Methods for identifying material boundaries in volumetric image data

Abstract

A method for identifying material boundaries in volumetric image data is provided. The method is based on the use of a model boundary transition function that models the expected progression of voxel values ​​across a material boundary as a function of distance. Each voxel is acquired in turn, and voxel values ​​in a subregion surrounding the voxel are fitted to the model function to obtain corresponding fitting parameters, in addition to parameters related to the quality of the model fit. Based on these parameters for each voxel, candidate spatial points estimated to lie on the material boundary within the 3D image dataset are identified for at least a subset of voxels. Optionally, an additional preliminary step can be applied to filter voxels to candidate voxels, for example, based on each voxel's estimated distance to the material boundary. This can be estimated based on a specific combination of model parameters for each voxel. Candidate points can be identified only for the identified candidate voxels. This results in a cloud of candidate spatial points that spatially correspond to the contour of the boundary wall. Based on these, a representation of the boundary wall (e.g., a surface mesh) can be generated.

Description

Technical Field

[0001] The present invention relates to an image processing method, and more particularly to an image processing method for identifying a substance boundary in volumetric image data.

Background Art

[0002] In the field of volumetric imaging, it is generally useful to be able to identify and visualize the boundaries or walls of specific structures within the imaged area. This need arises, for example, in the field of medical imaging where it is useful to generate a representation of the shape of the walls of specific anatomical structures such as organs or parts thereof. One particular field where this is useful is the field of virtual colonoscopy.

[0003] Virtual colonoscopy is a visual inspection of the colon wall using, for example, volumetric image data obtained using X-ray computed tomography (CT) imaging. Visualization of the colon wall is usually performed using volume rendering techniques. As a result, an image impression similar to that of an actual endoscope view during colonoscopy is obtained.

[0004] Colonoscopy is used to identify polyps within the colon. Polyps within the colon have the potential to develop into colon cancer. If removed early, cancer can be effectively prevented. Therefore, even asymptomatic patients over a certain age are recommended to undergo a colonoscopy (colonoscopy) to detect and evaluate the possibility of polyps. Unfortunately, this screening is not very well accepted mainly due to the discomfort associated with colonoscopy.

[0005] Therefore, a non-invasive virtual colonoscopy (VC) based on CT scans has been developed as an alternative.

[0006] VC typically uses direct volume rendering techniques. In direct volume rendering, a transfer function (TF) is used for each voxel in a CT volumetric image dataset, mapping its Houndsfield density value to opacity and color values. This opacity and color are projected onto the VC image to create a virtual rendering of the colon. [Overview of the project] [Problems that the invention aims to solve]

[0007] This approach has several drawbacks.

[0008] Firstly, depending on the form and parameters of the transfer function, the apparent position of the colon wall in the rendered image may differ and may not precisely match the actual position of the colon wall.

[0009] The second problem arises from the presence of residual stool or excrement in the colon. This has a contrast similar to colonic wall tissue in the CT scan data. Therefore, its presence can obscure the view of the colonic wall. To resolve this, a process called virtual cleansing is usually performed. This typically involves a pre-processing step of removing parts of the image related to residual stool or excrement. This image pre-processing is usually supported by administering an iodine-containing contrast agent before CT imaging to tag any remaining stool after oral administration of a laxative. This tagging is useful when digitally removing any remaining excrement from the images after data acquisition.

[0010] However, voxel-level virtual cleansing (VC) methods rely on knowing the Hounsfield values ​​of the pure substances. Typically, VC assumes there are three distinct classes of related substances: air, soft tissue, and tagged excrement. However, with the exception of air, the Hounsfield values ​​of soft tissue and tagged excrement differ not only from one CT image to another, but even within a single CT image. Therefore, cleansing methods that rely on knowledge of the Hounsfield values ​​of the pure substances can generate artifacts in the rendered images. Furthermore, even if the material properties of the pure substances are precisely known, voxel-level rendering methods can generate geometric artifacts at the transitions between the three substances. In VC, these appear as "water level" artifacts, which are the points where all three pure substances (air, tissue, and tagged excrement) come into contact with the colon wall.

[0011] An improved approach to identifying material boundaries within volumetric imaging datasets is useful not only for virtual colonoscopy but also for imaging any region where the boundaries of a target object of interest can be identified. [Means for solving the problem]

[0012] The present invention is defined by the claims.

[0013] According to an embodiment of one aspect of the present invention, a method is provided for processing volumetric image data to identify material boundaries within the data. This method is The steps include obtaining volumetric image data representing the anatomical regions of the subject, At least for each subset of voxels within an anatomical region, A step of identifying a volumetric subregion surrounding a voxel, wherein this subregion includes at least a subset of directly adjacent voxels. A step of obtaining a predetermined boundary transition function, wherein this boundary transition function represents a model of voxel values ​​expected as a function of distance across a specified material boundary, A fitting step, which includes fitting a predetermined boundary transition function to voxel values ​​contained within a volumetric subdomain, and which includes determining the fitting parameters of the model function; The steps include determining fitting quality parameters that indicate the quality of fitting the model function, The above method further includes the step of identifying candidate spatial points for the location of the material boundary based on the determined fitting parameters and fitting quality parameters for each voxel, The process includes at least the step of estimating material boundaries in the data based on a subset of identified candidate points.

[0014] Embodiments of the present invention are based on various approaches to obtaining a representation of a material boundary wall based on a material transition model. The proposed approaches use a previously generated model function, which models the progression of voxel values ​​across a material boundary. By fitting the region surrounding each voxel to this model and calculating fitting parameters and fitting quality, it is possible to identify, for at least a subset of voxels, candidate spatial points estimated to be on or near the boundary.

[0015] In some embodiments, an additional step can be performed to initially identify a specific subset of voxels most likely to be on the material boundary, based on the fitting parameters and fitting quality of each voxel, as candidate voxels. Then, for each of these candidate voxels, boundary space points are identified. These are further filtered to reduce the number based on model fitting parameters (e.g., the distance of each voxel to the material boundary).

[0016] Therefore, this is an approach to identifying material boundaries that is entirely different from, for example, more common volume rendering techniques. As mentioned above, volume rendering uses a transfer function (TF) to map each voxel from voxel density values ​​to opacity and color values. The approach according to embodiments of the present invention is specifically for identifying material boundaries and aims to identify spatial points corresponding to the locations of material boundaries based on the result of fitting a predetermined material transition function to a volumetric region surrounding each voxel. These spatial points are inter-voxel spatial points, that is, they lie between coordinate positions that indicate the center of each voxel.

