JP5133505B2 - Image determination apparatus and X-ray CT apparatus - Google Patents

Description

  The present invention relates to a method for outputting a result of logic of image determination / diagnosis / classification of an image determination apparatus using an image in a medical image diagnostic apparatus or an industrial image determination apparatus.

In a conventional image determination apparatus, when an image determination parameter is changed, a determination result based on a past determination logic cannot be reproduced from the determination result (see, for example, Patent Document 1). However, there are cases where it is desired to change the determination logic due to the evolution of the image determination algorithm (algorithm) and the discovery of new image feature parameters. However, at the same time, there is a desire to be able to reproduce the determination results accumulated in the past. Considering these, “a system capable of realizing a plurality of decision logics of a new decision logic and a previous decision logic (system)” and “a system capable of reproducing a decision result based on a previous decision logic from a result based on a new decision logic” Unless realized, a system that outputs only new decision logic was problematic because it may be unusable.
JP 2003-250794 A, (first page, FIG. 1)

  In the future, an image determination apparatus having a learning function and an image determination apparatus that advances image determination logic will be essential directions. However, on the other hand, based on the decision logic at a certain point in the process of evolution of the decision logic, there is a growing need for an image decision device that can reproduce the decision result of all accumulated images with the same decision logic. is there.

  Therefore, an object of the present invention is to enable determination based on a determination criterion at a certain past time or determination based on a current determination criterion even in an image determination device having a learning function that continues to evolve. Another object of the present invention is to realize an image determination apparatus capable of outputting a determination result based on one determination logic from a plurality of determination logics.

  Another object of the present invention is to realize an image determination apparatus capable of outputting a plurality of determination results based on a plurality of determination logics.

  In the image determination apparatus using an image according to the present invention, it is usually not possible to return to the previous determination logic when the determination / diagnosis / classification logic (hereinafter referred to as determination logic) is changed. However, if the new decision logic is built on the old decision logic, it is possible to reproduce the previous decision logic. In order to prevent the operator / user from knowing this, the image determination apparatus, the image determination method, and the X-ray are characterized by reproducing the past determination logic and the current determination logic internally. A CT apparatus is provided.

  In a first aspect, the present invention provides an image data input means for inputting image data, an image binarization means for pre-processing the image data and binarizing, and the binarized image. On the other hand, using continuous area numbering means for labeling each continuous area, image feature parameter measuring means for measuring and measuring image feature parameters for each labeled continuous area (segment), and using the image feature parameters, In the image determination apparatus, comprising: an image determination unit that performs determination / diagnosis / classification by determination logic, diagnosis logic, or classification logic; and a result output unit that outputs a result obtained by the determination / diagnosis / classification. The determination means provides an image determination apparatus comprising determination logic means for adding a new logic to a past determination logic when the determination logic is changed.

In the image determination apparatus according to the first aspect, since a new determination logic is added to the past determination logic, the determination result of the past determination logic can also be output.
In a second aspect, the present invention provides an image data input means for inputting image data, an image binarization means for performing preprocessing on the image data and binarizing, and the binarized image. On the other hand, using continuous area numbering means for labeling each continuous area, image feature parameter measuring means for measuring and measuring image feature parameters for each labeled continuous area (segment), and using the image feature parameters, Image determination means for performing determination / diagnosis / classification by determination logic, diagnosis logic or classification logic, determination logic condition input means for inputting a plurality of types / conditions of determination logic to the image determination means, and the image determination A result output means for outputting a result obtained by the means, wherein the image determination means is a determinism specified by the determination logic condition input means. Provided is an image determination apparatus characterized in that determination is performed based on a reason.

  In the image determination apparatus according to the second aspect, it is possible to switch to the specified determination logic and output the determination result because it is sufficient to replace only the determination logic unit without affecting the image feature parameter measurement means. It is.

  In a third aspect, the present invention provides the image determination apparatus according to the first or second aspect, wherein the result output means outputs a plurality of types of results obtained by the image determination means using a plurality of determination logics. An image determination apparatus characterized by the above is provided.

  In the image determination device according to the third aspect, it is possible to output a plurality of types of results obtained by a plurality of determination logics. Thereby, it is possible to improve the determination accuracy of the person making the diagnosis and the person making the final decision.

  In a fourth aspect, the present invention provides the image determination apparatus according to any one of the first to third aspects, wherein the result output unit includes a determination result database having a structure capable of holding a plurality of types of determination results. An image determination apparatus is provided.

  In the image determination apparatus according to the fourth aspect, it is possible to create a database having a structure that can have determination results based on a plurality of determination logics. As a result, a plurality of determination results can be output, and the determination accuracy of the person making the diagnosis and the person making the final determination can be increased.

  In a fifth aspect, the present invention provides the determination result database subdivided so that the result output means can output a plurality of types of determination results in the image determination apparatus according to any one of the first to fourth aspects. An image determination apparatus is provided.

