JP2007522826A - Volume visualization method using tissue mixing - Google Patents

Abstract

A method for calculating a relative contribution of different organizations to at least one data element in a medical object data set includes: the medical object data set includes a data element, and the data element includes two or more data elements. A method of assigning data values to relative positions in a multidimensional geometric space having tissue, wherein parameters are calculated for the at least one data element, and the parameters are in proximity of the at least one data element. The parameter for the data element in the data set of the medical object depending on the data value of the surrounding data element and the data value of the at least one data element, the parameter being present in a boundary region between tissues It is characterized by being compared with the combination of.

Description

  The present invention is a method for calculating a hypothetical contribution of an organization different from at least one data element in a medical object data set, the medical object data set having data elements, and two data elements. The present invention relates to a method for assigning data values to respective positions in a multidimensional geometric space having the above organization.

  Tissue mixing is described in, for example, the document “Partial Volume Tissue Segmentation using Gray-Level Gradient”, D.M. C. Williamson, N .; A. Thacker, S .; R. Williams and M.M. Pokric, MIUA 2002 discloses a method for the calculation and is well known in the art. Tissue mixing calculations can be used to solve some volume impact problems. The problem of the effect of some of the volumes is a misclassification of CT values in the voxels that appear at the boundary between different tissue types, due to the contribution of that voxel from an unknown proportion of surrounding tissues. The problem appears whenever different tissues or materials come into contact at the boundary, and therefore appears, for example, between bone and soft tissue, but also, for example, between the area of tissue and air in the colon and air It appears in combinations such as liquids and different types of tissues.

  It is an object of the present invention to provide an improved method for calculating tissue mixing. This is achieved in accordance with the object of the invention in which parameters are calculated for at least one data element, which parameters are the data value of the surrounding data element in the neighborhood of the at least one data element and at least one data element In addition, the parameters are compared with the parameter combinations for the data elements in the medical object data set appearing in the boundary region between the tissues.

  Therefore, the calculation of different tissue contributions in a single voxel is enhanced by adding additional measurements to the amount of data. These measurements include the gradient value of the tone value in the voxel and neighboring voxels around the voxel, the average tone value of the number of neighboring voxels around the voxel, each of the different sizes (multi-scale method), and the specific organization It includes the adaptation of various values of measured tone values and gradients to the model of predictive behavior for transition types and the use of an optimization model and goodness of fit as a means.

  This can be further explained, for example, using a division of three tissue types that provides a range of intensity values from high intensity to low intensity. For simplicity and clarity, the three selected tissue types are stool that emphasizes air, soft tissue and contrast. The method of calculating tissue mixture begins with some assumptions about the information in the data set of interest. The assumption on which the present invention is based is that the edge or boundary between the two types of tissue looks like, for example, a smoothed step edge of a Gaussian distribution between two materials each having a constant tone value.

  For tissue transitions, a model is constructed about how the tone values are derived, usually expressed as the magnitude of the gradient, and changes when it meets the edge in the direction of the gradient. This is done by sampling both gradient values and gradient values along the direction of the maximum gradient and plotting them to produce a graph of the gradient value-gradient magnitude relationship. This model gives the predictive behavior of measurements in the vicinity of different transitions and is a great help in determining robustly any transition in a voxel between one tissue and another.

In the three types of tissue classification, there are three different transitions possible between any two of the three different types of tissue. They are the boundary between tissue types 1 and 2, the boundary between tissue types 2 and 3, and the boundary between tissue types 1 and 3. The tone values are sampled along a gradient direction at the image boundary, which is perpendicular to the separation plane. More advanced assumptions lead to more complex models, but do not change the general concept of how this tissue mixing is calculated. The method can be further generalized for junctions between three tissues or transitions between four or more tissues.

  Using the above model, any voxel can now be classified as belonging to one of the three transitions. For every voxel in the image, the gradient direction is determined, along which the tone values and gradient magnitudes are sampled. It corresponds best to samples in the vicinity of the voxel, so that any voxel in the appropriate part of the plot of our tone value-gradient magnitude relationship for the boundaries of different tissue types The parameters of the model can be determined, which can position the voxel around it.

  It can also separate the voxels by specific tissue mixing within the voxel, thereby solving the problem of partial volume effects.

  The method of the present invention will be described with application to three tissues in the colon. The effect of partial volume can be understood as follows. Assume three types of materials in the area around the colon surface: gas, tissue and contrast enhancing material (tagged). The CT value φ in the voxel can be modeled as a linear combination of material contributions as follows:

φ = a. μg + b. μt + c. μc
Here, part a corresponds to gas, b corresponds to tissue and c corresponds to tagged material. Those mixed portions can be represented as the center of gravity of the triangle of FIG. These material portions are measured as a single CT value while being acquired. The present invention makes it possible to perform the reverse operation, i.e. to determine the substance mixture starting from the acquired CT values as shown in Fig. Ib. The transition between the two materials is modeled as a smoothed step edge of a Gaussian distribution with variance σ ² (cumulative Gaussian distribution). Assume that φ represents the CT value and | ∇φ | represents the magnitude of the gradient. By plotting | ∇φ | as a function of φ, an arch-shaped curve is obtained. Hereinafter, this curve is referred to as an A curve and is shown in FIG. The value φ depends linearly on the proportion of the substance.