[0017] This approach provides more accurate results for rendered boundary walls in terms of both position and shape, and reduces geometric artifacts that can occur when using volume rendering techniques.

[0018] This approach is advantageous in the field of virtual colonoscopy, but can also be used in a wide range of fields, both in and outside of medical imaging. It can be used with volumetric image data representing any region containing material boundaries whose location needs to be identified.

[0019] Preferably, the volumetric sub-region surrounding each voxel contains all directly adjacent voxels.

[0020] Identifying candidate spatial points teeth This may be done based on further (secondary) parameters derived from the fitting parameters and / or fitting quality parameters.

[0021] The boundary transition function is the minimum voxel value m on one side of the material boundary. high From there, the maximum voxel value m on the other side of the material boundary. low As a result of this process, we can model the values ​​of voxels.

[0022] The fitting parameters include: Minimum voxel value m low , Maximum voxel value m high , The vertical distance d of the voxel from the material boundary, and It contains one or more of the unit normal vectors of the material boundary.

[0023] Vertical distance refers to the distance along a direction perpendicular to the boundary.

[0024] The fitting parameters may also include the width σ of the point spread function of the data acquisition device (such as a CT scanner).

[0025] Some fitting parameters are known in advance (e.g., σ), while others are measured.

[0026] The identification of candidate spatial points may be based on the measured values ​​of one or more fitting parameters and / or on further parameters derived from the fitting parameters. These further parameters may include, for example, m high (i) and m low (i) includes the absolute difference from (i) and / or a parameter indicating the gradient of at least a portion of the function.

[0027] Fitting quality parameters are quantities that characterize the fitting quality of a model function. For example, fitting quality parameters may include one or more of the following: a measure of fitting optimality, goodness of fitting, surplus from fitting, a measure of fitting success, and acceptable range of results.

[0028] In some embodiments, the identification of candidate spatial points includes a preliminary step of identifying candidate voxels, each candidate point being identified based on the fitting parameters of each candidate voxel. In other words, this is a step of narrowing or filtering the entire set of voxels to the voxels most likely to yield accurate or reliable estimated boundary spatial points. The identification of candidate voxels may be based on the determined fitting parameters and / or fitting quality parameters of each voxel.

[0029] Candidate voxels can be identified based on reference values ​​of various parameters obtained for each voxel (model fitting parameters, fitting quality parameters, or any quantities derived therefrom). These reference values ​​may correspond to expected values ​​of these parameters for the material boundary of interest. For example, typical m on both sides of the target material boundary. high and m low For example, this may be known in advance from previous scan data of the region including its boundary. Therefore, candidate voxels can be identified as voxels whose parameter values ​​fall within a predefined tolerance range of the reference value.

[0030] In some embodiments, candidate voxels may be identified based on further (quadratic) parameters derived from fitting parameters and / or fitting quality parameters.

[0031] In some embodiments, the identification of candidate voxels may include performing voxel masking within the imaged anatomical region based on one or more fitting parameter values ​​and / or quality parameter values.

[0032] Separate masks may be generated for each of the different fitting parameters and for the fitting quality parameters. In some embodiments, multiple of these masks can be combined and applied to volumetric image data, and voxels in the overlapping regions of the sets of masks are identified as candidate voxels.

[0033] In some embodiments, a new image or image representation of the imaged anatomical region is generated based on each of the fitting parameters, the fitting quality parameters, and any secondary parameters derived therefrom. In particular, for each parameter, a new image representation of the imaged region is obtained. The voxel value of each voxel is set to be the same as the parameter value obtained for that voxel.

[0034] Subsequently, masking is applied to each of the newly constructed images based on a reference value of the corresponding parameter that the constructed image represents. In some embodiments, these masks are combined into one mask and applied to the original volumetric image data to identify candidate voxels as voxels in the overlapping region of all the masks.

[0035] In some embodiments, the criteria used for masking are calibrated to identify voxels in which a substance boundary is within the volumetric sub-region of the voxel. However, this is not essential.

[0036] The fitting quality parameter of each voxel, where each voxel is masked depending on the fitting quality parameter exceeding a predefined reference value of the fitting quality. The value of m of each voxel, where each voxel is masked depending on the values of m and being within a predefined tolerance range of a predefined pair of reference values of m and, the reference values corresponding to a particular substance boundary of interest, the values of m and, and high and m low of each voxel, where each voxel is masked depending on the values of m and being within a predefined tolerance range of a predefined pair of reference values of m and, the reference values corresponding to a particular substance boundary of interest, the values of m and, and high and m low of each voxel, where each voxel is masked depending on the values of m and being within a predefined tolerance range of a predefined pair of reference values of m and, the reference values corresponding to a particular substance boundary of interest, the values of m and, and high and m low of each voxel, where each voxel is masked depending on the values of m and being within a predefined tolerance range of a predefined pair of reference values of m and, the reference values corresponding to a particular substance boundary of interest, the values of m and, and high and m low of each voxel, where each voxel is masked depending on the values of m and being within a predefined tolerance range of a predefined pair of reference values of m and, the reference values corresponding to a particular substance boundary of interest, the values of m and, and The distance d of each voxel, where each voxel is masked depending on the distance value d being within a predefined tolerance range of a reference value of d. A separate mask is generated / applied based on one or more of these factors.

[0037] In advantageous embodiments, at least the fitting quality parameter, as well as m high and m low Masking is performed on the values ​​of m. This is because these most directly reflect the correspondence between voxels and the voxel regions closest to the boundary of the material of interest. high and m low A separate mask may be generated for each of them, or a single mask may be generated for each pair of values.

[0038] According to one or more embodiments, multiple masks are generated for different parameters and applied in combination to volumetric image data, and voxels in the overlapping regions of the mask combinations are identified as candidate voxels.

[0039] For each of these candidate voxels, a candidate spatial point is obtained. The set of candidate spatial points obtained from the candidate voxels typically represents a point group of candidate material boundaries. This point group defines the specified material boundary.

[0040] Instead of first identifying a set of candidate voxels and then determining a candidate spatial point for each candidate voxel, this process can be performed voxel by voxel. Here, each voxel is evaluated sequentially to determine whether it meets predefined criteria as a candidate voxel, and if so, the candidate spatial point is determined using the fitting parameters obtained for that voxel. In this alternative embodiment, a separate masking step is unnecessary. Instead, the same evaluation criteria used to obtain the mask described above can be applied to each voxel.