  In the image determination apparatus according to the fifth aspect, a combination of image feature parameters is multiplexed so as to correspond to a plurality of determination logics, and a subdivided database structure is realized. It can be output. Thereby, it is possible to improve the determination accuracy of the person making the diagnosis and the person making the final decision.

  In a sixth aspect, the present invention provides a determination result database subdivided so that the result output means can output a plurality of types of determination results in the image determination apparatus according to any one of the first to fifth aspects. From the above, an image determination apparatus is provided that outputs a plurality of types of determination results.

  In the image determination apparatus according to the sixth aspect, a combination of image feature parameters is multiplexed so that a plurality of determination results can be handled, so that a plurality of determination results can be obtained. It can be output. Thereby, it is possible to improve the determination accuracy of the person making the diagnosis and the person making the final decision.

  In a seventh aspect, the present invention is the image determination apparatus according to any one of the first to sixth aspects, wherein the result output unit is subdivided so that a plurality of types of determination results can be output. Therefore, an image determination apparatus is provided that outputs one type of determination result.

  In the image determination apparatus according to the seventh aspect, a plurality of types of determination results can be used, and a result of a logical operation or the like of these results can be used as one final result. Thereby, the accuracy of the determination result can be increased.

  In an eighth aspect, the present invention provides the image determination apparatus according to any one of the first to seventh aspects, wherein the result output unit calculates a plurality of determination results as a plurality of final determination results. An image determination apparatus is provided that outputs a determination result of a seed.

  In the image determination apparatus according to the eighth aspect, it is possible to use a plurality of types of determination results and to output a plurality of types of determination results with the result of logical operation or the like of the results. Thereby, it is possible to improve the determination accuracy of the person making the diagnosis and the person making the final decision.

  In a ninth aspect, the present invention provides the image determination apparatus according to any one of the first to eighth aspects, wherein the image determination unit can change the determination logic by a learning function, and the past determination logic of the change can be obtained. An image determination apparatus characterized in that the result output means selects a plurality of determination results from various stages learned from the past to the present from the image determination means, and outputs the results. provide.

  In the image determination device according to the ninth aspect, since a plurality of determination results in each learning process stage learning from the past to the present are selected and a plurality of determination results are output, It is possible to improve the determination accuracy of the person who makes the final determination.

  In a tenth aspect, the present invention provides the image determination apparatus according to any one of the first to ninth aspects, wherein the image determination unit can change the determination logic by a learning function, and the past determination logic of the change can be obtained. And the result output means selects one determination result from each learning process stage learned from the past to the present from the image determination means, and outputs the result. I will provide a.

  In the image determination apparatus according to the tenth aspect, a result such as a logical operation of a plurality of determination results in various stages learned from the past to the present can be used as one final result. For example, it is possible to predict in advance a direction that is likely to converge according to the progress of the learning process, and to output a determination result.

  In an eleventh aspect, the present invention provides the image determination apparatus according to any one of the first to tenth aspects, wherein the image data input means inputs X-ray image data. provide.

In the image determination apparatus according to the eleventh aspect, an X-ray image determination apparatus can be realized by inputting X-ray image data.
In a twelfth aspect, the present invention provides the image determination apparatus according to any one of the first to eleventh aspects, in which the image data input unit generates a three-dimensional X-ray tomographic image including a plurality of X-ray tomographic images. Provided is an image determination device characterized by inputting.

  In the image determination apparatus according to the twelfth aspect, a three-dimensional X-ray tomogram composed of a plurality of X-ray tomograms, that is, a three-dimensional X-ray tomogram that extends in the xyz coordinate space or an xy coordinate space that extends in the time axis direction 3 By inputting a three-dimensional X-ray tomographic image, a three-dimensional X-ray tomographic image determination device in xyz coordinate space and a three-dimensional X-ray tomographic image determination device in time-series X-ray tomographic image can be realized.

  In a thirteenth aspect, the present invention provides the image determination apparatus according to any one of the first to twelfth aspects, wherein the image data input unit generates a four-dimensional X-ray tomographic image including a plurality of X-ray tomographic images. Provided is an image determination device characterized by inputting.

  In the image determination apparatus according to the thirteenth aspect, a four-dimensional X-ray tomographic image can be formed from a plurality of X-ray tomographic images spread in the z-axis direction and the time-axis direction. As a result, a four-dimensional X-ray tomographic image determination apparatus using a time-series xyz coordinate space can be realized.

  In a fourteenth aspect, the present invention inputs image data, binarizes the image data after preprocessing, performs labeling on the binarized image for each continuous region, and performs the labeling The image feature parameters are measured and measured for each continuous region (segment), and the image feature parameters are used for determination / diagnosis / classification by determination logic, diagnosis logic or classification logic, and the determination / diagnosis / classification is performed. In the image determination method for outputting the result obtained by the above, the determination / diagnosis / classification provides an image determination method in which a new logic is added to a past determination logic when the determination logic is changed To do.

