JP2014161388A - Image processing device, image processing method, control program of image processing device, and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently and accurately extract oblique fissures of lung from a three-dimensional image of the lung area.SOLUTION: An image processing device is configured to: extract pixels of which maximum principal curvatures are equal to or more than a predetermined value, from a three-dimensional image; extract planar areas composed of the plurality of continuous pixels of which angle differences of direction vectors of the maximum principal curvatures are less than a first angle, among the extracted pixels; and assign first identification information to respective planar areas according to the direction vectors of maximum principal curvatures of the pixels.

Description

  The present invention relates to an image processing apparatus and an image processing method. More specifically, the present invention relates to an image processing apparatus and an image processing method for extracting an interlobar fissure in a three-dimensional tomographic image of a lung of a subject.

  Lungs are one of the most important organs for living life, and technology for accurately detecting and diagnosing lesions and abnormalities in the lungs of a subject is important in the fields of health, medical care, and research. is there.

  The human lung is generally composed of the upper lobe, middle lobe, and lower lobe of the right lung, and the upper lobe and lower lobe of the left lung, and there is a thin membranous tissue called interlobar fissure between each lobe. Interlobar fissure extraction is a basic process of lung structure analysis, and provides useful information for lesions per lung lobe or preoperative diagnosis. Knowing which lesion belongs to the lung lobe can provide, for example, important clues for predicting at what speed and how the lesion will expand in the future.

  A technique for acquiring an image (an X-ray transmission image and an X-ray tomographic image) in a subject (living body) imaged using X-rays is widely used mainly in the medical field. Furthermore, for early detection and early treatment of lesions occurring in the lung, it is expected to realize computer-aided diagnosis (CAD) using lung tomographic images that can reduce the burden on the doctor. CT (Computed Tomography) and MRI (Magnetic Resonance Imaging) play a central role in diagnostic imaging that can be used for CAD in modern medicine.

  In particular, X-ray tomograms obtained using a CT device have high spatial resolution, and various types of organs and tissues of the subject, including the lungs, can be observed in detail under non-invasive conditions. It is. The CT apparatus provides an image showing a CT value distribution proportional to the density of the internal tissue constituting the subject.

  The CT apparatus has an X-ray tube and a detector arranged to receive X-rays emitted from the X-ray tube. X-rays emitted from the X-ray tube toward the subject pass through the subject and reach the detector, and the detector detects the intensity of the transmitted X-ray. The CT apparatus analyzes and integrates the intensity of transmitted X-rays when X-rays are emitted from various directions toward the subject, thereby generating projection data (tomographic images) for each tomographic portion of the subject.

  Recently, spiral activated scans (called helical scans or spiral scans) have been developed, and multi-detector row CT (MDCT) devices that can generate 3D images by arranging 2D tomographic images continuously. Has appeared. This MDCT apparatus is a CT apparatus provided with a plurality of detector rows (also called an element in which a plurality of detectors (channels) are arranged). The MDCT apparatus is capable of imaging a wide range at high speed, and can reduce the burden on the subject, such as reducing the number of breath-holding times on which the subject (subject) is forced. Further, since the slice thickness of each tomographic image can be 1 mm or less, even a small lesion can be depicted in the generated three-dimensional image.

  Examples of the method for accurately extracting the interlobar fissure using the three-dimensional lung image generated by using the MDCT apparatus having the above-described characteristics include the following.

  Non-Patent Document 1 describes a method of extracting a foliar fissure by a morphological operation that extracts a surface shape after applying a two-dimensional line enhancement filter to a three-dimensional image.

  Non-Patent Document 2 describes a method of classifying each leaf blood vessel to set a search region for interlobar fissures and applying a surface enhancement filter to this region to extract interlobar fissures.

  Non-Patent Document 3 extracts interlobar fissures by evaluating hypoclinic fissures created by registration of oblique fissure 3D atlases and 3D images and oblique fissure candidates obtained by 2D ridge line extraction. The method is described.

  Non-Patent Document 4 describes a technique for emphasizing an interlobar fissure with a supervised emphasis filter and extracting the interlobar fissure by a threshold method.

  Non-Patent Document 5 uses the membranous nature of interlobar fissures, performs surface Laplacian equilibration for polygons obtained by the marching cube method, extracts surface shapes, and uses expanded Gaussian images for each surface shape. A technique is described in which the interlobar fissure is extracted by obtaining a large surface shape using the obtained normal vector.

Mitsuru Kubo, Noboru Niki, Kenji Eguchi, Masahiro Kaneko, Naoto Yamaguchi, “Leaving fissures from thin-section CT images using two-dimensional enhancement”, IEICE Transactions (D-II), 2000 January, Vol. J83-D-II, no. 1, p.175-182 Shinsuke Shinada, Mitsuru Kubo, Yoshiki Kawada, Noboru Niki, Hironobu Omatsu, Noriyuki Moriyama, “Interlobar Lattice Extraction Algorithm for Screening Multislice CT Images”, IEICE Transactions (D-II), January 2004, Vol. J87-D-II, no. 1, p.134-145 L. Zhang, E. A. Hoffman, and J. M. Reinhardt, “Atlas-Driven Lung Lobe Segmentation in Volumetric X-Ray CT Images”, IEEE Transactions on Medical Imaging. Vol.25, No.1, pp1-16, 2006 EM van Rikxoort, B. de Hoop, S. van de Vorst, and van Ginneken, “Automatic Segmentation of Pulmonary Segmentation From Volumetric Chest CT Scans”, IEEE Transactions on Medical Imaging. Vol.28, No.4, pp.621- 630, 2009 J. Pu, C. Fuhrman, J. Duric, JK Leader, A. Klym, FC Sciurba, and D. Gur, “Computerized Assessment Pulmonary Fissure Integrity Using High Resolution CT”, Medical Physics, Vol. 37, No. 9, pp.4661-4672, 2010 M. Matsuhiro, Y. Kawata, N. Niki, Y. Nakano, M. Mishima, H. Ohmatsu, T. Tsuchida, K. Eguchi, M. Kaneko, N. Moriyama, “Classification algorithm of lung lobe for lung disease cases based on multi-slice CT images ”, Proc. SPIE Medical Imaging, Vol.7963, pp.796331-1-6, 2011 P. K. Saha, “3D Digital Topology under Binary Transformation with Applications”, Computer Vision And Image Understanding, Vol.63, No.3, pp.418-429, 1996.

  However, the conventional technology as described above has a problem that a surface shape other than the interlobar fissure is erroneously extracted as an interlobar fissure. In the first place, the interlobar fissure is a thin and smooth membrane-like tissue, and the interlobar fissure is imaged with a low contrast on the lung image. Furthermore, it is known that many planes other than interlobar fissures are generated and exist in lungs suffering from lung cancer, severe emphysema, interstitial pneumonia and the like. Therefore, in the prior art, in the lung image obtained by imaging the lung suffering from lung cancer, severe emphysema, interstitial pneumonia, etc. It was difficult to extract the fissure accurately.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an image processing apparatus capable of efficiently and accurately extracting a planar region such as an interlobar fissure in a three-dimensional image such as a lung. Is to provide etc.

  In order to solve the above-described problem, an image processing apparatus according to an aspect of the present invention is an image processing apparatus that extracts a planar region from a three-dimensional image having pixels arranged in three dimensions. A pixel extracting means for extracting a pixel whose maximum principal curvature of a four-dimensional hypersurface having a pixel value of the fourth dimension as a fourth dimension is a predetermined value or more, and a continuous pixel among the pixels extracted by the pixel extracting means A surface that is constituted by a plurality of the pixels and that extracts a planar region from the three-dimensional image in which the angle difference of the maximum principal curvature direction vector on the four-dimensional hypersurface of any two pixels included is less than the first angle First identification information corresponding to the maximum principal curvature direction vector of one or more pixels constituting the planar area is assigned to each planar area extracted by the planar area extracting means and the planar area extracting means First identification information And it includes a grant means.

  In order to solve the above problem, an image processing method according to an aspect of the present invention is an image processing method for extracting a planar region from a three-dimensional image having pixels arranged in a three-dimensional manner. A pixel extraction step of extracting a pixel having a maximum principal curvature of a four-dimensional hypersurface having a fourth dimension of the pixel value of each pixel of the two-dimensional image, and the pixel extracted in the pixel extraction step A planar region composed of a plurality of successive pixels and having an angular difference of the maximum principal curvature direction vector on the four-dimensional hypersurface of any two pixels included is less than the first angle from the three-dimensional image. A planar area extracting step to extract, and a first area corresponding to the maximum principal curvature direction vector of one or more pixels constituting the planar area in each planar area extracted in the planar area extracting step. Add identification information A first identification information addition step that includes.