[0041] In some embodiments, the fitting parameters for each voxel include the vertical distance d of the voxel from the material boundary and the unit normal vector of the material boundary. Identifying candidate space points for a voxel may involve identifying the coordinates of a point located at a distance d from the voxel location along the unit normal direction.

[0042] According to one or more embodiments, the estimation of a material boundary further includes generating a surface mesh representing the material boundary, the surface mesh including at least a subset of identified candidate spatial points.

[0043] Mesh generation involves connecting spatial points from at least a subset of spatial points in a way that forms triangles.

[0044] This makes the surface rendering clearer.

[0045] However, mesh generation is not mandatory; instead, a different representation of the material boundary may be generated. For example, an array of spatial point coordinates identified as corresponding to the location of the material boundary could simply be used as the representation.

[0046] According to one or more embodiments, the acquired volumetric image data is X-ray CT volumetric image data.

[0047] In some embodiments, the above method includes receiving CT projection data and performing image reconstruction to obtain volumetric image data. In other embodiments, the above method includes receiving reconstructed volumetric image data.

[0048] Alternatives to CT data include, for example, MRI image data or volumetric ultrasound data.

[0049] According to one or more embodiments, the obtained volumetric image data is spectral image data. This data includes multiple image datasets, each set formed from data corresponding to different spectral data channels. The method according to the present invention is performed on each dataset, and for each spectral channel, there exists a different predetermined material transition function.

[0050] One example is the use of X-ray spectral CT data. X-ray spectral CT is an imaging modality that extends the functionality of conventional X-ray CT systems by acquiring projection data at multiple X-ray energies. This can be achieved by incorporating detectors that can distinguish between different X-ray energies (such as energy-distinguishing photon count detectors or energy-integrating detectors) or by sequentially changing the X-ray energy spectrum and acquiring corresponding detector data sequentially. Spectral X-ray data allows for the identification and quantification of substances contained in scanned objects. This is because different substances may have different X-ray absorption spectra. For example, it is possible to use X-ray energy spectra that are known to be absorbed most or least by the substance of interest.

[0051] Supplemental information is provided by generating different material boundary representations using data from different spectral channels, because the image data from different spectral channels differs depending on the material composition of the object being imaged. For example, identifying and combining boundaries between multiple pairs of different materials can enhance the identification of organ boundary walls marked by different material transitions.

[0052] As an example, spectral CT imaging can be used in the field of CT colonography to generate a number of (at least two) spectral channels corresponding to low and high energy levels of projected X-ray energy. These spectral channels can be converted into various representations, including iodine concentration maps, which allow for visualization of contrast agent distribution within the imaged colon. In the context of colonography, in some embodiments, multiple spectral channel data can be used to assist in performing the virtual cleansing described above.

[0053] According to one or more embodiments, the identified material boundary represents a structural wall of an anatomical structure.

[0054] In some embodiments, the substance boundary represents the wall of an anatomical structure that defines the lumen, such as the colon.

[0055] Embodiments according to further aspects of the present invention provide a computer program product including computer program code. The computer program code is executable on a processor and causes the processor to perform a method according to any embodiment or model described above or below, or according to any claim of this application.

[0056] A further embodiment of the present invention also provides a processing apparatus that processes volumetric image data to identify material boundaries within the data. This processing apparatus, Obtaining volumetric image data representing the anatomical regions of the subject, For each voxel within an anatomical region, Identifying a volumetric subregion surrounding a voxel, wherein this subregion includes at least a subset of directly adjacent voxels. This involves obtaining a predetermined boundary transition function, which represents a model of the expected voxel values ​​as a function of the distance across the specified material boundary. Fitting a predetermined boundary transition function to voxel values ​​contained within a volumetric subdomain, including determining the fitting parameters of the model function, This involves determining fitting quality parameters that indicate the quality of fitting the model function, and performing the following: The above processing device further identifies candidate spatial points of the material boundary based on the determined fitting parameters and fitting quality parameters for each voxel. At a minimum, the material boundaries within the above data are estimated based on a subset of the identified candidate points.

[0057] According to one or more embodiments, the boundary transition function is the minimum voxel value m on one side of the material boundary. low From there, the maximum voxel value m on the other side of the material boundary. high As a result of this process, we can model the voxel values, and the fitting parameters are: Minimum voxel value m low , Maximum voxel value m high , The vertical distance d from the boundary to the voxel, and Includes one or more of the boundary unit normal vectors.

[0058] In some embodiments, the identification of candidate spatial points includes a preliminary step of identifying candidate voxels, each candidate point being identified based on the fitting parameters of each candidate voxel. The identification of candidate voxels may include performing masking of voxels within the imaged anatomical region based on one or more fitting parameter values ​​and / or fitting quality parameter values.

[0059] In some embodiments, A fitting quality parameter for each voxel, where each voxel is masked depending on whether the fitting quality parameter exceeds a predefined reference value for fitting quality. m high and ml ow The value is m high and m low The value of m high and m low The masking depends on the reference values ​​being within a predefined tolerance range of a predetermined pair of reference values, and the reference values ​​correspond to the specific material boundary of interest, m high and ml ow The value, and The distance d for each voxel is such that each voxel is masked depending on whether the distance value d is within a predefined tolerance range of the reference value of d. A separate mask is applied based on one or more of these factors.

[0060] These and other aspects of the present invention will become apparent from the embodiments described below and will be explained with reference to those embodiments. [Brief explanation of the drawing]

[0061] To gain a deeper understanding of the present invention and to more clearly illustrate how it is implemented, please refer to the attached drawings as just one example.

[0062] [Figure 1] Figure 1 outlines the steps of an example method according to one or more embodiments. [Figure 2] Figure 2 schematically shows the modeling parameters used to fit the model boundary transition function to each voxel in one or more embodiments. [Figure 3] Figure 3 schematically shows examples of boundary transition functions according to one or more embodiments. [Figure 4] Figure 4 shows a slice of an example volumetric image dataset containing the material boundary of a specific target. [Figure 5(a)] Figure 5(a) shows the image representation of the model function parameters (the minimum voxel value mlow on one side of the material boundary) calculated for each voxel in the image dataset. [Figure 5(b)] Figure 5(b) shows the image representation of the model function parameter (mhigh, the maximum voxel value on the other side of the material boundary) calculated for each voxel in the image dataset. [Figure 6] Figure 6 shows the image representation of the model fitting quality parameters obtained for each voxel. [Figure 7] Figure 7 shows an image representation of the determined absolute distance from the material boundary of interest to each voxel. [Figure 8] Figure 8 shows the masking of voxels based on the Mlow model parameter values ​​determined for each voxel. [Figure 9] Figure 9 shows the masking of voxels based on the mhigh model parameter values ​​determined for each voxel. [Figure 10] Figure 10 shows the result of overlaying the masks from Figures 8 and 9 onto the original volumetric image dataset. [Figure 11] Figure 11 shows the extracted candidate point set obtained from voxels located in the overlapping region of the applied mask combinations. [Figure 12] Figure 12 shows the extracted candidate point set obtained from voxels located in the overlapping region of the applied mask combinations. [Figure 13] Figure 13 shows an example of a surface mesh constructed based on candidate spatial points obtained from candidate voxels. [Figure 14] Figure 14 shows an example of a surface mesh constructed based on candidate spatial points obtained from candidate voxels. [Figure 15] Figure 15 shows an example of a further surface mesh. [Figure 16] Figure 16 shows an example of a further surface mesh. [Modes for carrying out the invention]

[0063] The present invention will be described with reference to the figures.

