JP5353876B2 - Image processing device - Google Patents

Description

  The present invention relates to an image processing apparatus.

  In diagnosing bone diseases such as osteoporosis, the density of trabecular bone is an important finding.

  2. Description of the Related Art Conventionally, a system for determining the density of trabecular bone from an X-ray image obtained by imaging a bone site has been developed. For example, a method is disclosed in which a wavelet transform is performed on an X-ray image, and the density of the trabecular bone is obtained from the magnitude of the reduction rate of the basic wavelet function (see, for example, Patent Document 1).

In addition, an energy subtraction process is performed to obtain a bone part image at each photographing time point, and a skeleton process based on a morphological operation is performed on the bone part image to obtain an enhanced image in which the trabecular bone is emphasized. And the method of calculating | requiring the information which shows the temporal change of a trabecular bone using this emphasized image is also disclosed (for example, refer patent document 2, 3, 4).
Japanese Patent No. 302434 Japanese Patent Laid-Open No. 10-108073 JP-A-10-155115 Japanese Patent No. 3799603

  However, a wavelet called a Gabor wavelet is used in the method of Patent Document 1, but the Gabor wavelet has a high processing load. Further, trabecular bone corresponds to the medium frequency component, but Gabor wavelet has low resolution of the medium frequency component, and accurate information on the trabecular bone cannot be obtained.

  On the other hand, according to the methods of Patent Documents 2 and 3, the energy subtraction requires imaging twice, and low-energy X-ray irradiation is performed, resulting in an increase in imaging time and exposure dose.

  Furthermore, the morphological process used for emphasizing the trabecular component in Patent Documents 2 to 4 is characterized in that artifacts are likely to appear depending on the shape of the template used. Thus, accurate trabecular information cannot be obtained.

  The subject of this invention is calculating | requiring the feature-value of a trabecular bone accurately and simply.

According to the invention described in claim 1,
Perform frequency decomposition by binomial wavelet transform on the medical image of the bone, weight the frequency component at the level corresponding to the trabecular bone, reconstruct the image, create an emphasized image that emphasizes the trabecular bone, An image processing apparatus including an image processing unit that calculates a trabecular feature amount using an emphasized image is provided.

According to the invention described in claim 2,
The image processing unit reconstructs an image using a frequency component at a level other than a high frequency component at a lowest level and / or a low frequency component at a highest level from each frequency component subjected to frequency decomposition. An image processing apparatus according to item 1 is provided.

According to the invention described in claim 3,
The level corresponding to the trabecular bone is a level in which the frequency intensity peak of the wavelet waveform is in a range where the wavelength when converted into the actual size of the subject is 100 μm or more and 500 μm or less. The described image processing apparatus is provided.

According to the invention described in claim 4,
The trabecular feature amount is the density of the trabecular bone,
The image processing unit creates a weighted image in which a frequency component at a level corresponding to a trabecular bone is weighted in the main scanning direction and the sub scanning direction of the image, determines an image portion corresponding to the trabecular bone in the emphasized image, The image processing device according to any one of claims 1 to 3, wherein the density of the trabecular bone is calculated from the area ratio between the image determined to be the trabecular bone and the entire emphasized image. .

According to the invention described in claim 5,
The feature amount of the trabecular bone is the number of trabecular bones,
The image processing unit creates an emphasized image obtained by weighting only a frequency component of a level corresponding to a trabecular bone only in the main scanning direction of the image and an emphasized image obtained by weighting only the sub-scanning direction of the image. The image processing apparatus according to any one of claims 1 to 3, wherein an image portion corresponding to the above is determined, and the number of trabeculae is calculated based on the number of pixels of the image determined to be a trabecular bone Is provided.

According to the invention described in claim 6,
The image processing unit further determines an image portion corresponding to the trabecular bone in the enhanced image obtained by weighting the frequency component of the level corresponding to the trabecular bone in the main scanning direction and the sub scanning direction of the image, and is determined to be the trabecular bone Only the image portion where the image portion and the image portion determined to be trabecular in the weighted image weighted only in the main scanning direction or only the sub-scanning direction of the image are subject to calculation of the number of trabeculae. An image processing apparatus according to item 5 is provided.

According to the invention described in claim 7,
The image processing unit performs anisotropic thinning on each of the emphasized image weighted only in the main scanning direction and the emphasized image weighted only in the sub-scanning direction, and the trabecular bone in the emphasized image subjected to the thinning processing. The image processing apparatus according to claim 5 or 6, wherein the number of trabeculae is calculated based on the number of pixels of the image determined to be.

According to the invention described in claim 8,
Prior to the anisotropic thinning process, the image processing unit performs anisotropic expansion / contraction processing on each of the emphasized image weighted only in the main scanning direction and the emphasized image weighted only in the sub-scanning direction. An image processing apparatus according to claim 7 is provided.

According to the invention described in claim 9,
The image processing unit extracts an image portion for calculating a trabecular feature amount from the created enhanced image, and calculates the trabecular feature amount using the extracted image. An image processing apparatus according to any one of the items is provided.

  According to the inventions described in claims 1 and 3, it is possible to create a high-resolution enhanced image even in the middle frequency range corresponding to the trabeculae by the two-wavelet. By using such an enhanced image, the feature amount of the trabecular bone having a fine structure whose outline is not clear can be obtained accurately and easily.

  According to the invention described in claim 2, it is possible to create an enhanced image excluding the lowest level high frequency component and the highest level low frequency component that do not correspond to the trabecular bone. Can be obtained. Thereby, the trabecular feature amount can be calculated with high accuracy.

  According to the fourth aspect of the present invention, it is possible to accurately grasp a trabecular bone having a fine structure by emphasizing the image portion of the trabecular bone, and to accurately calculate the density thereof.