In the image determination method according to the fourteenth aspect, since a new determination logic is added to the past determination logic, the determination result of the past determination logic can also be output.
In a fifteenth aspect, the present invention inputs image data, binarizes the image data after preprocessing, performs labeling on the binarized image for each continuous region, and performs the labeling The image feature parameters are measured and measured for each continuous region (segment), and the image feature parameters are used for determination / diagnosis / classification by determination logic, diagnosis logic or classification logic, and the determination / diagnosis / classification On the other hand, in the image determination method for inputting a plurality of types / conditions of determination logic and outputting the results obtained by the determination / diagnosis / classification, the determination / diagnosis / classification is specified by the type / condition Provided is an image determination method characterized by performing determination based on determination logic.

  In the image judging method according to the fourteenth aspect described above, it is possible to switch to the designated judgment logic and output the judgment result because only the judgment logic unit needs to be replaced without affecting the image feature parameter measuring means. It is.

  In a sixteenth aspect, the present invention provides image data input means for inputting image data, image binarization means for performing preprocessing on the image data and binarizing, and binarized images. On the other hand, using continuous area numbering means for labeling each continuous area, image feature parameter measuring means for measuring and measuring image feature parameters for each labeled continuous area (segment), and using the image feature parameters, An X-ray CT apparatus comprising: an image determination unit that performs determination / diagnosis / classification by determination logic, diagnosis logic, or classification logic; and a result output unit that outputs a result obtained by the determination / diagnosis / classification, The image determination means provides an X-ray CT apparatus comprising determination logic means for adding a new logic to a past determination logic when the determination logic is changed. .

In the X-ray CT apparatus according to the sixteenth aspect, since a new determination logic is added to the past determination logic, the determination result of the past determination logic can also be output.
In a seventeenth aspect, the present invention provides image data input means for inputting image data, image binarization means for performing preprocessing on the image data and binarizing, and binarized images. On the other hand, using continuous area numbering means for labeling each continuous area, image feature parameter measuring means for measuring and measuring image feature parameters for each labeled continuous area (segment), and using the image feature parameters, Image determination means for performing determination / diagnosis / classification by determination logic, diagnosis logic or classification logic, determination logic condition input means for inputting a plurality of types / conditions of determination logic to the image determination means, and the image determination And a result output means for outputting a result obtained by the means, wherein the image determination means is a determination specified by the determination logic condition input means. Provided is an X-ray CT apparatus characterized by making a determination based on logic.

  In the X-ray CT apparatus according to the seventeenth aspect, switching to the designated determination logic and outputting the determination result does not affect the image feature parameter measuring means, and only the determination logic unit needs to be replaced. Is possible.

  As an effect of the image determination apparatus and method of the present invention, even an image determination apparatus having a learning function that is constantly evolving can be determined based on a determination criterion at a past time point, or based on a current determination criterion It is possible to output a determination result based on one determination logic from a plurality of determination logics that can be determined by Another advantage of the present invention is that it is possible to realize an image determination apparatus that can output a plurality of determination results based on a plurality of determination logics.

Hereinafter, the present invention will be described in more detail with reference to embodiments shown in the drawings. Note that the present invention is not limited thereby.
In one embodiment of the present invention described below, a case where image determination is performed on a lung field tomogram obtained by an X-ray CT apparatus is shown.

  FIG. 1 is a configuration block diagram of an X-ray CT apparatus according to an embodiment of the present invention. The X-ray CT apparatus 100 includes an operation console 1, an imaging table 10, and a scanning gantry 20.

  The operation console 1 includes an input device 2 that receives input from an operator, a central processing unit 3 that executes preprocessing, image reconstruction processing, postprocessing, and the like, and X-ray detector data (data) collected by the scanning gantry 20. A data collection buffer 5 for collecting image data, a monitor 6 for displaying a tomographic image reconstructed from projection data obtained by preprocessing the X-ray detector data, a program (program) and an X And a storage device 7 for storing line detector data, projection data, and X-ray tomographic images. The central processing unit 3 includes an image determination device described later.

  The imaging table 10 includes a cradle 12 on which a subject is placed and put into and out of the opening of the scanning gantry 20. The cradle 12 is moved up and down and moved linearly by a motor built in the imaging table 10.

  The scanning gantry 20 includes an X-ray tube 21, an X-ray controller 22, a collimator 23, a multi-row X-ray detector 24, a DAS (Data Acquisition System) 25, and the body axis of the subject. A rotating unit controller 26 for controlling the X-ray tube 21 rotating around, and a control controller 29 for exchanging control signals and the like with the operation console 1 and the imaging table 10. The scanning gantry tilt controller 27 can tilt the scanning gantry 20 forward and backward in the z direction by about ± 30 degrees.

  FIG. 2 is an explanatory diagram of the geometric arrangement of the X-ray tube 21 and the multi-row X-ray detector 24. The X-ray tube 21 and the multi-row X-ray detector 24 rotate around the rotation center IC. When the vertical direction is the y direction, the horizontal direction is the x direction, and the table traveling direction perpendicular thereto is the z direction, the rotation plane of the X-ray tube 21 and the multi-row X-ray detector 24 is the xy plane. The moving direction of the cradle 12 is the z direction.