[0064] The detailed descriptions and specific examples illustrate exemplary embodiments of the apparatus, system, and method, but should be understood to be for illustrative purposes only and not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the present invention will be better understood from the following description, the appended claims, and the appended drawings. It should be understood that the figures are schematic diagrams and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to indicate the same or similar parts.

[0065] This invention provides a novel approach to identifying material boundaries in volumetric image data. This approach is based on the use of a model boundary transition function that models the expected progression of voxel values ​​across a material boundary as a function of distance. Each voxel is acquired sequentially, and the voxel values ​​within the sub-region surrounding the voxel are fitted to the model function, yielding corresponding fitting parameters in addition to parameters related to the quality of the model fitting. Based on these parameters for each voxel, candidate spatial points estimated to lie on the material boundary in the 3D image dataset are identified for at least a subset of voxels. Optionally, additional preliminary steps can be applied to filter voxels down to candidate voxels based, for example, on the estimated distance of each voxel to the material boundary. This can be estimated based on a specific combination of model parameters for each voxel. Candidate points can only be identified for the identified candidate voxels. This results in a set of candidate spatial points that spatially correspond to the outline of the boundary wall. Based on these, a representation of the boundary wall (e.g., a surface mesh) can be generated.

[0066] Figure 1 shows a block diagram illustrating the basic steps of one or more method examples according to one or more embodiments. Method 10 is a computer-aided method. This method is for processing volumetric image data to identify material boundaries within volumetric image data.

[0067] Method 10 includes step 12 of obtaining volumetric image data representing anatomical regions of the subject.

[0068] Method 10 further includes performing the following steps for each subset of voxels within at least an anatomical region: namely, Step 14 of identifying a volumetric subregion surrounding a voxel, wherein this subregion includes at least a subset of directly adjacent voxels. Step 16 involves obtaining a predetermined boundary transition function that represents a predetermined model function of voxel values ​​with respect to the distance across the specified material boundary, Step 18 involves fitting a predetermined boundary transition function to voxel values ​​contained within a volumetric sub-domain, wherein this fitting includes determining the fitting parameters of the model function. Step 20 is performed to determine fitting quality parameters that indicate the quality of fitting the model function.

[0069] In some embodiments, the above steps are performed on all voxels within the imaged region. However, in other embodiments, one or more preprocessing steps may be performed to remove or filter out specific voxels (which are presumed to be less likely to coincide with boundary locations) before applying the main method.

[0070] Once the above steps have been performed for each voxel, the method further includes a process 22 to identify candidate spatial points of the boundary wall based on the determined fitting parameters and fitting quality parameters for each voxel. For example, in one set of embodiments, the fitting parameters include the perpendicular distance d of the voxel from the material boundary and the unit normal vector of the material boundary, and determining candidate spatial points includes determining the coordinates of a point that is at a distance d from the voxel location along the unit normal direction.

[0071] After candidate spatial points have been identified (22), the method further includes estimating material boundaries in the data based on the identified candidate spatial points (24).

[0072] In some embodiments, a surface mesh is generated based on identified candidate spatial points or an extracted subset thereof. In other embodiments, a representation of the boundary containing simply an array of coordinate locations of identified candidate spatial points or an extracted subset thereof is generated.

[0073] The above method is performed by a single processor or a processing unit comprising one or more processors. Acquiring volumetric image data includes receiving volumetric image data from an external source such as an imaging device (e.g., a CT or MRI scanning device) or from a data store. In further embodiments, obtaining volumetric image data includes receiving unreconstructed scan data (such as CT projection data) and performing the step of reconstructing the volumetric image data.

[0074] Embodiments of the present invention are based on a mathematical model of the emergence of material transitions between discrete pure material classes in volumetric image data (e.g., CT image data). However, the same principle can be applied to other forms of volumetric image data, such as MRI.

[0075] Depending on the specific material transition in the problem, different model material transition functions can be applied. For example, different model functions depend on the tissue density on each side of the material boundary, the typical distance over which the transition occurs, and the potential geometric properties of the boundary.

[0076] Figure 2 schematically shows the model used for the model boundary transition function. This represents a two-material transition. However, the same principle can be applied to other transitions (e.g., three-material transitions).

[0077] Referring to Figure 2, the two-material transition is represented by the model function g(m high , m low It can be modeled by d, n, σ. Here, m low m is the expected low voxel 32 value of the two pure substances at the first side 38a and the second side 38b of the two-substance transition boundary 34, high is the expected high voxel value. For example, if the image data is CT image data, the voxel value will be the Hounsfield value. Parameter d represents the vertical distance from the material boundary 34 to the voxel i to which the function is fitted, and n represents the unit normal vector of the material boundary. Parameter σ represents the width of the point spread function of the scanning device (such as a CT scanner).

[0078] For each voxel i, the model function is fitted to the values ​​of voxels 32 within the volumetric subregion 36 surrounding the voxel. This is a fixed-size subregion, for example, extending one voxel away from the voxel in question on all sides; that is, a cube with a length of 3 voxels. This includes all directly adjacent voxels. It may also include more voxels than directly adjacent voxels, for example, a cube containing 5x5x5 or 7x7x7 voxels, or any other number of voxels.

[0079] Figure 3 schematically shows an example of a boundary transition function. The boundary transition function is the maximum voxel value m on one side of the material boundary 34. high Therefore, the minimum voxel value m on the other side of the material boundary 34 lowAs a result of this process, the values ​​of voxel 32 are modeled. The function is nonlinear. Figure 3 shows the location of example voxel i to which the model can be fitted, and the model fitting parameter d, which represents the distance of the voxel from the material boundary 34. In some embodiments, there is a predefined spatial coordinate system for the imaged anatomical region of the subject, and d is expressed in units of this coordinate system (e.g., mm). In other embodiments, it may be expressed in units of voxel space (i.e., the width of one voxel corresponds to a unit length).