  According to this configuration and method, pixels whose maximum principal curvature is a predetermined value or more are extracted from the three-dimensional image. A pixel having a large maximum principal curvature is considered to be a pixel corresponding to the boundary surface of the structure represented by the three-dimensional image. The maximum principal curvature direction vector corresponds to the normal vector at the pixel of the boundary surface. Therefore, among the extracted pixels, a plurality of continuous pixels in which the angle difference between the maximum principal curvature direction vectors is less than the first angle is a pixel constituting one planar area.

  Therefore, according to the above configuration, each planar region can be extracted from the three-dimensional image, and the first identification information corresponding to the maximum principal curvature direction vector of the pixel can be given to each planar region. Thereby, even if the contrast of the planar tissue in the three-dimensional image is low, the planar area can be accurately and individually extracted. Therefore, for example, an interlobar fissure that is a planar tissue having a low contrast in a three-dimensional image of a lung field region of a subject can be extracted efficiently and accurately.

  The image processing apparatus according to the present invention may further include a normal vector having an angular difference of a second angle between two adjacent planar areas among the plurality of planar areas to which the first identification information is assigned. If it is less, the first integration means for updating the first identification information given to each pixel constituting the two planar regions to the same value is provided.

  According to this configuration, when the angle difference between the normal vectors of two adjacent planar regions is small, the two planar regions are considered to be one continuous planar region.

  Therefore, the first identification information having the same value can be given to two adjacent planar regions extracted as different planar regions by the planar region extracting means according to the angle difference between the normal vectors. Therefore, the two adjacent planar regions that are substantially continuous planar regions can be identified as one planar region. In addition, separate first identification information can be left to be applied to two planar regions that are considered to be planar regions having greatly different normal vector angular differences. Therefore, even when a plurality of extracted planar regions intersect each other, one continuous planar region can be accurately identified. Therefore, a relatively large planar tissue in the three-dimensional image can be extracted efficiently and accurately.

  The image processing apparatus according to the present invention may further include two planar areas adjacent to each other with a predetermined number of pixels or less sandwiched between the edges among the plurality of planar areas given the first identification information. When the angle difference between the normal vectors is less than the second angle, the pixels assigned to the pixels sandwiched by the pixels constituting the two planar regions and the edges of the two planar regions are described above. Second integration means for updating the first identification information to the same value is provided.

  According to this configuration, the same first identification information as that of the two planar regions can be given to the linear region corresponding to the intersection of the surfaces sandwiched between the two planar regions. Therefore, the two planar regions and the linear region corresponding to the intersection can be identified as one continuous planar region. Therefore, even when a plurality of extracted planar regions intersect each other, it is possible to accurately identify substantially one planar region.

  The image processing apparatus according to the present invention may further include the three-dimensional image representing a lung field region of the subject, including one of the planar regions to which the first identification information having the same value is given, and When the ratio of the volume of the smaller divided area to the volume of the entire lung field area is equal to or greater than a predetermined value by the division plane virtually set to divide the lung field area into two divided areas, An interlobar fissure selection means is provided for selecting the planar region as an interlobar fissured surface.

  Interlobar fissures in the lung field region are relatively large planar tissues, and greatly longitudinally or cross the lung field region. Therefore, when the planar area represents an interlobar fissure, the two divided areas defined by the dividing plane representing the planar area have a certain amount of volume.

  According to said structure, it can be determined appropriately whether this planar area | region is an interlobar fissured surface by the ratio of the volume of the division area with respect to the whole. Therefore, a planar region corresponding to the interlobar fissured surface can be appropriately selected from a plurality of planar regions.

  The image processing apparatus according to the present invention further includes, among the planar regions to which the first identification information having the same value is assigned, the planar region configured by a predetermined number of pixels or more. A candidate plane extracting unit that extracts a region as an interlobar fissure candidate surface, wherein the interlobar fissure selecting unit selects the interlobar fissured surface from the interlobar fissure candidate surface extracted by the candidate plane extracting unit; It is.

  Interlobar fissures in the lung field are relatively large planar tissues.

  According to the above configuration, a relatively small planar region can be determined as a planar region that is not an interlobar fissure, and a relatively large planar region can be selected as an interlobar fissure candidate surface. Therefore, it is possible to appropriately determine a planar region that is a candidate for an interlobar fissure and a planar region that is not an interlobar fissure.

  The image processing apparatus according to the present invention further divides the pixels constituting the three-dimensional image into first pixel group units each including a first number of adjacent pixels, and includes each of the interlobar fissured surfaces. 1st edge pixel which is the 2nd adjacent 1st pixel group which assign | provides one 2nd identification information according to the normal vector of the said interlobar fissured surface to a 1st pixel group, and the said division surface crosses When the angle difference of the sum of the maximum principal curvature direction vectors of all the pixels included in each of the first target pixel group and the first target pixel group is less than a fourth angle, the first target pixel group includes the first end pixel group. The first target pixel group after the first extension processing is performed until the angle difference becomes equal to or larger than the fourth angle, in the first extension processing that gives the second identification information having the same value as the second identification information. Is set to the new first end pixel group and adjacent to the first target pixel group. An expansion unit that repeats while setting a new first target pixel group to which the second identification information is not assigned, and the expansion unit is configured to select the first end pixel group in the first first expansion process. Setting the position including the end of the interlobar fissure plane, setting the first target pixel group on the side not including the interlobar fissure plane adjacent to the first end pixel group, and the first Instead of the sum of the maximum principal curvature direction vectors of all the pixels included in the one end pixel group, the normal vector of the interlobar fissured region included in the first end pixel group is used.

  According to said structure, in the area | region located in the edge part of an interlobar fissured surface, the area | region considered to be a part of an interlobar fissured surface WHEREIN: 2nd identification information of the same value as the pixel group of an interlobar fissured surface Can be granted. Therefore, the interlobar fissure surface can be expanded and the interlobar fissure surface can be extracted more accurately.

  Further, the image processing apparatus according to the present invention further includes the interlobar fissure plane selected by the interlobar fissure selection unit for each first pixel group to which the second identification information having the same value is assigned. For the first pixel group including the pixel value of each pixel, the angle difference between the maximum principal curvature direction vector of the pixel and the normal vector of the interlobar fissured region included in the first pixel group is small. For the first pixel group that is corrected by a large amount and does not include the interlobar fissure plane selected by the interlobar fissure selection means, the pixel value of each pixel is set to the maximum principal curvature direction vector of the pixel and the Emphasis means is further provided for correcting by a larger amount as the angle difference from the average of the maximum principal curvature direction vectors of all the pixels of the first pixel group is smaller.

  According to said structure, the maximum principal curvature of the pixel which has the largest principal curvature direction vector with a small angle difference with the normal vector of an interlobar fissure is emphasized greatly. As a result, the maximum principal curvature of the pixels constituting the interlobar fissures of the interlobar fissure surface is emphasized more than that of the surrounding pixels. The influence from curvature can be reduced.

  Further, in the image processing device according to the present invention, for the first pixel group including the interlobar fissure plane, the enhancement means further includes the maximum principal curvature direction vector of the pixel and the pixel value of each pixel. An operation of multiplying the inner product with the normal vector of the region of the interlobar fissured surface included in the first pixel group is performed, and for the first pixel group not including the interlobar fissured surface, the pixel value of each pixel is calculated. Then, an operation of multiplying the inner product of the maximum main curvature direction vector of the pixel and the average of the maximum main curvature direction vectors of all the pixels of the first pixel group is performed.

  According to the above configuration, the maximum principal curvature of the pixel having the maximum principal curvature direction vector having a large angular difference from the interlobar fissure normal vector is reduced, and the relative difference between the normal vector of the interlobar fissure is relatively small. The maximum principal curvature of pixels having a small maximum principal curvature direction vector is greatly emphasized.

  Further, in the image processing apparatus according to the present invention, the expansion unit further includes a second pixel group unit including pixels constituting the three-dimensional image, the second number of adjacent pixels smaller than the first number. In the second end pixel group and the second target pixel group, which are two adjacent second pixel groups, which are divided by the dividing plane, and the sum of the maximum principal curvature direction vectors of all pixels included in each of the second end pixel group and the second target pixel group When the angle difference is less than a fifth angle, the second extension processing for giving the second identification information having the same value as the second identification information of the second end pixel group to the second target pixel group, The second target pixel group after the second expansion process is set as the new second end pixel group and the second target pixel group is adjacent to the second target pixel group until the angle difference becomes equal to or greater than the fifth angle. The new second target image to which the second identification information is not attached In addition, in the first second expansion process, the end portion that does not include the interlobar fissured surface among the plurality of first pixel groups that are continuous with the second end pixel group. Is set inside the first end pixel group, and the second target pixel group is set adjacent to the second end pixel group on the side to which the second identification information is not given.