The X-ray tube 21 generates an X-ray beam called a cone beam (CB) CB. A view angle of 0 degree is defined when the central axis direction of the cone beam CB is parallel to the y direction.
The multi-row X-ray detector 24 has, for example, 256 X-ray detector rows. Each X-ray detector array has, for example, 1024 channel X-ray detector channels.

  The projection data collected by the X-ray irradiation is A / D converted from the multi-row X-ray detector 24 by the DAS 25 and input to the data acquisition buffer 5 via the slip ring 30. Data input to the data collection buffer 5 is processed by the central processing unit 3 according to a program in the storage device 7, reconstructed into a tomographic image, and displayed on the monitor 6. Further, the tomographic image can be stored in the storage device 7 or transferred to another system via a network.

FIG. 3 is a flow chart showing an outline of the operation of the X-ray CT apparatus 100 of the present invention.
In step S1, first, a helical scan operation is performed while rotating the X-ray tube 21 and the multi-row X-ray detector 24 around the subject and moving the cradle 12 on the imaging table 10 linearly. The table linear movement z-direction position Ztable (view) is added to the X-ray detector data D0 (view, j, i) represented by the view angle view, the detector row number j, and the channel number i. X-ray detector data is collected. Alternatively, in a conventional scan (also called an axial scan) or a cine scan, X-ray detector data is collected while the cradle 12 on the imaging table 10 is fixed.

  In step S2, the X-ray detector data D0 (view, j, i) is preprocessed and converted into projection data. As shown in FIG. 4, the preprocessing includes step S21 offset correction, step S22 logarithmic conversion, step S23 X-ray dose correction, and step S24 sensitivity correction.

  In step S3, beam hardening correction is performed on the preprocessed projection data D1 (view, j, i). In the beam hardening correction S3, the projection data subjected to the sensitivity correction S24 in the pre-processing S2 is D1 (view, j, i), and the data after the beam hardening correction S3 is D11 (view, j, i). The beam hardening correction S3 is expressed, for example, in a polynomial form as follows.

  In step S4, a z-filter convolution process for applying a filter in the z direction (column direction) to the projection data D11 (view, j, i) subjected to beam hardening correction is performed.

  In step S4, the projection of the multi-row X-ray detector D11 (ch, row) (ch = 1 to CH, row = 1 to ROW) subjected to beam hardening correction after pre-processing in each view angle and each data acquisition system. For example, a filter having a column filter size of 5 columns as shown below is applied to the data in the column direction.

(W ₁ (ch), w ₂ (ch), w ₃ (ch), w ₄ (ch), w ₅ (ch)),
However,

And
The corrected detector data D12 (ch, row) is as follows.

It becomes. If the maximum channel value is CH and the maximum column value is ROW,

And
In step S5, reconstruction function superimposition processing is performed. That is, Fourier transform, multiplication with a reconstruction function, and inverse Fourier transform are performed. In the reconstruction function superimposing process S5, assuming that the data after the z filter convolution process is D12, the data after the reconstruction function convolution process is D13, and the reconstruction function to be superimposed is Kernel (j), the reconstruction function convolution process is as follows. It is expressed as

Note that * represents a superimposition operation.
That is, since the reconstruction function Kernel (j) can perform an independent reconstruction function convolution process for each j column of the detector, the difference in noise (noise) characteristics and resolution characteristics for each column can be corrected.

  In step S6, three-dimensional backprojection processing is performed on the projection data D13 (view, j, i) subjected to reconstruction function superimposition processing to obtain backprojection data D3 (x, y). At this time, the image to be reconstructed is reconstructed into a three-dimensional image on a plane perpendicular to the z axis and an xy plane.

In step S7, post-processing such as image filter superposition and CT value conversion is performed on the backprojection data D3 (x, y, z) to obtain a tomographic image D31 (x, y).
In post-processing image filter convolution processing, the tomographic image after three-dimensional backprojection is D31 (x, y, z), the data after image filter convolution is D32 (x, y, z), and the image filter is Filter (z )

  That is, since an independent image filter convolution process can be performed for each j column of the detector, a difference in noise characteristics and resolution characteristics for each column can be corrected. The obtained tomographic image is displayed on the monitor 6.

  In the X-ray CT apparatus 100 described above, continuous tomographic images are obtained in the z direction. Image determination is performed on the tomographic images continuous in the z direction of the X-ray CT apparatus thus obtained according to the example of the processing flow of the X-ray CT tomographic image shown in FIG. This image determination is performed in an image determination apparatus included in the central processing unit 3. Then, the image determination device includes, for example, an image data input unit that inputs an image from the storage device 7, an image binarization unit that preprocesses and binarizes the image, and labels the binarized image for each continuous region. Continuous area numbering means for performing image feature parameter measuring means for measuring and measuring image feature parameters for each labeled continuous area (segment), using this image feature parameter, by decision logic, diagnosis logic or classification logic, For example, image determination means for performing determination / diagnosis / classification and result output means for outputting the result obtained by the determination / diagnosis / classification to the monitor 6 are included.