[0080] One example workflow of the method will be outlined in detail according to one or more embodiments. From there, the general steps of the method will become clearer. The method described herein uses abdominal CT images as input. Optionally, residual excrement may be tagged with iodine contrast agent before imaging.

[0081] The method described below is for two-substance transitions (air-tissue, air-feces, or feces-tissue), but it can be similarly applied to other transitions (such as three-substance transitions).

[0082] This method begins with acquiring volumetric image data representing relevant anatomical regions of the subject. Figure 4 shows one image slice from an example of a CT volumetric image dataset. In this example, the image dataset represents a region including the colon of the subject.

[0083] A specific model boundary transition function g, unique to the target substance boundary, is read from, for example, a local or remote data store. In this particular embodiment, the identified substance boundary is the colon wall. The substance boundary is, for example, the boundary between fecal residue and colon wall tissue, or the boundary between air in the colon lumen and colon wall tissue.

[0084] Once the model boundary transition function is obtained, the model function is fitted to the voxel values ​​contained within the volumetric sub-regions surrounding each voxel in the volumetric image dataset.

[0085] This model fitting process is performed sequentially on each voxel (or at least a specified subset of voxels) within a 3D image dataset. For example, this method applies model fitting sequentially by iterating through or indexing each voxel one by one.

[0086] For each related voxel i, a volumetric subregion 36 (or local environment) surrounding the voxel is identified. This subregion contains at least a subset of directly adjacent voxels. As mentioned above, the volumetric subregion can be a cubic subregion of a predefined size. For example, the subregion contains all voxels immediately adjacent to voxel i. The subregion may have a size k, for example. 3 In the cubic subimage, voxel i is in the center.

[0087] Next, the obtained boundary transition model function g is fitted to the voxel values ​​in sub-region 36 (for example, Hounsfield values ​​in this case). The fitting is performed using, for example, the least squares fitting method. However, those skilled in the art will recognize that various model fitting methods can be applied instead. For example, almost all regression methods are suitable, such as polynomial regression, logistic regression, and multivariable regression.

[0088] As a result of model function fitting, the fitted model parameters (referred to as fitting parameters in this specification) for each voxel i are obtained: m high (i), m low (i), d(i), n(i), σ(i) (explained in more detail above).

[0089] Furthermore, one or more additional parameters are determined that represent the quality of the fitting of the model function to the sub-region of each voxel i. These will vary depending on the fitting method used. For example, fitting quality parameters may include one or more of the following: optimality of the model fitting, remainder of the least-squares fitting, success of the fitting, tolerance of the fitting result, or any other appropriate parameters that would be obvious to those skilled in the art.

[0090] Further quadratic parameters or quantities can be obtained based on the fitting parameters and fitting quality parameters of each voxel i. These are not parameters of the model function itself, but are calculated from the estimated model parameters. For example, these include the absolute distance of voxel i from the material transition boundary 24, m high (i) and m low (i) The absolute difference from and / or the gradient or steepness of the model transition function are included.

[0091] In some embodiments, new images or representations of the imaged anatomical region are generated, each representing a fitting parameter, a fitting quality parameter, and any parameters derived therefrom. In particular, a new representation of the imaged region is obtained for each parameter. The voxel value of each voxel is set to the same as the parameter value obtained for that voxel.

[0092] For example, Figure 5(a) shows the fitted model parameter m for each voxel. low Figure 5(b) shows an image representing the value of parameter m. high Figure 6 shows an image representing the values ​​of the fitting quality parameters for each voxel. Figure 7 shows an image representing the quadratic parameter value of the absolute distance of voxel i from the material transition boundary.

[0093] Voxel masking is performed based on various parameters obtained for each voxel i. This is based on the further use of reference values ​​or criteria for the parameter values. In some embodiments, masking is based on images obtained for model parameters and fitting quality parameters. For example, reference values ​​are stored in advance in a local or remote data store. Reference values ​​are values ​​known from previous scan data associated with voxels near the material boundary of interest. For example, reference values ​​may be used as thresholds, or the decision criteria may include whether the voxel parameter value is within a predefined tolerance range of the reference value.

[0094] A separate mask may be generated for at least a different set of fitting parameters and fitting quality parameters.

[0095] In some embodiments, at least one mask is generated based on the fitting quality parameter of each voxel. In this case, each voxel is masked depending on whether the fitting quality parameter exceeds a predefined reference value or threshold for fitting quality.

[0096] Furthermore, or alternatively, the m of each voxel high and m low Based on the value, it is also possible to generate at least one mask. Here, each voxel is m high and m low The value of m high and m low Masking relies on the reference values ​​of a predetermined pair falling within a predefined tolerance range. The reference values ​​correspond to specific substance boundaries of interest. Thus, here, masking is based on identifying specific transitions between two (or more) known substances using prior knowledge of typical voxel values ​​(e.g., Hounsfield values) within these two substances. For example, the Hounsfield value for air is known to be approximately -1000, and the Hounsfield value for soft tissue is approximately 0. In this case, voxel i belonging to the air-tissue transition is mhigh (i) ≈ 0 and m low (i) The condition ≈ -1000 should be satisfied, m high (i) and m low (i) Only voxels whose values ​​fall within the predefined tolerance range of these prior values ​​are masked. As an example, Figures 8 and 9 show the m of air-texture transitions. low (i) and m high Examples of each mask in (i) are shown.

[0097] Furthermore, or alternatively, at least one mask can be generated based on the distance d of each voxel. In this case, each voxel is masked depending on whether the distance value d is within a predefined tolerance range of the reference value of d. Thus, here, only voxels within a predefined minimum proximity to the material boundary are masked.

[0098] According to one or more embodiments, multiple masks are generated for different parameters and applied in combination to volumetric image data. Voxels in the overlapping regions of the mask combinations are identified as candidate voxels.

[0099] Figure 10 schematically shows an example of the result of superimposing the two mask combinations shown in Figures 8 and 9. The voxels in the overlapping mask regions are shown in white in Figure 10, and it can be seen that these follow the colon wall shown in the image data. These voxels form a set of candidate voxels; that is, these are the candidate voxels that best match the colon wall in the image data, or that provide the optimal model fitting.