  The general definitions of the Gaussian function and the error function are given by

Ａ曲線の閉論理式は、次式のように与えられ、
Ａ（φ；（Ｌ，Ｈ，σ））＝（（Ｈ−Ｌ）ｇ／σ）｛（√２）ｅｒｆ^−１（２（ψ−Ｌ）／（Ｈ−Ｌ）−１）｝
限定された数のパラメータ、即ち、ステップエッジにおける２つの一定のＣＴ値により定義され、それらの値は、図２ａに示すＨ及びガウス分布のσである。

The closed logical expression of the A curve is given by
A (φ; (L, H, σ)) = ((HL) g / σ) {(√2) erf ⁻¹ (2 (ψ−L) / (HL) −1)}
A limited number of parameters are defined by two constant CT values at the step edge, which are H and Gaussian distribution σ shown in FIG. 2a.

  To focus on the example profile traversing the transition between the two two types of tissue shown in FIG. 3, the position in the A curve is tracked by proceeding along the line from top to bottom. There are three A curves, each showing a curve at the transition between the two types of tissue. Since there are three types of transitions in this example, there are three types of such A curves. In this case, the three A curves show the combination of gradation φ and gradient | ∇φ | for the boundary adjacent voxels in the tissue-gas transition, gas-tagged material transition and tagged material-tissue transition. ing.

The three models are defined by choosing L and H values that best fit the three types of transitions. These values are determined by the following method.
1. For all voxels, the CT value of the gradation value φ and the gradient | ∇φ | are sampled in the direction of the gradient.
Note that in the present invention, it is only necessary to sample at the nearest transition.
2. As shown in FIG. 4, the parameters of the A curve that best fit the CT measurements are determined and L and H parameters are collected in a two-dimensional histogram. The L parameter is plotted on the horizontal axis and the H parameter is plotted on the vertical axis. Since the partial volume value does not appear near the φ axis of the A curve, it is not plotted.
3. The average values for gas, tissue and contrast enhancing fluid are determined from the two-dimensional histogram of FIG. 4b. It can be understood from the comparison with FIG. 4a that the first plot regions 1a and 1b correspond to air in the L direction. By limiting the range of values in the H direction for this value, it can be understood that the region labeled with 5 corresponds to the tissue, and the region labeled with 3 corresponds to the tagged substance. . The value of σ is evaluated by averaging σ of all A curves. Assume that the point spread function is isotropic.

  A model for the CT value characteristics for each of the three transitions is now available, and the contribution of material per voxel can be determined as follows. First, the CT measurements φ, | ∇φ | are sampled in the direction of the gradient at the closest voxel. The positions that best fit the A curves and their local measurements are selected and read as shown in FIG. This allows the exact transition represented by the voxel to be identified.

The two-material transition model is extended to further enable the solution of the tissue mixing classification problem where all three materials are in contact, in this case gas, tissue and tagged materials. be able to. The contents of the fiber model of the three kinds of substances are similar to the contents of the transition model of the two kinds of substances.
In this case, the transition between the gas-tagged material is modeled to cross the tissue transition at an angle α. The extended model models the PSF and shows the behavior of the first derivative φ ′, the second derivative φ ″ and the third derivative φ ′ ″ as a function of the local coordinates x, y, angles α and σ. Here, φ ′ = | ∇φ |, φ ″ = | ∇φ ′ |, and φ ″ ″ = | ∇φ ″ |. First, as shown in FIG. 6, the step edge of the joint is convolved with a Gaussian distribution σ. Each position around the joint corresponds to the proportion of material captured by the center of gravity shown in FIG. Convolution of the step edge at the junction with a Gaussian differential filter gives the magnitude of the gradient, as shown in FIG. 6b. By plotting the magnitude of the CT gradient | ∇φ | against the coordinates of the center of gravity, a parachute-type surface called a P-plane is obtained as shown in FIG. 6c. In almost the same way, both φ ″ and φ ″ ″ are determined for each centroid.

  The model is created using a method similar to that described above. As shown in FIG. 6a, CT values L, M and H corresponding to air, tissue and tagged material are determined using the A curve as described above. Note that the data has fewer joints than edges. The angle is set to 135 ° that is substantially matched with the adhesion of the joint. Tissue mixing in the voxel is determined conventionally. In principle, CT measurements of CT measurements φ, φ ′, φ ″ and φ ′ ″ are sampled in the direction of the gradient at the voxel closest. The position of the P plane that best fits these local measurements (least squares) then determines the centroid.

  By using the fiber model, it is possible to give a high robustness to the influence of disturbance, for example, noise. Furthermore, only the sample at the nearest voxel is required. CT measurements at voxel positions are obtained by convolution with a Gaussian kernel and interpolated using cubic spline interpolation. A sufficiently large size value σ and a sampling trajectory in the gradient direction make the method robust to noise. The resolution of critical tissue surface details (set to diameter ≧ 5 mm) determines the maximum size of the kernel. A typical value of σ is in the range of 1 to 2 mm, and the distance in the direction of the gradient is 0.3 mm.