  According to said structure, in the area | region located in the edge part of an interlobar fissured surface, the area | region considered to be a part of an interlobar fissured surface WHEREIN: 2nd identification information of the same value as the pixel group of an interlobar fissured surface Can be granted. Therefore, the interlobar fissure surface can be expanded and the interlobar fissure surface can be extracted more accurately.

  In the image processing apparatus according to the present invention, it is further preferable that the three-dimensional image is one in which a lesion area is removed.

  According to this configuration, an interlobar fissured surface is extracted from an area excluding a lesion area in which many planar objects other than interlobar fissures are often generated.

  As a result, since there are many surfaces that are not interlobar fissures, it is possible to reduce the computational resources and time required to extract the interlobar fissures, and the lesion in the area of the lesion area is mistaken for an interlobar fissure. To prevent extraction.

  In the image processing apparatus according to the present invention, it is preferable that the pixel value of the pixel of the three-dimensional image corresponds to the X-ray absorption rate in the subject.

  According to this configuration, a planar region is extracted from a three-dimensional image captured using X-rays.

  As a result, in a three-dimensional image obtained by an MDCT apparatus, an MRI apparatus or the like often used for image diagnosis, it is possible to observe a membranous tissue such as an interlobar fissure that is low in contrast and not clear. Therefore, it can be widely applied to image diagnosis.

  The image processing apparatus according to each aspect of the present invention may be realized by a computer. In this case, the image processing apparatus is realized by the computer by causing the computer to operate as each unit included in the image processing apparatus. A control program for an image processing apparatus and a computer-readable recording medium that records the control program also fall within the scope of the present invention.

  According to one aspect of the present invention, for example, when a three-dimensional image of a lung field of a subject is used, a three-dimensional CT image of a lung having a lesion due to lung cancer, severe emphysema, interstitial pneumonia, etc. The interlobar fissure can be efficiently and accurately extracted.

1 is a block diagram illustrating a configuration example of an image diagnosis support system including an image processing apparatus according to an embodiment of the present invention. It is a flowchart which shows the outline of the flow of the interlobar fissure extraction process which the image processing apparatus which concerns on embodiment of this invention performs. It is a flowchart which shows the flow of the rough extraction of the interlobar fissure which the image processing apparatus which concerns on Embodiment 1 of this invention performs. It is a flowchart which shows the flow of the detailed extraction of the interlobar fissure which the image processing apparatus which concerns on Embodiment 1 of this invention performs. It is a flowchart which shows the flow of the correction process of the interlobar fissure which the image processing apparatus which concerns on Embodiment 1 of this invention performs. (A) an example of a three-dimensional lung image (axial axial tomographic image) acquired by the image processing apparatus according to the first embodiment of the present invention, and (b) a maximum of four-dimensional curvature in pixels included in the lung field region It is an example of the maximum principal curvature image produced | generated using the principal curvature. It is a figure which shows the classification | category and the coupling | bonding of the surface in the rough extraction of the interlobar fissures which concern on Embodiment 1 of this invention, (a) is the result of having extracted the surface used as the interlobar fissure candidate in the maximum principal curvature image. It is an image superimposed on a tomographic image of a field region, (b) is a diagram expressing an image at a stage where a label is given to the extracted candidate surface for interlobar fission as a line drawing, (c) is It is the figure in the stage which updated the label and integrated the surface. It is a figure explaining the method to determine the surface of an interlobar fissure among the surfaces which divide a lung field area, (a) shows an example of a surface judged to be an interlobar fissure, (b) is an interlobe space It is a figure which shows the example of the surface determined not to be a crack. In the detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention, (a) a diagram showing the interlobar fissure extracted by rough extraction, and (b) the region including the interlobar fissure is the target region It is a figure which shows having set as. Leaves before and after (a) and (b) performing processing for emphasizing the maximum principal curvature for pixels in the target region in the detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention It is a figure which shows a fissure. The detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention is performed (a) once (b) twice (c) three times so that the extracted interlobar fissure is enlarged. FIG. In the detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention, it is a diagram showing the result of the process of extracting the interlobar fissure that contacts a large lesion, (a) original image, (b (1) Interlobar fissure candidates around the lesion area, (c) Interlobar fissures extracted around the lesion area. It is a figure for demonstrating the pixel of 26 vicinity and 6 vicinity which are located so that a certain pixel (voxel) may be surrounded in a three-dimensional tomographic image. It is a figure for demonstrating the (a) axial (axial) tomographic image and (b) coronal (tomical) tomographic image of a lung when making a human subject into a subject.

  Hereinafter, an embodiment in which the image processing apparatus 10 according to an aspect of the present invention is applied to an image diagnosis support system 100 will be described in detail as an example. In particular, an image processing device 10 is used to detect an interlobar fissure of a lung of a subject using a three-dimensional lung image generated from a chest X-ray tomographic image including the lung of the subject generated by a multi-slice CT (MDCT) apparatus. The case of extracting by will be described in detail.

(Image diagnosis support system)
FIG. 1 is a block diagram illustrating a configuration example of an image diagnosis support system including an image processing apparatus according to an embodiment of the present invention. As shown in FIG. 1, the diagnostic imaging support system 100 includes at least an image processing apparatus 10, an MDCT apparatus 20, an image data storage unit 30, and a display apparatus 40 according to the present invention. The schematic configuration and function of the image processing apparatus 10 will be described in detail later.

  The MDCT apparatus 20 captures a plurality of chest X-ray tomographic images of a subject, and generates a three-dimensional image by continuously connecting adjacent tomographic images. The three-dimensional image generated by the MDCT apparatus 20 is generated using a tomographic image based on the absorption rate at which X-rays are absorbed by the tissue in the body of the subject. Accordingly, each pixel value of the pixels constituting the image corresponds to the X-ray absorption rate. In general, an area where the X-ray absorption rate is high is displayed bright (white), and a low area is displayed dark (black).

In order to evaluate the X-ray absorption rate, a unit system called Hounsfield Unit (abbreviated as HU) is generally used. In this unit system, the X-ray absorption rate of pure water (H ₂ O) is 0 HU, and the X-ray absorption rate of air (air) is defined as −1000 HU. Since the lung of a living body is a thin membrane-like, thin-film organ containing air, unlike the bone cortex (+1000 HU) and organ parenchyma other than the lung (40-60 HU), the X-ray absorption rate is low. In the present embodiment, the lung X-ray absorption rate is used to extract a lung field region from a chest X-ray tomographic image of the subject. Details of this will be described later.

  The image data storage unit 30 stores the three-dimensional image generated by the MDCT apparatus 20. The display device 40 displays the display image generated by the display image generation unit 19 of the image processing device 10 described later. The display device 40 displays an image stored in the image data storage unit 30, an image generated by the MDCT device 20, or a GUI (Graphical User Interface) that guides an instruction input from the user of the image processing device 10. You may comprise.

  In addition, the image processing apparatus 10, the MDCT apparatus 20, the image data storage unit 30, and the display apparatus 40 of the image diagnosis support system 100 in FIG. 1 do not need to be directly connected to each other, for example, configured via a network. Also good.

  Note that FIG. 1 does not show an input device (keyboard, mouse, etc.), printer, and the like included in the image diagnosis support system 100 that are not directly related to the features of the image processing apparatus 10 of the present invention.

(Schematic configuration of image processing apparatus)
The image processing apparatus 10 includes at least an image data acquisition unit 11, a lung field extraction unit 12, a region of interest determination unit 13, a maximum principal curvature calculation unit (pixel extraction unit, planar region extraction unit, candidate plane extraction unit) 14, and identification Information providing unit (first identification information providing unit) 15, surface integration unit (first integration unit, second integration unit) 16, representative plane determination unit 17 (expansion unit, enhancement unit), interlobar fissure determination unit ( And a display image generation unit 19.

  The image data acquisition unit 11 acquires from the image data storage unit 30 a three-dimensional image that is a target for extraction of interlobar fissures.

  The lung field extraction unit 12 extracts a lung region (corresponding to the lung field region) from the three-dimensional image acquired by the image data acquisition unit 11 based on the pixel value (corresponding to the X-ray absorption rate).

  The region-of-interest determining unit 13 extracts a spherical small region having a radius of 5 mm from the lung field region image extracted by the lung field region extracting unit 12, and the pixel values of the pixels included in the small region are used to extract the lung field region. The proportion of pixels that are less than or equal to the used pixel value is calculated for each small region. A small region with a low ratio of the calculated pixels is removed from the lung field region as a region considered to be a large lesion, and a region of interest (region to be subjected to interlobar fissure extraction) is determined.

