JP2013255808A - Medical image processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a medical image processing device and a medical image processing method which can be applied even to the case of a mucosal surface structure having a complicated shape such as a blood vessel which branches, meanders or changes in a diameter and to the case having a complicated shape with a density change, and can accurately extract the structure.SOLUTION: For input images for which noise is eliminated or the like from endoscopic images, a filtering processing part 32, with a plurality of one-dimensional filters prepared according to the width and running direction of a blood vessel, takes a plurality of pixels of a local area including a pixel of interest, performs a product-sum operation, and extracts the pixel having the featured value of the blood vessel from a filter whose output value takes the maximum value. Further, by a determination part 34, comparison with the threshold of a determination reference is made and whether or not it is a blood vessel pixel is determined.

Description

  The present invention relates to a medical image processing apparatus that extracts a mucosal pattern such as a blood vessel structure by image processing on a medical image.

  Conventionally, in a technique for extracting a target shape from an image, for example, a medical image such as a fundus fluoroscopic image or an endoscopic image, two threshold values using a color tone difference (also referred to as a density difference) caused by light absorption of an object are used. The valuation process and the following filtering are performed.

  Filtering includes an edge extraction filter that extracts the target edge, a digital filter, for example, a low-pass filter (LPF) that uses the frequency characteristics of the target, and a band-pass filter designed based on the width information of the target ( A method of extracting a target shape using a band-pass filter (BPF) is used.

  For example, in Japanese Unexamined Patent Publication No. 9-62836 (Patent Document 1) as a first conventional example, a difference image between a medical image and a smoothed image obtained by performing a smoothing process on a medical image is obtained, and an edge is obtained. Extract regions. Edge enhancement is not performed on edge regions with a degree of edge greater than a predetermined threshold, and edge enhancement is performed on edge regions with a degree of edge greater than a predetermined threshold, and over-emphasis is performed. To generate an easy-to-view image.

  Japanese Patent Laying-Open No. 2007-325656 (Patent Document 2) as a second conventional example describes a dark pixel that is a pixel having a relatively smaller luminance value than peripheral pixels in a near-infrared image obtained by imaging a living body. And a dark pixel image data representing the dark pixel is generated, and then a continuous dark region that is a dark pixel in which the pixels are continuous is extracted using a two-dimensional mask pattern, and the continuous dark region is indicated. Create dark area image data.

  And the specific area | region which consists of the pixel in which the blood vessel exists with respect to continuous dark area | region image data is extracted, and the specific area image data which shows the said specific area is produced.

  However, in the first conventional method, for example, when a blood vessel is to be extracted from an endoscopic image, an image in which the color difference between the blood vessel and a region other than the blood vessel is unclear, or a shading due to a difference in depth is used. An accurate extraction result cannot be obtained for an image in which different blood vessels are present or an image in which the amount of light is not uniformly applied and the background brightness changes due to the curved surface of the living body in the entire image.

JP-A-9-62836 JP 2007-325656 A

  When the above-described image is smoothed and extracted by the edge using the first conventional method, variations occur due to differences in position and color tone in the image, and the desired target shape is obtained. It is not possible to obtain a highly accurate extraction result such as extracting not only the shape of the blood vessel but also the portion that is not the target shape.

  The second conventional example uses a two-dimensional mask pattern indicating the continuity of pixels to facilitate the extraction of the shape that the blood vessel travels. For example, the blood vessel can be bent or meandered in various ways. Accuracy cannot be improved.

  For this reason, it can be applied to a mucosal surface structure that travels in various branched or meandering shapes or has a complex shape such as a blood vessel whose tube diameter (width) changes, and the structure can be extracted with high accuracy. Therefore, a medical image processing apparatus capable of performing the above is desired.

  The present invention has been made in view of the above-described points, and can be applied to a case where a complicated shape in which a branch or meander or a pipe diameter is changed can be applied, and the structure can be extracted with high accuracy. The purpose is to provide.

  The medical image processing apparatus according to one aspect of the present invention includes an input unit that inputs medical image information including a plurality of pixels obtained by imaging a biological mucous membrane, and the target pixel in the plurality of pixels of the medical image information. By applying a first filter that includes at least one of the traveling direction and width of the surface structure of the biological mucous membrane, the first feature information of the pixel of interest is obtained, and the surface structure By applying a second filter including at least one of the traveling direction and width information, the second feature information of the pixel of interest is obtained, and further, the first pixel is added to a neighboring pixel located in the vicinity of the pixel of interest. The third feature information of the neighboring pixels is obtained by applying a filter, and the fourth feature information of the neighboring pixels is obtained by applying the second filter to the neighboring pixels. And a filtering unit that selects a filter that is more responsive to the target pixel from the first filter and the second filter based on the first feature information of the target pixel and the second feature information of the target pixel. And a selection unit that selects a filter that reacts more in the neighboring pixel out of the first filter and the second filter based on the third feature information of the neighboring pixel and the fourth feature information of the neighboring pixel; Determining whether the pixel of interest is a surface structure candidate pixel or a non-mucosal surface structure candidate pixel based on feature information corresponding to the filter selected by the selection unit in the pixel of interest, and further, Based on the feature information corresponding to the filter selected by the selection unit, the neighboring pixel is a surface structure candidate pixel or non-mucosal A comparison result between a determination unit that determines whether the pixel is a surface structure candidate pixel, and at least one information of a neighboring pixel that is determined to be the surface structure candidate pixel by the determination unit and at least one information of the pixel of interest And a correction unit that corrects the determination result made on the target pixel.

FIG. 1 is a diagram showing a configuration of an endoscope system according to a first embodiment of the present invention. FIG. 2 is a block diagram showing a functional configuration of a CPU constituting the image processing apparatus. FIG. 3 is a flowchart showing the contents of processing by the image processing apparatus. FIG. 4 is a view showing an endoscopic image of a biological mucosa including blood vessels. FIG. 5 is an explanatory diagram schematically showing a peripheral region including a blood vessel to be extracted in an endoscopic image. FIG. 6 is a diagram showing a cross-sectional shape of a blood vessel. FIG. 7 is a diagram showing the outline of a one-dimensional filter corresponding to the cross-sectional shape of a blood vessel. FIG. 8 is a flowchart showing filtering processing using a plurality of one-dimensional filters. FIG. 9 is a diagram showing the shape of a plurality of one-dimensional filters. FIG. 10 is a flowchart showing the contents of extraction of feature amounts and determination of whether or not the feature amount is a blood vessel. FIG. 11 is a diagram illustrating a display example in which a processed image is displayed together with an input image. FIG. 12 is a diagram showing information stored in the filter storage unit according to the second embodiment of the present invention. FIG. 13 is a flowchart showing the processing contents in the second embodiment. FIG. 14 is a flowchart showing the processing contents in the third embodiment of the present invention. FIG. 15 is a flowchart showing the processing contents of the determination in FIG. FIG. 16 is a flowchart showing the processing contents of the determination in FIG. FIG. 17 is a flowchart showing the processing contents of the correction in FIGS. 15 and 16.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(First embodiment)
An endoscope system 1 according to the first embodiment of the present invention shown in FIG. 1 includes an endoscope device 2 as a medical device and an endoscope image as a medical image obtained by the endoscope device 2. A medical image processing apparatus (hereinafter simply abbreviated as “image processing apparatus”) 3 composed of a personal computer that performs image processing on the image processing apparatus 3 and a monitor 4 that displays an image processed by the image processing apparatus 3. .

  The endoscope device 2 is inserted into a body cavity and forms an endoscope 6 that forms an in-vivo imaging device that images the inside of the endoscope, a light source device 7 that supplies illumination light to the endoscope 6, and an endoscope 6 A camera control unit (abbreviated as CCU) 8 that performs signal processing on the imaging means, and a monitor 9 that displays an image captured by the imaging device as an endoscopic image by inputting an image signal output from the CCU 8; Have

  The endoscope 6 includes an insertion portion 11 that is inserted into a body cavity and an operation portion 12 that is provided at the rear end of the insertion portion 11. A light guide 13 that transmits illumination light is inserted into the insertion portion 11.

  The rear end of the light guide 13 is connected to the light source device 7. Then, the illumination light supplied from the light source device 7 is transmitted by the light guide 13 and emitted from the distal end surface attached to the illumination window provided at the distal end portion 14 of the insertion portion 11 (the transmitted illumination light) is inspected. Or illuminating a living tissue to be diagnosed. The light source device 7 supplies the light guide 13 with white illumination light for normal observation or frame sequential illumination light of red (R), green (G), and blue (B) covering white.