In step P1, a determination logic output method is selected or designated.
For example, the judgment result candidate of the lung field lesion based on the current latest judgment logic and the judgment result candidate of the lung field lesion based on the judgment logic six months ago are different from the judgment logic of the current six months ago. Output part. As described above, it is possible to output not only the current determination result but also the result of six months ago and compare it with the result of the examination six months ago. As a result, it can be used as a material for determining which lesion is growing compared to the result of image determination six months ago.

In step P2, an image is input from the image data input means.
In step P3, the image input from the image data input means is preprocessed and binarized.

  In this case, binarization is performed with a threshold that can cut out the inside of the lung field, and preprocessing is performed by adjusting the contour of the binarized image by expansion / contraction of the logical filter, and a mask area. As above, the region is cut out from the original tomographic image. In addition, in order to improve the accuracy of the region cut out by this region cut-out, a labeling process is performed on the cut-out region, and the lung region is automatically determined by the size and shape feature parameters of the continuous region. You may choose. Binarization is performed with a threshold value that can extract a candidate for a lesion in the lung in the extracted region.

  In step P4, the binarized image is labeled for each continuous area, and consecutive area numbers are assigned. Then, a lesion candidate in the lung field or a peripheral blood vessel can be divided for each continuous region.

  In Step P5, an image feature parameter is obtained for each labeled continuous region (segment). Here, for example, the following geometric feature parameters are obtained for each labeled continuous region.

1. Area S
2. Concentration sum (CT value sum) D
3. X-direction length Lx and y-direction length Ly of the circumscribed rectangle (Lx and Ly are also referred to as ferret diameters)
4). X-direction length Lx, y-direction length Ly (ferret diameter ratio) of the circumscribed rectangle
5. Perimeter L
6). Circularity S / (4πL ² )
7). When the ellipse is approximated, the major axis RL, minor axis RS, inclination θ
8). Number of holes in the circle (Euler number)
In step P6, the determination is performed according to the determination logic selected or designated in advance based on the value of the obtained image feature parameter.

As shown in FIG. 5, each classified space region, that is, a result output region, is determined as a part of a parameter space (Mahalanobis space) defined by the characteristic parameters P _{1 to} P _m . The decision logic defines each classification space area and result output area.

In Step P7, the result is output according to the output method selected or designated in advance.
In other words, only the latest output result is output, or the output result of the stage in the middle of learning or parameter change is also output, or they are processed into one result output, and the result output is selected in advance, The result is output based on a method specified in advance.

  In step P8, it is determined whether all the images have been completed. If there is a tomographic image that has not yet been processed, the process returns to step P2 to repeat the process. In this way, two-dimensional image processing may be performed for each tomographic image on the xy plane of the X-ray CT apparatus. Alternatively, a tomographic image continuous in the z direction may be input from an X-ray CT apparatus and processed as a three-dimensional image. An example of image determination in the case of a three-dimensional image is shown in FIG.

In FIG. 7, the processing proceeds in the same manner as the two-dimensional image processing of FIG.
In step P11, a determination logic output method is selected or designated.
Similar to step P1, not only the current determination result but also the result of six months ago can be output together and compared with the result of inspection six months ago. As a result, it can be used as a material for determining which lesion is growing compared to the result of image determination six months ago.

In Step P12, a tomographic image continuous in the z direction is input and input as a three-dimensional image.
When inputting a continuous tomographic image in the z direction, in an X-ray CT apparatus having a multi-row X-ray detector, if there is sufficient S / N, the slice image having as thin a slice thickness as possible. If image reconstruction is performed at fine image reconstruction intervals in the z direction, the amount of information as a three-dimensional image increases and the accuracy increases. However, image processing and measurement time are required. In order to shorten the image processing time and increase the throughput (through-put), it is preferable to optimize the slice thickness and the image reconstruction interval in consideration of the size of the lesion to be searched.

Further, it is preferable to reduce the exposure by optimizing the X-ray dose for obtaining a sufficient S / N ratio corresponding to the slice thickness and reducing the X-ray dose.
In step P13, the image input from the three-dimensional image data input means is preprocessed and binarized.

  In this case, binarization is performed with a threshold that can sufficiently extract the inside of the lung field, and preprocessing includes a three-dimensional logic filter so that arteries, veins, hearts, etc. that are not to be examined can be removed. A target lung field region is cut out from a binarized region using a three-dimensional morphology filter or an original image cut out using the binarized region as a mask. Further, in order to further improve the accuracy of selection of the three-dimensional region cut out by cutting out the three-dimensional region (three-dimensional segment), the three-dimensional labeling process is performed on the cut-out three-dimensional region, and the three-dimensional continuous region is obtained. The three-dimensional region may be selected by automatically determining the feature parameters such as the size and shape of the image. In this way, binarization is performed with a threshold value that can extract candidate lung lesions in the extracted target lung field region.