  According to the invention of claim 5, it is possible to accurately grasp the trabecular line structure extending in the main scanning direction and the trabecular line structure extending in the sub-scanning direction. The number can be calculated for each direction in which the line structure extends. Therefore, the number can be obtained with higher accuracy.

  According to the invention described in claim 6, it is possible to calculate the feature amount by excluding the image portion emphasized more than necessary from the image portion determined to be the trabecular bone, and to calculate with higher accuracy. it can.

  According to the invention described in claim 7, when thinning the trabecular image, the trabecular line image extending in the main scanning direction or the sub-scanning direction is reduced in the extending direction. Can be prevented. Thereby, the disappearance of the trabecular line image can be prevented and the number thereof can be accurately calculated.

  According to the invention described in claim 8, with respect to the trabecular line image extending in the main scanning direction or the sub-scanning direction, the image portions that should originally constitute one line structure are connected, and It is a line structure and can be processed so as not to connect image portions that should not be connected. Therefore, the number of trabeculae can be accurately calculated.

  According to the invention described in claim 9, the feature amount can be calculated by removing a noise component that is not a trabecular bone, and calculation with higher accuracy is possible. In addition, the processing time required for calculation can be shortened.

It is a figure which shows the functional structure of the image processing apparatus in this embodiment. It is a figure which shows an example of the imaging device which performs expansion imaging or phase contrast imaging. It is a figure explaining that a picked-up image is expanded by expansion photography or phase contrast photography. It is a figure explaining the edge emphasis effect by phase contrast photography. It is a flowchart which shows the flow of the process performed in an image processing apparatus when calculating the feature-value of a trabecular bone. It is a flowchart which shows the flow of the calculation process of the density of a trabecular bone. It is a flowchart which shows the flow of a calculation process of the number of bone mass. It is a figure which shows the example of the image extracted as ROI. It is a graph which shows the frequency characteristic of a binomial wavelet. It is a figure which shows the frequency decomposition by a binomial wavelet, and the reconstruction of an image. It is a figure which shows the emphasis image example. It is a figure which shows the example of the emphasis image binarized. It is a figure which shows the example of a process result which performed the normal expansion / contraction process. It is a figure which shows the example of a template used by anisotropic expansion / contraction processing. It is a figure which shows the example of a process result which performed the anisotropic expansion / contraction process. It is a figure which shows the example of a process result which performed the normal thinning process. It is a figure which shows the example of a template used by anisotropic thinning process. It is a figure which shows the example of a process result which performed the anisotropic thinning process. It is a figure explaining calculation of the number of trabeculae.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Image processing apparatus 11 Control part 12 Operation part 13 Display part 14 Communication part 15 Storage part 16 Image processing part 20 Imaging device 33 X-ray source 35 X-ray detection part

  First, the configuration will be described.

  FIG. 1 shows a functional configuration of an image processing apparatus 10 in the present embodiment.

  As shown in FIG. 1, the image processing apparatus 10 includes a control unit 11, an operation unit 12, a display unit 13, a communication unit 14, a storage unit 15, and an image processing unit 16.

  The control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and the like. The control unit 11 performs various calculations in cooperation with a control program stored in the storage unit 15 and performs operations of the units. Centralized control.

  The operation unit 12 includes a keyboard and the like, generates operation signals according to these operations, and outputs them to the control unit 11.

  The display unit 13 includes a display, and displays various operation screens and medical images according to display control of the control unit 11.

  The communication unit 14 includes a communication interface, and communicates with an external device such as a server that stores and distributes medical images and an imaging device that captures medical images via a network. For example, the communication unit 14 receives a medical image for image processing from a server or an imaging device, or transmits a medical image that has been subjected to image processing to the server.

  The storage unit 15 stores a control program, an image processing program, parameters necessary for executing the program, and the like. The storage unit 15 stores medical images to be image processed. A hard disk or the like can be applied as the storage unit 15.

  The image processing unit 16 performs necessary image processing on the medical image. Examples of image processing include sharpening processing and gradation conversion processing. Since the type and conditions of the required image processing differ depending on the imaging region, the image processing unit 16 determines the necessary image processing type and image processing condition information for each imaging region, and performs image processing according to the information.

  In addition, the image processing unit 16 performs image processing on a medical image obtained by imaging a bone part, and calculates a feature amount of the trabecular bone. The calculation result is output to the control unit 11.

  Next, the operation will be described.

  In the image processing apparatus 10, when a medical image to be image-processed is received from a server or the like via the communication unit 14, the image processing unit 16 performs necessary image processing according to the imaging part. For the medical image of the bone part, the image processing unit 16 executes a process of calculating the trabecular feature amount. The trabecular feature amount is information that is an index of the progress and degree of disease such as osteoporosis, and is information on the density of the trabecula and the number of trabeculae. The trabecular bone forms a network structure in which several line images intersect on the image. The number of trabecular bones refers to the number of line images.

  Since the trabecular bone has a fine structure, it is preferable that the medical image of the bone part has a high resolution in order to calculate an accurate feature amount. Specifically, the size of one pixel when converted to the actual size of the subject is preferably 50 μm or less. More preferably, it is 30 μm or less. Such a medical image may be an image captured and generated by an imaging device capable of realizing a high resolution, or may be an image obtained by reading a medical image recorded on a film.

  When the medical image of the bone part is an X-ray image, it is preferably an image obtained by magnified imaging. In particular, an image obtained by phase contrast imaging is preferable. Phase contrast imaging refers to an X-ray source and a subject, and a subject and an X-ray in enlarged imaging, as described in JP-A-2001-91479, JP-A-2001-311701, JP-A-2003-180607, and the like. This is an imaging method in which the distance between the detection units, the X-ray irradiation conditions, and the like are predetermined, and has the effect of enhancing the edges of the structure included in the image. By enlarging a fine trabecular portion by magnified imaging or phase contrast imaging, the feature amount can be calculated with high accuracy.

  Hereinafter, enlarged photographing and phase contrast photographing will be described.

  FIG. 2 is a diagram illustrating an example of an imaging device that images a bone part of a hand.

  As shown in FIG. 2, the photographing apparatus 20 includes a photographing unit 3 and a main body unit 4 that performs photographing control and the like. The imaging unit 3 is configured such that an arm 31 attached to a support column 32 can be moved up and down along the support column 32. The arm 31 is provided with an X-ray source 33, a subject table 34, and an X-ray detector 35. The X-ray detector 35 is a cassette or the like containing a flat panel detector (FPD) or a phosphor plate. The subject table 34 and the X-ray detection unit 35 are configured to be movable up and down along the support shaft of the arm 31. The arm 31, the subject table 34, or the X-ray detection unit 35 is moved up and down to adjust its height position. By doing so, it is possible to adjust the distance between the X-ray source 33 and the subject H and between the subject H and the X-ray detector 35. The arrows shown in FIG. 2 indicate the movement directions of the arm 31, the subject table 34, and the X-ray detection unit 35.

  The main body unit 4 is a computer device that includes a display, operation buttons, a communication unit, and the like, and is used for photographing operations. For example, the main body unit 4 performs imaging control by the imaging unit 3, reads a medical image from the X-ray detection unit 35, and transfers the read medical image to a server.

  Enlarged photographing is performed by providing a distance between the subject H and the X-ray detector 35 as shown in FIG. By providing a distance between the subject H and the X-ray detection unit 35, the X-rays emitted from the X-ray source 33 in the shape of a cone beam as shown in FIG. Since the advance X-ray detection unit 35 is reached, the obtained medical image (b) is an image having an enlarged size compared to the close-contact photographed image (a) close to the actual size of the subject H. At this time, the magnification M is R1 from the X-ray source 33 to the subject H, R2 from the subject H to the X-ray detector 35, and R3 from the X-ray source 33 to the X-ray detector 35. Assuming that (R3 = R1 + R2), it can be obtained by the following equation 1.

M = R3 / R1 (1)
The enlargement ratio M can be adjusted by changing the ratio of the distances R1 and R2.

  In phase contrast imaging, the X-ray focal diameter (μm), the tube voltage applied to the X-ray source 33, and the like are set in a predetermined range, and the distances R1, R2, and R3 are set as predetermined distances. For example, when the distance R1 is 65 (cm), R2 is 49 (cm), and the enlargement ratio is 1.75 times, the focal range is 30 μm or more and 350 μm or less at the nominal focal diameter at the time of IEC compliance evaluation. The focal range is particularly preferably from 50 cm to 280 cm. Further, the reading pitch at this time is 43.75 (μm).

  An edge enhancement effect can be obtained in a medical image photographed under these conditions.

  The edge enhancement effect will be described.

  As shown in FIG. 4, the X-rays refracted by passing through the edge of the subject H overlap with the X-rays passed without passing through the subject H, and the X-ray intensity of the overlapped portion is increased. On the other hand, a phenomenon occurs in which the X-ray intensity is weakened in a portion inside the edge of the subject H. As described above, as a result of the X-ray intensity difference being widened at the border of the subject H, the edge of the structure of the subject H is emphasized. This is the edge enhancement effect. By obtaining a medical image of the bone by phase contrast imaging, it is possible to obtain a medical image in which the edge of the trabecular bone is emphasized and sharply depicted.

  Note that the imaging apparatus differs in mechanical structure and the like depending on which part the bone part to be imaged belongs to, but the basic imaging method is the same as described above.

  Next, the flow of processing in which the image processing unit 16 calculates the trabecular feature amount will be described with reference to FIGS. These processes are software processes realized by the cooperation of the feature amount calculation program stored in the storage unit 15 and the image processing unit 16.

  As shown in FIG. 5, first, the image processing unit 16 extracts an image part (this is called ROI: Region Of Interest) to be diagnosed from the medical image of the bone part (step S1). This is because a portion unrelated to the trabecular bone is excluded from the processing target and the feature amount is calculated with high accuracy.

  The size of the ROI may be determined as appropriate. For example, as shown in FIG. 8, the extracted image G2 having a size of 200 pixels × 200 pixels extracted from the medical image G1 of the bone part is used. Since only the image portion of the trabecular bone is extracted as much as possible, it is necessary to determine the position of the ROI while avoiding the tissue called the growth line F that appears inside the bone and the edge E of the bone. This is because the growth line F or the edge portion E having a signal value smaller than that of the trabecular bone can be noise. When extracting, a position that does not include the growth line F or the edge portion E in advance is empirically determined in advance, and a region centered on that position may be extracted. For example, when the bone portion is a rib, as shown in FIG. 8, the edge E of the distal end portion of the rib is detected by an edge detection filter, template matching, or the like, and the center 200 is located about 400 pixels away from the distal end portion. An area of pixels × 200 pixels is extracted as an ROI image portion.

  Next, the image processing unit 16 performs frequency decomposition on the extracted image using a Dyadic Wavelet to create an emphasized image in which the trabecular bone is emphasized (step S2). The enhancement image to be created is an enhancement image in which only the frequency component in the x direction of the trabecular bone is emphasized, and only the frequency component in the y direction when the main scanning direction of the medical image is represented in the x direction and the sub scanning direction in the y direction. The enhanced image and the enhanced image in which both frequency components in the x direction and the y direction are enhanced.

  The enhanced image in which both frequency components in the x direction and the y direction are enhanced is used for processing for calculating a trabecular density feature amount. The enhanced image in which only the frequency component in the x direction is emphasized and the enhanced image in which only the frequency component in the y direction is enhanced are used for processing for calculating the feature quantity of the number of trabeculae. The types of emphasized images used for processing differ from each other because the density of the trabecular bone shows the proportion of the trabecular bone occupying in the bone portion. Therefore, it is necessary to grasp the entire trabecular bone regardless of the x and y directions. On the other hand, the trabecular bone has a mesh structure in which line structures (linear structures) intersect, and the number of trabecular bones needs to grasp the linear images extending in the x direction and the y direction, respectively. It depends.

  Here, frequency decomposition by binomial wavelet will be described.

The wavelet function of the binomial wavelet is expressed by the following equation 2. In Expression 2, i represents the scale of the wavelet function, and the larger the value of _{i, the} smaller the frequency of the wavelet function Ψ _{i, j} (x). That is, the value of i indicates a level at which frequency decomposition is performed. j indicates the position of the wavelet function. The position where the wavelet function Ψ _{i, j} (x) oscillates moves according to the value of _j .

  The characteristic of the binomial wavelet is that the minimum moving unit of the position of the wavelet function is a constant value j regardless of the level i. In general, in the image processing field, since the calculation is fast, orthogonal wavelets and biorthogonal wavelets as shown in the following Equation 3 are widely used.

  The structure of the trabecular bone is a very fine structure, and high resolution of the image is required for quantifying the trabecular bone. Further, since the trabecular bone has a spongy structure and the structure itself is very thin, the trabecular bone depicted in the image is not clear in comparison with other tissues. That is, the contour of the trabecular bone cannot be captured in the high frequency range.

  FIG. 9 is a graph showing the frequency characteristics of the binomial wavelet, and shows the waveforms from level 1 to level 5 of the wavelet function group. The horizontal axis represents the wavelength (unit: μm, expressed as a wavelength obtained by converting the size of one pixel into the actual size of the subject), and the vertical axis represents the wave intensity.

  In general, the thickness of the trabecular bone is said to be 100 μm or more and 200 μm or less, and the distance between the trabecular bone and the trabecular bone is about the same. Assuming that the repetitive structure of the line structure intersecting the trabeculae, the repetitive period of the trabecular structure is estimated to be about 200 μm to 400 μm. Therefore, it can be said that a wavelet function having a wavelength component comparable to the range from the lower limit 100 to the upper limit 400 (μm) is suitable for detecting a trabecular bone.

  Therefore, it can be said that the level corresponding to the trabecular bone is a waveform level having a peak value in a range of 100 μm or more and 400 μm or less, but as shown in FIG. 9, a high level having a peak value in the vicinity of 500 (μm). However, since the component of wavelength 400 (μm) is sufficient, the upper limit is expanded in this embodiment, and a level having a waveform having a peak value in the range of wavelength 100 to 500 μm is set as a level corresponding to the trabecular bone. deal with. The level of the waveform preferably has a peak value in a range of 200 μm and 300 μm or less. From the graph shown in FIG. 9, it is the wavelet functions Ψ3 and Ψ4 of levels 3 and 4 that satisfy such conditions, and the waveform of Ψ3 + Ψ4 that combines them, so that the trabeculae have medium frequencies such as levels 3 and 4 It can be seen that it has a frequency component in the range.

However, as shown in Equation 3, in the orthogonal wavelet and the bi-orthogonal wavelet, the minimum movement unit of the wavelet function is 2 ⁱ and is defined discretely. As level i increases, the number of moving units increases at an accelerated rate, so that the resolution is very low in the middle frequency range such as level 3 or level 4.

  On the other hand, in the binomial wavelet, the minimum moving unit j of the wavelet function can be arbitrarily specified. Therefore, by specifying j to a small value, regardless of the level i, that is, frequency resolution with high resolution at any level. Is possible. Therefore, in order to emphasize a trabecular bone having a fine structure whose outline is not clear, a binomial wavelet capable of realizing a high resolution even in the middle frequency range is particularly effective.

  Next, a method for creating an emphasized image by binomial wavelet transform will be described.

  The level n wavelet transform can be obtained by a filter process as shown in FIG. In FIG. 10, HPF and HPF ′ are high-pass filters, and LPF and LPF ′ are low-pass filters. Further, x attached to the HPF and LPF indicates processing in the x direction, and y indicates processing in the y direction.

  Preferred examples of filter coefficients used here are those described in the document `` Stephan Mallat and Sifen Characterization of Signals from Multiscale Edges, IEEE Transaction on Pattern Analysis and Machine Intelligence, 14, (7), 710-732, 1992 ''. Is mentioned.

  As shown in FIG. 9, first, in the level 1 wavelet transform, the original image S0 is decomposed into a low-frequency component S1 and high-frequency components Wx1 and Wy1. The low frequency component S1 is further decomposed into a low frequency component S2 and high frequency components Wx2 and Wy2 by a level 2 wavelet transform. By repeatedly wavelet transforming the low frequency components in this way, finally, the low frequency component Sn and the high frequency components Wx1 to Wxn and Wy1 to Wyn are decomposed. In FIG. 10, processing from level 3 to level (n-2) is omitted.

  Each of the decomposed frequency components Sn, Wx1 to Wxn, and Wy1 to Wyn can completely reconstruct the original image S0 by inverse wavelet transform. That is, when each frequency component to be reconfigured is represented by Sn ′, Wxn ′, and Wyn ′, as shown in FIG. 10, the filter processing HPF′nx, LPF′nx, LPF′ny, and HPF′ny for inverse conversion are represented by these. , LPFnx, LPFny and inversely transform, and combine the obtained components to obtain Sn-1 ′. The same processing is repeated at each level, and finally the original image S0 is reconstructed.

  Before performing the inverse transformation, the image processing unit 16 weights the frequency components Sn, Wx1 to Wxn, and Wy1 to Wyn of each level in order to emphasize the frequency components corresponding to the trabecular bone. As described above, the trabecular bone corresponds to a medium frequency component such as levels 3, 4 and the like, and the low frequency component at the highest level and the high frequency component at the lowest level are considered to be at least components of the trabecular bone. Therefore, the weighting is performed so that the image is reconstructed only with the frequency components of other levels except the highest level low frequency component and / or the lowest level high frequency component.

  Hereinafter, an example in which frequency decomposition from level 1 to level 4 is performed and weighting is performed so as to perform reconstruction by removing high frequency components (Wx1, Wy1) of the lowest level 1 will be described.

  The weighting condition is determined according to the feature quantity to be calculated.

  For example, when creating an enhanced image used to calculate the density of trabecular bone, one or more weights are assigned to level 3 and level 4 frequency components Wx3, Wx4, Wy3, and Wy4 among level 1 to level 4 frequency components. The other frequency components Wx1, Wx2, Wy1, and Wy2 are multiplied by a weighting coefficient of 0 to obtain Wx1 ′ to Wx4 ′ and Wy1 ′ to Wy4 ′ for each level from level 1 to level 4. S4 is adjusted to a constant value (for example, an intermediate gray value) Sn ′. By performing image reconstruction using Wx1 ′ to Wx4 ′, Wy1 ′ to Wy4 ′, and Sn ′, the enhanced image S0 ′ in which the frequency components at levels 3 and 4 are enhanced in both the x direction and the y direction. Is reproduced. Note that the low-frequency component Sn at the highest level 4 may be subjected to image reconstruction by excluding Sn by multiplying by a weighting coefficient of 0.

  On the other hand, when calculating the number of trabeculae, an emphasized image emphasized only in the x direction and an emphasized image emphasized only in the y direction are created.

  When emphasizing only in the x direction, the frequency components Wx3 and Wx4 in the x direction of levels 3 and 4 are multiplied by one or more weighting coefficients, and the other frequency components Wx1, Wx2, Wy1, Wy2, Wy3, and Wy4 are weighted with 0. Multiply the coefficients to obtain Wx1 ′ to Wx4 ′ and Wy1 ′ to Wy4 ′ for each level from level 1 to level 4. Further, S4 is adjusted to a constant value (for example, a value that becomes intermediate gray) Sn ′. By performing image reconstruction using the obtained Wx1 ′ to Wx4 ′, Wy1 ′ to Wy4 ′, and Sn ′, the enhanced image S0 ′ in which the frequency components of level 3 and 4 are enhanced only in the x direction is reproduced. Is done.

  When emphasizing only in the y direction, the frequency components Wy3 and Wy4 in the y direction at levels 3 and 4 are multiplied by one or more weighting coefficients, and the other frequency components Wx1, Wx2, Wx3, Wx4, Wy1, and Wy2 are weighted with 0. Multiply the coefficients to obtain Wx1 ′ to Wx4 ′ and Wy1 ′ to Wy4 ′. S4 is similar to the case where only the x direction is emphasized. By reconstructing the image using the obtained Wx1 ′ to Wx4 ′, Wy1 ′ to Wy4 ′, and Sn ′, the enhanced image S0 ′ in which the frequency components of level 3 and 4 are enhanced only in the y direction is reproduced. Is done.

  An example of the emphasized image created as described above is shown in FIG. The enhanced image G4 shown in FIG. 11 is an image enhanced in both the x direction and the y direction, the enhanced image G5 is an image enhanced only in the x direction, and the enhanced image G6 is an image enhanced only in the y direction. It can be seen from FIG. 11 that the structure of the trabecular bone is clearer in the enhanced image G4 than before the frequency decomposition (extracted image G2 shown in FIG. 8). Further, it can be seen that only the linear trabecular structure extending in the x direction is emphasized in the emphasized image G5, and conversely, only the linear trabecular structure extending in the y direction is emphasized in the emphasized image G6.

  When the enhanced image is created, the image processing unit 16 performs a trabecular density calculation process (step S3) and a trabecular number calculation process (step S4) using the enhanced image.

  First, the trabecular density calculation processing will be described with reference to FIG. What is used for calculating the density is an enhanced image (emphasized image G4 in FIG. 11) in which both the x direction and the y direction are enhanced.

  As shown in FIG. 6, the image processing unit 16 first binarizes the emphasized image G4 (step S31). FIG. 12 shows an example of a binarized enhanced image G41.

  Next, the image processing unit 16 determines the image portion of the trabecular bone in the binarized enhanced image G41, and calculates the area (step S32). In the medical image obtained by X-ray imaging, the trabecular portion has a low signal, so it is considered that the signal value becomes 0 by binarization. Therefore, the image processing unit 16 determines that the pixel having the signal value 0 (white pixel in FIG. 12) in the binarized enhanced image G41 is the image portion of the trabecular bone. Then, the area of the image of the trabecular bone is calculated. Here, the number of pixels of the image portion determined to be the trabecular bone is counted as indicating the area.

  Next, the image processing unit 16 calculates the density of the trabecular bone according to the following equation 4 from the calculated area (number of pixels) of the trabecular bone and the area of the entire image G2 extracted as the ROI (indicated by the number of pixels) (step S4). S33).

Trabecular density = area of image determined to be trabecular / area of extracted image of ROI (4)
Next, processing for calculating the number of trabeculae will be described with reference to FIG. What is used for calculating the number is an enhanced image G5 in which only the x direction is enhanced, and an enhanced image G6 in which only the y direction is enhanced.

  As shown in FIG. 7, the image processing unit 16 binarizes each of the enhanced image G5 and the enhanced image G6 (step S41).

  Next, mask processing is performed on each of the binarized emphasis images G5 and G6 to determine the image portion of the trabecular bone (step S42). In the mask processing, the image processing unit 16 primarily determines the image portion of the trabecular bone in each of the binarized enhanced images G5 and G6. The determination method is the same as described above, and a white pixel having a signal value of 0 is determined as a trabecular bone. Then, the image portion determined as the trabecular bone is determined also in the enhanced image G4 in which both the x direction and the y direction are emphasized among the image portions determined primarily. That is, only the image portion determined to be a trabecular bone is finally determined to be a trabecular bone in both the emphasized image G4 and the emphasized image G5 or G6. The image processing unit 16 replaces the signal value with 1 for an image portion that is not determined to be trabecular in the enhanced image G4 among image portions that are primarily determined as trabecular in the enhanced images G5 and G6.

  When emphasizing only one direction of only x direction or only y direction, the frequency component of trabeculae may be emphasized more than necessary. Since the emphasized image G4 in which both the x direction and the y direction are emphasized is close to the state of the trabecular bone when viewed, the trabecular determination accuracy is close to that of the visual image. Only the determined image portion is a target for calculating the number of trabeculae.

  When the image portion of the trabecular bone is determined, the image processing unit 16 performs anisotropic expansion and contraction processing on the binarized enhanced images G5 and G6 (step S43). The expansion / contraction process is performed by combining the expansion process and the contraction process. For example, a method described in “Practical image processing first version learned from C language” (Maki Inoue et al., Ohmsha, 1998) or the like can be applied. The dilation processing refers to pixels around a certain pixel (referred to as a pixel of interest), and if there is at least one pixel with a signal value of 1, the signal value of that pixel of interest is set to 1, and the signal values of other surrounding pixels Is a process of setting 0 to 0. The contraction process is the opposite, and if there is at least one pixel with a signal value of 0 in the periphery, the signal value of the pixel of interest is set to 0, and the signal values of other peripheral pixels are set to 1.

  Since the trabecular bone has a network structure, the image portion of the trabecular bone should have many long line images (linear images) extending in the x direction or the y direction. Later, the number of lines will be calculated by converting the line image into an image having a pixel width of 1 by thinning processing, but the images forming the same line structure are connected as much as possible, and the image parts forming another line structure are not connected. By doing so, the number of trabeculae can be calculated with high accuracy. Therefore, an expansion / contraction process is performed prior to the thinning process, and the image portions of the trabeculae that should originally constitute one line structure are connected.

  FIG. 13 shows an example of the result of processing the expansion process four times and the contraction process four times. The processed image G51 is a processing result of the enhanced image G5, and the processed image G61 is a processed result of the enhanced image G6. If the expansion process and the contraction process are simply repeated, the image portion that is supposed to be connected to form the same line as the part surrounded by a circle in FIG. The image parts of different lines that should not be connected are connected.

  Therefore, in the present embodiment, anisotropic expansion / contraction processing is performed that expands and contracts only in either the x direction or the y direction. That is, for the enhanced image G5 in which only the x direction is enhanced, the image portion of the trabecular bone that extends linearly in the x direction is to be connected, so that it expands or contracts only in the x direction. Similarly, the enhanced image G6 only in the y direction is expanded or contracted only in the y direction so as to connect the image portions of the trabeculae extending linearly in the y direction.

  Specifically, templates T11 to T14 shown in FIG. 14 are used. In the case of performing expansion processing only in the x direction, the template T11 is matched so that the target pixel having the signal value 0 is located in the center, and any one of pixels adjacent to the x direction (pixels indicated by asterisks in FIG. 14). However, if there is a signal value of 1, the signal value of the target pixel is set to 1. When the expansion process is performed only in the y direction, the same process may be performed using the template T13. When the reduction process is performed only in the x direction, the template T12 is used for matching so that the target pixel having the signal value 1 is located in the center, and any one of the adjacent pixels in the x direction has a signal value of 0. If there is, the signal value of the target pixel is set to zero. When the reduction process is performed only in the y direction, the same process is performed using the template T14.

  Note that the number and order of the expansion process and the contraction process may be appropriately determined so that an optimum result is obtained depending on the site where the bone part is imaged.

  FIG. 15 shows a processing result of the anisotropic expansion / contraction processing.

  In FIG. 15, the processed image G52 is the processing result of the enhanced image G5, and the processed image G62 is the processed result of the enhanced image G6. In the processed image G51 of FIG. 13, the image portions of different lines are connected, but in the processed image G52 of FIG. 15, such connection is not made. Further, in the processed image G61 in FIG. 13, the image portion of the trabecular bone extending in the y direction could not be connected well and was disconnected, but in the processed image G62 in FIG. 15, the connection was successful.

  When the expansion / contraction process is thus completed, the image processing unit 16 performs anisotropic thinning processing on the processed images G52 and G62 (step S44). The thinning process is a process of converting a trabecular image into a line image having a width of 1 pixel. The thinning process can employ a technique described in “Digital Image Processing for Image Understanding (II)” (by Junichiro Toriwaki, Shogodo, 1999). For example, pay attention to a pixel having a signal value of 1, and refer to 3 pixels × 3 pixels centered on the target pixel. Prepare several templates to detect the area where the signal value is 1 in half of the 3 pixels x 3 pixels and the signal value is 0 in the remaining half of the area. For example, the signal value of the target pixel is set to 0.

  The thinning process is a process of reducing the line width so that the processed line image is positioned at the center of the original image. Therefore, if the thinning process is simply performed, the line width in the x direction and the y direction is reduced. Will be reduced. As a result, as shown by arrows (a) and (b) in FIG. 16, in the processed image G63 after the thinning process of the image G62, the trabecular line image is shorter than the trabecular image in the original processed image G62, The image itself will be missing. In this case, the significance of connecting the image portions of the trabeculae by the expansion / contraction process is lost, and the number of trabeculae cannot be accurately calculated.

  Therefore, in the present embodiment, anisotropic thinning processing is performed. That is, the processing image G52 that is emphasized only in the x direction is thinned only in the y direction, and the processing image G62 that is emphasized only in the y direction is thinned only in the x direction. As a result, in the processed image G52, the trabecular image is reduced only in the y direction, and in the processed image G62, the trabecular image is reduced only in the x direction. Therefore, the lengths of the line images of the plurality of trabeculae extending in the x direction or the y direction, each of which has been enhanced, are not shortened by thinning.

  Specifically, an anisotropic thinning method will be described.

  For example, when thinning only in the x direction, templates T21 and T22 shown in FIG. 17 are used. Templates T21 and T22 in FIG. 17 are templates for detecting a line structure extending in the y direction. That is, an image of 3 pixels × 3 pixels centering on the target pixel of signal value 1 is extracted, and this extracted image and templates T21 and T22 are collated. If the signal values 0 and 1 defined at the 3 × 3 positions of the templates T21 and T22 match the signal values at the positions of the extracted image, it is determined that the template T21 and T22 are matched, and Change signal value 1 to 0. When thinning only in the y direction, the same processing may be performed using the templates T23 and T24 shown in FIG. Templates T23 and T24 are templates for detecting a line structure extending in the x direction.

  Note that the templates T21 to T24 are thinned by four connections, but thinning may be performed in an oblique direction using an eight-connected template. The concept of 4-link and 8-link is described in the document “Digital Signal Processing Series Vol. 7 Digital Image Processing (II) for Image Understanding (p12 to p14, Junichiro Toriwaki et al., Shosodo, 1994)”. Are known.

  FIG. 18 is a diagram showing a processed image G64 obtained by performing anisotropic thinning processing on the processed image G62. Since the processed image G64 is thinned only in the x direction, the image of the trabecular bone disappeared in the processed image G63 in FIG. 16 without reducing the trabecular image in the y direction (FIG. 16). The part surrounded by a frame also remains as a line image in the processed image G64 of FIG.

  When anisotropic thinning processing is performed in this way, the image processing unit 16 determines a pixel having a signal value of 0 as an image portion of the trabecular in the processed image, and determines the number of pixels of the image portion of the trabecular bone. Count (step S45). The total number of pixels counted corresponds to the sum of the lengths of a plurality of trabeculae included in the image extracted as the ROI.

  Next, the image processing unit 16 calculates the number of trabeculae from Expression 5 below, assuming that the length of one trabecular bone extending in the x direction or the y direction corresponds to the length of one side of the image G2 extracted as the ROI. (Step S46). That is, the number of Expression 5 is an index indicating how many line images cross the extracted image G2 in the x direction or the y direction. Note that the length of one side of the extracted image G2 used in Equation 5 is the length in the x direction when the number is obtained from the processed image that is emphasized only in the x direction, and is the length of the processed image that is emphasized only in the y direction. When obtaining the number, the length in the y direction is adopted.

Number of trabeculae = number of pixels in trabecular image / length of one side of ROI extraction image (5)
The processing image G64 shown in FIG. 19 will be described as an example. The length of one side of the image G2 extracted as the ROI is 200 pixels on one side in both the x and y directions. Since the processed image G64 is an image that is emphasized only in the y direction, the length of one side of the extracted image is 200 pixels in the y direction. Here, when the number of pixels in the image portion of the trabecular bone is 600 pixels, the number of trabecular bones can be calculated from Equation 5 as 600 pixels / 200 pixels = 3.

  When the trabecular density and the number of feature quantities are calculated in this way, the image processing unit 16 adds the calculated feature quantity information to the header of the original medical image. Then, the feature amount information is transmitted to an external server or the like together with the medical image by the communication control of the control unit 11.

  As described above, according to the present embodiment, the image processing unit 16 performs frequency decomposition by binomial wavelet transform on the medical image of the bone, and weights the frequency component at a level corresponding to the trabecular bone. The levels corresponding to the trabeculae are levels 3 and 4 in which the peak of the frequency intensity of the wavelet waveform is in the range of the wavelength of 100 μm or more and 500 μm or less converted to the actual size of the subject. Thereafter, the image is reconstructed to create an emphasized image in which the trabecular bone is emphasized. Then, the feature amount of the trabecular bone is calculated using the created enhanced image.

  Since the binomial wavelet can realize high resolution regardless of the level, it is possible to create a high-resolution enhanced image even in the middle frequency range of levels 3 and 4 corresponding to the trabecular bone. By using such an enhanced image, the feature amount of the trabecular bone having a fine structure whose outline is not clear can be obtained accurately and easily.

  Also, by multiplying the lowest level high frequency component and / or the highest level low frequency component by a weighting coefficient of 0, the image is reconstructed using only frequency components other than the frequency component. Since the trabecular bone corresponds to the level of the medium frequency component, the lowest level high frequency component and the highest level low frequency component that do not correspond to the trabecular bone can be excluded, and an enhanced image that emphasizes only the trabecular bone is created. Can do. Thereby, the trabecular feature amount can be calculated with high accuracy.

  The image processing unit 16 calculates the trabecular density as the trabecular feature amount. The image processing unit 16 creates a reconstructed enhanced image G4 by weighting the frequency components of levels 3 and 4 corresponding to the trabecular bone in the x direction and the y direction, and the image portion of the trabecular bone in the enhanced image G4. judge. Then, the area ratio between the area of the image portion determined to be a trabecular bone and the area of the entire extracted image G2 from the emphasized image is calculated as a density. By using an emphasized image in which a trabecular bone with an unclear outline is emphasized, the image portion of the trabecular bone can be grasped with high accuracy, and the density thereof can be calculated with high accuracy.

  Further, the image processing unit 16 calculates the number of trabeculae as the trabecular feature amount. The image processing unit 16 creates an enhanced image G5 reconstructed by weighting only the x direction in the level 3 and 4 frequency components corresponding to the trabecular bone, and an enhanced image G6 reconstructed by weighting only the y direction. Then, the image portion of the trabecular bone is determined in the emphasized images G5 and G6, and the number of trabecular bones is calculated from the number of pixels.

  In determining the trabecular bone, masking is applied to the emphasized images G5 and G6 to determine the image portion of the trabecular bone. Thereby, a trabecular bone can be determined by removing an image portion emphasized more than necessary.

  In calculating the number of trabeculae, anisotropic thinning processing is performed. Thereby, the length of the line image extending in the x direction or the y direction can be reduced without being reduced in the extending direction.

  Further, an anisotropic expansion / contraction process is performed prior to the anisotropic thinning process. Thereby, it is possible to process the line images extending in the x direction or the y direction so as to connect image portions that should originally constitute one line and not connect image portions that should not be connected. Therefore, the number of trabeculae can be calculated with high accuracy.

  In addition, the above-mentioned embodiment is a suitable example of this invention, and is not limited to this.

  For example, in the above-described embodiment, the calculation of the trabecular feature amount is described using a medical image obtained by X-ray imaging as an example. However, if the trabecular image is included in the captured image of the bone, Medical images obtained by MRI (Magnetic Resonance Imaging) or the like may be targeted.

  In addition, the configuration in which the frequency decomposition is performed up to levels 3 and 4 corresponding to the trabeculae has been described, but a configuration in which the frequency decomposition is performed at a higher level than the levels 3 and 4 corresponding to the trabeculae may be employed. For example, frequency components from level 1 to level 5 may be performed, and weighting may be performed so that image reconstruction is performed by removing frequency components of the lowest level 1 and the highest level 5 among them.

  In addition, the effect of performing anisotropic expansion / contraction processing or anisotropic thinning processing is greater when the direction of the trabecular bone forming the network structure matches the x direction and the y direction as much as possible. At the time of imaging, it is necessary to secure the position of the site to be imaged so that the direction of the trabecular bone matches the x and y directions of the image as much as possible. If the position at the time of photographing is clearly deviated, processing such as rotation of the medical image may be performed so that the trabecular direction matches the x and y directions as much as possible. For example, an imaging region (hand, finger, etc.) is detected from a medical image, and the angle of how much is displaced is detected. Then, a process for rotating the medical image by the angle of deviation is performed.

  Or, since the bone may be bent depending on the progress of a medical condition such as rheumatism, it may not be photographed at an assumed angle unlike a healthy person even when photographing at the same position. In this case as well, the feature amount may be calculated after detecting the angle shifted as described above and performing the rotation process.

  In addition, as a computer-readable medium for the program according to the present invention, a non-volatile memory such as a ROM and a flash memory, and a portable recording medium such as a CD-ROM can be applied.

  Further, a carrier wave (carrier wave) is also applied to the present invention as a medium for providing program data according to the present invention via a communication line.

Claims (9)

  Perform frequency decomposition by binomial wavelet transform on the medical image of the bone, weight the frequency component at the level corresponding to the trabecular bone, reconstruct the image, create an emphasized image that emphasizes the trabecular bone, An image processing apparatus including an image processing unit that calculates a feature amount of a trabecular bone using an emphasized image.   The image processing unit reconstructs an image using a frequency component at a level other than a high frequency component at a lowest level and / or a low frequency component at a highest level from each frequency component subjected to frequency decomposition. The image processing apparatus according to item 1.   The level corresponding to the trabecular bone is a level in which the frequency intensity peak of the wavelet waveform is in a range where the wavelength when converted into the actual size of the subject is 100 μm or more and 500 μm or less. The image processing apparatus described. The trabecular feature amount is the density of the trabecular bone,
The image processing unit creates a weighted image in which a frequency component at a level corresponding to a trabecular bone is weighted in the main scanning direction and the sub scanning direction of the image, determines an image portion corresponding to the trabecular bone in the emphasized image, The image processing apparatus according to any one of claims 1 to 3, wherein the density of the trabecular bone is calculated from an area ratio between the image determined to be the trabecular bone and the entire emphasized image. The feature amount of the trabecular bone is the number of trabecular bones,
The image processing unit creates an emphasized image obtained by weighting only a frequency component of a level corresponding to a trabecular bone only in the main scanning direction of the image and an emphasized image obtained by weighting only the sub-scanning direction of the image. The image processing apparatus according to any one of claims 1 to 3, wherein an image portion corresponding to the above is determined, and the number of trabeculae is calculated based on the number of pixels of the image determined to be a trabecular bone .   The image processing unit further determines an image portion corresponding to the trabecular bone in the enhanced image obtained by weighting the frequency component of the level corresponding to the trabecular bone in the main scanning direction and the sub scanning direction of the image, and is determined to be the trabecular bone Only the image portion where the image portion and the image portion determined to be trabecular in the weighted image weighted only in the main scanning direction or only the sub-scanning direction of the image are subject to calculation of the number of trabeculae. 6. The image processing apparatus according to item 5.   The image processing unit performs anisotropic thinning on each of the emphasized image weighted only in the main scanning direction and the emphasized image weighted only in the sub-scanning direction, and the trabecular bone in the emphasized image subjected to the thinning processing. The image processing apparatus according to claim 5 or 6, wherein the number of trabecular bones is calculated based on the number of pixels of the image determined as.   Prior to the anisotropic thinning process, the image processing unit performs anisotropic expansion / contraction processing on each of the emphasized image weighted only in the main scanning direction and the emphasized image weighted only in the sub-scanning direction. The image processing device according to claim 7.   The image processing unit extracts an image portion for calculating a trabecular feature amount from the created enhanced image, and calculates the trabecular feature amount using the extracted image. The image processing apparatus according to any one of the items.