  The light source device 7 may be configured to supply narrow-band illumination light to the light guide 13 and illuminate the living tissue to be examined with narrow-band light.

  An objective lens 15 is attached to an observation window adjacent to the illumination window. For example, a charge coupled device (abbreviated as CCD) 16 serving as a solid-state image sensor is disposed at the image forming position of the objective lens 15. An imaging unit 17 that performs imaging is formed by the objective lens 15 and the CCD 16. The optical image formed on the imaging surface of the CCD 16 is photoelectrically converted by the CCD 16.

  The CCD 16 is connected to the CCU 8 via a signal line, and the CCD 16 outputs a photoelectrically converted image signal when a CCD drive signal from the CCU 8 is applied. This image signal is signal-processed by an image processing circuit in the CCU 8 and converted into an image signal to be displayed on the display device. This image signal is output to the monitor 9, and an endoscopic image is displayed on the display surface of the monitor 9. This image signal is also input to the image processing device 3.

  The image processing apparatus 3 includes an image input unit 21 as an image input unit that inputs an endoscopic image as a medical image generated by the endoscope apparatus 2, and an endoscope input from the image input unit 21. A CPU 22 as a central processing unit that performs image processing on a digitized image signal corresponding to an image, and a processing program storage unit 23 that stores a processing program (control program) that causes the CPU 22 to perform image processing; Have Note that the image input unit 21 may be provided on the endoscope apparatus 2 side.

  The image processing apparatus 3 is used when the image processing is performed by the CPU 22 and the image storage unit 24 that stores the image signal of the input image input from the image input unit 21, the image signal that is being processed, and the processing is completed. A processing information storage unit 25 that stores information such as a plurality of one-dimensional filters as processing information; a hard disk 27 as a storage device that stores image signals processed by the CPU 22 and processing information and the like via a storage device interface 26; An input operation unit 29 including a display processing unit 28 that performs a display process for displaying an image processed by the CPU 22, and a keyboard that allows a user such as an operator to input, select, and instruct an image processing parameter. And have.

  As described above, the processing information storage unit 25 includes the filter storage unit 39 that stores (stores) information on a plurality of one-dimensional filters used when the CPU 22 performs image processing.

  A plurality of one-dimensional filters are stored in the filter storage unit 39 in advance, and a user such as an operator who uses the image processing apparatus 3 can input from the input operation unit 29 to a living tissue to be diagnosed in a medical image. Depending on the target shape of the contained blood vessel, a one-dimensional filter that makes it easy to extract the target shape or its feature amount can be selected and an input operation used for actual image processing can be performed.

  Although a plurality of one-dimensional filters are used in the present embodiment, the present invention is not limited to the case of a plurality of one-dimensional filters, as will be described later, and includes a plurality of one-dimensional filters or filters corresponding thereto. A plurality of filters may be used.

  The image signal generated by the display processing unit 28 is output to the monitor 4, and a processed image or the like subjected to image processing is displayed on the display surface of the monitor 4. The image input unit 21, CPU 22, processing program storage unit 23, image storage unit 24, processing information storage unit 25, storage device interface 26, display processing unit 28, and input operation unit 29 are connected to each other via a data bus 30. Has been.

  In general, a normal tissue and a diseased tissue often have different shapes including the state of blood vessel running in an endoscopic image. For this reason, when the surgeon diagnoses whether the tissue is a normal tissue or a diseased lesion tissue from the endoscopic image on the surface of the living mucosa, the shape including the state of the blood vessel running in the endoscopic image is used. In some cases, it may be used as an indicator for diagnosis or as a base. Therefore, if the image processing is performed so that the shape including the state of the blood vessel running in the endoscopic image can be extracted and displayed so as to be more easily visible, diagnosis by the operator can be effectively supported.

  In this embodiment, an extraction object having a complicated shape (desired to be extracted) like a blood vessel in the mucosal surface structure can be accurately extracted by image processing with a small amount of calculation, and can be displayed easily. The main purpose is to support.

In the following description, a case where a blood vessel is extracted as an extraction target from an endoscopic image as a medical image will be described, but the present invention may be applied to a mucosal surface structure of a living body other than a blood vessel.
In the present embodiment, for example, as illustrated in FIG. 2, the CPU 22 includes a preprocessing unit 31 that performs preprocessing such as noise suppression and inverse gamma correction on an image signal as a medical image input via the image input unit 21. It has a function.

  As will be described later, the pre-processing unit 31 includes a noise suppression unit 31a that performs processing for suppressing noise, an inverse gamma correction unit 31b that performs reverse gamma correction, and a ratio of two color signals in order to reduce the influence of light quantity. A color signal ratio calculation unit 31c that calculates an image signal is included.

  Further, the CPU 22 extracts feature amounts of the target shape to be extracted from the preprocessed image signal using a plurality of one-dimensional filters as a plurality of filters stored in the filter storage unit 39. The filtering processing unit (or filtering operation unit) 32 functions as a filtering unit that performs an operation to perform the operation.

  Further, the CPU 22 outputs (generates) a filter that reacts most to the pixel among a plurality of filter outputs of the pixel by the filtering processing unit 32, specifically, the maximum value (in the broad sense, the absolute value is the maximum value). The function of the feature quantity selection unit 33 (as a feature quantity selection unit) is selected (or extracted) as a feature quantity of the target shape to be extracted, and a target pixel corresponding to the one-dimensional filter is obtained. Have.

  The feature quantity selection unit 33 includes a maximum value search unit 33 a that searches for the maximum value from the output value of the filtering processing unit 32. Then, the target pixel corresponding to the one-dimensional filter that outputs the maximum value as described above is extracted as a feature amount of the target shape, that is, a blood vessel feature amount having a blood vessel feature amount.

  Further, the maximum value search unit 33a includes a width estimation unit 33b that estimates the width (or width candidate) of the blood vessel at the target pixel from information on the width and direction of the one-dimensional filter that outputs the maximum value, and the target pixel. It has the function of the traveling direction estimation part 33c which estimates the traveling direction (or traveling direction candidate) as the direction which is running on the surface of the blood vessel. In addition, the CPU 22 is a determination unit as a determination unit that determines whether or not the feature amount extracted by the feature amount extraction unit 33 matches a determination criterion for determining whether or not the feature amount includes a feature of the target shape. 34 functions.

  The determination unit 34 makes a determination using a threshold prepared as a determination criterion for determining whether or not the extracted blood vessel feature amount includes a blood vessel feature. In addition, as illustrated by a dotted line in order to explain in another embodiment, the CPU 22 determines a blood vessel pixel (or a blood vessel candidate pixel when there are a plurality of blood vessel pixels), a non-blood vessel pixel (or a blood vessel pixel). It has a function of a correction unit 35 for correcting non-blood vessel pixel candidate pixels when correcting non-blood vessel pixels.

  The correction unit 35 includes a width correction unit 35a that corrects the estimated width and a travel direction correction unit 35b that corrects the estimated travel direction. In addition, as will be described later, the CPU 22 includes a statistic calculation unit 36 that calculates a statistic by a plurality of pixels in a local region including the target pixel, and a correction unit 36a that corrects the feature using the calculated statistic.

  Note that the preprocessing unit 31, the filtering processing unit 32, the feature amount selection unit 33, the determination unit 34, the correction unit 35, and the statistic calculation unit 36 are formed by hardware without using the CPU 22, or a part thereof. You may do it. Further, the present invention is not limited to the separated configuration as shown in FIG. 1, and one processing function may be a combination including other functions.

  The image processing apparatus 3 according to the present embodiment has an image input unit 21 serving as an image input unit that inputs a medical image captured by a medical device, and a local region including a plurality of pixels including the target pixel from the medical image. Set according to the mucosal surface structure to be extracted and prepared for detecting a change in density with respect to the cross-section of the mucosal surface structure, and a plurality of different cross-sectional sizes and the surface of the mucosal surface structure running A plurality of filters corresponding to a plurality of different directions prepared for detecting a direction in which the target area is detected, and the mucosal surface structure for the local region including the pixel of interest using the plurality of filters. The filtering processing unit 32 as a filtering unit that performs a calculation for extracting a feature amount corresponding to the A feature quantity selection unit 33 as feature quantity selection means for selecting a feature quantity of the mucosal surface structure from a filter that is most responsive to the pixel among a plurality of filter outputs in the image, and the feature for the selected feature quantity And a determination unit 34 as a determination unit for determining whether or not the amount includes a feature of the mucosal surface structure, and the plurality of filters, where N is an integer of 1 or more and M is an integer of 2 or more, N × M prepared by arranging N filters corresponding to the size of density change with respect to the cross section in M directions prepared corresponding to directions orthogonal to the direction of density change with respect to the cross section. It consists of pieces.