  In Step P14, the binarized three-dimensional image is subjected to three-dimensional labeling for each continuous region. The labeling is performed by dividing into candidates for lesions in the target lung field of the three-dimensional region and three-dimensional continuous regions of peripheral blood vessels. The neighborhood condition (6 neighborhood / 18 neighborhood / 26 neighborhood) that is the sensitivity of the labeling may be selected based on the S / N of the image at this time.

In Step P15, a three-dimensional region feature parameter is obtained for each three-dimensionally labeled three-dimensional continuous region (three-dimensional segment).
For example, the following geometric feature parameters are obtained for each three-dimensionally labeled three-dimensional region.

1.3 Volume of the three-dimensional region V
2. Concentration sum (CT value sum) D in a three-dimensional region
3. X-direction length Lx, y-direction length Ly, z-direction length Lz (Ferret diameter) of the circumscribed cuboid of the three-dimensional continuous region
4. Ferret diameter ratio of circumscribed rectangular parallelepiped in 3D continuous area, Ly / Lx, Lx / Lz, Lz / Ly
5. Surface area S of 3D region
6). Sphericality 6π ^1/2・ V / S ^3/2
7). Diameter RL, RM, RS slope θx, θy, θz when approximated by ellipsoidal sphere
In Step P16, determination is performed according to the determination logic selected or designated in advance based on the value of the obtained three-dimensional region feature parameter.

Similar to step P6, as shown in FIG. 5, the spatial regions classified as part of the parameter space (Mahalanobis space) defined by the characteristic parameters P _{1 to} P _m , that is, the result output regions are determined. The decision logic defines each classification space area and result output area.

In Step P17, the result is output according to the output method selected or designated in advance.
As in Step P7, that is, only the latest output result is output, or the output result of the stage in the middle of learning or parameter change is also output, or the results are output into a single result output. The result is output based on a method designated in advance and designated in advance.

In this way, the geometric feature parameters used in the determination logic for the two-dimensional image processing and the three-dimensional image processing have been described. Hereinafter, the determination logic will be described.
The decision logic that has been known for a long time includes the value of each geometric feature parameter value in the decision tree structure of “If then ~, else ~ type” as shown in FIG. The logic can be advanced in comparison with the threshold value, and finally can be divided into each classified class (for example, the type of lung lesion). Alternatively, as shown in FIG. 8B, the logic of the decision tree structure is advanced while comparing the linearly combined value of each weighted geometric feature parameter with a threshold value, and each class is assigned to each class. It can also be classified.

  Also, as shown in FIG. 8 (c), parameter weighting determination for classifying each class 1 to class m by applying a weighting coefficient matrix to the parameter value sequence and comparing it with the determination threshold sequence without taking a determination tree structure. You can also. If you want to achieve a particularly simple supervised learning function,

There is a learning method in which errors are distributed to θ ₁ to θ _n , W _{11 to} W _1m ,... W _{n1 to} W _nm and converged so that the errors of Δθ ₁ to Δθ _n become positive, for example. .
In this way, classification can be performed based on the determination logic, and the determination logic varies depending on a determination threshold, a weighting coefficient, a determination tree structure, or the like.

In this case, in the parameter space, as shown in FIGS. 9 (a) and 9 (c), what is divided into each of the classes 1, 2 and 3 by a certain decision logic is changed by learning or adjustment of the decision logic, As shown in FIGS. 9B and 9D, the areas of classes 1, 2, and 3 move. In other words, this means that the weighting coefficient and the threshold value, which are parameters of the linear weighting coefficient W _i1 · P ₁ + W _i2 · P ₂ +... + W _im · P _m > θ _i (i = 1 to n) serving as the boundary line, change. As a result, these boundaries change and the region moves.

As shown in FIG. 10, briefly and took parameter space in the horizontal axis direction, indicating that were classified into Class A _1, B ₁ in J ₁ at decision logic time T ₁ in FIG. 10 (a). Decision logic in time T _2, is changed to J _2, when it is classified into class _{A 2, B 2, C 2} , which was the decision logic J ₂ Class A ₂ enters the decision logic J ₁ in Class A ₁ I understand that. Similarly, it can also be seen that what has entered class B ₂ in decision logic J ₂ enters class B ₁ in decision logic J ₁ . However, it is unclear whether a decision logic J ₂ that has entered class C ₂ will enter decision logic J ₁ in class A ₁ or class B ₁ . That is, the determination result of the determination logic J ₁ cannot be restored from the determination logic J ₂ .