[0100] Obtaining a representation of the material boundary further involves identifying candidate points estimated to lie on the material boundary from each candidate voxel, based on the distance d obtained by the model of each voxel from the boundary, and based on the unit normal vector obtained by the model of the boundary.

[0101] In particular, identifying candidate spatial points for each candidate voxel involves determining the coordinates of points located at a distance d from the voxel location along the boundary unit normal direction obtained from the voxel fitting parameters.

[0102] In particular, in most cases, a flat plane is a good approximation of the boundary shape in the local region closest to a given voxel. A flat plane can be parameterized, for example, by the Hessian normal form and a unit vector (n x , n y , n z It consists of the voxel's position (i,j,k) and the distance d from the boundary. Since both of these parameters are fitting parameters for the model boundary transition function, they are pre-calculated for each candidate voxel. Thus, candidate spatial points of the material boundary can be identified as spatial points at a distance d from the voxel along the boundary unit normal direction. Therefore, for a voxel with coordinate position (i,j,k), the coordinate position p of the candidate point is p=(i,j,k)+d * (n x ,n y ,n z ) is identified as such.

[0103] For a full set of candidate voxels, the resulting set of candidate space points is a list (or group) of points. These points all lie on the material boundary, but do not necessarily coincide with points on the voxel grid.

[0104] Figure 11 shows an image representation of an example of a candidate point cloud generated in this way. Figure 12 shows a close-up view of the point cloud.

[0105] For each point in the point cloud, the unit normal vector n is saved for later use. This is useful for further rendering steps.

[0106] As an example, according to one or more embodiments, a surface mesh representing a material boundary is further generated. This surface mesh contains a reduced subset of candidate points that form a point cloud. Obtaining the surface mesh involves connecting points from the point cloud to form triangles. These can then be used for rendering.

[0107] Figure 13 shows an example of a surface mesh constructed based on the candidate point cloud in Figure 11. Figure 14 shows a close-up magnified view of the surface mesh. Figures 16–18 show further examples of surface mesh renderings in which triangles obtained by connecting points are filled to provide a smooth surface representation of the colon wall. Figure 16 shows a section of the colon lumen.

[0108] As described above, in some embodiments, a new image is generated for each voxel, representing the fitting parameter, the quadratic parameter, and the fitting quality parameter, respectively. In some embodiments, some or all of these images are further used to color the surface mesh in order to visualize more information. The images may be rendered semi-transparently and overlaid on the mesh. Multiple images may be overlaid on the mesh together in different colors, or the user may selectively switch between the parameter images overlaid on the mesh using a user interface.

[0109] Although the above examples are described in relation to two-material boundaries, the same principles can be applied to more complex boundaries. One example is a multilayer boundary. This example is a thin layer of material (material 2) embedded within another region of material (material 1). In this case, the multilayer boundary is simply treated as a pair of two-material boundaries: a first boundary from material 1 to the thin layer of material 2, and a second boundary from the layer of material 2 to material 1. A single boundary transition function may be used that encompasses both two-material boundaries, or separate two-material transition functions may be applied sequentially to volumetric image data to identify pairs of boundaries on both sides of the thin layer. In either case, the principle for estimating pairs of material boundaries using boundary transition functions is the same as outlined above for a single two-material boundary.

[0110] A further example involves a three-material boundary (where three material regions meet along a common line). Here again, the same principle applies as in two-material transitions, but the model of the expected voxel value as a function of the distance across the boundary (represented by the boundary transition function) has an additional directional dependence (the direction of approach to the boundary), taking into account the fact that the three material regions meet along a common boundary line. However, the boundary transition function still models the voxel value as a function of position relative to the boundary. The model includes additional model parameters (fitting parameters) related to the angles between the various material regions surrounding the boundary line, and / or the angles of the voxel relative to one or more different material regions or the boundary line. This allows the boundary transition model to fit a given voxel to a volumetric sub-region surrounding it, regardless of its position within or relative to the various material regions.

[0111] As described above, the volumetric image data obtained according to one or more embodiments is spectral image data. This data includes multiple image datasets, each set formed from data corresponding to different spectral data channels. The method according to the present invention is performed on each dataset, and for each spectral channel, there exists a different predetermined material transition function.

[0112] One example is the use of X-ray spectral CT data. X-ray spectral CT is an imaging modality that extends the functionality of conventional X-ray CT systems by acquiring projection data at multiple X-ray energies. This can be achieved by incorporating detectors that can distinguish between different X-ray energies (such as energy-distinguishing photon count detectors or energy-integrating detectors) or by sequentially changing the X-ray energy spectrum and acquiring corresponding detector data sequentially. Spectral X-ray data allows for the identification and quantification of substances contained in scanned objects. This is because different substances may have different X-ray absorption spectra. For example, it is possible to use X-ray energy spectra that are known to be absorbed most or least by the substance of interest.

[0113] Another example is the use of MRI data. In multi-parameter MRI, a similar principle applies, where two or more MR sequences are repeated, and a set of image data is acquired for each. Different substances are represented with higher or lower contrast for different MR sequences (channels).

[0114] The above example workflow can be extended to be applied to spectral volumetric image data.

[0115] This section outlines an example of spectral CT data. However, the same principles can be applied to spectral image data obtained from other modalities such as MRI and ultrasound.

[0116] When using spectral CT data, for example, it is possible to create a single-energy image synthesized at different keV (X-ray tube voltage) values. Within these single-energy images, different materials are shown with different contrasts.

[0117] With a few minor modifications, the above workflow can be adapted to process data from two or more different spectral channels. This allows for more robust rendering of material boundaries.

[0118] This method involves spectral image data I i This involves obtaining n distinct sets of images, where n is any integer greater than 1. Each image dataset corresponds to image data acquired from different spectral channels. Different spectral channels mean that the projection data (from which the image datasets are reconstructed) was obtained using different X-ray tube voltages (or different X-ray energy spectra).

[0119] As an example, two image datasets I1 and I2 are received (n=2), each corresponding to a different X-ray energy channel (X-ray tube voltage) (for example, each corresponding to a different single-energy image dataset). For example, I1 is at 45 keV (tube voltage) and I2 is at 125 keV.

[0120] Here, the two-material transition is represented by n different material boundary transition functions g. i (m i,high , m i,low It is modeled as (i=1, ..., n) with respect to material parameters m. i,high and m i,low This is each spectral image dataset I i The geometric parameters d, n, and σ differ, but are the same for each dataset.