  Next, processing contents by the image processing apparatus 3 of the present embodiment will be described with reference to FIG. FIG. 4 shows an example of an endoscopic image of the biological mucosal surface structure captured by the endoscope apparatus 2 and displayed on the monitor 9. The blood vessel in the endoscopic image has a density value lower than that of the background mucosa around the blood vessel because hemoglobin absorbs the G component.

  For example, the blood vessel which is a portion having a low density value (dark portion) in FIG. 4 travels in various directions on the image, and has a cross-sectional structure in a direction perpendicular to the traveling direction as the surface traveling direction. Is considered to have a downwardly convex shape (this will be referred to as a valley structure). The valley structure as a feature amount of a blood vessel having such a shape has various width sizes (the number of pixels forming a valley).

  Such an endoscopic image signal is read into the image processing apparatus 3 by the image input unit 21 in step S1 of FIG. 3 and stored in the image storage unit 24.

  In the next step S2, the preprocessing unit 31 of the CPU 22 reads the image signal temporarily stored in the image storage unit 24 and performs preprocessing on the image signal.

   As this preprocessing, the noise suppression unit 31a performs noise suppression on the image signal using, for example, a 3 × 3 median filter, and the inverse gamma correction unit 31b further performs inverse gamma on the image signal after the noise suppression processing. Make corrections.

   Further, in order to make the image signal subjected to inverse gamma correction less susceptible to the influence of the light amount or the like, the color signal ratio calculation unit 31c includes, for example, pixel signals of G signal and R signal in the three color signals in each pixel. A process of calculating a ratio value between the values g and r is performed. A G / R image composed of the calculated color signal value ratio (also referred to as g / r value) is used as an input image for the next step S3.

   In step S <b> 3, the filtering processing unit 32 performs a filtering process for extracting blood vessel feature amounts using a plurality of one-dimensional filters as a plurality of filters stored in the filter storage unit 39 on the input image.

   By this filtering process, a plurality of filtered processed values equal to the number of one-dimensional filters (however, the same one-dimensional filter is prepared in different directions as described below) are generated as output values. Is done.

   In the next step S4, the feature value selection unit 33 sets the processing value that is the maximum among the plurality of output values generated for each pixel as the blood vessel feature value having the blood vessel feature value. That is, a blood vessel feature amount is selected or extracted from a filter having the maximum output value among a plurality of filter processing results in a certain pixel.

  In the next step S5, the determination unit 34 determines whether or not the pixel having the extracted blood vessel feature amount is a blood vessel pixel that actually has the blood vessel feature using, for example, a threshold value as a determination criterion. A process of determining whether or not a pixel having a blood vessel feature amount is a blood vessel pixel having a blood vessel feature is performed. By the determination process in step S5, a blood vessel pixel having a blood vessel characteristic amount is determined as shown in step S6, and a blood vessel region including the blood vessel pixel is determined.

  Information on the processing result determined by the determination unit 34 in the next step S7 is recorded, and a processing image of the processing result (processing image in which the blood vessel region is determined) is displayed on the monitor 4 through the display processing unit 28, and FIG. 3 image processing ends.

  Next, a filtering process for extracting the blood vessel feature amount in FIG. 3 will be described in detail. FIG. 5 shows a model of a mucosal region 52 including a target-shaped blood vessel 51 to be extracted in an endoscopic image.

  A pixel of interest P for extracting the blood vessel 51 is set in the mucous membrane region 52. FIG. 5 shows a state where the target pixel P is set in the blood vessel 51 portion. Actually, when the processing target region as a two-dimensional target region to which the operator pays attention is designated in the endoscopic image, the target pixel P scans (scans) the entire processing target pixel in the processing target region. Thus, the target pixel P is set. For example, when the surgeon sets the mucosal region 52 shown in FIG. 5 as the processing target region, the target pixel P is set to sequentially scan in the horizontal direction from the upper left pixel Pa, and the lower right last pixel It is set up to Pz.

  For example, as shown in FIG. 6, the cross section of the blood vessel 51 has a downwardly convex shape, that is, a valley structure, and there are different valley widths, that is, different sizes of the cross section.

  For this reason, in the present embodiment, in order to extract the blood vessel 51 having a shape corresponding to the concentration change with respect to the cross section of the surface structure shown in FIG. 6, for example, the shape shown in FIG. A plurality of one-dimensional shape filters are prepared. The horizontal axis in FIG. 7 is a direction set corresponding to the direction of the cross section of the blood vessel 51, and the vertical axis represents the value of the one-dimensional shape filter, that is, the filter coefficient value.

  Also, in order to be able to cope with the case where the traveling direction of the blood vessel 51 is different, a plurality of directions corresponding to the plurality of traveling directions (traveling direction candidates) are prepared. It should be noted that the embodiments described herein can be applied to epithelial structures such as pit patterns other than blood vessels. And not only when the cross section is convex downward, but if it is a cross section with a characteristic shape to be extracted, such as convex upward, by preparing a filter according to the cross section, The same applies to cases other than blood vessels.

Therefore, a plurality of one-dimensional filters in the present embodiment are prepared by arranging a plurality of one-dimensional shape filters corresponding to the width W of the blood vessel 51 to be extracted in the image in a plurality of directions. The dotted lines in FIG. 5 indicate the four directions d1, d2, d3, and d4 that form the horizontal direction, 45 ° with the horizontal direction, 90 ° with the horizontal direction, and 135 ° with the horizontal direction as a plurality of directions.
FIG. 8 shows a filtering process for extracting the target shape in step S3 of FIG. When a processing target region as a target region of interest is set in the input image, the filtering processing unit 32 has a feature that the target pixel P has a desired target shape with respect to the target pixel P in the processing target region. A filtering process is performed to determine whether the number of pixels.

  In order to represent the target pixel P in the processing target region, the target pixel Pk (0 ≦ k ≦ number of processing target pixels−1) is set using the parameter k.

  Further, an XY coordinate expression as a two-dimensional coordinate orthogonal to the parameter k is (i, j). Here, 0 ≦ i <width, 0 ≦ j <height (where width and height indicate the number of pixels (number of pixels) of the image in the horizontal direction and the vertical direction of the processing target region. Also, the pixel of interest P (i, The observed value of j) is assumed to be p (i, j), and the pixel of interest P (i, j) and Pk, pk simply representing the observed value p (i, j) are also used. The observed value p (i, j) is the g / r value at the target pixel P (i, j).

    FIG. 8 shows an image obtained by performing the preprocessing in step S2 as an input, and a plurality of one-dimensional filters created in advance corresponding to a plurality of widths W and a plurality of directions D are expressed as N × M filter coefficients fdw (d1 ≦ d ≦ dN, w1 ≦ w ≦ wM) are applied.

  Then, as shown in Equation (3) described later, a product-sum operation is performed on the observed value p (i, j) and the filter coefficient fdw.

  When performing this product-sum operation, assuming that the number of coefficient elements (coefficient length) of the filter coefficient fdw is FN (= 2K + 1) and the j-th element is expressed as fdw (j), the filter coefficient fdw is expressed as follows. be able to.

fdw = {fdw (-K), fdw (-(K-1)), ..., fdw (0), fdw (1), ..., fdw (K-1), fdw (K)} (1)
The coefficient element number FN of the filter coefficient fdw is appropriately set according to the filter shape (particularly the width size) prepared (set), as shown in a specific example described later.

  In addition, a plurality of directions in the filter coefficient fdw are set corresponding to an angle when the horizontal direction of the processing target region is 0 ° (in this embodiment, 0 ° to 360 ° counterclockwise). The

  When performing the above product-sum operation, the pixel of interest P is set at the center position of the coefficient element number FN in the above equation (1) in the filter coefficient fdw, and the pixel of interest along the one-dimensional array direction of the filter coefficient fdw A plurality of pixels are set as a local region including (product-sum operation).

  Further, in the present embodiment, a plurality of filter coefficients fdw (fdw1, fdw2,..., FdwM) having different width sizes corresponding to the cross-sectional shapes of the plurality of blood vessels 51 having different widths are prepared, and product-sum operation is performed. When performed, the filter coefficient value of the filter coefficient fdw is set so that the output value as the calculated value using the filter coefficient fdw corresponding to the actual size of the blood vessel width is maximized.

  For example, the one-dimensional filter (filter coefficient) shown in FIG. 7 has a size corresponding to the actual width of the valley of the blood vessel 51 and is negative at the center and positive at both sides of the center. And the filtering process part 32 performs a filter coefficient and a product-sum operation using the observed value pk which imaged the blood vessel 51. FIG.