  For example, when evaluating an automatic diagnosis algorithm for a lung lesion in lung cancer screening, the evaluation is performed over a long period of time, and a lung lesion is found only in a very small number of subjects. In such a case, if the algorithm of the determination logic is changed, the evaluation of the previous determination logic and the determination result in the previous determination logic are not known, and the evaluation returns to the start. In addition, when there are very few objects to be detected as described above, there are many cases where it is desired to restore the algorithm of the determination logic by trial and error. For this reason, an automatic diagnosis system that can easily perform the original determination even if the determination logic is updated is very convenient for performing evaluation results simultaneously with a plurality of determination logics and evaluating the determination results. .

Therefore, the determination logic means of the image determination means for adding new logic to the past determination logic when changing the determination logic will be described below.
For example, in FIG. 10B, it is assumed that classification can be made into classes A ₁ and B ₁ by the determination logic J ₁ at time T ₁ . It is assumed that the decision logic J ₂ at time T ₂ is classified into classes A ₂ , B ₂ , CA ₂ , and CB ₂ . In this case, the boundary line of the judgment logic J ₁ because it is stored in the boundary line of the class CA _2, CB _2, Class A _2, CA ₂ of the decision logic J ₂ are classified into the determination logic J ₁ in Class A ₁ , class B _2, CB ₂ decision logic J ₂ is readily seen to be classified in the decision logic J ₁ in class B _1. That is, in this case, it can be said that the result of the determination logic J ₁ can be restored from the determination logic J ₂ . In this way, it is understood that the past judgment logic can be restored if the boundary line used in the past judgment logic is continuously held.

For example, FIG. 11 has the following decision logic history.
(1) Judgment logic J _{1 at} time T ₁
(2) Judgment logic J _{2 at} time T ₂
(3) determining at time T ₃ logical J _3,
(4) determining at time T ₄ logical J _4,
In this case, the classification as a result of each determination is as follows.

(1) In decision logic J ₁ , B ₁ : normal, R ₁ : abnormal (2) In decision logic J ₂ , B ₂ : normal, K ₂ : doubtful, R ₂ : abnormal (3) In decision logic J ₃ B _3, G _3: normal, K _3: doubtful, R _3: In abnormal (4) decision logic _{J 4, B 4, G 4} : normal, K _4: suspicious, R _4, P _4: the abnormality determination result for this If the database is classified in detail into classes C _{1 to} C _11, it can return to any decision logic at any time. This determination result database is not divided into classes C _{1 to} C ₁₁ from the beginning, and is subdivided at each time.

At time T ₁ , C _{1 to} C ₃ and C _{4 to} C ₁₁
At time T _2, and C ₁ -C _3, a C _4, and _{C 5 ~C 8, C 9 ~C} 11
At time T ₃ , C _{1 to} C ₂ , C ₃ , C ₄ , C _{5 to} C ₆ , C _{7 to} C ₈ , C ₉ and C _{10 to} C ₁₁
At time T ₄ , one by one is classified from C _{1 to} C ₁₁ .

  In this way, it can be seen that a database that is always fragmented grows with decision logic. Such a database is “a database capable of returning the determination result to the previous determination logic immediately”.

  Further, when the image determination apparatus is moved, it is possible to select, set and input a determination output pattern in advance, which is convenient for operation. In that case, as shown in FIG. 12, the decision logic algorithm used for the decision output or the calculation method of the result can be selected. As a result, a plurality of types of determinations can be made between the past algorithm and the current algorithm, and the difference between the respective determination logics can be known. In particular, this function is effective in the comparison stage during the evaluation of the image determination algorithm, and is also effective when it is desired to make a diagnosis with reference to a plurality of determination results.

In order to realize such a database, the following proposals can be considered.
1. Each feature parameter amount is held as a continuous value. (See Fig. 13 (a))
2. Each feature parameter amount is classified with sufficient resolution. (See FIG. 13 (b))
3. Each feature parameter amount is classified according to the resolution at that time. (See FIG. 13 (c))
Since the plans 1 and 2 are recorded with a sufficient resolution of the feature parameter space, it is possible to determine the determination target example determined at the old time with the latest determination logic. However, as in the case of the third proposal, if the determination example of the old time is recorded only with the resolution of the rough feature parameter space, the determination logic at the new time may not be applied. For this reason, it is preferable to record in the database with an appropriate resolution of the feature parameter space.

  In the above image determination apparatus, according to the image determination apparatus and the method of the present invention, even an image determination apparatus having a learning function that is constantly evolving can be determined based on a determination criterion at a certain point in the past. Thus, it is possible to realize an image determination apparatus that can output a determination result based on one determination logic from a plurality of determination logics that can be determined based on the current determination criterion. In addition, it is possible to realize an image determination apparatus that can output a plurality of determination results based on a plurality of determination logics.

Note that the image reconstruction method may be a three-dimensional image reconstruction method by a conventionally known Feldkamp method. Furthermore, other three-dimensional image reconstruction methods may be used. Also, two-dimensional image reconstruction may be used.
Although this embodiment is written based on a medical X-ray CT apparatus, it is used in an X-ray CT-PET apparatus, an X-ray CT-SPECT apparatus or the like combined with an industrial X-ray CT apparatus or another apparatus. it can.

