according to a large amount of more complicated method shown in the document and some commercially available wave filter bags (package) is excellent, and the more simple algorithm (algorism) that relates to of embodiment moves quite at high speed simultaneously.

As shown in Figure 2, " input picture " I in image space zone ₀In order to generate corresponding input picture k spatial data array K ₀, carry out the contrary fourier transform of two dimension and get final product.Perhaps, certainly, for the untreated N * M array of image data Kraw in the k space of the storage part (for example, can visit the MR image storage part 46 of the MRI data processing division 42 of Fig. 1) of drawing the optimum that is stored in the MRI system, input image data array K ₀Also can obtain sequence and directly obtain identical with the common MRI data of carrying out via system through Fig. 1.In any case; Unnecessary complicated in order to avoid; Also to have carried out that all recovered part Fourier (perhaps " half (half) Fourier (Fourier) ", " AFI ", other) rebuild and the required all basic data processing step of whole untreatment data scopes such as parallel (parallel) imaging reconstruction (for example, SENSE, GRAPPA, SPEEDER, other), the product that has included them in N * M " untreatment data " array is a prerequisite.

No matter according to which path (route), k space input picture complex data array K ₀Can both be included in bigger (P * Q, wherein, the k space of being obtained in the k spatial data array of P＞N and Q＞M) (the untreatment data array K of N * M) _RawIn this k spatial data array, the value of the extraneous all key element of N * M untreatment data all is zero, and (self-evident for a person skilled in the art) is the general knowledge that is called as the MRI of zero filling.Generally the untreated k spatial data zero filling that is obtained is arrived more than " 200% ", this means P >=2N and Q >=2M.For example, have that low-resolution image that little N * M obtains array sizes is handled via zero filling mostly or other interpolation is handled and shown with bigger array of display size P * Q.Under such situation, the input picture array sizes of that kind that the MRI system is stored has been P * Q.

If according to N * M untreatment data array K _RawObtain bigger P * Q input picture array I ₀, or the k space input data array K of bigger P * Q ₀And all pre-treatment steps that use, as be linear the situation of the direct zero filling of the k spatial data that is untreated, then have only the k space input data array K of P * Q size ₀Center N * M position be beyond zero.In addition, from directly being combined in the k space by the signal of the up-to-date employed a plurality of coil key elements of MRI scanner mostly, thus, also there is the case that makes untreated k spatial data size constancyization.

Generally speaking, a plurality of key element signals, combine in image space after individually carrying out two-dimentional fourier transform being in the non-linear process (for example, the combination of general quadratic sum (SOS (Sum Of Squares))).Even when untreated signal is not the signal from a plurality of coil key elements, simple absolute value action (complanation or the process of thresholding that perhaps comprise the immobilized pattern matrix of value) also is non-linear.Under the nonlinear situation of that kind, after the contrary fourier transform of the two dimension in k space, the input k spatial data array K in the outside at original N * M untreatment data position ₀In possibly have little " spilling ", that is, possibly have little, the plural number beyond zero in the outside at the k spatial data position that is positioned at (separately) untreated N * M.

In a word, its that k spatial windowization (windowing) (based on multiplying each other of each pixel of window function) is handled with respect to using, actual (untreated k spatial data array K of N * M) that is obtained according to the embodiment of this LP wave filter and BP wave filter _RawScope.And, in this embodiment, the whole P in the k space * Q input data array K ₀The middle HP wave filter of using, possibly exist in original untreatment data N * M scope but this is outer by high spatial frequency inscape generation, useful of Nonlinear Processing step in the past.Generally speaking, the k space untreatment data array K that is being obtained _RawIn, N with obtain in the MRI data that the significant figure in phase encoding (PE (phase encoded)) the phase encoding increment that direction obtained equates between the sequence.On the other hand, M equates with the significant figure of obtaining in the MRI data between the scanning sequence in reading the sampling of (RO (readout)) frequency coding that direction obtained.As shown in Figure 2, this is at first through low pass, then through with the logical untreated k area of space (that is spatial frequency zone) that carries out Filtering Processing, this N * M array.Hi-pass filter is contained whole k space array K ₀And use.