  That is, when the product-sum operation is performed when the relationship between the observed value pk and the one-dimensional filter is a relationship corresponding to the width and a relationship corresponding to the direction (matched relationship), The product of the observed value p and the negative value becomes a positive value in almost all products, and the maximum output value is obtained as a large operation value by the sum of those products.

  On the other hand, when the product-sum operation is performed with the observation value pk and the one-dimensional filter that do not have the corresponding relationship, for example, the product in the portion in the valley in FIG. 7 includes positive and negative, or the product in the outside of the valley is positive and negative. In other words, the product sum of them does not result in an output value that is a large operation value.

  In the following description of the present embodiment, the number of types in a plurality of directions N = 4 and the number of types of width M = 3 will be described. For example, for a blood vessel 51 having a width from a maximum value to a minimum value, the number of types M is set to three types: a maximum value, a median value between the maximum value and the minimum value, and a minimum value.

  The value is not limited to the three types, and the value can be changed, for example, by increasing the value of M according to the mucosal region to be detected. The number N of types in a plurality of directions is not limited to four types, and may be set to five or more types.

  As described above, when the number of types of width M is 3, for example, the filter coefficient fdw1 is set to a width (defined as an adaptive width) that can best extract the blood vessel 51 having a width of 5 pixels. Suppose that

  Similarly, the adaptive width of the filter coefficient fdw2 is 7 pixels, and the adaptive width of the filter coefficient fdw3 is 9 pixels.

In a specific example, the filter coefficient when the direction is d is
fdw1 = {0.18238, 0.32356, 0.1682, -0.3481, -0.652, -0.3481, 0.1682, 0.32356, 0.18238},
fdw2 = {0.19347, 0.28177, 0.24509, -0.0356, -0.4009, -0.5676, -0.4009, -0.0356, 0.24509, 0.28177,. 19347},
fdw3 = {0.16261, 0.18215, 0.2109, 0.20337, 0.08723, -0.1554, -0.4214, -0.5389, -0.4214, -0.1554, 0. 08723, 0.20337, 0.2109, 0.18215, 0.16261} (2)
Is set.

  Note that the sum of the coefficient values (coefficient element values) for the filter coefficients fdw1, fdw2, and fdw3 is set to be 0. The contours of the shapes of these filter coefficients fdw1, fdw2, and fdw3 are as shown in FIG. Note that the contour of FIG. 9 is set corresponding to the size of the density change with respect to the cross section of the blood vessel.

  The directions in which the filter coefficients fd1w, fd2w, fd3w, and fd4w are set are respectively set to 0 °, 45 °, 90 °, and 135 ° (corresponding to d1, d2, d3, and d4 in FIG. 5), respectively. It will be described as a thing. In the lower part of FIG. 9, one of the one-dimensional arrangement directions (0 °) forming a local region including a plurality of pixels including the filter coefficients fdw1, fdw2, and fdw3 and the target pixel Pk on which the product-sum operation is performed. In the case of

  Since the filter coefficients are set as described above, N × M pieces of the observed value pk of the target pixel Pk are obtained as shown in step S32 by the N × M pieces of filtering processing shown in step S31 of FIG. The output value of the product-sum operation value is obtained.

  As described above, when the number of coefficient elements of the filter coefficient fdw is FN, the filtered output value FOUTwd can be expressed by the following equation (3).

FOUTwd1 (i, j) = Σkp (i + k, j) fd1w (k)
FOUTwd2 (i, j) = Σkp (i + k, j) fd2w (k)
FOUTwd3 (i, j) = Σkp (i + k, j) fd3w (k) (3)
FOUTwd4i, j) = Σkp (i + k, j) fd4w (k)
In Equation (3), Σk is a product p (i + k, j) of the observed value p (i + k, j) and the filter coefficient fdw (k) from k = −FN / 2 to k = FN / 2 including 0. Indicates that fdw (k) is added (added).

  In the expression (3), in the 45 ° and 90 ° direction filtering processing (product-sum operation), the distance between the pixels is √2 (representing the square root of 2), and the filtering processing in the 0 ° or 90 ° direction is performed. This is different from the case where the distance between pixels is 1.

  For this reason, in this embodiment, in order to be able to compare the output in any direction under the same conditions, a pixel with a distance between the pixels of 1 is obtained using the Bicubic interpolation method, and the observation values in the 45 ° and 90 ° directions are obtained. . However, the observation value may be set by interpolation using another interpolation method, or a filter coefficient corresponding to the observation value having a distance between pixels of √2 may be separately prepared.

  With the above processing, N × M output values Vkdw of the filtering processing are obtained for the observation value pk of each pixel of interest Pk.

  The processing shown in FIG. 10 is performed on the output value Vkdw obtained in this way as the feature amount selection (extraction) processing in step S4 in FIG.

  As described above, when the target pixel Pk in the filtering process is a pixel constituting the blood vessel 51, the output of the filter coefficient fDW that most closely matches the width of the blood vessel and the traveling direction at the pixel position is N × by a plurality of filter coefficients. It becomes the maximum among the calculation results of the M filtering processes.

  Using this characteristic, in step S41 of FIG. 10, the maximum value search unit 33a searches for the filter coefficient fDW that is the maximum value from the N × M output values Vkdw.

  Since the filter coefficient fDW having the maximum output value is the filter most fitted to the target pixel Pk, the maximum value search unit 33a determines the filter coefficient fDW having the maximum output value Vkdw subjected to the filtering process in step S41. The searched filter coefficient fDW and the output value VkDW at the target pixel Pk used for the calculation of the filtering process are set as the output value of the target pixel k.

  Further, the adaptive width of the filter coefficient fDW and the set direction become information on the width and running direction of the blood vessel in the target pixel Pk. This information is also referred to as information on the width and running direction of the blood vessel of the pixel of interest Pk.

  In the next step S42, the determination unit 34 compares the predetermined threshold T set (prepared) in advance to determine whether or not it is a blood vessel with the output value VkDW searched in the previous step S41. It is determined whether or not the target pixel Pk (represented by the observation value pk or p (i, j)) is a blood vessel pixel, that is, a blood vessel pixel. In the present embodiment, the predetermined threshold T is set to T = 0.003.

  In the determination process of step S42, if VkDW> T, the determination unit 34 proceeds to step S43, and determines that the pixel of interest Pk (the pixel of the observation value pk) is a blood vessel pixel in a blood vessel region that forms a blood vessel.

  On the other hand, in the determination process of step S42, the determination unit 34 proceeds to step S44 if VkDW ≦ T, and the pixel of interest Pk (the pixel of the observation value pk) is not a pixel of the blood vessel region constituting the blood vessel (that is, a non-blood vessel). Pixel). Note that the above-described processing is repeated until there is no pixel of interest Pk in the processing target region.

  By performing the above-described processing on the processing target region, whether or not the pixel of interest Pk is a pixel of the blood vessel region constituting the blood vessel is determined as shown in step S6.

Through the above processing, blood vessels having various valley structures can be extracted and a blood vessel region can be determined. Further, when the processing for all the target pixels Pk in the processing target region is completed, for example, the processing information storage unit 25 stores information of the determination result in step S45. Note that the determination result information may be stored in the hard disk 27.
In step S46, the display processing unit 28 outputs a processed image reflecting the information of the determination result in step S45 to the monitor 4, and the monitor 4 displays the processed image. When displaying a processed image on the monitor 4, for example, a pixel signal value determined as a blood vessel pixel is emphasized with a luminance value or a specific color so as to be easily distinguished from a pixel signal value determined as a non-blood vessel posterior element. To do.
Further, the monitor 4 can be easily compared with the input image input to the image processing apparatus 3 and the processed image determined to be a blood vessel or a non-blood vessel by the processing of FIG. 3 or 10 as shown in FIG. To display. Then, the processing in FIG. 10 or FIG. 3 is terminated.

  The medical image processing method according to the present embodiment includes a step S1 for reading the input image of FIG. 3 as an image input step for inputting a medical image captured by a medical device, and a plurality of pixels including a target pixel from the medical image. For the local region to be set according to the mucosal surface structure of the extraction target, the concentration change size for a plurality of different cross-sections prepared for detecting the concentration change with respect to the cross-section of the mucosal surface structure, Using the plurality of filters corresponding to the plurality of different directions prepared for detecting the surface running direction of the mucosal surface structure, the local region including the pixel of interest with respect to the local region Step S3 as a filtering step for performing an operation for extracting a characteristic amount corresponding to the mucosal surface structure, and the filtering step The step S4 as a feature selection step for selecting the feature quantity of the mucosal surface structure from the filter that most reacted to the pixel among a plurality of filter outputs in a certain pixel, and the selected feature quantity. Step S5 as a determination step for determining whether or not the feature amount includes a feature of the mucosal surface structure, and the plurality of filters, wherein N is 1 or more and M is an integer of 2 or more, N × M prepared by arranging N filters corresponding to the size of density change with respect to the cross section in M directions prepared corresponding to directions orthogonal to the direction of density change with respect to the cross section. It consists of pieces.