[0121] Perform a fitting of the model function g i and all spectral images I iSimultaneously optimize the matching with. That is, the model function is fitted to each spectral dataset using a cost function that attempts to optimize the fitting of all models simultaneously and at the same time for each spectral dataset. For example, a spectral image dataset can be represented as a single dataset with multiple voxel values for each voxel. For example, if k is the index of the voxel, i = 0, 1 is the index of the spectral image dataset, and I ik is the voxel value of the spectral image i at voxel k, for example, the cost function Σ k (I 0k - g 0k (m 0high , m 0low , d, n, sigma)) 2 +(I 1k - g 1k (m 1high , m 1low , d, n, sigma)) 2 is optimized.

[0122] From the fitting of each of the two model functions for each voxel of the spectral image dataset, the model function fitting parameters for each voxel j are estimated. The fitting parameters include the material parameters m i of each spectral image dataset I i,high (j) and m i,low (j) (where m i,low is the minimum voxel value on one side of the material boundary and m i,high is the maximum voxel value on the other side of the material boundary). The fitting parameters also include the geometric parameters d(j), n(j), σ(j), which are the same for each image dataset I i .

[0123] Furthermore, based on the fitting parameters, additional secondary parameters are obtained for each spectral image dataset I i . The secondary parameters include, for example, the absolute difference between m i,high (j) and m i,low (j).

[0124] Candidate voxels are determined based on fitting parameters, quadratic parameters, and fitting quality parameters for each voxel in each image dataset. As described in the workflow above, this is based on the use of voxel masking.

[0125] In particular, masking is performed based on model fitting parameters determined for each spectral dataset and based on one or more pre-determined reference values ​​or criteria. Masking can be applied to image data of a single image dataset, but it uses masks obtained based on parameter values ​​from multiple spectral channels. That is, candidate voxels are identified using information obtained from multiple spectral channels. For example, a first mask is generated based on fitting parameters from one spectral channel, and a second mask is generated based on fitting parameters from a second spectral channel, each based on a predefined reference criterion. Both masks are combined and applied to image data of either of the two spectral channels or a third spectral channel. Thus, information from multiple spectral channels can be taken into account, and as a result, the identification of candidate voxels is greatly enhanced. Candidate voxels may be identified as voxels in the overlapping region of two or more masks. In this way, erroneous fitting results from a single spectral channel can be more reliably excluded from consideration.

[0126] Once candidate voxels are identified, the remaining steps of the method are the same as those outlined in the workflow described for non-spectral data. In particular, candidate spatial points of the material boundary are identified for each candidate voxel based on the model fitting parameters of the candidate voxel. This forms a set of candidate points for each. A surface mesh is created from these candidate points, for example, by connecting the points to form triangles.

[0127] By using different spectral datasets, boundaries between different substance pairs or sets can be identified, so that the point clouds in each spectral dataset define the boundaries of transitions between different substance pairs or sets. For example, these include one point cloud for the transition from air to tissue, a point cloud for the transition from contrast agent to tissue, and a point cloud for the transition from air to contrast agent. For example, by combining the point clouds for air to tissue and the point clouds for contrast agent to tissue, a composite point cloud of the entire tissue boundary can be obtained. This is what is needed, for example, when trying to identify the boundary of the colon wall. Thus, the virtual cleansing process (mentioned above) can be performed.

[0128] The following modifications can be applied according to any of the embodiments described above.

[0129] In the above embodiment, it is assumed that the point spread function of the image scanning device used to acquire image data is isotropic. However, in a further embodiment, the point spread function of the scanner may be anisotropic. In this case, σ is replaced by a vector σ.

[0130] The scanner's point spread function (isotropy or anisotropy) may be known a priori. In this case, estimation is unnecessary, and it can be a fixed value in the algorithm.

[0131] According to one or more embodiments, a preprocessing step is applied to each voxel before fitting the model function. The preprocessing is for filtering the voxels into a reduced voxel set, based on identifying the voxels most likely to be close to the material boundary. Voxels that are close to the material boundary and are estimated to be less likely to be candidate voxels are excluded from consideration. Then, the fitting of the model function is applied only to the voxels that remain after filtering.

[0132] Filtering is performed based on one of several factors. One option is to determine the predicted m on either side of the desired material boundary. high Value and m lowOnly voxels whose voxel values ​​fall within a predefined range are selected. Other options detect a representation of the voxel value gradient across the entire imaged volumetric region and select only voxels in regions with relatively high gradients.

[0133] Further embodiments of the present invention provide a processing apparatus that performs a method according to any embodiment or model described above or later, or according to any claim of this application.

[0134] A processing unit generally includes a single processor or multiple processors. A processing unit may be installed within a single containing device, structure, or unit, or it may be distributed among multiple different devices, structures, or units. Therefore, a reference to a processing unit adapted or configured to perform a particular step or task corresponds to that step or task being performed, either alone or in combination, by any one or more of multiple processing components. Those skilled in the art will understand how such a distributed processing unit can be implemented. A processing unit includes a communication module or input / output unit for receiving data and outputting data to further components.

[0135] One or more processors in a processing unit can be implemented in various ways using software and hardware to perform a variety of required functions. Typically, a processor uses one or more microprocessors programmed using software (e.g., microcode) to perform the required functions. A processor can be implemented as a combination of dedicated hardware for some functions and one or more programmed microprocessors and associated circuits for other functions.

[0136] Examples of circuits that may be used in various embodiments of this disclosure include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0137] In various implementations, a processor may be associated with one or more storage media, such as volatile and non-volatile computer memory, including RAM, PROM, EPROM, and EEPROM®. The storage media may be encoded with one or more programs that, when executed on one or more processors and / or controllers, perform the necessary functions. The various storage media may be fixed within a processor or controller, or they may be transportable so that one or more programs stored therein can be loaded into the processor.

[0138] A further aspect of the present invention provides a system comprising the above-described processing apparatus and a display unit operably coupled with the processing apparatus used to display a generated representation of a material boundary in image data. The processing apparatus generates a display control output for displaying the image representation of the identified material boundary on the display unit. The system may further include a user interface that allows a user to change properties and settings of how the operation is performed and to configure settings for the display of the material boundary on the display unit.

[0139] Modifications of the disclosed embodiments can be understood and implemented by those skilled in the art in carrying out the claimed invention, based on a review of the drawings, disclosures, and appended claims. In the claims, the word “including” does not exclude other elements or steps, and singular elements do not exclude plural elements.

[0140] A single processor or other unit may perform the functions of some of the items described in the claims.

[0141] The mere fact that certain means are described in mutually different dependent claims does not mean that combinations of these means cannot be used advantageously.

