The content of the invention
The technical problems to be solved by the invention are to provide a kind of part tissue of eye localization method, can remove doctor's hand fromThe work of dynamic positioning, accurate positioning, computing is quick and stability is good.
The present invention solve the technical scheme that is used of above-mentioned technical problem for:A kind of localization method of part tissue of eye,Comprise the following steps：
The pending head image of input, is divided into air by the head image according to pixel gray value and coversFilm and skull mask；
Based on morphological feature, the convex closure profile of the skull mask is obtained, and generates skull convex closure mask；
According to the skull convex closure mask, internal cavity mask is determined in the air mask；
Merge the skull mask and the internal cavity mask, form skull filling mask；
The skull filling mask is subtracted using the skull convex closure mask, the position of part tissue of eye is obtained.
Further, in addition to, extract the skull convex closure mask and subtract the three-dimensional that the skull fills maskConnected domain, and further determine that according to the form in three-dimensional communication domain the position of the part tissue of eye.
Further, the form according to three-dimensional communication domain further determines that the position tool of the part tissue of eyeBody is：According to eye socket in the morphological feature of three dimensions, circular cone Measure Operation is carried out to the three-dimensional communication domain,Screen the three-dimensional communication domain where part tissue of eye.
Further, the detailed process that the circular cone is estimated is：
Obtain the detection volume V in the three-dimensional communication domain；
The floor space S in the three-dimensional communication domain and height h of the connected domain is obtained, then the three-dimensional communicationThe calculating volume V'=S × h/3 in domain；
Calculate the similar of the detection volume V in the three-dimensional communication domain and calculating volume V' in the three-dimensional communication domainDegree, if the similarity is in setting range, judges the position in the three-dimensional communication domain as part tissue of eyePosition；Otherwise, it is determined that the position in the three-dimensional communication domain is not the position of part tissue of eye.
Further, the skull convex closure mask is the minimum external convex closure comprising the skull mask profile.
Further, the external air mask is located at outside the skull convex closure mask, the internal cavityMask is located inside the skull convex closure mask.
Further, in addition to through merging skull mask and internal cavity mask make morphological dilations andFilling is handled.
Further, the part tissue of eye position is additionally included in detect using many radius spheroids or blood vessel detectionMethod carries out fine segmentation, obtains in eyeball mask, crystalline lens mask, extraocular muscle mask, optic nerve maskAt least one.
The present invention also proposes a kind of positioner of part tissue of eye, including：
Image input module, the pending medical image for obtaining, the medical image is head image,And the head image is divided into by air mask and skull mask according to pixel gray value；
Part tissue of eye locating module, for based on morphological feature, obtaining the convex closure profile of the skull mask,Generate skull convex closure mask；According to the skull convex closure mask, internal cavity is determined in the air maskMask；Merge the skull mask and the internal cavity mask, form skull filling mask；Using describedSkull convex closure mask subtracts the skull filling mask, obtains the position of part tissue of eye；
Split module, for using the detection of many radius spheroids or blood vessel detection method in the part tissue of eye positionCarry out fine segmentation and obtain soft tissue, the soft tissue is eyeball, crystalline lens, extraocular muscle, optic nerveAt least one of.
Further, in addition to head image acquisition module, the head image acquisition module is used to obtain CTHead image or MR head images.
Compared with prior art, the advantage of the invention is that：It is crinial bone to take full advantage of eye socket in positioningThe morphological feature of most deep two depressions in surface, and all eye soft tissues are all located at the priori in eye socketInformation, by carrying out convex closure operation to skull mask, generates skull convex closure mask, utilizes skull convex closure maskThe skull mask of filling internal cavities is subtracted, so that the position of eye soft tissue is obtained, without atlas registration,The easy and effective and speed of service is fast,；According to the form of eye soft tissue, the method estimated using circular cone is furtherThe accuracy of position location is determined, the stability of detection method is improved.
Embodiment
The above objects, features and advantages of the present invention is set to become apparent understandable, it is below in conjunction with the accompanying drawings and realExample is applied to be described in detail the embodiment of the present invention.