  Therefore, it is not necessary to sample until a homogeneous tissue region is reached. This provides an exceptional advantage. Other methods of calculating tissue blends known in the art are actually based on different sampling techniques that pass the net very widely and draw voxels from regions in the image that are not closest to the boundaries of the tissue. . This has the disadvantage that voxels outside those ranges have strong tone values from the type of tissue that do not contribute to the original voxel being examined.

  It can be seen that the above method gives information about what type of transition the original voxel was positioned in when positioned within uniform tissue. The exact location of voxel points in the model curve for tone values and gradient magnitudes allows you to identify not only what tissue contributes to that voxel, but what their relative contributions are To do.

  Here, the method of the present invention, described for three types of organizations, can be applied to any number of types of organizations, depending on how complex the application is.

  This method of calculating tissue mixing can be applied to visualization, quantification and segmentation where high accuracy is required. Examples of its use include visualization. For example, in the study of a double contrast colon (the contrast material present in the air and the colon), it is desirable to visualize the air-tissue boundary and the contrast material-tissue boundary. This is not possible using only tone value information (e.g., iso-surface rendering) because tone values cannot be distinguished between three types of tissue transitions. It is necessary to know the tissue mixing so that only the desired transition can be visualized. Also, in lung nodule and tumor quantification, threshold-based methods are often used to distinguish between lesioned and non-lesional voxels, which are counted to determine volume. It becomes. Since voxels near the interface around the lesion also belong to the lesion in particular, accuracy can be improved by using tissue mixing information. This can lead to significant differences in small lesions. Further, in segmentation, it is possible to use simple techniques such as thresholding to obtain sub-voxel accuracy in segmentation by using mixed density. As a result of segmentation, it is no longer a binary volume (as is the case when using simple thresholding with respect to tone values), but more fuzzy than all voxels have values between 0 and 1 representing their tissue density. The right volume. A pure binary volume (only 0 and 1 values exist) is a special case of this fuzzy volume.

  The invention also relates to a computer program and a workstation for carrying out the above method.

  The present invention is a novel method of calculating tissue mixing that can be used independently for applications to volume visualization.

It is a figure which shows (a) projection of the substance mixture regarding CT value, and (b) back projection of CT value. It is a figure which shows the model or A curve of (a) CT value of two substance transitions, and (b) the magnitude | size of a gradient. It is a figure which shows the profile in (a) organization transition, and (b) corresponding A curve. (A) Typical CT slice and (b) L and H histograms corresponding to CT values. FIG. 6 shows a calculation of tissue mixing according to the present invention with (a) a voxel located at the boundary between two substances and (b) a corresponding A curve with appropriate values for tone and gradient. . It is a figure which shows a mode that this invention was applied to three substance transition which shows the magnitude | size of (a) gradient and (b) the parachute type | mold surface P of the model of the magnitude | size of the gradient with respect to the coordinate of a gravity center.

Claims (6)

A method for calculating the relative contribution of different tissues to at least one data element in a medical subject data set:
The medical object data set has data elements, which assign data values to relative positions in a multidimensional geometric space having two or more tissues;
Is the way
Parameters are calculated for the at least one data element, the parameters depending on the data values of surrounding data elements in the proximity of the at least one data element and the data values of the at least one data element; The parameters are compared to the combination of parameters for data elements in the medical object data set present in a border region between tissues;
A method characterized by that.   A method for calculating the relative contribution of different tissues to at least one data element in a medical object data set according to claim 1, wherein the effective value of said parameter for all data elements representing boundaries between tissues. A combination is coupled to a model that represents a relationship between the parameters.   A method for calculating the relative contribution of different tissues to at least one data element in a medical object data set according to claim 1 or 2, wherein the parameters are of two types represented by numbers, A method, characterized in that, for each data element, a gradation value of that data element and a gradient value along the direction of the maximum gradient for that data element.   A method for calculating the relative contribution of different tissues to at least one data element in a medical subject data set according to claim 1, wherein the surrounding data elements in the proximity of said at least one data element are A data element having a data value representing a boundary transition between one organization and another. A computer program for calculating the relative contribution of different tissues to at least one data element in a medical subject data set:
The medical object data set has data elements, which assign data values to relative positions in a multidimensional geometric space having two or more tissues;
A computer program,
The computer program further calculates parameters for the at least one data element, the parameters being the data value of the surrounding data element in the vicinity of the at least one data element and the data value of the at least one data element The computer program compares the parameter with the combination of parameters for data elements in the data set of medical objects existing in a border region between tissues;
A computer program characterized by the above. A workstation for calculating the relative contribution of different tissues to at least one data element in a medical subject data set:
The medical object data set has data elements, which assign data values to relative positions in a multidimensional geometric space having two or more tissues;
A workstation,
The workstation further calculates parameters for the at least one data element, the parameters being a data value of a surrounding data element and a data value of the at least one data element in the vicinity of the at least one data element. The computer program compares the parameter with the combination of parameters for data elements in the data set of medical objects existing in a border region between tissues;
A workstation characterized by that.

Images