[0142] Computer programs may be stored and distributed on any suitable medium, such as optical storage media or solid-state media, supplied together with or as part of other hardware, but they may also be distributed in other forms, such as via the Internet or other wired or wireless communication systems.

[0143] Please note that when the term "adapted to..." is used in the claims or description, it is intended to be equivalent to the term "configured to...".

[0144] Any reference numeral in the claims should not be construed as limiting the scope.

Claims

1. A method for processing volumetric image data to identify material boundaries within the volumetric image data, The steps include obtaining volumetric image data representing the anatomical regions of the subject, At least for each subset of voxels within the anatomical region, A step of identifying a volumetric subregion surrounding a voxel, wherein the volumetric subregion includes at least a subset of directly adjacent voxels; A step of obtaining a predetermined boundary transition function, wherein the predetermined boundary transition function represents a model of voxel values ​​expected as a function of the distance across the specified material boundary, A fitting step of fitting the predetermined boundary transition function to the voxel values ​​contained within the volumetric sub-region, the fitting step of determining the fitting parameters of the model function, The steps include determining fitting quality parameters that indicate the quality of fitting the aforementioned model function, Includes, The aforementioned method further, A step of identifying candidate spatial points for the location of the material boundary for each voxel, based on the determined fitting parameters and fitting quality parameters, The steps include at least estimating the material boundary in the volumetric image data based on a subset of identified candidate spatial points, Methods that include...

2. The boundary transition function is defined by the minimum voxel value m on one side of the material boundary. ｌｏｗ Therefore, the maximum voxel value m on the other side of the material boundary ｈｉｇｈ As a result of the process up to this point, the voxel values ​​are modeled, and the fitting parameters are, The minimum voxel value m ｌｏｗ , The aforementioned maximum voxel value m ｈｉｇｈ , The vertical distance d from the material boundary of the voxel, and The unit normal vector of the aforementioned material boundary, The method according to claim 1, comprising one or more of the following.

3. The method according to claim 1 or 2, wherein the step of identifying candidate spatial points includes a preliminary step of identifying candidate voxels, each candidate spatial point being identified for each candidate voxel based on the fitting parameters of the candidate voxel, and the preliminary step of identifying candidate voxels includes a step of performing masking of voxels in the imaged anatomical region based on one or more of the values ​​of the fitting parameters and / or the values ​​of the fitting quality parameters.

4. The fitting quality parameter of each voxel, wherein each voxel is masked depending on whether the fitting quality parameter exceeds a predefined reference value for the quality of the fitting, The maximum voxel value m of each voxel ｈｉｇｈ and the minimum voxel value ml ｏｗ such that each voxel is masked depending on the values of the maximum voxel value m ｈｉｇｈ and the minimum voxel value m ｌｏｗ being within a predefined tolerance range of a predefined pair of reference values of the maximum voxel value m ｈｉｇｈ and the minimum voxel value m ｌｏｗ where the reference values correspond to a specific material boundary of interest, the maximum voxel value m ｈｉｇｈ and the minimum voxel value ml ｏｗ of said values, and The vertical distance d of each voxel, where each voxel is masked depending on whether the value of the vertical distance d is within a predefined tolerance range of a reference value of the vertical distance d, The method according to claim 3, wherein a separate mask is generated based on one or more of the following.

5. The method according to claim 3 or 4, wherein a plurality of masks are generated for different fitting parameters, are applied in combination to the volumetric image data, and voxels in the overlapping regions of the combination of masks are identified as candidate voxels.

6. The fitting parameters include the vertical distance d of the voxel from the material boundary and the unit normal vector of the material boundary. The method according to any one of claims 1 to 5, wherein the step of identifying the candidate spatial point includes the step of identifying the coordinates of a point located at a distance d from a voxel position along the unit normal direction.

7. The method according to any one of claims 1 to 6, wherein the step of estimating the material boundary further includes the step of generating a surface mesh representing the material boundary, the surface mesh comprising at least a subset of the identified candidate spatial points.

8. The method according to any one of claims 1 to 7, wherein the acquired volumetric image data is an X-ray CT volumetric image.

9. The method according to any one of claims 1 to 8, wherein the acquired volumetric image data is spectral image data comprising a plurality of image datasets, each image dataset being formed from data corresponding to different spectral data channels, the method being performed on each of the image datasets, and for each spectral channel, there exists a different predetermined material boundary transition function.

10. The method according to any one of claims 1 to 9, wherein the material boundary represents a structural wall of an anatomical structure.

11. The method according to claim 10, wherein the substance boundary represents, for example, the wall of an anatomical structure that defines the boundary of the lumen, which is the colon.

12. A computer program comprising computer program code, wherein the computer program code is executable on a processor, and causes the processor to perform the method according to any one of claims 1 to 11.

13. A processing apparatus for processing volumetric image data and identifying material boundaries within the volumetric image data, Obtaining volumetric image data representing the anatomical regions of the subject, At least for each subset of voxels within the anatomical region, Identifying a volumetric subregion surrounding a voxel, wherein the volumetric subregion includes at least a subset of directly adjacent voxels. Obtaining a predetermined boundary transition function, wherein the predetermined boundary transition function represents a model of expected voxel values ​​as a function of the distance across the specified material boundary, Fitting the predetermined boundary transition function to the voxel values ​​contained within the volumetric sub-region, including determining the fitting parameters of the model function, Determine fitting quality parameters that indicate the quality of fitting the aforementioned model function, Execute, The aforementioned processing apparatus further, Identifying candidate spatial points of the material boundary for each voxel based on the determined fitting parameters and fitting quality parameters, At a minimum, the material boundary in the volumetric image data is estimated based on a subset of the identified candidate spatial points. A processing unit that executes this process.

14. The boundary transition function is defined by the minimum voxel value m on one side of the material boundary. ｌｏｗ Therefore, the maximum voxel value m on the other side of the material boundary ｈｉｇｈ As a result of the process up to this point, the voxel values ​​are modeled, and the fitting parameters are, The minimum voxel value m ｌｏｗ , The aforementioned maximum voxel value m ｈｉｇｈ , The vertical distance d from the material boundary of the voxel, and The unit normal vector of the aforementioned material boundary, The apparatus according to claim 13, comprising one or more of the following.

15. The apparatus according to claim 13 or 14, wherein identifying the candidate spatial points includes identifying candidate voxels, each candidate spatial point being identified based on the fitting parameters of each candidate voxel, and identifying the candidate voxels includes performing masking of voxels in the imaged anatomical region based on one or more of the values ​​of the fitting parameters and / or the values ​​of the fitting quality parameters.

Images