The wall of eye socket four constitutes the deep chamber of taper bone shape, inside also have eyeball (vitreum and crystalline lens), optic nerve,The soft tissue such as extraocular muscle and fat or content.The imageological examinations such as CT, MRI can usually use eye groupKnit segmentation.Common image segmentation generally includes two parts：Coarse segmentation and fine segmentation.The former is typically alsoFirst positioning including treating segmentation organ, is the basis for carrying out Accurate Segmentation.Due to the difference of human body voxel densitiesDifferent, image can slightly have difference formed by different body parts.By taking CT images as an example, the CT of pixel in imageIt is worth size to depend in voxel (the corresponding rectangle institutional framework of each pixel in the CT image scannings visual field) wrappingThe institutional framework contained is to the number of X-ray absorption, and voxel averag density is bigger, and attenuation coefficient is higher；DensitySmaller, attenuation coefficient is smaller.For specific institutional framework, its density collects difference less in Different Individual,Corresponding CT values are relatively fixed within the specific limits.Typical cerebral CT image can be divided into two big regions：I.e.Region CT values one outside region outside the circular scan visual field in image center and scan vision, scan visionAs be：- 2048HU (Heng Shi units), only the image in scan vision is only the influence in tested region, justFor cerebral CT image, in scan vision in addition to the head image of examinee, in addition to the sky around headEach main compositing area CT values substantially distribution is in gas, scan vision：Head surrounding air CT values- 1000HU, skull+200~+1000HU, cerebrospinal fluid -20~+20HU, brain tissue+20~+40HU.
The present invention proposes a kind of localization method of part tissue of eye, and it is based on three dimensional morphology feature, extracts headSkull and air position part tissue of eye position jointly, while the anatomy priori based on part tissue of eye periphery is knownKnow, to obtaining part tissue of eye position further screening and fine segmentation, concretely comprise the following steps：
S10, the pending head image of input, air is divided into according to pixel gray value by head medicine imageMask and skull mask.Head image can be CT images or MRI image, and the present embodiment is with CT headsImage is that the gray value of each pixel in process object, CT images is generally represented with CT values, and CT valuesThe pad value after X-ray is absorbed through tissue is characterized, the CT values of something are equal to the decay system of the materialThe difference of the attenuation coefficient of number and water, then with being multiplied by 1000 after the ratio between the attenuation coefficient of water.The CT of different tissuesValue is different, wherein the CT value highests of skull, is 1000HU, the CT values scope of soft tissue is 20-70HU,The CT values of water are 0HU, and fatty CT values are below -50~-100HU, and the CT values of air are -1000HU.During human body CT scan, the CT values with peripheral air are widely different, can be by given threshold by head imageIn bone and air part extract, first threshold K is set in the present embodiment in the present embodiment1(K1=300HU) and Second Threshold K2(K2=-500HU), all gray values (CT values) are searched more than theOne threshold k1Pixel, and extract maximum three-dimensional communication domain, obtain the skull mask (bone in such as Fig. 2Bone region 11,12 and 13), searches all gray values and is less than Second Threshold K2Pixel obtain air mask(nose chamber air 21 and external air part in such as Fig. 2), and between first threshold and Second Threshold itBetween pixel be fat, muscle and a variety of soft tissues.
It should be noted that before handling original image, in order to strengthen the useful letter of imageBreath, can carry out smooth and noise suppression preprocessing in above-mentioned before processing to original image.Conventional image smoothing and de-noisingMethod mainly has the methods such as medium filtering, gaussian filtering, bilateral filtering and mean filter.During the present invention is usedValue filtering carries out the smoothing denoising of image, is specially that the grey scale pixel value in a sliding window is sorted, usesMesophyticum replaces the gray value of window center pixel, i.e. i (x, y)=med [x1,x2,…,xn], wherein x1, x2...,xnFor the gray value of each pixel in sliding window, i (x, y) denotation coordination is the sliding window center pixel of (x, y)Gray value.