  According to this embodiment, the surgeon refers to the input image displayed on the monitor 4 and the processed image, so that the image is compared with a case where the blood vessel portion is identified from only the input image and diagnosis is performed. By referring to the processed image determined as a blood vessel and a non-blood vessel by the processing, it is possible to grasp the blood vessel in a state where it is easier to identify the blood vessel. For this reason, the diagnosis based on the structure of the blood vessel part can be performed in a shorter time and smoothly.

  In addition, according to the present embodiment, since image processing for extracting feature amounts is performed using a plurality of one-dimensional filters, bending or meandering is performed by image processing that is simple and requires a small amount of calculation. Even when it has a complicated shape (structure) such as a blood vessel whose tube diameter changes, the shape (structure) can be extracted with high accuracy. Also, since image processing with a small amount of computation is sufficient, it takes a short time to display the processed image.

  In the above-described embodiment, it has been described that the blood vessel candidate extraction process includes a maximum value search process for the output value of the filtering process. However, the blood vessel candidate extraction process includes a filtering process and a maximum value search process. You may define that it is comprised by.

(Second Embodiment)
Next, a second embodiment of the present invention will be described. This embodiment has the same hardware configuration as the first embodiment (the same applies to third and fourth embodiments described later). In this embodiment, a function for extracting a portion where a blood vessel is branched is added to the first embodiment.

  For this reason, the filter storage unit 39 in the present embodiment, in addition to the information storage unit 39a of the one-dimensional filter corresponding to the one storing information of the plurality of one-dimensional filters in the first embodiment as shown in FIG. Further, a branch detection filter information storage unit 39b for storing information of a branch detection filter configured by a one-dimensional filter for detecting (extracting) a branch portion is provided. Hereinafter, the plurality of one-dimensional filters in the first embodiment are referred to as valley structure filters having a shape corresponding to the valley structure of the cross section of the blood vessel.

  Note that the wide valley structure filter shown in parentheses in FIG. 12 is a modified one-dimensional filter that can be used instead of the branch detection filter.

  In the first embodiment, the valley structure filter having a shape corresponding to the valley structure of the cross section of the blood vessel when the branch is not taken into consideration is used. However, in this embodiment, the branched portion is detected (extracted). Further, a branch detection filter is used.

  As the branch detection filter, for example, n × M filter coefficients fdb in which n one-dimensional branch shape filters are arranged in M traveling directions are added to the above-described filter coefficient fdw.

  And this embodiment performs the process for extracting the blood vessel in a process sequence as shown in FIG.

  Steps S1 and S2 in FIG. 13 are the same processing as in FIG. In the filtering process in step S3 ′ in this embodiment, a branch detection filter is used in addition to the filtering process using the plurality of one-dimensional filters (valley structure filters) described in the first embodiment. But do the same.

  Further supplementarily, in the first embodiment, the filtering process is performed using N × M filter coefficients fdw in step S31 shown in FIG. 8 as the filtering process. Filtering processing is performed using × M filter coefficients fdb. That is, (n + N) × M filtering processing is performed. When N in FIG. 8 is replaced with (n + N), the same processing as in FIG. 8 is performed.

  Also in the next step S4 ', in addition to the processing in step S4 described above, a blood vessel feature amount is selected (extracted) from the output value Vkdb when the branch detection filter is used.

  In this way, in addition to the valley structure filter, a branch detection filter is prepared, and a pixel whose output value filtered by the branch detection filter is larger than, for example, a second predetermined value is a blood vessel pixel. By adding and extracting to the result of filtering processing by the filter, correction can be made so as to extract blood vessel pixels that could not be extracted by the valley structure filter due to branching or the like.

  In step S5 ′, the blood vessel feature amount, that is, whether it is a blood vessel pixel or a non-blood vessel pixel is determined based on whether or not the blood vessel feature amount extracted in this manner meets the determination criterion.

  In the determination process in step S5 ′, as described in the first embodiment, whether the target pixel is a blood vessel pixel or a non-blood vessel pixel depending on whether the output value by the valley structure filter exceeds the threshold value or not. Was judged.

  On the other hand, in the case of the branch detection filter, a criterion for determining whether the target pixel is a blood vessel pixel or a non-blood vessel pixel is used depending on whether the output value exceeds a threshold value in a plurality of directions. For example, if the output result of the branch detection filter is an output value that exceeds a preset threshold value Tw (for example, Tw = 0.003) in a plurality of directions, that pixel is used as the blood vessel pixel, and the blood vessel pixel by the valley structure filter is selected. Make additional corrections. In this way, the blood vessel region is determined as shown in step S6. The determination criterion is not limited to the threshold value Tw, and may be set to a different threshold value, for example, a second threshold value different from the threshold value Tw.

  Then, the process proceeds to step S7 in FIG. 3 to record and display the processing result.

  According to the present embodiment described above, the same effect as that of the first embodiment can be obtained, and the portion where a blood vessel that is difficult to extract in the first embodiment can be accurately extracted. . Since this embodiment also employs a filtering process using a plurality of one-dimensional filters, a small amount of calculation is sufficient.

  As a modification of the present embodiment, instead of the branch detection filter, a wide valley structure filter (or wide valley shape filter) for extracting (detecting) a valley structure having a wider width than that of the valley structure filter. May be used. Further, the present invention can be applied to other than a valley structure that protrudes downward (for example, a structure that protrudes upward).

  The processing when the wide valley structure filter is used instead of the branch detection filter is the same as that when the above-described branch detection filter is used.

  For example, as the filtering process in step S31 of FIG. 8, in addition to the valley structure filter, the above wide valley structure filter for extracting a valley structure wider than the width of the blood vessel assumed to exist in the image is used. Perform filtering processing.

  Then, by using the output value of the wide valley structure filter in addition to the output value of the valley structure filter in the first embodiment, the selection result of the blood vessel feature amount in the first embodiment can be corrected.

  For example, if the output result of the wide valley structure filter is an output that is equal to or greater than a preset threshold value Tw (eg, Tw = 0.003) in a plurality of directions, that pixel is added as a blood vessel pixel.

  In addition, when the output value of the valley concept filter is the maximum and the detection result of the wide valley structure detection filter indicates the same direction, the pixel is not extracted as a blood vessel pixel from the valley concept filter. In the case of a pixel, it may be added to a blood vessel pixel.

  In addition, prepare a wide valley structure filter wider than the width of the cross-sectional width in the valley structure filter or a filter having a shape that matches the shape of the branched part, and select both or one of them as the above branch detection filter You may make it usable. By applying both, the branched portion can be extracted more reliably. In addition to the branched portion, a meandering detection filter having a meandering shape for detecting the meandering portion may be prepared.

  Further, as a branch detection filter or a meandering detection filter (that is, a branching / meandering detection filter), for example, besides a one-dimensional filter, the one-dimensional filter has a plurality of coefficient elements in one line. If it is arranged, a two-dimensional filter close to a one-dimensional filter within two or three rows may be used. This two-dimensional filter is called a filter according to the one-dimensional filter.

  For example, as a branching / meandering detection filter, for example, when the direction of a one-dimensional filter is d, the filter coefficients are {0, 0.1, −0.2, −0.6, −0.2, 0. 1,0}, the filter coefficients of the filter according to the one-dimensional filter are {0, 0.05, −0.1, −0.3, −0.1, −0.05, consisting of three rows. 0}, {0, 0.1, -0.2, -0.6, -0.2, 0.1, 0}, {0, 0.05, -0.1, -0.3,- It is also possible to prepare an extension to 0.1, -0.05, 0} so that the branched or meandering part can be extracted more reliably.

  Also, prepare a one-dimensional filter and a two-dimensional filter similar to the one-dimensional filter so that the operator can select one when actually using it as a branch / meander detection filter. Also good.

  In addition, although the case where the thing of the two-dimensional filter as a filter according to a one-dimensional filter is actually used as a branching / meandering detection filter has been described, it is limited to the case of detecting a branching or meandering part Instead, a two-dimensional filter as a filter according to the one-dimensional filter may be applied to the mucosal surface structure to be extracted.