  In the present embodiment, the determination logic is performed using the geometric feature parameter and the parameter value. However, a plurality of standard templates (templates) of typical lesion sites registered in advance are prepared, The same effect can be obtained even if the determination logic is set based on the degree of correlation with various templates. In addition, a determination logic condition input unit may be separately provided in the image determination apparatus, and the determination logic may be performed using a parameter value input from the determination logic condition input unit.

  In the present embodiment, in the case of three-dimensional image processing, an xy tomographic image that is continuous in the z direction is used, but depending on the application, it may be applied to an xy tomographic image that is continuous in time in the t direction in the continuous time direction. Similar effects can be expected. The same applies to a four-dimensional image continuous in the z direction and the t direction.

  In this embodiment, the decision logic of the decision tree structure and the decision logic of the weighted learning method are used, but the same effect can be obtained in the decision method of the back propagation method (Back Propagation method).

  In the present embodiment, several two-dimensional geometric feature parameter amounts and three-dimensional geometric feature parameter amounts are used, but the same effect can be obtained by using other feature parameter amounts.

  In the present embodiment, an example of lung field inspection is used, but it is also used for inspection of other parts, and other than the medical image determination, other industrial image determination It can be used in the same manner for an inspection object or an image determination apparatus.

  In the present embodiment, an example of an image determination apparatus based on two-dimensional or three-dimensional image processing has been described. Can be used similarly.

1 is a block diagram showing an X-ray CT apparatus according to an embodiment of the present invention. It is explanatory drawing which shows rotation of a X-ray generator (X-ray tube) and a multi-row X-ray detector. It is a flowchart which shows schematic operation | movement of the X-ray CT apparatus concerning one embodiment of this invention. It is a flowchart which shows the detail of pre-processing. It is a figure which shows the result output area | region in parameter space. It is a flowchart which shows the example of the flow of a process of a X-ray CT tomogram. It is a flowchart which shows the example of the flow of a process of the three-dimensional image by X-ray CT. (A) It is a figure which shows the example 1 of a judgment tree. (B) It is a figure which shows the example 2 of a judgment tree. (C) It is a figure which shows the example of parameter weighting determination. (A), (b), (c), (d) It is a figure which shows the change of the characteristic parameter space area | region of each classified class by the change of determination logic. (A) It is a figure which shows the classification method which cannot be restored | restored. (B) It is a figure which shows the classification | category method which can be decompress | restored. It is a diagram illustrating a database corresponding to each stage of evolution by learning function (T ₁ ~T _4). (A) It is a figure which shows the example of input of determination logic. (B) It is a figure which shows the example of a determination result. (A) It is a figure which shows the database which waits for each characteristic parameter by a continuous value. (B) It is a figure which shows the database which classifies each characteristic parameter with sufficient resolution beforehand. (C) It is a figure which shows the database which classifies each feature parameter by the resolution of the feature parameter space at that time.

Explanation of symbols

1 Operation Console 2 Input Device 3 Central Processing Unit 5 Data Collection Buffer 6 Monitor 7 Storage Device 10 Imaging Table 12 Cradle 20 Scanning Gantry 21 X-ray Tube 22 X-ray Controller 23 Collimator 24 Multi-row X-ray Detector 25 DAS (Data Collection Device) )
26 Rotating part controller 27 Scanning gantry tilt controller 29 Control controller 30 Slip ring 100 X-ray CT apparatus

Claims (7)

In an image determination apparatus for performing determination based on predetermined determination logic and obtaining a determination result for image data,
When processing based on different decision logic is performed at different time points in time series, the classification indicating the determination result based on the determination logic used in the past and the determination result based on the new determination logic are used for the past. An image determination apparatus comprising a determination result database for subdividing a classification indicating a determination result into a common classification corresponding to all classifications different from a classification indicating a determination result based on the determination logic based on the determination logic . The image determination apparatus according to claim 1, further comprising a result output unit that outputs a classification indicating a plurality of types of determination results based on different determination logics . The image determination apparatus according to claim 1, further comprising a result output unit that outputs a selected one of classifications indicating a plurality of types of determination results based on different determination logics. . The image determination apparatus, by changing the determination logic by the learning function, when at different times series, to any one of claim 1 to 3, characterized in that different determination logic is obtained Image determination device. The image determination device includes:
Image data input means for inputting image data;
Image binarization means for preprocessing and binarizing the image data;
Continuous area numbering means for labeling the binarized image for each continuous area;
Image feature parameter measurement means for measuring and measuring image feature parameters for each labeled continuous region (segment);
Using said image characteristic parameter, the image determination apparatus according to any one of claim 1 to 4, characterized in that an image determination means for determining based on a predetermined decision logic. The image data, the image determining apparatus to any one of claim 1 to 5, characterized in that an X-ray image data. X-ray CT apparatus characterized by having the image determining apparatus according to any one of claims 1-6.

Images