The maximum principal curvature calculation unit 14 calculates the maximum principal curvature of the four-dimensional curvature and the maximum principal curvature direction vector with the pixel value (corresponding to the X-ray absorption factor) of each pixel constituting the three-dimensional image as the fourth dimension. To do. Maximum principal curvature calculating unit 14 further maximum principal curvatures are calculated, and using the maximum principal curvature direction vector to generate the maximum principal curvature image, the maximum principal curvature and maximum principal curvatures for each pixel thresholds T ₁ or more Extracting the region, thinning the obtained region, and calculating a three-dimensional topology (Non-Patent Document 7) extracts a plurality of surfaces in the maximum principal curvature image as interlobar fissure candidates. .

  Based on the normal vector of the surface (planar region) extracted in the region of interest, the identification information adding unit 15 assigns and classifies each surface with a label (identification information). The surfaces given the same label mean that the difference in the direction of the normal vector is within a predetermined angle. When another surface exists around the intersection (proximity portion) of a certain surface, a normal vector is obtained for each surface, and the same label is assigned to each surface. In addition, the label provided at this time may be anything as long as each surface can be identified. The case where the labels applied to the pixels constituting the surface are “1”, “1 ′”, etc. will be described in detail later.

  The surface integration unit 16 assigns the same label to the surfaces including the area where the surface deformation is large and the region where the surface intersection exists, and integrates them as one surface. The integrated region to which the same label is assigned is defined as an interlobar fissure candidate (interlobar fissure candidate surface).

  The representative plane determination unit 17 represents a plane defined by the plane generated by the plane integration unit 16 using an expression indicating a plane in (x, y, z) space, ax + by + cz−d = 0. The respective parameters (a, b, c, d) are calculated for each pixel on the surface integrated as each interlobar fissure candidate, and the average value (a ′, b ′, c ′, d ′) is calculated. Then, an expression a′x + b′y + c′z−d ′ = 0 representing a plane corresponding to each interlobar fissure candidate surface is determined.

  The interlobar fissure determination unit 18 calculates what volume ratio the plane defined by the interlobar fissure candidate plane divides the lung field region, and uses the calculated volume ratio as anatomical knowledge. By comparing, it is determined which of the plurality of interlobar candidate surfaces is an interlobar fissure. In addition, the interlobar fissure determination unit 18 extracts a surface constituted by pixels having a predetermined number or more of pixels constituting the surface extracted as the interlobar fissure candidate surface.

  The display image generation unit 19 acquires the same image as the image acquired by the image data acquisition unit 11 from the image data storage unit 30, and the image and the image of the surface determined by the interlobar fissure determination unit 18 as an interlobar fissure. Are combined to generate a display image.

(Flow of fissure extraction processing)
Next, the operation flow of the image processing apparatus 10 according to the present invention for extracting lung interlobar fissures will be described with reference to FIGS. 2 to 5 while associating with the functions of the respective parts of the image processing apparatus 10 shown in FIG. explain.

  FIG. 2 is a flowchart showing an outline of the flow of the interlobar fissure extraction process according to Embodiment 1 of the present invention. Here, a process from when the image processing apparatus 10 acquires a three-dimensional lung image to when an image showing an interlobar fissure is generated as a display image will be described.

  The image data acquisition unit 11 of the image processing apparatus 10 according to one aspect of the present invention acquires a three-dimensional image of the subject's lung from the image data storage unit 30 as an original image from which the interlobar fissure is extracted (step 1; , Simplified and expressed as S1).

  First, rough extraction of the interlobar fissure (S100) is performed on the three-dimensional image. Next, detailed extraction of the interlobar fissure is performed (S200), and finally, interlobar fissure correction processing (S300) is performed. The image processing apparatus 10 extracts the interlobar fissure by each of these processing steps.

  In the display image generation unit 19, a combined display image is generated by accurately superimposing the three-dimensional lung image (original image) and the extracted interlobar fissure (S2).

  In the following, the interlobar fissure extraction process performed by the image processing apparatus 10 will be described in the order of the interlobar fissure rough extraction (S100), the interlobar fissure detail extraction (S200), and the interlobar fissure correction process (S300).

1. Rough extraction of the interlobar fissure In order to extract the interlobar fissure from the three-dimensional image (chest X-ray tomographic image) captured by the MDCT apparatus 20, first, rough extraction of the interlobar fissure is performed (S100 in FIG. 2). This rough extraction of the interlobar fissure is composed of the following four processes. 1. Image preprocessing, 2. 2. Extraction of interlobar fissure candidates; 3. Classification and integration of faces; Selection of interlobar fissures.

  FIG. 3 is a flowchart showing a flow of rough extraction of an interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention. Hereinafter, each process of rough extraction of the interlobar fissure will be described in detail with reference to FIG.

1-1. Image Preprocessing First, the image data acquisition unit 11 acquires a chest X-ray CT tomographic image (three-dimensional image) from the image data storage unit 30 (S1), and the lung field region extraction unit 12 extracts the lung field from the acquired three-dimensional image. An area is extracted (S110). The lung field region extraction unit 12 extracts a region having an X-ray absorption rate equal to or higher than a predetermined value T ₀ (for example, −700 HU) from the extracted lung field region, and a minute amount included in the extracted region by Closing processing. To fill holes. Closing processing is one of image processing methods in which expansion processing (dilation) and degeneration processing (erosion) are repeated the same number of times.

In addition, in order to extract the interlobar fissure of the lung from the lung field region, T ₀ is preferably set to −750 HU or more and less than −650 HU. In an example described later, T ₀ is set to −700 HU.

Next, the region-of-interest determination unit 13 extracts a spherical small region having a radius of 5 mm from the lung field region, and the pixel value of the pixel included in the small region is equal to or less than the pixel value used for the lung field region extraction. A ratio is calculated for each small region. Calculated pixel low ratio R ₁ (e.g., 30% or less) a small area, as a region thought that large lesions, were removed from the lung region, the region of interest (the region of interest for interlobular裂抽unloading) Is determined (S120).

  In the rough extraction of the interlobar fissure, the process of removing the lesion area is not an indispensable process. However, since there are many surface shapes other than the interlobar fissures in the lesion area, it is desirable to perform the process of removing the lesion area as a pre-process on the image for rough extraction of the interlobar fissures.

  The region of interest extracted by the image preprocessing is used for the next processing (2. Extraction of interlobar fissure candidates).

1-2. Extraction of interlobar fissure candidates In order to extract the interlobar fissure candidate (plane) region from the region of interest in the lung field region extracted as described above (S130: pixel extraction step, planar region extraction step) Then, the method described in Non-Patent Document 6 is used. That is, the image processing apparatus 10 calculates the maximum principal curvature of the four-dimensional curvature having the pixel value of each pixel as the fourth dimension from the three-dimensional image having the three-dimensionally arranged pixels representing the lung field region of the subject. A maximum principal curvature image having pixel values is generated, and lung interlobar fissures are extracted from the maximum principal curvature image.

  The maximum principal curvature calculation unit 14 takes the pixel value of each pixel on the fourth axis in addition to the three spatial axes of the three-dimensional image to the pixels constituting the region of interest, thereby converting the hypersurface in the four-dimensional space. Treat as. As a result, a plurality of planar regions included in the region of interest can be extracted based on the spatial spread and connection in consideration of the information regarding the contrast of the shade when the curved surface is imaged.

  In an image in which the density value at the point (x, y, z) in the three-dimensional space is given by w = f (x, y, z), the pixel value of each pixel constituting the three-dimensional image indicates the respective density value. Yes.

1 partial derivatives in the region of interest is expressed as _{f x, f y, f z} , the 2 partial derivatives _{f xx, f yy, f zz} , f xy, f yz, denoted as _{f xz.} At this time, the first basic format is

  And the second basic form

Then, the principal curvature of the four-dimensional hypersurface is expressed by the following equation.

An eigenvalue principal curvature of the product W of the inverse matrix and the F ₂ of the F _1, the main direction of curvature vector of the four-dimensional hypersurface is the eigenvector of the dot product W. The maximum one of the three main curvatures obtained is the maximum main curvature, and the eigenvector corresponding to the maximum main curvature is the maximum main curvature direction vector.

  FIG. 6 shows (a) an example of a three-dimensional image of an lung (axial axial tomographic image) acquired by the image processing apparatus according to Embodiment 1 of the present invention, and (b) 4 in pixels included in a lung field region. It is an example of the maximum principal curvature image produced | generated using the maximum principal curvature of a dimensional curvature.