  In the maximum value search process of step S41 shown in FIG. 10, the g / r value and the valley structure may be transformed into a feature quantity and used as an auxiliary filter output value. For example, for a pixel having a small filter output value, a pixel having a g / r value equal to or smaller than a preset threshold Tgr (eg, Tgr = 0.01), or a pixel having a valley structure by comparison with the vicinity. May be added as a blood vessel pixel.

(Third embodiment)
Next, a third embodiment of the present invention will be described. In the present embodiment, the determination result of the feature amount with respect to the target pixel is corrected using the information of the surrounding pixels. As described in the above-described embodiment, the width and direction of the filter coefficient fDW that maximizes the filtered output value are determined as the blood vessel width and travel direction information of the pixel of interest.

  Thereby, it is possible to determine the blood vessel width and the traveling direction of the target pixel with a small amount of calculation. However, if noise that could not be removed is superimposed on the pixel signal value (g / r value) of the target pixel in the input image, a determination result may be obtained in which the blood vessel width or the traveling direction changes rapidly in the target pixel.

  The width and running direction of adjacent blood vessel pixels on a single blood vessel are, for example, a width 9 only at one point where the width was 5 continuously, or a blood vessel that was running in the direction of d4. The width and direction are corrected on the assumption that there is no sudden fluctuation such as that the direction becomes d2 only at a certain point.

  In order to perform this correction, in the present embodiment, the determination result of the feature amount is corrected using the result around the target pixel. In order to perform this correction, the CPU 22 in the present embodiment performs the above-described correction using the correction unit 35 illustrated in FIG. 2 in the configuration of the first embodiment, for example.

  The processing content in this embodiment is as shown in FIG.

  Steps S1-S3 in FIG. 14 are the same as the processes described in FIG. Further, step S41 and step S42 are the same as the processing for performing the comparison using the threshold value described in FIG. The processing for the comparison result in step S42 is slightly different from steps S43 and S44 as shown in steps S43 'and S44'.

  In the first embodiment described above, depending on whether or not the maximum value of the filtered output value exceeds the threshold value, as shown in steps S43 and S44 in FIG. In this embodiment, as shown in steps S43 ′ and S44 ′ in FIG. 14, the target pixel is a blood vessel pixel before correction (also referred to as a blood vessel candidate pixel) based on the determination result. ) Or non-blood vessel pixels before correction (also referred to as non-blood vessel candidate pixels).

  Note that the threshold T in step S42 in FIG. 14 is set to a value different from the threshold T in the first embodiment, that is, a second determination criterion different from the above-described determination criterion. A comparison (for determining that the pixel is a candidate pixel) may be performed. In other words, step S42 in FIG. 14 may be performed using a second determination criterion different from the determination criterion of the first embodiment.

  In step S43 ′ and step S44 ′, a blood vessel candidate pixel and a non-blood vessel candidate pixel are determined. As shown in step S51, a blood vessel candidate pixel (a blood vessel pixel before correction) and a non-blood vessel candidate pixel (a non-blood vessel pixel before correction) are obtained. Extracted.

  In the next step S52, the CPU 22 (by the correction unit 35 shown in FIG. 2) performs a determination process for determining whether the blood vessel candidate pixel and the non-blood vessel candidate pixel are the blood vessel pixel and the non-blood vessel pixel. Do. Since the process of step S52 in this embodiment substantially corrects the result determined as a blood vessel pixel or a non-blood vessel pixel in the first embodiment, the determination process can also be said to be a correction process.

  After the candidate determination process in step S52, the feature amount is calculated (a blood vessel region is determined) as shown in step S6, and the process of step S7 is further performed. Steps S6 and S7 are the same processing as in FIG.

  Next, as the determination process in step S52, the traveling direction correction unit 35b constituting the correction unit 35 substantially performs the traveling direction correction process, for example, by the following process.

  First, a blood vessel candidate line is extracted by setting a blood vessel candidate region composed of blood vessel candidate pixels (blood vessel pixels before correction) to 1, and setting a region other than the blood vessel candidate pixels to 0, and performing thinning processing.

  Next, the target pixel Pk is selected from the blood vessel core line, information on the travel direction of the blood vessel of the target pixel Pk (observed value pk) is set as dK1, and information on the travel direction of the adjacent core line is set as dK2.

  When the angle formed by the traveling direction information dK1 and dK2 is 90 °, the direction of the filter coefficient having the second largest output value of the filtering process is performed on the blood vessel traveling direction information dK1 or dK2 of the target pixel Pk or the adjacent core line pixel. Correction processing is performed.

  The extraction of the blood vessel core line is not limited to the case of determining using the output value of the filtering process, but is determined by comparing the boundary when the region is divided using the extreme value convergence method with the extracted blood vessel candidate region. Also good.

  Through the above processing, misrecognition of the traveling direction on a single blood vessel can be corrected. Other effects are the same as those of the first embodiment.

  Note that, for example, dK2l and dK2r are dK2l and dK2r, where dK2l and dK2r are information on the traveling direction of the blood vessels of two adjacent pixels (in a broad sense, a plurality of neighboring pixels) that are adjacent to the target pixel Pk along the traveling direction of the core line of the target pixel Pk. Are satisfied, that is, when dK2l = dK2r (= dK2) is satisfied, if the angle formed by the information dK1 of the pixel of interest Pk and dK2 is 90 °, the information dK1 of the blood vessel traveling direction of the pixel of interest Pk is filtered. Correction processing may be performed in which the output value of the processing is corrected as the direction of the second largest filter coefficient.

  In this way, when the information on the traveling direction of the blood vessel in the pixel of interest differs from the information on the traveling direction of the blood vessel in a plurality of neighboring pixels adjacent to the blood vessel in the core line direction by a predetermined threshold or more, the traveling of the blood vessel in the pixel of interest. You may make it correct | amend the information of a direction.

  The case where the traveling direction of the blood vessel is corrected has been described above, but the width correcting unit 35a constituting the correcting unit 35 in FIG. 2 corrects the width of the blood vessel as follows.

  For example, information on the width of a blood vessel of, for example, two adjacent pixels (in a broader sense, a plurality of neighboring pixels) along the traveling direction of the core line in the pixel of interest Pk is wK2l, wK2r, and the condition that wK2l, wK2r match. When wK2l = wK2r (= wK2) is satisfied, if the absolute value of the difference between the width information wK1 and wK2 of the pixel of interest Pk is equal to or greater than a preset threshold value, the blood vessel width information wK1 of the pixel of interest Pk is Correction processing for correcting the value of the adjacent pixel may be performed.

  As described above, when the blood vessel width information in the pixel of interest differs from the blood vessel width information in a plurality of neighboring pixels adjacent to the blood vessel in the center line direction by a predetermined threshold or more, the blood vessel width information in the pixel of interest May be corrected.

  Note that, in S43 ′ and S44 ′, the target pixel Pk is determined as a blood vessel candidate pixel and a non-blood vessel candidate pixel in S43 ′ and S44 ′ in accordance with the determination processing in step S42 of FIG. Instead of such determination, the blood vessel pixel and the non-blood vessel pixel are determined as in the first embodiment, and the determined blood vessel pixel and the non-blood vessel pixel (however, the blood vessel pixel before correction and the non-blood vessel pixel before correction) are determined. On the other hand, corrected blood vessel pixels and non-blood vessel pixels before correction may be determined in step S52 to determine corrected blood vessel pixels and non-blood vessel pixels (corresponding to step S6 in FIG. 14). .

(Fourth embodiment)
Next, a fourth embodiment of the present invention will be described. In the present embodiment, the blood vessel pixels and the non-blood vessel pixels determined by the feature amount determination are corrected using information on a plurality of pixels in the vicinity of the target pixel. Since the entire processing content of the present embodiment is the same as that of FIG. 14 described in the third embodiment, the entire processing content of the present embodiment is not illustrated. However, in the present embodiment, the processing content of the determination in step S52 in FIG. 14 is different from that in the third embodiment.

  In the present embodiment, the blood vessel and non-blood vessel candidate pixels are determined by the processing in steps S43 ′ and S44 ′ in FIG. 14 in the same manner as the third embodiment described above, as shown in step S51. Extract non-blood vessel candidate pixels. In other words, a blood vessel candidate region made up of blood vessel candidate pixels and a non-blood vessel candidate region made up of non-blood vessel candidate pixels are extracted.