S20, based on morphological feature, obtain the convex closure profile of skull mask profile, generate skull convex closure mask.Convex closure refers to that the point on the minimum simple polygon for encasing given convex set, polygon is salient point, i.e., only convexThe point of collection outermost is only possible to referred to as salient point, and the point at convex set center can not possibly be referred to as salient point.Convex closure(ConvexHull) acquisition methods can (detailed process be referred to using Ge Liheng (Graham) scanning methodGraham R L.An efficient algorith for determining the convex hull of a finite planarset[J].Information processing letters,1972,1(4):132-133), Jarvis step-by-step methods can also be used(detailed process refers to Jarvis R A.On the identification of the convex hull of a finite setof points in the plane[J].Information Processing Letters,1973,2(1):18-21), it can also adoptWith quick convex closure method, (detailed process is referring to Eddy W F.A new convex hull algorithm for planarsets[J].ACM Transactions on Mathematical Software(TOMS),1977,3(4):398-403).In one embodiment of the invention, the acquisition methods of convex closure, which are used, is based on Harris angle point algorithms, according toAngle steel joint is ranked up score from high to low, the forward angle point of member-retaining portion score, and is found the score and leaned onThe minimum external convex closure of preceding angle point, the minimum external convex closure is the convex closure of skull mask profile, i.e. skullConvex closure mask.Make convex closure computing to bony areas, a plurality of boundary line 31-34 as shown in Figure 2 can be obtained and surroundedPolygonal region (the minimum external convex closure of the profile of skull mask) be skull convex closure mask, this is polygonThe skull mask being made up of region 11-13 is included in the inside by shape region, while including internal cavity (areaDomain 21), cheek (region 22), eye socket (region 41), the soft tissue such as brain (region 42).
S30, according to skull convex closure mask, internal cavity mask is determined in air mask, is covered by skull convex closureThe region that film and air mask are covered simultaneously, as internal cavity mask.In actual operation, by craniumBone convex closure mask and air mask carry out friendship operation (common factors of two masks) and can determine that internal cavity mask.BodyInternal cavity mask (region 21) mainly includes oral cavity and nasal cavity etc..
S40, merging skull mask and internal cavity mask, form skull filling mask.It should be noted thatTo make skull mask and internal cavity mask fully combine, the present invention is also to the skull mask after merging and in vivoCavity mask makees morphological dilations and filling processing, and carrying out the filling of morphology cavity using internal cavity mask graspsMake, the skull filling mask of internal closing can be obtained.By aforesaid operations, skull filling mask both includes craniumBone mask, again including internal cavity, the region such as brain.
S50, eye socket are crinial bone surface most deep two this anatomy prioris that are recessed, and are given birth to using skullInto convex closure body subtract skull fill out hole body, then can be accurately positioned eye socket position.Therefore covered using skull convex closureFilm subtracts skull filling mask, obtains the position of part tissue of eye.As shown in Fig. 2 skull convex closure mask is includedSkull, internal cavity, brain, cheek and eye soft tissue, wherein cheek and eye soft tissue belong to skullOuter soft tissue.And skull filling mask hole is all then filled out into the region in skull, its comprising skull, in vivoCavity and brain.Therefore above-mentioned two mask, which is made to obtain after reducing, includes the position of eye soft tissue.NeedIllustrate, the position of the part tissue of eye (soft tissue) now obtained may also contain the cranium of both sides cheekGu Wai soft tissue areas (22), such as facial muscle.Therefore, the present invention further, extracts skull convex closure and coveredFilm subtracts the remaining three-dimensional communication domain that skull fills mask, and the connected domain is determined according to the form in three-dimensional communication domainPosition whether be part tissue of eye position.Because the eye socket of normal person is in three-dimensional cone shape, therefore it can adoptThe form in three-dimensional communication domain is detected with circular cone Measurement Method, whether the position for determining three-dimensional communication domain is object machineThe position of official's (part tissue of eye).Specially：
Statistics skull convex closure mask subtracts all pixels point in the three-dimensional communication domain obtained after skull filling maskNumber, so as to obtain the detection volume V of the connected domain；
The floor space S in the above-mentioned three-dimensional communication domain and height h in three-dimensional communication domain is obtained, according to floor space SThe calculating volume V'=S of connected domain × h/3, wherein three-dimensional communication domain bottom is obtained with the height h in three-dimensional communication domainArea is the set for the pixel for being all 1 apart from extraneous air mask distance in all connected domain point sets, and threeThe height of dimension connected domain is the pixel in the connected domain point set apart from extraneous air mask distance maximum and outsideThe distance of air mask.