The maximum principal curvature calculation unit 14 further extracts a region having a predetermined threshold T ₁ (for example, 0.05) or more with respect to the maximum principal curvature in the region of interest, and performs thinning on the obtained region. Compute a three-dimensional topology. The calculation of the three-dimensional topology can be performed by a known technique (see Non-Patent Document 7). When a line shape is extracted in addition to the surface shape (planar region) in the maximum principal curvature image, the line shape is removed before performing a rough extraction process of the interlobar fissures thereafter. Further, based on the three-dimensional topology, each pixel is classified into one of a surface, an edge of the surface, and an intersection of the surfaces.

Specifically, the maximum principal curvature calculating unit 14, a collection of pixels to calculate the maximum principal curvature and the maximum principal curvature direction vector for each pixel, which are adjacent to each other maximum principal curvature is the threshold value above T ₁ in the region of interest Are extracted as a region of one shape. The region constituted by the extracted pixels is classified into a surface shape, a line shape, and the like according to the shape. Of the regions extracted in this way, the linear region is removed, while the planar region is classified into a surface, an edge of the surface, and an intersection of the surfaces. This plane-shaped region is set as a candidate leaf fissure region.

1-3. Classification and integration of faces A plurality of faces (planar regions) extracted as interlobar fissure candidates are classified using normal vectors of each face. The identification information adding unit 15 uses the maximum principal curvature direction vector of the pixels constituting the interlobar fissure candidate as a normal vector in the pixels of the surface, and assigns a label (first identification information) to each surface (S140). : 1st identification information provision step). Specifically, the process in which the identification information adding unit 15 classifies a surface by applying a label is performed by the following four steps. Here, a pixel classified into a plane is referred to as a plane pixel, and a pixel classified into a crossing portion between the planes is referred to as a crossing pixel.

  Step 1: The initial value of the label is set to 1. In the initial state, all the pixels are not yet labeled.

  Step 2: Search for intersection pixels in the region of interest. Step 3 is performed for each intersection pixel. When the process for all the intersection pixels is completed, the identification information providing unit 15 ends the process.

  Step 3: A label is assigned to one of the pixels that have not yet been labeled among the surface pixels in the vicinity of the intersection pixel of interest (for example, “near 26”: see FIG. 13), Let it be a point of interest. Step 4 is performed for the point of interest. When there is no surface pixel to which no label is applied in the vicinity of the focused intersection pixel, the process returns to step 2.

Step 4: For a surface pixel that has not yet been labeled and exists in the vicinity of the point of interest, the difference in angle between the normal vector of the surface pixel and the normal vector of the point of interest is a predetermined angle Q ₁ (first If the angle condition of less than 25 degrees (for example, 25 degrees) is satisfied, a label is assigned to the surface pixel. That is, the same label as the label of the attention point is given to the pixel. This process is executed for all the surface pixels that are present in the vicinity of the point of interest and have not yet been labeled. Step 4 is executed by setting each surface pixel to which the same label is assigned as a new attention point. If there is no surface pixel to which the same label is to be applied in the vicinity of a plurality of surface pixels to which the same label is applied (no surface pixel that satisfies the above angle condition), the label is increased from 1 to 2, for example. Return to step 3 (updating).

The planar area extracted by the above steps is constituted by a plurality of continuous pixels among the pixels extracted by the maximum principal curvature calculation unit 14, and the above-described two pixels included in the planar area angular difference of the maximum principal curvature direction vector in the four-dimensional hypersurface is also less than a predetermined angle Q _1.

  As a result of the above steps, the surface pixels to which the same label is assigned are considered to constitute one surface. However, if a region having a large surface deformation or a region where a surface intersects exists, the surface pixels are not given the same label. Therefore, the surface integration unit 16 assigns the same label to the surfaces including the intersection, and integrates the surfaces having the normal vector in the same direction across the intersection of the surfaces. The surface integration is performed by the following three steps.

  Step 1: Search for pixels at intersections of surfaces in the region of interest. Step 2 is performed for each intersection pixel. Continue until all the intersections of the faces have been searched.

  Step 2: Normal vectors are compared between all the planes in the vicinity of the pixel 26 at the intersection of the planes.

Step 3: When there is a combination of surfaces where the angle difference between the normal vectors is less than a predetermined angle Q ₂ (second angle; for example, 25 degrees), they are the same at the intersection of those surfaces and the surfaces sandwiched between them. The label is given (updated). This is done until labels are applied to the intersections of all surfaces.

  The above-described surface classification and integration will be specifically described with reference to FIG.

  FIG. 7 is a diagram showing surface classification / combination in rough extraction of interlobar fissures according to Embodiment 1 of the present invention. FIG. 7A illustrates extraction of a surface that is a candidate for interlobar fissures in the maximum principal curvature image. (B) is a diagram expressing the image at the stage where a label is assigned to the extracted interlobar fissure candidate surface as a line drawing, (C) is a figure in the stage which updated the label and integrated the surface.

  As shown in (a) of FIG. 7, when a surface that is a candidate for interlobar fissures is extracted from the maximum principal curvature image, the identification information adding unit 15 labels the interferon fissure candidate surfaces in order from 1 Is granted. As a result, as shown in (b) of FIG. 7, separate labels 1 to 6 are assigned to the surfaces of candidate interlobar fissures and classified.

  The surface to which the label 1 is applied, the surface to which the label 2 is applied, the surface to which the label 2 is applied and the surface to which the label 3 is applied, and the surface to which the label 5 is applied and the surface to which the label 6 is applied are It has an intersection.

Therefore, the surface integration unit 16 compares the normal vectors with respect to the pixels at the intersection between the edge of the surface to which the label 1 is assigned and the edge of the surface to which the label 2 is assigned. As a surface angle difference is less than a predetermined angle Q ₂ between normal vectors to identify a face that is granted granted surface and label 2 the label 1, given the same label 1 'thereto. Label 1 granted faces a, or angular difference between normal vectors of the label 2 the granted surface is a predetermined angle Q ₂ or more, the surface that has been granted the label 5, the label 1 'grant Not.

Next, the surface integration unit 16 applies the pixel at the intersecting portion existing between the surface to which the label 1 ′ is applied (the surface to which the label 2 is initially applied) and the surface to which the label 3 is applied, to Compare normal vectors. As a surface angle difference is less than a predetermined angle Q ₂ between normal vectors, 'identifies granted surface labeled 3 a granted faces the same label 1 of these' labels 1 to grant.

  Thereafter, the same processing is performed, and as shown in FIG. 7C, the surface to which the labels 1 to 4 are initially attached is integrated as one surface to which the label 1 ′ is assigned. The surfaces to which the labels 5 and 6 are assigned are integrated as one surface to which the label 5 ′ is assigned.

  That is, for each surface extracted as an interlobar fissure candidate, a label corresponding to the maximum principal curvature direction vector of the pixels constituting the surface is assigned to classify each surface, and a plurality of surfaces to which labels are assigned Among these, for surfaces whose normal vector angle difference is within a predetermined angle, the labels assigned to each pixel on the surface are updated to the same value, thereby classifying the surfaces that are candidates for interlobar fissures. Integrate.

  Next, the interlobar fissure determination unit 18 extracts a surface composed of a predetermined number or more of the classified and integrated surfaces. That is, in the rough extraction of the interlobar fissures according to the present embodiment, among the surfaces extracted as the interlobar fissure candidate surfaces by classification / integration, the surfaces that are less than a predetermined size as a result of classification / integration As not a fissure, it is removed before making an interlobar fissure selection.

1-4. Selection of Interlobar Fissure Next, the representative plane determination unit 17 determines a plane (divided plane) defined by each plane obtained as a result of classification and integration. Then, the interlobar fissure determination unit 18 calculates what volume ratio of the lung field region the determined plane (division plane) divides, and calculates the plane from the calculated volume ratio. The interlobar fissure is determined by determining whether the prescribed surface is an interlobar fissure.

The plane equation passing through the point (x ₀ , y ₀ , z ₀ ) and having the normal vector as (a, b, c) is
d = ax + by + cz (1)
It is represented by However, d = ax ₀ + by ₀ + cz ₀ . That is, the parameters (a, b, c, d) are parameters that define the plane. For a plane that is a candidate for interlobar fissure, a plane on the point is defined from coordinates (x ₀ , y ₀ , z ₀ ) at each point on the plane and a normal vector (maximum principal curvature direction vector) at the point. Find the parameters. An average value (a ′, b ′, c ′, d ′) of the parameters (a, b, c, d) in the plane that is a candidate for interlobar fissure is obtained. A plane defined by the average parameter (a ′, b ′, c ′, d ′) is a plane representing (approximate) the plane that is a candidate for interlobar fissure.

Calculating an average value (a ′, b ′, c ′, d ′) of the parameters (a, b, c, d) of the formula (1) corresponding to the plurality of surfaces integrated by the integration unit;
f (x, y, z) = a′x + b′y + c′z−d ′ (2)
A plane satisfying f (x, y, z) = 0 in the above formula (2) is defined.

  Step 1: The average of each parameter of the plane formula of all the pixels constituting the plane is obtained and set as a plane formula defined by the plane.

Step 2: When the lung field region is divided into two divided regions by the plane expressed using the parameters obtained in Step 1, the larger volume V ₁ and the smaller volume V ₂ are obtained.

Step 3: If V ₂ / (V ₁ + V ₂ ) is smaller than a predetermined value R ₂ (for example, 0.25), it is determined that the surface is not an interlobar fissure and is removed. This is because an interlobar fissure is a planar tissue that spreads in a longitudinal direction or across a lung field region as an anatomical property.

FIG. 8 is a diagram for explaining a method for determining a plane of an interlobar fissure among planes that divide a lung field region. FIG. 8A illustrates an example of a plane that is determined to be an interlobar fissure. ) Is a diagram illustrating an example of a surface that is determined not to be an interlobar fissure. A case where the plane defined by the plane A shown in FIG. 8A divides the lung field region and a case where the plane defined by the plane B shown in FIG. 8B divides the lung field region. Compare. The plane A is determined as an interlobar fissure because V ₂ / (V ₁ + V ₂ ) is equal to or greater than a predetermined value R ₂ . On the other hand, the surface B is determined not to be an interlobar fissure because V ₂ / (V ₁ + V ₂ ) is smaller than the predetermined value R ₂ and is removed.

  As described above, using the interlobar fissures extracted by the rough extraction of interlobar fissures (the surface obtained as a result of the coarse extraction), detailed extraction of interlobar fissures is performed next.

2. Detailed extraction of interlobar fissures In order to extract interlobar fissures from a three-dimensional image (chest X-ray tomographic image) imaged by the MDCT apparatus 20, following the above-described rough extraction of interlobar fissures (S100 in FIG. 2). Detailed extraction of interlobar fissures. This detailed extraction of interlobar fissures is composed of the following five processes. 1. 1. setting of the target area where the interlobar fissure exists; 2. Expansion of target area 3. Emphasis on maximum principal curvature. 4. Re-extraction of interlobar fissures, and Extraction of interlobar fissures that contact lesions.

  FIG. 4 is a flowchart showing a flow of detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention. Below, each process of the detailed extraction of an interlobar fissure is demonstrated in detail, referring FIG.

2-1. Setting the target region where the interlobar fissure exists The detailed extraction of the interlobar fissure is a process of setting the region where the interlobar fissure exists and extracting the interlobar fissure in more detail. Here, a target region where an interlobar fissure exists is set (S210).

  The lung field region is divided into a plurality of small regions (first pixel group unit) each having a size of 10 (pixels) × 10 (pixels) × 10 (pixels), and the interlobar fissure plane obtained as a result of rough extraction Is set as a target area, and the following processing is performed for each target area.

The maximum principal curvature calculation unit 14 calculates a normal vector of a surface in the target region (an interlobar fissured surface obtained as a result of rough extraction) and an average of angles of main curvature direction vectors of pixels in the target region. In comparison, if it is larger than Q ₃ (for example, 15 degrees), the target region is removed.

  FIG. 9 includes (a) a diagram illustrating a leaf fissure extracted by rough extraction and (b) an interlobe fissure in detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention. It is a figure which shows having set the area | region as an object area | region. In FIG. 9, the area | region containing the interlobar fissure seen in (a) is set as the some object area | region C in FIG.9 (b).

  Each parameter of the expression of the plane of the interlobar fissure existing in the target region is obtained by averaging. The target area is classified by the normal vector obtained from the equation of each plane. The process of assigning a label (second identification information) to the target area by the identification information adding unit 15 and classifying the target area is specifically performed by the following four steps.

  Step 1: One label is assigned to each target area.

  Step 2: A target area without a label is searched, and a label is given to the target area obtained as a result of the search to make it an attention area. If there is no target area without a label, the process ends.

Step 3: The normal vector of the attention area (first end pixel group) is compared with the normal vector of the target area (first target pixel group) having no label connected to the attention area, and the normal vectors are compared. If the angle is within Q ₄ (fourth angle; for example, 20 degrees), a label is given to the target area to make it an attention area (first extension process). Thereafter, this process is repeated.

  Step 4: If there is no target area to which a label is assigned, the value of the label is increased (updated) and the process returns to Step 2 described above.

  As a result of the above steps, the surfaces to which the same label (second identification information) is assigned are considered to constitute one region. Subsequent processing is performed for each target region to which each label is assigned.

2-2. Expansion of the target area The expansion of the target area is a process of extending the target area to extend (S220).

  Specifically, among the 10 (pixel) × 10 (pixel) × 10 (pixel) small areas, an area including the periphery of the target area is set as an extension area candidate. A selection is made using an expression of a plane included in an adjacent target area from the extension area candidates. This substitutes each pixel of the extension area candidate into the equation (1) obtained by the representative plane determination unit 17 to obtain the value of f (x, y, z) as the pixel value. A pixel for which f (x, y, z) = 0 is on a plane included in the target region, and a pixel whose value of f (x, y, z) is positive or negative is not on the plane. Means that.

  When the pixel value in each extension region candidate includes both positive and negative pixel values in one extension region candidate, the target region is extended by adding the extension region candidate to the target region. For example, a value obtained by averaging parameters of surrounding target regions may be used as the parameter of the expression of the plane of the target region added.

2-3. Emphasis on maximum principal curvature Here, the maximum principal curvature is obtained by emphasizing the interlobar fissured portion of the 3D image by adding the second-order differential coefficient in the direction of the average normal vector to the 3D image in the target region. Recalculate, find the inner product of the direction vector of the maximum principal curvature of the pixel in the target area and the average normal vector in the target area, and emphasize the maximum main curvature by multiplying the maximum main curvature of each pixel by the inner product This is performed (S230).

  This increases the maximum principal curvature of the interlobar fissure, increases the maximum principal curvature around the interlobar fissures having the maximum principal curvature direction vector in the same direction as the interlobar fissure, and the maximum principal curvature direction different from the interlobar fissure The maximum principal curvature around the interlobar fissure with the vector is small.

  FIG. 10 is a detailed extraction of the interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention. Before performing (a) and (b) performing processing for enhancing the maximum principal curvature for the pixels in the target region. FIG. The maximum principal curvature emphasis process described here is performed on the interlobar fissure D in the target region shown in FIG. 10 (a), so that the maximum principal of the interlobar fissure E is obtained as shown in FIG. 10 (b). Compared to the curvature, it can be seen that the maximum principal curvature of the region decreases so as to sandwich the interlobar fissure, and the interlobar fissure E becomes clearer.

  Note that the inner product is not limited to the calculation. The maximum principal curvature of a pixel having the maximum principal curvature direction vector in the same direction as the interlobar fissure plane is increased, and the maximum principal curvature of a pixel having a maximum principal curvature direction vector different from the interlobar fissure plane is decreased. That's fine.

2-4. Re-extraction of interlobar fissures Next, a region with a threshold value T ₂ (for example, 0.04) or more is extracted with respect to the emphasized maximum principal curvature. Integration is performed and the interlobar fissure is extracted again (S240).

  Using the extracted interlobar fissure, the target region is expanded until the target region does not increase in the region of interest (YES in S250). When the target region is increasing in the region of interest (No in S250), the setting from the target region where the interlobar fissure exists (S210) to the re-extraction of the interlobar fissure in the target region (S240) Repeat the process.

  FIG. 11 shows the interlobar fissures extracted by performing (a) once (b) twice (c) three times the detailed extraction of the interlobar fissures performed by the image processing apparatus according to Embodiment 1 of the present invention. It is a figure which shows a mode that it expands. The target region indicated by the arrow in FIG. 11 is enlarged, and the interlobar fissures that have not been extracted by one extraction are reextracted twice (FIG. 11 (b)) and three times (FIG. 11). 11 (c)), and the larger is extracted.

2-5. Extraction of interlobar fissures that contact a lesion When the interlobar fissures obtained as a result of detailed extraction and the lesion area extracted by the image preprocessing (S120 in FIG. 3) performed at the beginning of rough extraction are in contact (YES in S260), Extract lesions and interlobar fissures around lesions in more detail. Specifically, the process after the extraction of the leaf fissure candidates (S130 in FIG. 3) after the rough extraction of the interlobar fissure is performed on the lesion that is in contact with the interlobar fissure and the vicinity of the lesion (detail extraction result). S270).

  Around the lesion, an interlobar fissure may be drawn into the lesion area and deformed. Therefore, the target region used for the detailed extraction has a size of 5 (pixels) × 5 (pixels) × 5 (pixels) smaller than the small region of 10 (pixels) × 10 (pixels) × 10 (pixels) described above. It is set in a plurality of small regions (second pixel group unit). Then, the average of the normal vectors of the pixels constituting the lesion and the surface existing around it is set as the normal vector of the surface.

A region in contact with the interlobar fissures extracted as a result of detailed extraction is selected as an interlobar fissure candidate. The surface where the normal vector direction of the surface in contact with the interlobar fissure candidate differs from the normal vector direction of the interlobar fissure candidate is within Q ₅ (fifth angle; for example, 20 degrees). The interlobar fissures that come into contact with (second expansion process).

  FIG. 12 is a diagram illustrating a result of a process of extracting an interlobar fissure that contacts a large lesion in the detailed extraction of an interlobar fissure performed by the image processing apparatus according to the first embodiment of the present invention. An image, (b) candidate interlobar fissures around the lesion area, and (c) interlobar fissures extracted around the lesion area, respectively.

3. Interlobar fissure correction processing The interlobar fissure correction processing (S300 in FIG. 2) is performed between the interlobar fissures in order to perform distraction and filling of the interlobar fissures extracted by the detailed extraction of the interlobar fissures (S200 in FIG. 2). This is a process of interpolating missing pixels to be attributed to the fissure and a process of correcting the interpolation result.

  FIG. 5 is a flowchart illustrating a flow of correction processing for interlobar fissures performed by the image processing apparatus according to the first embodiment of the present invention.

  The coefficient of the plane equation is obtained by linear interpolation between the target region of each label created by the detailed extraction of the interlobar fissure and the surrounding pixels, and the pixel for which the plane equation (1) is 0 is interpolated for the interlobar fissure. The result is taken (S310). Of the interlobar fissure interpolation results, interpolated fissures that are not adjacent to the interlobar fissures obtained as a result of the detailed extraction of the interlobar fissures are deleted to correct the interlobar fissures (S320).

  Specifically, first, the target region obtained by the detailed extraction of the interlobar fissure (S200 in FIG. 2) is expanded by the method described above. Of the 10 (pixel) × 10 (pixel) × 10 (pixel) small regions, a region including the periphery of the target region is set as an interpolation candidate region.

  Formula (1) of the plane of the target area adjacent to each pixel in the interpolation candidate area is calculated, and the value of f (x, y, z) is obtained for each pixel in the interpolation candidate area. When this pixel value is 0, or when the sign is different from the pixel value in the vicinity of 6 (see FIG. 13), the pixel is extracted as an interpolation plane.

  Interpolation plane is interleaved and filled. However, in the case of lobular failure, the edge of the interlobar fissure inside the lung field is over-interpolated. For this purpose, a portion that becomes the edge of the expanded target region is searched, and the interpolation plane inside the lung field region is deleted.

  Edge extraction of the target region is performed by thinning the target region and classifying the surface shape, and a target region within the outline of the lung field region is selected from among them. Of the interpolation planes in the interpolation candidate area around the selected target area, those having a distance of 3 mm or more from the extraction result of the interlobar fissure are removed. Also, the interpolation plane outside the lung field region is removed.

〔Example〕
(MDCT equipment imaging conditions)
The imaging conditions of 40 normal cases and 40 abnormal cases of lung cancer, emphysema, and interstitial pneumonia used in the experiment are shown below. The MDCT apparatus used was Toshiba Aquilion (registered trademark). A normal example has a slice thickness of 1.0 mm, a slice interval of 1.0 mm, a tube current of 100 to 500 mA, a tube voltage of 120 kV, a pixel size of 0.625 to 0.647 mm, a resolution of 512 × 512 pixels, and an average number of images of 300.

  On the other hand, abnormal cases are slice thickness 0.5-1.0mm, slice interval 0.5-1.0mm, tube current 100-500mA, tube voltage 120kV, pixel size 0.625-0.683mm, resolution 512 × 512pixel. The number of images is 300-700. In 19 cases, a slice thickness of 0.5 mm was used, and in 21 cases, a slice thickness of 1.0 mm was used.

(Evaluation method)
The gold standard in this example used data in which an interlobar fissure and an excessively divided leaf can be visually observed on an MDCT image, and the interlobar fissure was extracted based on the guidance of a doctor. The interlobar fissure was extracted using the image processing apparatus 10 or the image processing method of the present invention. And the ratio S of the gold standard in which the distance from the extracted interlobar fissure was contained within 2 mm was calculated | required. The larger the ratio S (closer to 100%), the smaller the overextraction of the extraction result.

(Evaluation results)
20 normal cases and 10 abnormal cases were used as training data, and the remaining 20 normal cases and 30 abnormal cases were used as test data. Each suitable parameter was determined from the training data, and rough extraction (S100), detailed extraction (S200), and correction processing (S300) were performed.

  As a result, we succeeded in extracting interlobar fissures with high accuracy and sensitivity in both normal and abnormal cases. Furthermore, each time through the steps of rough extraction, detailed extraction, and correction processing, the accuracy is gradually reduced, but the sensitivity is gradually improved.

  Specifically, the gold standard ratio S1 (mean ± standard deviation) within a distance of 2 mm from the extraction result by interlobar fissures using the extraction method of the present invention is 91.2 ± 3.6 in normal cases. %, Abnormal cases were 89.7 ± 4.9%. The ratio S2 (mean ± standard deviation) of the results of the extraction method of the present invention that is within 2 mm from the gold standard is 95.5 ± 3.7 for normal cases and 93.6 ± 4.8% for abnormal cases. Met. Therefore, good results with few erroneous extractions (overextraction and non-extraction) were obtained not only in normal cases but also in abnormal cases. In addition, interlobar fissures could be extracted in various cases.

(Explanation of 26 neighborhoods, 6 neighborhoods)
FIG. 13 is a diagram for explaining pixels in the vicinity of 26 and 6 that are located so as to surround a certain pixel (voxel) in a three-dimensional tomographic image. Here, each pixel constituting the image is indicated by a cube.

  The three-dimensional tomographic image is an image formed by connecting adjacent tomographic images as shown in FIG. In this figure, three tomographic images parallel to the xz plane, that is, a tomographic image P (n−1), a tomographic image P (n), and a tomographic image P (n + 1) are connected in the y-axis direction.

  If attention is paid to the pixel F in the tomographic image P (n), the pixel F includes eight pixels of the tomographic image P (n), 18 in the tomographic image P (n−1) and tomographic image P (n + 1). The pixels (9 pixels + 9 pixels) are surrounded by 26 pixels, and the 26 pixels are referred to as 26 neighborhoods with respect to the pixel F of the tomographic image P (n).

  Next, the six neighboring pixels of the pixel F are the six pixels that are in direct contact with the six surfaces of the pixel F, and the six neighborhoods of the pixel F in the tomographic image P (n). Call.

(Description of lung field image)
FIG. 14 is a diagram for explaining a lung (a) axial tomographic image and (b) coronal tomographic image when a human is the subject.

  A cross-sectional tomographic image parallel to the subject's counter-axis direction is called an axial tomographic image, and a cross-sectional tomographic image perpendicular to the subject's counter-axis direction is called a coronal tomographic image.

(Addendum)
Note that the image processing apparatus 10 of the present invention can be particularly preferably applied to a three-dimensional image used for image diagnosis in the medical field. For example, the image processing apparatus 10 of the present invention is not limited to the present example of extracting lung interlobar fissures, and can be applied to images obtained by imaging arbitrary tissues and organs used for image diagnosis. Further, the image processing apparatus 10 of the present invention can be suitably used for any image other than a tissue / organ.

  Accordingly, the diagnostic imaging support system 100 may include devices such as three-dimensional MRI, PET (Positron Emission Tomography), and SQUID (Superconducting Quantum Interference Device) as an alternative to or in addition to the MDCT device 10. . The image processing apparatus 10 according to the present invention is not limited to the three-dimensional image generated by the MDCT apparatus, and for any three-dimensional image generated in units of pixels using various imaging units and image generation units. The image processing apparatus 10 according to one embodiment of the present invention can be applied.

[Example of software implementation]
Control blocks of the image processing apparatus 10 (particularly, a lung field region extraction unit 12, a region of interest determination unit 13, a maximum principal curvature calculation unit 14, an identification information addition unit 15, a surface integration unit 16, a representative plane determination unit 17, and an interlobar fissure) The determination unit 18) may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or may be realized by software using a CPU (Central Processing Unit).

  In the latter case, the image processing apparatus 10 includes a CPU that executes instructions of a program that is software that implements each function, and a ROM (Read Only Memory) in which the program and various data are recorded so as to be readable by a computer (or CPU). Alternatively, a storage device (these are referred to as “recording media”), a RAM (Random Access Memory) that expands the program, and the like are provided. And the objective of this invention is achieved when a computer (or CPU) reads the said program from the said recording medium and runs it. As the recording medium, a “non-temporary tangible medium” such as a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. The program may be supplied to the computer via an arbitrary transmission medium (such as a communication network or a broadcast wave) that can transmit the program. The present invention can also be realized in the form of a data signal embedded in a carrier wave in which the program is embodied by electronic transmission.

  The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope shown in the claims, and embodiments obtained by appropriately combining technical means disclosed in different embodiments. Is also included in the technical scope of the present invention.

  The present invention can be used for medical image diagnosis based on a three-dimensional image of a lung or the like generated by an MDCT apparatus or the like.

DESCRIPTION OF SYMBOLS 10 Image processing apparatus 11 Image data acquisition part 12 Lung area extraction part 13 Region of interest determination part 14 Maximum principal curvature calculation part (Pixel extraction means, planar area extraction means, candidate surface extraction means)
15 Identification information giving unit (first identification information giving means)
16 plane integration section (first integration means, second integration means)
17 Representative plane determination unit (expansion means, enhancement means)
18 Interlobar fissure determination part (interlobar fissure selection means)
DESCRIPTION OF SYMBOLS 19 Display image generation part 20 MDCT apparatus 30 Image data memory | storage part 40 Display apparatus 100 Image diagnosis assistance system S130 Pixel extraction step, planar area extraction step S140 1st identification information provision step

Claims (14)

An image processing apparatus for extracting a planar region from a three-dimensional image having pixels arranged in three dimensions,
Pixel extraction means for extracting pixels having a maximum principal curvature of a four-dimensional hypersurface having a fourth dimension of the pixel value of each pixel of the three-dimensional image, and a predetermined value or more;
Among the pixels extracted by the pixel extraction means, the angle difference between the maximum principal curvature direction vectors on the four-dimensional hypersurface of any two pixels included and included is less than the first angle. A planar area extracting means for extracting the planar area from the three-dimensional image;
1st identification information provision which assign | provides 1st identification information according to the said largest principal curvature direction vector of the 1 or more pixel which comprises this planar area to each planar area extracted by the said planar area extraction means And an image processing apparatus.   Among the plurality of planar areas to which the first identification information is assigned, when two adjacent planar areas have an angle difference of normal vectors that is less than the second angle, the two planar areas are configured. The image processing apparatus according to claim 1, further comprising a first integration unit that updates the first identification information given to each pixel to be the same value.   Among the plurality of planar regions to which the first identification information is assigned, the normal vector angle difference is a second angle between two planar regions adjacent to each other with a predetermined number of pixels sandwiched between the edges. If it is less than the second, the first identification information given to each pixel constituting the two planar regions and the pixel sandwiched between the edges of the two planar regions is updated to the same value. The image processing apparatus according to claim 1, further comprising: an integrating unit. The three-dimensional image represents the lung field of the subject,
The lung field is divided by a dividing plane that includes one of the planar areas to which the first identification information of the same value is assigned and is virtually set to divide the lung field area into two divided areas. The interlobar fissure selection means is provided for selecting the planar region as the interlobar fissured surface when the ratio of the volume of the smaller divided region to the entire region is greater than or equal to a predetermined value. The image processing apparatus according to any one of 1 to 3. Candidate surface extraction for extracting a planar region constituted by a predetermined number of pixels or more among the planar regions to which the first identification information having the same value is assigned as the interlobar fissure candidate surface of the lung field region With means,
5. The image processing apparatus according to claim 4, wherein the interlobar fissure selection unit selects the interlobar fissure plane from the interlobar fissure candidate plane extracted by the candidate plane extraction unit. . Dividing the pixels constituting the three-dimensional image into first pixel group units each including a first number of adjacent pixels;
One second identification information corresponding to the normal vector of the interlobar fissured surface is given to each first pixel group including the interlobar fissured surface,
In the first end pixel group and the first target pixel group, which are two adjacent first pixel groups crossed by the dividing plane, the angular difference of the sum of the maximum principal curvature direction vectors of all the pixels included in each of them is the first. When the angle is less than 4, a first extension process for providing the first target pixel group with second identification information having the same value as the second identification information of the first end pixel group,
The first target pixel group after the first expansion process is set as a new first end pixel group and is adjacent to the first target pixel group until the angle difference becomes equal to or greater than the fourth angle. And an expansion unit that repeats while setting a new first target pixel group to which the second identification information is not given,
The expansion means sets the first end pixel group to a position including the end of the interlobar fissure plane and sets the first target pixel group to the first end in the first first expansion process. The first end pixel is set on the side not including the interlobar fissure plane adjacent to the pixel group, and instead of the sum of the maximum principal curvature direction vectors of all the pixels included in the first end pixel group. 6. The image processing apparatus according to claim 4, wherein a normal vector of the interlobar fissured region included in a group is used. For each first pixel group to which the second identification information of the same value is given,
For the first pixel group including the interlobar fissure plane selected by the interlobar fissure selection means, the pixel value of each pixel is included in the maximum principal curvature direction vector of the pixel and the first pixel group. The smaller the angle difference from the normal vector of the interlobar fissure area, the larger the amount,
For the first pixel group that does not include the interlobar fissure plane selected by the interlobar fissure selection means, the pixel value of each pixel is set to the maximum principal curvature direction vector of the pixel and all pixels of the first pixel group. The image processing apparatus according to claim 6, further comprising an emphasizing unit that corrects by a larger amount as the angle difference from the average of the maximum principal curvature direction vector becomes smaller. The above emphasis means is
For the first pixel group including the interlobar fissure plane, the pixel vector value of each pixel includes a maximum principal curvature direction vector of the pixel and a normal vector of the interlobar fissure region included in the first pixel group. Multiplying the inner product with
For the first pixel group that does not include the interlobar fissure plane, the pixel value of each pixel includes the maximum principal curvature direction vector of the pixel and the average of the maximum principal curvature direction vectors of all the pixels of the first pixel group. The image processing apparatus according to claim 7, wherein an operation of multiplying an inner product is performed. The expansion means is
Dividing the pixels constituting the three-dimensional image into a second pixel group unit composed of a second number of adjacent pixels smaller than the first number;
In the second end pixel group and the second target pixel group which are two adjacent second pixel groups traversed by the dividing plane, the angular difference of the sum of the maximum principal curvature direction vectors of all the pixels included in each of the second end pixel group and the second target pixel group is When the angle is less than 5, a second expansion process for providing the second target pixel group with second identification information having the same value as the second identification information of the second end pixel group,
The second target pixel group after the second expansion process is set as a new second end pixel group and is adjacent to the second target pixel group until the angle difference becomes equal to or greater than the fifth angle. And repeating while setting a new second target pixel group to which the second identification information is not given,
Furthermore, in the first second extension process, the first end pixel group that is an end portion that does not include the interlobar fissured surface among the plurality of first pixel groups that are continuous with the second end pixel group. 9. The device according to claim 6, wherein the second target pixel group is set on a side adjacent to the second end pixel group and not provided with the second identification information. The image processing apparatus according to any one of the above.   The image processing apparatus according to claim 4, wherein the three-dimensional image is obtained by removing a lesion area.   The image processing apparatus according to claim 4, wherein a pixel value of a pixel of the three-dimensional image corresponds to an X-ray absorption rate in the subject. An image processing method for extracting a planar region from a three-dimensional image having pixels arranged in three dimensions,
A pixel extraction step of extracting a pixel having a maximum principal curvature of a four-dimensional hypersurface having a pixel value of each pixel of the three-dimensional image as a fourth dimension that is a predetermined value or more;
Among the pixels extracted in the pixel extraction step, the angle difference between the maximum principal curvature direction vectors on the four-dimensional hypersurface of any two pixels included and included is a first angle. A planar region extracting step of extracting a planar region that is less than the three-dimensional image;
1st identification information which provides the 1st identification information according to the said maximum principal curvature direction vector of the 1 or more pixel which comprises this planar area to each planar area extracted at the said planar area extraction step An image processing method comprising: an assigning step.   A control program for causing a computer to function as the image processing apparatus according to any one of claims 1 to 11, wherein the control program causes the computer to function as each of the means.   A computer-readable recording medium on which the control program according to claim 13 is recorded.