  In the present embodiment, in the candidate determination process in step S52, a neighboring region including the target pixel Pk of the blood vessel candidate region (the blood vessel region before correction) and the non-blood vessel candidate region (the non-blood vessel region before correction) in the center. As shown in FIGS. 15 and 16, the number of pixels of blood vessel candidates and non-blood vessel candidates is counted (counted) for each pixel in the (local region), and the target pixel Pk ( Processing for determining whether the observation value pk) is a blood vessel or a non-blood vessel is performed.

FIG. 15 and FIG. 16 show processing for counting (counting) the number of pixels of blood vessel candidates and non-blood vessel candidates in the vicinity region (the specific processing content is FIG. 17). The outline of the processing according to FIG. 15 will be described.
(I) When the information on the traveling direction of the blood vessel is known,
(A) When the neighboring pixel is a blood vessel candidate pixel, if the pixel of interest is located in the running direction of the neighboring pixel and the information on the running direction of the pixel of interest matches the running direction information of the neighboring pixel, the neighborhood Pixels are counted as blood vessel candidate pixels.

  (B) When the neighboring pixel is a non-blood vessel candidate pixel, if the pixel of interest is located in the running direction of the neighboring pixel and the information on the running direction of the pixel of interest does not match the running direction of the neighboring pixel, the neighboring pixel Are counted as non-blood vessel candidate pixels.

Also, the outline of the processing according to FIG.
(Ii) when the width information of the blood vessel is known,
(C) When the neighboring pixel is a blood vessel candidate pixel, if the difference between the width of the pixel of interest and the width of the neighboring pixel is less than the threshold T2, the neighboring pixel is counted as a blood vessel candidate pixel.

  (D) When the neighboring pixel is a non-blood vessel candidate pixel, if the difference between the width of the pixel of interest and the width of the neighboring pixel is equal to or greater than the threshold T2, the neighboring pixel is counted as a non-blood vessel candidate pixel.

Furthermore, when each of the above conditions (a) to (d) is satisfied with one or more pixels, the blood vessel candidate pixels / non-blood vessel candidates are given priority in the order of (a), (c), (b), (d). Pixels are counted, and blood vessel pixels and non-blood vessel pixels are corrected as shown in FIG.
Since (a) and (b) or (c) and (d) described above are in an exclusive relationship with each other, a combination such as adopting a pixel having a large number of pixels that satisfies a condition in the vicinity is adopted. It is also possible to use the output when the number of matching conditions is the largest.

  Next, the processing of FIG. 15 will be described in detail. In the first step S61, the CPU 22 determines whether or not the information on the traveling direction of the blood vessel is known in a neighboring region including the pixel of interest (for example, a 3 × 3 pixel size arranged around the pixel of interest). Judgment is made.

  This determination is performed when, for example, there are a plurality of pixels belonging to the blood vessel candidate region in the vicinity region (local region), and the information on the traveling direction of the blood vessel in the plurality of pixels falls within a variation within a preset threshold value. It is assumed that information on the traveling direction of a blood vessel in a neighboring pixel included in the neighboring region is known.

  On the other hand, if the variation does not fall within the threshold value, the CPU 22 determines that the information on the traveling direction of the blood vessel is not known, and ends the processing of FIG.

  When it is determined that the traveling direction of the blood vessel is known, in the next step S62, the CPU 22 determines whether or not the pixel near the target pixel is a blood vessel candidate pixel. If the determination result indicates that the neighboring pixel of the target pixel is a blood vessel candidate pixel, in the next step S63, the CPU 22 determines whether or not the target pixel is present in the running direction of the neighboring pixel.

  In the case of the determination result that the target pixel is present in the traveling direction of the neighboring pixel, in the next step S64, the CPU 22 determines whether or not the traveling direction of the neighboring pixel matches (is the same as) the traveling direction of the target pixel. Judgment is made.

  In the case of a determination result in which the traveling direction of the neighboring pixels matches the traveling direction of the target pixel, in the next step S65, the CPU 22 sets the count number of the blood vessel candidate pixels (abbreviated as the blood vessel candidate number in FIG. 15). Add one. That is, the neighboring pixels are counted as blood vessel candidate pixels. The initial value of the count number of the blood vessel candidate pixels is set to zero.

  On the other hand, in Steps S62, S63, and S64, the determination result that the neighboring pixel of the target pixel is not a blood vessel candidate pixel, the determination result that there is no target pixel in the traveling direction of the neighboring pixel, the traveling direction of the neighboring pixel and the traveling direction of the target pixel are In the case of a determination result that does not match, in step S66, the CPU 22 adds 1 to the count number of the non-blood vessel candidate pixels (abbreviated as the non-blood vessel candidate number in FIG. 15). That is, the neighboring pixels are counted as non-blood vessel candidate pixels. The initial value of the count number of the non-blood vessel candidate pixels is set to zero.

  After the processing of steps S65 and S66, in step S67, the CPU 22 determines whether or not all the pixels in the vicinity area of the target pixel have been searched. Steps S61 to S67 are repeated for the neighboring pixels.

  On the other hand, when the above-described processing of steps S61 to S67 is performed on all the pixels in the vicinity region, in step S68, the CPU 22 performs the correction processing and ends the processing of FIG. The processing in step S68 will be described later with reference to FIG.

  If the determination process in step S64 results in a determination result that the traveling direction of the neighboring pixel does not match the traveling direction of the target pixel, whether or not the traveling direction of the neighboring pixel is further orthogonal to the traveling direction of the target pixel. If it is determined and orthogonal, 1 may be added to the number of non-blood vessel candidate pixels.

  In FIG. 16, the following processing is performed. In the first step S71, the CPU 22 determines whether or not information on the width of the blood vessel in the neighboring pixels is known. This determination is made when, for example, there are a plurality of pixels belonging to the blood vessel candidate region in the vicinity region (local region), and the blood vessel width information in the plurality of pixels falls within a variation within a preset threshold value. The width of is known.

  On the other hand, if the variation does not fall within the threshold, the CPU 22 determines that the width of the blood vessel is not known, and ends the process of FIG.

  If it is determined that the blood vessel width is known, in the next step S72, the CPU 22 determines whether or not the neighboring pixel of the pixel of interest is a blood vessel candidate pixel. In the case of the determination result that the neighboring pixel of the target pixel is a blood vessel candidate pixel, in the next step S73, the CPU 22 calculates the blood vessel width (information) of the neighboring pixel and the blood vessel width (information) of the target pixel. It is determined whether or not the difference (abbreviated as the difference between the width of the neighboring pixel and the target pixel) is less than the threshold T2.

  In the case of the determination result that the difference between the width of the neighboring pixel and the target pixel is less than the threshold value T2, in the next step S74, the CPU 22 adds 1 to the count number of the blood vessel candidate pixels. That is, the neighboring pixels are counted as blood vessel candidate pixels. Note that the initial value of the count number of the blood vessel candidate pixels when the processing of FIG. 16 is started is set to zero.

  On the other hand, if it is determined in steps S72 and S73 that the neighboring pixel of the target pixel is not a blood vessel candidate pixel, and the determination result is that the difference between the width of the neighboring pixel and the target pixel is not less than the threshold T2, the CPU 22 in step S75. Adds 1 to the count of non-blood vessel candidate pixels. That is, the neighboring pixels are counted as non-blood vessel candidate pixels. Note that the initial value of the count number of non-blood vessel candidate pixels when the processing of FIG. 16 is started is set to zero.

  After the processing of steps S74 and S75, in step S76, the CPU 22 determines whether or not all pixels in the vicinity region of the target pixel have been searched. The processes in steps S71 to S76 are repeated for the neighboring pixels.

  On the other hand, when the processes of steps S71 to S76 described above are performed on all the pixels in the vicinity region, the CPU 22 performs the candidate correction process in step S77, and ends the process of FIG.

  Next, the correction processing in step S68 in FIG. 15 and step S77 in FIG. 16 will be described with reference to FIG.

  When the correction process starts, in the first step S81, the CPU 22 determines whether or not the number of blood vessel candidates in the vicinity region of the target pixel is equal to or greater than the number of non-blood vessel candidates. When the number of pixels of the blood vessel candidate is Nv and the number of pixels of the non-blood vessel candidate is Nnv, the CPU 22 determines whether or not the condition of Nv ≧ Nnv is satisfied. When the number of blood vessel candidate pixels is equal to or greater than the number of non-blood vessel candidate pixels, in the next step S82, the CPU 22 determines that the pixel of interest Pk is a blood vessel pixel.

  On the other hand, if the number of pixels of the blood vessel candidate does not satisfy the condition equal to or greater than the number of pixels of the non-blood vessel candidate, the CPU 22 determines that the pixel of interest Pk is a non-blood vessel pixel in step S83. Then, the process of FIG. 17 ends.