As illustrating in greater detail; In order to meet in fact (as mentioned; After the treatment step that has carried out part Fourier and side by side imaging, comprise the recovered data scope) the original array of image data that obtains of N * M data value, perhaps; To the HP wave filter, meet bigger input data array K ₀(usually hope be strict, even but not strict), low pass, band is logical and filtering window high pass all is adjusted into the size of regulation.As Fig. 2 (in fact as 3D shape) was roughly illustrated, the wave filter window function met the original k area of space that carry out Filtering Processing or imports the scope one side of data array.In by the data array of Filtering Processing, there is the array (that is, N=M, perhaps P=Q) of " square ", then, make filter kernel meet square or circular xsect (along the tangent situation of wave filter window function axis of Fig. 2) basically.On the other hand, N ≠ M, perhaps under the situation of P ≠ Q, then, and the xsect that low-pass filter is rectangle basically, on the other hand, band passes to and the wave filter of the high pass xsect of ovalize basically.

It should be noted especially; In embodiment; The BPF. nuclear (kernel) in k space in the Tuscany type edge extracting filtering of prior art, generally uses both sides tropism's filtering core of two quadratures in image-region with respect to extracting the edge effectively in whole directions; Afterwards, this product is combined in image-region.

Original array of image data K as shown in Figure 2, as to be imported ₀Use LPF window (W respectively _LP), bandpass filtering window (W _BP) and high-pass filtering window (W _HP) carry out high pass, be with logical and the substantial low pass filtering processing.Original view data K ₀In N * M or P * Q array correspondence key element for generate with low pass, band is logical and high pass is carried out the k spatial data array K of the P * Q after the Filtering Processing _LP, K _BP, and K _BP, respectively each pixel multiply by the suitable weighting coefficient that Filtering Processing is accomplished the correspondence of window function (that is " highly " of 3D shape shown in Figure 2).Then; The Filtering Processing that has obtained LP and HP after the fourier transform accomplish product absolute value (magnitudes), with fourier transform after the Filtering Processing of HP accomplish the real part of product after, these Filtering Processing are accomplished data arrays and are reflected in corresponding respectively P * Q digital picture area array I so that these Filtering Processing are accomplished data arrays respectively _LP, I _BP, and I _HPMode, carry out two-dimentional fourier transform (2D FT) respectively.As shown in Figure 2, low-pass pictures area array I _LPRepresent original raw image I ₀(under the situation of its actual generation) or its effective homologue (do not use pattern matrix I in image-region ₀, and use the untreated view data K that is obtained ₀Situation under) smoothed version.Image-region array I after the bandpass filtering treatment _BPThe edge that expression is extracted, on the other hand, the image-region array I after high-pass filtering is handled _HPExpression sharpening mask.This sharpening mask dashes and overshoot under generating at the [that is untreated, and increases the consciousness acutance of image, and its result plays the effect of the consciousness contrast that increases the edge.

Self-evident for a person skilled in the art, simple and clear the multiplying each other in Fourier frequency zone (that is k space) is equivalent to the convolution in the actual spatial image zone.But, in addition, as know, the convolution process is complicated, and carries out than spended time.In addition, in image-region, be difficult to obtain the filtering convolution kernel in optimal image zone of the employed k of the being equivalent to spatial window of the embodiment function in k space sometimes.Because the derivation of appropriate image zone convolution kernel generally defines Laplce (Laplacian) type sharpening mask and uses, and therefore, for the high-pass filtering process, we can say more and possibly carry out.But (owing to need the unnecessary strategies such as reduction of processing speed, clearly, not being preferred embodiment although be considered to) can be carried out all filter functions at image-region at least in logic.

As shown in Figure 2, the image-region array I at the edge that expression is extracted _BPFor (" gray scale ") P * Q edge mask array I that generates successive value _EM, and the process of carrying out thresholding and emergence.Then, as Fig. 2 202 shown in, P * Q sharpening mask (if hope to obtain optimum achievement, preferably after the soft-thresholdization) to each pixel, multiply by P * Q edge mask array.Then, in order to obtain final P * Q image-region array I _F, its multiplied result is to each pixel, add Fig. 2 204 shown in P * Q low pass smoothing pattern matrix ILP.

Fig. 3 representes the applied computer program code structure of embodiment shown in Figure 2.In at least one data processing division that suitably constitutes (for example, shown in Figure 1 MRI data processing division 42), use this computer program code structure.In addition, this computer program code structure is used the input picture array I that is stored in the intrasystem suitable storage part (for example, carrying out data processing division 42 and the MR image storage part 46 of communicating by letter) of MRI shown in Figure 1 ₀And input data array K ₀Both sides or a side.

In S300, data processing division begins the filter module of the ring artifact/noise artifacts of gibbs.In S302, data processing division is the k spatial data array K of untreatment data (raw data) in order to carry out data processing ₀Generation and the both sides that accept or a side (for example, via suitable operation-interface).In step S304, shown in dotted line, (for example, because the MRI system data is obtained the workflow design of processing) when by obtaining data being provided the data processing time spent, as the input to filtering, imported data array K ₀In order to generate the k spatial data array K of suitable substance P * Q ₀, also can be with the raw image array I that had both deposited in the image-region ₀Carry out the contrary fourier transform (and zero filling at random) of two dimension.

In addition, the input data array K of the object of Filtering Processing ₀Sometimes being the k spatial data itself that arrives through the MRI systematic collection, is to carry out the k spatial data that contrary fourier transform generates according to view data sometimes.Situation to the latter describes, and the MRI system possesses the array coil with a plurality of key element coils sometimes, collects data through array coil.At this moment; Through collect the data in k space by a plurality of key element coils; And respectively the data in the k space of each key element coil are carried out fourier transform; Generate the data of the image space of each key element coil, and the data of the image space of synthetic each the key element coil that is generated, thereby original image generated.At this moment, the MRI system also can generate the input data array K of the object of Filtering Processing through original image (composograph) being carried out contrary fourier transform ₀And the shortening data processing time, perhaps also can generate the input data array K of the object of Filtering Processing through respectively the view data of each key element coil being carried out contrary fourier transform ₀, and reflect the influence of each view data of each key element coil.In addition; Even passing through under the situation of k spatial data itself that the MRI systematic collection arrives with the former as the object of Filtering Processing; Also can implement Filtering Processing, also can implement Filtering Processing the k spatial data separately of each key element coil for the generated data of the k spatial data that synthesizes each key element coil.

In S306, in order to generate the k space array KLP of P * Q that low-pass filtering treatment accomplishes, data processing division is for the k spatial data array K of P * Q ₀, application of low-pass filters windowed (windowing) (multiply by window function W _LP).Window function W _LPThe four corner ground of containing the input k spatial data of P * Q defines, but window function W _LPZero beyond scope preferably strictly, or consistent with the valid window beggar array of N * M in fact.Likewise, in S308, in order to generate the k space array K of P * Q that bandpass filtering treatment accomplishes _BP, data processing division is for the k spatial data array K of P * Q ₀, use the BPF. windowed and (multiply by window function W _BP).Window function W _BPZero beyond scope strictly, or consistent with the valid window beggar array of N * M in fact.In addition, in S310, in order to generate the k space array K of P * Q that high-pass filtering finishes dealing with _HP, data processing division is for the k spatial data array K of P * Q ₀, application strictly, perhaps the Hi-pass filter windowed of unanimity (multiply by window function W in fact _HP).Certainly, the order of this Filtering Processing can at random change as required.

In S312, the k space array that data processing division is accomplished Filtering Processing, the two-dimentional fourier transform of application of optimal.In addition, then, data processing division is in image-region, in order to generate image-region P * Q array I that Filtering Processing is accomplished _LPAnd I _BP, get the absolute value (magnitudes) of the k space array that Filtering Processing accomplishes.In addition, data processing division is in order to generate image-region P * Q array I that Filtering Processing is accomplished _BP, from being to obtain real part the product of fourier transform of plural number.Thereby, the image-region array I that its result generates _LPAnd I _BPAll delimited and be just (all array key element values are all more than zero).To this, the image-region array I that the result generates _HP, the HP sharpening mask of soft-threshold before handling be included as certainly near zero the mean value just and negative both sides' value.

Then, in S314, data processing division is increased to suitable level for " through the scope " that makes the edge mask, for the image-region array I of bandpass filtering treatment completion _BPCarry out suitable hard-threshold and handle, generate P * Q threshold value edge mask array I _BPTHAt this moment, as threshold value, data processing division uses according to being untreated edge mask I _BPPeak value and mean value and the numerical value that automatically calculates.At this, it is benchmark that hard (hard) threshold process is called with the threshold value, and the processing that value is all reduced for example, are called for the value more than the threshold value and value are changed to " threshold value ", value are changed to " zero " such processing for the value below the threshold value.For example, in step S314, if the above value of threshold value, then data processing division carries out value is changed to the processing of " threshold value ".In addition, soft (soft) threshold process is the threshold process beyond hard-threshold is handled.

In S316, in order to generate the complex data array K of corresponding P * Q _BPTH, data processing division is for threshold value edge mask array I _BPTH, use to the contrary fourier transform of the two dimension in k space.In S318, accomplish the complex data array K of k spatial edge mask for the low-pass filtering treatment that generates P * Q _EM, data processing division is for this K _BPTH, multiply by P * Q gauss low frequency filter window function W to each pixel _GaussianIn S320, data processing division passes through for k space array K _EMUse two-dimentional fourier transform and turn back to image space, and get its absolute value, thereby obtain the emergence blurred picture edges of regions mask array IEM of P * Q.

In S322, in order to obtain soft-threshold image-region array I _HPST, as required, data processing division can adopt for the high-pass filtering of the P * Q pattern matrix I that finishes dealing with _HPSimple soft-threshold handle.Nonlinear soft-threshold is handled through the value of making to zero side minimizing, thereby only to less than very little threshold value T _STInitial point around data exert an influence.Thereby, handle according to soft-threshold, can reduce the random noise and the gibbs ring that possibly in the sharpening mask, take place.

In S324, the edge mask array I that data processing division generates for the bandpass filtering through P * Q _EM, multiply by through the soft-threshold of P * Q and handle the sharpening mask array I that high-pass filtering generates _HPST, then, with the smoothed image area array I of this product with the LPF generation of passing through P * Q _LPFinal image I is formed in addition _FAll calculating is all carried out in image-region according to each pixel.In step 326, after the filtering of this module that is through with, carry out return (return) to the processor control of calling (calling) program.

Self-evident, with two-dimensional array embodiment has been described relatively, but 3-D view can (for example, through Filtering Processing is carried out in the adjacent section of said three-dimensional body continuously) carry out Filtering Processing too.

Below, explain two dimension (2D) type that comprises the practicality of basic one dimension (1D) type with can use the time, the window function nuclear in preferred, several k space now.

[I a. low pass (LP) window function: basic 1D formulism]

N=k obtains in the space size of data (in 2D, being N and M)

N=k space pixel index (from 1 to N) (among 2D, is n _xAnd n _y)

R, w, a, p: preferred (upper) value and the parameter of the 2nd preferred value for as marking, providing

[mathematical formulae 1]

$W{(n,N)}_{\mathrm{HBRR}\_1}=2\sin \left(\frac{\mathrm{\πn}}{N}\right)-{\sin }^2\left(\frac{\mathrm{\πn}}{N}\right)$

[mathematical formulae 2]

$W{(n,N)}_{\mathrm{Fermi}}=\frac{1}{1+\exp \left(\frac{|n-N/2|-r}{w}\right)}$

R=(3/8) N, w=(10/128) N or r=(3/8) N, w=(15/128) N

[mathematical formulae 3]

$W{(n,N)}_{\mathrm{Exponential}}=\exp [-\mathrm{\α}{\left(\frac{|n-\frac{N}{2}|}{N}\right)}^p]$

α=32, p=4 or α=96, p=6

[Ib. low pass window function: basic " 1Dx1D " type 2D formulism]

The magnitude range that window is contained the untreatment data of the N * M that is obtained delimited.When N=M, according to all values that are set at zero the scope and the outside with the situation of 1D under the parameter value of the identical setting of the value that provides, can become square or rectangle.

[mathematical formulae 4]

W _2D(n _x，N，n _y，N _y，M)＝W(n _x，N)W(n _y，M)

[mathematical formulae 5]

$W{(n_x,N,n_y,M)}_{\mathrm{HBRR}\_1}=[2\sin \left(\frac{\mathrm{\π}{n}_x}{N}\right)-{\sin }^2\left(\frac{{\mathrm{\πn}}_x}{N}\right)][2\sin \left(\frac{{\mathrm{\πn}}_y}{M}\right)-{\sin }^2\left(\frac{{\mathrm{\πn}}_y}{M}\right)]$

[mathematical formulae 6]

$W{(n_x,N,n_y,N_y)}_{\mathrm{Fermi}}=\left[\frac{1}{1+\exp \left(\frac{|n_x-N_x/2|-r}{w}\right)}\right]\left[\frac{1}{1+\exp \left(\frac{|n_y-N_y/2|-r}{w}\right)}\right]$

[mathematical formulae 7]

$W{(n_x,N_x,n_y,N_y)}_{\mathrm{Expponential}}=\exp [-\mathrm{\α}{\left(\frac{|n_x-\frac{N_x}{2}|}{N_x}\right)}^p]\exp [-\mathrm{\α}{\left(\frac{|n_y-\frac{N_y}{2}|}{N_y}\right)}^p]$

[logical (BP) window function of II. band]

" γ variable " function is used as the BP window function.

[mathematical formulae 8]

W _BP(|k|)＝|k| ^αe- ^|k|β

Wherein, α=6 and β=3, and | k| is normalized to | k|=25.6K _n/ C.Wherein, K _nBegin " k space length " from initial point, C is along " effectively by scope " according to the specific direction of untreatment data array sizes.The magnitude range of the untreatment data of the N * M that only obtains through containing delimited, and when N=M (perhaps ε=1), according to being set at zero scope and all values in the outside, window can become circular or oval.

In one embodiment, W (n _x, n _y)=t ⁶Exp [t/3]

Wherein, t=25.6K _n/ C

K _n＝√(X ₀ ²+Y ₀ ²)

C＝(N _x/2)K _n/(X ₀ ²+ε ²Y ₀ ²)

ε＝(N/M)

X ₀＝n _x-(N/2)-1

Y ₀＝n _y-(M/2)-1

Annotate: the BP window function is at K _nHad 0/0 specificity at=0 o'clock.Wherein, need specificity be set at 0 clearly.In addition, the peak value of BP window function should be normalized to WBP=1/|max (WBP) | that kind.

[III. high pass (HP) window function]

K Space H P window function is as in order to carry out linear space constant (LSI (Linear Spatially Invariant)) smoothing filtering, and the two dimension of the LSI Laplce sharpening image-region convolution kernel corresponding with " intensity " that defined is processed against the quantity of fourier transform.At this, exist for the sharpening nuclear after several normalization of 1.5,2.0,2.5 and 3.0 " LSI intensity ", initial set value preferably is set to 2.5.The intensity number is high more, has used after the windowed, and it is sharp-pointed more that sharpening becomes.

[mathematical formulae 9]

$\mathrm{Kernel}\left(\mathrm{LSI\; strength}1.5\right)=\begin{array}{ccc}|-0.0568& -0.0947& -0.0568|\\ |-0.0947& 0.6061& -0.0947|\\ |-0.0568& -0.0947& -0.0568|\end{array}$

[mathematical formulae 10]

$\mathrm{Kernel}(\mathrm{LSI\; strength}.2.0)=\begin{array}{ccc}|-0.0915& -0.1524& -0.0915|\\ |-0.1524& 0.9756& -0.1524|\\ |-0.0915& -0.1524& -0.0915|\end{array}$

[mathematical formulae 11]

$\mathrm{Kernel}(\mathrm{LSI\; strength}.2.5)=\begin{array}{ccc}|-0.1148& -0.1913& -0.1148|\\ |-0.1913& 1.2245& -0.1913|\\ |-0.1148& -0.1913& -0.1148|\end{array}$

[mathematical formulae 12]

$\mathrm{Kernel}\left(\mathrm{LSI\; strength}3.0\right)=\begin{array}{ccc}|-0.1316& -0.2193& -0.1316|\\ |-0.2193& 1.4035& -0.2193|\\ |-0.1316& -0.2193& -0.1316|\end{array}$

If select image-region Laplce sharpening convolution kernel, then corresponding k Space H P window function is processed as following.

[mathematical formulae 13]

W _HP＝|FT(Kernel _Lapiacian ^*)|

This mathematical formulae is contained whole input k spatial data K ₀Scope defines.Wherein, Kernel _LaPlacian ^*As above-mentioned, be (that is, the 3x3 zone at the center of removing is obtained the position note and do 0) corresponding 3x3 laplacian image zone sharpening convolution kernel that k space container dimensional with P * Q carries out zero filling.

[IV.HP sharpening mask soft-thresholdization formulism]

The input HP sharpening mask that obtains when the high-pass filtering according to the input picture after the normalization is I _HP, and output soft-threshold sharpening mask be I _HPSTThe time, when hoping optimum achievement, can carry out soft-thresholdization.

| I _HP| under the situation less than t,

[mathematical formulae 14]

IHPST＝I _HP×(|I _HP|/t?)^p

In embodiment, under the situation of t=0.02 and P=1, this mathematical formulae is reduced to sharpening intensity below the threshold value t gradually.

[V. edge mask thresholding formulism]

BP derives the automatic adaptation hard-thresholdization of edge mask and can as following, carry out.

[mathematical formulae 15]

I _BPTH＝min(I _BP，mean(I _BP)/h)^2

Wherein, h is redefined for 0.36 constant in the embodiment of optimum.On the other hand, mean (I _BP)/h item is represented automatic threshold.Through changing the h value, carry out the control of the reach of edge mask.LSI HP filter strength is controlled in this control, controls the scope and the spatial dimension of the integral body of input picture edge sharpening on every side simultaneously.After this control, continuation is reduced to zero step and normalization step with lower value.

I _BPTH＝I _BPTH-min(I _BPTH)

I _BPTH＝I _BPTH/max(|I _BPTH|)

[mask emergence k space, VI. edge low pass formulism]

Threshold value edge mask is sprouted wings (from I _BPTHTo I _EM) emergence (edge softening, perhaps edge level) use low pass Gauss window to be performed in the k space.

[mathematical formulae 16]

W _Gaussian＝exp[-(n _x-(N/2))/(σ*(N/2))]exp[-(n _y-(M/2))/(σ*(M/2))]

Wherein, in the embodiment of optimum, σ=σ ₀* (min (N, M)/128) and σ ₀=0.16.The edge mask I that is generated after the two dimension fourier transform _EM| FT ^-1[W _Gaussian* FT (I _BPTH)] | be normalized to I once more _EM=I _EM/ max (| I _EM|).

In embodiment, there are 3 main inscapes.(1) as the version I of the level (smoothing) of image _LPIt realizes the basic reduction of ring artifact and noise artifacts.(2) sharpening mask I _HPSTIts accentuated edges acutance and small detail conspicuousness.(3) (" gray scale ") edge mask I of the successive value of fuzzy (completion of sprouting wings) _EMIt is in order to generate final product, all sidedly, or partly use, or do not use sharpening mask I _HPSTThe edge around, select I _LPA part.

In order to carry out the k low pass spatial filtering, we can say the window function that the most suitable use is suitable.This is because this function lowers for the ring artifact of equivalent, is easy to keep the image detail (level of integral body still less) of Duoing than other window function.

Image sharpening method in the image-region can roughly be simplified to the convolution of in image-region, using Laplce's sharpening nuclear of equal value, but the processing speed change is very slow through image-region laplace kernel convolution being replaced as equal k spatial window process.

Under the situation of using incorrect threshold value, do not keep edge or details (for example, threshold value is crossed when hanging down) sometimes, perhaps some ring artifacts and noise are not removed (when for example, threshold value is too high) from product sometimes.Though the edge mask threshold value of can manually set, inching is optimum; But in most cases; The edge mask threshold value of robotization derives algorithm through deriving the threshold value that very is similar to manual setting value; Come reasonably well to play a role, especially, stress and the main ring artifact of the image of direction of principal axis head and the sagittal cervical vertebra of T2 after stressing lowers target and comes correctly to play a role for the T1 that comprises low resolution and intermediate-resolution.

Self-evident; Flame Image Process/filtering method based on embodiment is cut down the ring artifact of gibbs and some noises through low pass (LP) spatial filter; On the other hand, used the edge mask of threshold value, kept and stress the details of imaging and the delimitation at edge through use.In embodiment, high pass (HP) wave filter that the edge mask is made into optionally to use through containing " transparent " zone is stressed details and edge.The characteristic that 1 important situation of leading to success exists with ... the wave filter window function that carefully makes LP, HP and edge extracting BP is consistent, thus, and those functions play a role continuously each other (that is, replenishing each other) with effective size and intensity.

Embodiment exists with ... the correct specification of the original k spatial data array size that is obtained.For example, in N * M array, specify improperly at untreatment data the PE matrix size (for example, in the time of N), the deterioration of image quality that the result generates.

In a word, high speed low pass k spatial filtering (windowed) method with the parameter after the optimization at first is used for cutting down according to data truncation the ring artifact of gibbs, and cuts down the noise of the integral body in the image.The logical edge detection filter (window) of simple and comprehensive k space band at a high speed is used to generate the edge mask at thresholding and after sprouting wings at image-region.Edge in the original image from the edge mask Zone Full accomplish to low-pass filtering treatment that image is optionally contrary to be pasted.

Product also can carry out two-dimentional fourier transform in comprising the k spatial dimension container extrapolation k spatial data that generates naturally through above-mentioned process, after the expansion.Then, this product uses the linear transformation function (single or a plurality of) of 1D or 2D to mix with most of maintained original outside k space segment of obtaining k space untreatment data.Then; This product is in order for example to generate; Fan etc. (Amartur et al.), 1991 and 1991; And Constable etc. (Constable et al.), 1991, the method for being narrated similarly other Filtering Processing is accomplished image artifacts, carries out two dimension against fourier transform and return to image-region.Its result, this product generates the impression of whole subjectivity acutance, more not whole, and spectators can observe more naturally.

For the ring that meets gibbs is not the large-scale purposes such as both sides or a side that main problem, pure (high-resolution) image noise are removed and contained the general image beyond the MRI inspection area, the parameter that can individually regulate the window function in low pass, high pass and rim detection k space.

Perhaps, Fourier space two dimension LPF can be replaced as image-region smoothing (weighted mean) convolution.But k spatial window method is more effective on computers, in addition, because all MRI untreatment datas are all obtained in the k space, therefore, is easy to meet better the character of MR data stream.In addition, the filtering of Fourier space two-dimensional high-pass can be replaced as image-region Laplce convolution or non-sharpening protected type sharpening convolution and calculating.But though repeat, in most cases, k spatial window method is quicker, and the flow process of work meets the character of MR data stream better.

In addition, the filtering of Fourier space two dimension edge extracting does not have the refinement at edge and the common final step of bindingization, can be replaced as Tuscany type image-region edge extracting.But though repeat again once, k spatial window method is quicker, and is more effective on computers, meets the character of MR data stream better.Even its performance has the tendency of the Effects of Noise that receives input picture more easily.For image, we can say also preferred Tuscany type edge extracting with low-down S (Signal)/N (Noise) ratio.Tuscany type edge extracting still is 2 k spatial window steps, in addition, makes up as the image-region of product and to carry out.

Perhaps, as the final step that mask generates, the Fourier space two dimension LPF that is used for the edge mask is sprouted wings can be replaced as the image-region smoothing, but space-spatial window method more at a high speed and more efficient on computers.

In embodiment, in order to overcome edge and the fuzzy some problems of details in the LPF method, stress edge and details in the past, the ring artifact attenuating and the noise that enroll gibbs simultaneously continuously reduce.Different with Sigma's filtering and anisotropy method of diffusion, embodiment is formed in the image of nature that non-edge keeps the beauty of trickle level.In other words, in embodiment, the MR image looks and is not " distinguishing definitely " perhaps " as CT ", and looks as the appearance of the reality of its MR inspection area.Through improving the conspicuousness of real edge and low contrast small detail, thereby the major part of removing the ring artifact of gibbs and reduces the level of the integral body of image noise, the doctor can be better with pathology, infraction and reflection unusually.

Embodiment has proposed, be used to reduce gibbs ring artifact and noise artifacts, optimum or near definition optimum, two-dimentional Fourier space low pass window function and be used for to the details of the object form of the MRI of low resolution and intermediate-resolution image stress and edge sharpening, optimum or near the definition of optimum, fixing Fourier space two-dimensional high-pass spectral window function.

Method that this instructions is put down in writing and system can with meet more well than the embodiment of being put down in writing specific Flame Image Process purpose, for example possess the various generation images that create various " tastes " low pass arbitrarily, band logical (directivity, both sides tropism, multidirectional and comprehensive) and high pass window function, various other forms specialize.

Basic process also can comprise following combination step.In combination step, as input k spatial data K ₀The improvement version, input k spatial data K ₀A part be replaced as final image I _FCarry out the data (data that completely or partially produce) after the contrary fourier transform.In addition, can select alternatively is optionally to replace, still partly displacement.In addition, final image I _FMathematics property ground through with (I _EM) * I _Sharpen+ (1-I _EM) * I _LP, perhaps (I _EM) * I _Sharpen+ (1-I _EM) * I _LPRoughly the same combined method shows.Final image I _FBe input picture I ₀Sharpening version I _Sharpen, successive value gray scale edge mask image I _EM, and low-pass filtering treatment accomplish image I _LPCombination.K spatial data after the improvement can be in the outside of the k spatial data scope of the original N that obtains * M size, comprises the inscape of the high spatial frequency after the reinforcement.Its extrapolation naturally between above-mentioned filtering.

Image processing apparatus, image processing method and MR imaging apparatus according to above-described at least one embodiment can improve image quality.

Although clear several embodiments of the present invention, but these embodiments are to point out as an example, are not intended to limit scope of the present invention.These embodiments can be implemented with other various forms, in the scope of the main idea that does not break away from invention, can carry out various omissions, displacement, change.These embodiments or its distortion be contained in scope of invention or main idea in the same, be contained in the scope of invention that claims put down in writing and equalization thereof.

Claims (8)

1. image processing apparatus is characterized in that possessing:

Application of low pass filters portion is to that obtain through magnetic resonance imaging, corresponding with the original image k space or the data application of low-pass filters of image space;

BPF. application portion uses BPF. to above-mentioned data;

Hi-pass filter application portion uses Hi-pass filter to above-mentioned data;

Edge mask generation portion according to having used the data that above-mentioned BPF. gets, generates the edge that extracts the object in the above-mentioned original image and the edge mask that obtains;

Synthetic portion, use used above-mentioned Hi-pass filter and data carry out mask process based on above-mentioned edge mask, and use carried out above-mentioned mask process data with having used above-mentioned low-pass filter data synthesize processing.

2. image processing apparatus according to claim 1 is characterized in that,

Above-mentioned edge mask generation portion carries out fourier transform through the data in the k space that will use above-mentioned BPF. and get, and the data of the image space after the fourier transform are carried out threshold process and sprouted wings and handle, thereby generates above-mentioned edge mask.

3. image processing apparatus according to claim 1 and 2 is characterized in that,

Above-mentioned synthetic portion will use above-mentioned Hi-pass filter and the data in k space carry out fourier transform, and the data of the image space after the fourier transform are carried out threshold process, carry out mask process then based on above-mentioned edge mask.

4. according to each described image processing apparatus in the claim 1～3, it is characterized in that,

Above-mentioned application of low pass filters portion and above-mentioned BPF. application portion use the above-mentioned low-pass filter and the above-mentioned BPF. of the size that is adjusted to the data that are fit to obtain through magnetic resonance imaging,

Above-mentioned Hi-pass filter application portion uses and is adjusted to the above-mentioned Hi-pass filter that is fit to than the size of the data array container through the big k space of data that above-mentioned magnetic resonance imaging obtained.

5. according to each described image processing apparatus in the claim 1～4, it is characterized in that,

This image processing apparatus also possesses contrary fourier transform portion; Should be when having generated above-mentioned original image against fourier transform portion; Through collect the data in k space by a plurality of key element coils, respectively the data in the k space of each key element coil are carried out the data that fourier transform generates the image space of each key element coil, and the data of the image space of synthetic each the key element coil that is generated; Thereby the data of this original image are carried out the data that contrary fourier transform generates the k space

Above-mentioned application of low pass filters portion, above-mentioned BPF. application portion and above-mentioned Hi-pass filter application portion are to the data in the k space that generates through above-mentioned contrary fourier transform portion, application of low-pass filters, BPF. and Hi-pass filter respectively.

6. image processing apparatus is characterized in that possessing:

Application of low pass filters portion is to the raw data application of low-pass filters in that obtain through magnetic resonance imaging, corresponding with original image k space;

BPF. application portion uses BPF. to above-mentioned raw data;

Hi-pass filter application portion uses Hi-pass filter to above-mentioned raw data;

Edge mask generation portion according to having used the data that above-mentioned BPF. gets, generates the edge that extracts the object in the above-mentioned original image and the edge mask that obtains;

Synthetic portion, use used above-mentioned Hi-pass filter and data carry out mask process based on above-mentioned edge mask, and use carried out above-mentioned mask process data with having used above-mentioned low-pass filter data synthesize processing.

7. an image processing method is carried out by image processing apparatus, it is characterized in that, comprises:

The application of low pass filters step is to that obtain through magnetic resonance imaging, corresponding with the original image k space or the data application of low-pass filters of image space;

The BPF. applying step is used BPF. to above-mentioned data;

The Hi-pass filter applying step is used Hi-pass filter to above-mentioned data;

The edge mask generates step, according to having used the data that above-mentioned BPF. gets, generates the edge that extracts the object in the above-mentioned original image and the edge mask that obtains;

Synthesis step, use used above-mentioned Hi-pass filter and data carry out mask process based on above-mentioned edge mask, and use carried out above-mentioned mask process data with having used above-mentioned low-pass filter data synthesize processing.

8. MR imaging apparatus is characterized in that possessing:

Application of low pass filters portion is to that obtain through magnetic resonance imaging, corresponding with the original image k space or the data application of low-pass filters of image space;

BPF. application portion uses BPF. to above-mentioned data;

Hi-pass filter application portion uses Hi-pass filter to above-mentioned data;

Edge mask generation portion according to having used the data that above-mentioned BPF. gets, generates the edge that extracts the object in the above-mentioned original image and the edge mask that obtains;

Synthetic portion, use used above-mentioned Hi-pass filter and data carry out mask process based on above-mentioned edge mask, and use carried out above-mentioned mask process data with having used above-mentioned low-pass filter data synthesize processing.

Images