Calculate the detection volume V in three-dimensional communication domain and the calculating volume V' in three-dimensional communication domain similarity or twoPerson's difference, wherein similarity can compare by calculating both ratio and calculate obtained similarity or difference and beNo in setting range (ratio close to 1), if similarity or difference are in setting range, judges the companyThe position in logical domain is the position of part tissue of eye；Otherwise, the connected domain may be the soft tissue area of cheek both sides,The position for judging the connected domain is not part tissue of eye position but belongs to cheek soft tissue position.Pass through above-mentioned behaviourMake, can interference of the exclusionary zone 22 to positioning result, obtain two conical areas as shown in Figure 3, beThe position of eye soft tissue.
The position of soft tissue in eye socket in CT images can be accurately positioned using aforesaid operations, 24 groups are randomly selectedHead image whole accurate positionings, and for 100 groups of lower half brain data (orbital portion in the image of lower half brainExisting defects, only containing part eye socket, its profile is not cone) be properly positioned for 97 groups, accuracy97% is still can reach, with preferable stability.In addition, can also obtain blood vessel inspection using the detection of many radius spheroidsInclude multiple devices such as eyeball, crystalline lens, extraocular muscle, optic nerve in the further Accurate Segmentation eye sockets of method such as surveyOfficial organizes.The eye of simultaneously automatic Medical Image Segmentation can be efficiently positioned using head image processing method of the present inventionSoft tissue, without manually clicking seed point, it is to avoid use atlas registration, improves positioning and splitting speed,And dependable performance.
In addition, the present invention also provides a kind of positioner of part tissue of eye, as shown in figure 4, including：
Image input module 100, the pending medical image for obtaining, the medical image is CT heads figureAs or MR head images, and head image is divided into by air mask according to pixel gray value and skull is coveredFilm.
Part tissue of eye locating module 200, is connected with image input module 100, for based on morphological feature,The convex closure profile of skull mask is obtained, skull convex closure mask is generated；According to skull convex closure mask, covered in airInternal cavity mask is determined in film；Merge skull mask and internal cavity mask, form skull filling mask；Skull filling mask is subtracted using skull convex closure mask, the position of part tissue of eye is obtained.
Part tissue of eye locating module 200 can also realize following screening function：The skull convex closure mask is extracted to subtractGo the skull to fill the three-dimensional communication domain of mask, the detection volume V in three-dimensional communication domain is obtained respectively with calculatingVolume V', and the detection volume V is calculated with calculating volume V' similarity, connected according to similarity in three-dimensionalThe position of the part tissue of eye is determined in logical domain.
Split module 300, be connected with part tissue of eye locating module 200, for being obtained using the detection of many radius spheroidsThe method that blood vessel is estimated carries out fine segmentation in part tissue of eye position and obtains soft tissue, wherein, soft tissueCan be eyeball, crystalline lens, extraocular muscle, optic nerve etc., CT head images or MRI head images are by upperThat states that device processing can show part tissue of eye organ on the display apparatus is accurately positioned result.It should be noted thatA kind of positioner of part tissue of eye of the present invention also includes head image acquisition module, connects with image input moduleConnect, image input module is used as obtaining CT head images or MR head images, and using the image of acquisition100 input.
Obviously, those skilled in the art can carry out various changes to the present invention and deform without departing from this hairBright spirit and scope.So, if the present invention these modifications and variations belong to the claims in the present invention andWithin the scope of its equivalent technologies, then the present invention is also intended to comprising including these changes and modification.