  When the number of pixels Nv and Nnv by the process of FIG. 15 and the number of pixels Nv and Nnv by the process of FIG. 16 are different, the above-described condition for determining whether or not Nv ≧ Nnv is satisfied sufficiently ( Specifically, one having a large difference in the magnitude relationship between the number of pixels is employed.

  In this embodiment, which is processed in this way, the pixel of interest is corrected using information on the number of blood vessel candidates and the number of non-blood vessel candidates that are the overall feature amounts in the vicinity region of the pixel of interest.

  Therefore, using a trough structure filter, it is possible to improve the discontinuity of the blood vessel when the blood vessel is extracted, to extract the contour of the blood vessel end, or in a line structure where the width changes abruptly. It is possible to suppress the extraction of a non-existing region. Other effects are the same as those of the first embodiment.

  Note that in this embodiment, correction is performed on blood vessel pixels and non-blood vessel pixels as described in the third embodiment, only when correction is performed on blood vessel candidate pixels and non-blood vessel candidate pixels. You may make it do.

  Further, the neighborhood area or the local area described above is not limited to a 3 × 3 rectangular pixel size arranged with the pixel of interest at the center, but may have different area sizes and shapes.

  In addition, when performing the process of step S52 of FIG. 14, it is not restricted to performing both FIG.16 and FIG.17, You may make it perform only one.

  In addition, when performing the determination process of FIGS. 16 and 17, the target pixel is selectively set so that the determination process is performed only for the target pixel of interest when performing the determination process. May be.

  In the above-described embodiment, for example, when determining a blood vessel and a non-blood vessel in step S5 in FIG. 3 or in steps S42 to S44 in FIG. 10, local in an input image used for such determination is used. There is a case where it is easily affected by a change in color tone and a function of extracting blood vessel feature values is lowered. Correction for suppressing such a decrease will be described below.

  For example, as shown in FIG. 2, a statistic calculating unit (more specifically, calculating an average value) as a statistic calculating unit that calculates a statistic regarding pixel signal values of a plurality of pixels in a local region including the target pixel Pk by the CPU 22. Part) 36 and a correction part 36a as correction means for correcting feature quantity candidates and the like using statistics.

  In the above-described embodiment, the g / r value in the G / R image is adopted. However, since the value in each pixel is adopted, it is easily affected by noise. Therefore, the influence of the color tone change can be reduced by performing correction using the correction value using the average value (in other words, the statistic) of the color tone in a plurality of pixels in the local region including the target pixel.

  For example, the statistic calculation unit 36 calculates a G / R average value (G / Rave) of a rectangular area of, for example, 51 × 51 pixels centered on an arbitrarily selected target pixel Pk, for example, the following equation (4) By substituting into the correction formula shown in FIG. 8, the correction unit 36a obtains a correction feature amount that reduces the influence of the color tone change, and corrects the feature amount.

Corrected feature value = feature value / (− 0.0271 · G / Rave2 + 0.02 · G / Rave−0.0008) (4)
However, the feature quantity to be corrected may be an output value of filtering processing or may be applied to the g / r value of the input image for calculating the feature quantity. In addition, each coefficient in Expression (4) may be changed according to the feature quantity to be corrected. Further, the rectangular area is not limited to 51 × 51 pixels, and may be a smaller rectangular area. In addition, as a method of correcting using a statistic, a correction table for correcting a determination criterion for determining an extraction pixel from a feature value using a statistic may be used.

  Further, the present invention is not limited to a local region such as a two-dimensional rectangular region, and may be a local region including a one-dimensional plurality of pixels.

  In addition to this, the threshold value for determining the blood vessel candidate / non-blood vessel candidate is corrected by a correction formula using the average value of the biological mucous membrane, or a threshold value correction table is prepared and stored in this threshold correction table. You may make it correct | amend using a threshold value.

  Further, in the above-described embodiments and the like, the case of the valley structure has been described, but the present invention can also be applied to the case of a mountain structure that is convex upward. In this case, a plurality of one-dimensional filters corresponding to the shape of the mountain structure may be prepared. Then, as described in the first embodiment, it is only necessary to search for the one having the maximum output value of the filtering process.

  Moreover, it can also be applied to the case of a mountain structure that is convex upward using a plurality of one-dimensional filters corresponding to the valley structure. In this case, it is only necessary to search for a filter whose absolute value of the output value of the filtering process is the maximum value. In addition, when searching or extracting only an upwardly convex structure, the output of the filtering process is a minimum value (the absolute value when the output is negative is the maximum value), or a value obtained by inverting the sign of the output Use it. Therefore, the maximum value or the absolute value may be set so that the plurality of filter outputs in the certain pixel described above are the best match.

  Furthermore, it is not limited to selecting a feature value from a filter when the absolute value of a plurality of filter outputs when the feature value selection unit 33 performs the filtering process is the maximum, but other selection criteria are adopted. You may do it. The reason is as follows.

  For example, if the value of filter A is the best match, the value is 1 and the value of filter B is 2, and the output of filter A actually calculated by the filtering process is 0.8. Consider the case of .2. When the maximum value selection criterion is applied, the filter B is selected because 0.8 <1.2. However, considering the match rate as a selection criterion, filter B is 1.2 / 2 = 0.6 <0.8 in a simple calculation, so that filter A matches the filter to be selected.

  For this reason, the feature value selection unit 33 compares the plurality of filter outputs calculated by the filtering means, and the feature value of the biological mucosa structure based on the set selection criterion (the comparison result is the most responsive or suitable for the selection criterion) May be selected.

  The mucosal surface structure to be extracted is not limited to blood vessels and epithelial structures, but the surface of the large intestine, pattern (surface structure), pit pattern, stomach, MCL, light blue crest, pit, interpit It can be widely applied to various cases such as parts.

  Note that embodiments configured by partially combining the above-described embodiments and the like also belong to the present invention.

  This application is filed on the basis of the priority claim of Japanese Patent Application No. 2011-099850 filed on April 27, 2011, and the above disclosure is cited in the present specification and claims. Shall be.

Claims (4)

An input unit for inputting medical image information composed of a plurality of pixels obtained by imaging a biological mucous membrane;
By applying a first filter including at least one of the traveling direction and width of the surface structure of the biological mucosa to the target pixel in the plurality of pixels of the medical image information, the first feature of the target pixel Information is acquired, and further, second characteristic information of the target pixel is acquired by applying a second filter including at least one of the traveling direction and width of the surface structure to the target pixel, and By applying the first filter to neighboring pixels located in the vicinity of the target pixel, the third feature information of the neighboring pixels is obtained, and further, by applying the second filter to the neighboring pixels, A filtering unit that acquires fourth feature information of neighboring pixels;
Based on the first feature information of the pixel of interest and the second feature information of the pixel of interest, a filter that has reacted more in the pixel of interest is selected from the first filter and the second filter, and the neighboring pixels A selection unit that selects a filter that has reacted more in the neighboring pixels out of the first filter and the second filter based on the third feature information of the first and fourth feature information of the neighboring pixels;
It is determined whether the target pixel is a surface structure candidate pixel or a non-mucosal surface structure candidate pixel based on feature information corresponding to the filter selected by the selection unit in the target pixel, and the selection is performed in the neighboring pixels. A determination unit that determines whether the neighboring pixel is a surface structure candidate pixel or a non-mucosal surface structure candidate pixel based on feature information corresponding to the filter selected by the unit;
Correction for correcting the determination result made on the target pixel based on a comparison result between at least one information on a neighboring pixel determined to be the surface structure candidate pixel by the determination unit and at least one information on the target pixel And
A medical image processing apparatus comprising: The correction unit is the surface structure candidate pixel in which a difference between at least one information in a neighboring pixel determined to be the surface structure candidate pixel and at least one information in the target pixel is less than a predetermined threshold value. The medical image processing apparatus according to claim 1, wherein the determination result made on the target pixel is corrected based on the determined number of neighboring pixels. The correction unit is determined to be the surface structure candidate pixel by the determination unit, and the target pixel is located on a travel direction included in the filter selected by the selection unit, and the travel direction is the target pixel. 2. The medical use according to claim 1, wherein the determination result made to the target pixel is corrected based on a result of calculating the number of neighboring pixels matching the traveling direction included in the filter selected by the selection unit. Image processing device. The correction unit is determined to be the surface structure candidate pixel in the determination unit, and a difference between a width included in the filter selected by the selection unit and a width included in the filter selected by the selection unit in the target pixel Correcting the determination result made to the pixel of interest based on the result of calculating the number of neighboring pixels that is less than a predetermined threshold;
2. The medical image processing apparatus according to claim 1, wherein: