JP2005352571A - Image processing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily analyze a structure, especially a concentric structure, of a living tissue sample. <P>SOLUTION: A glandular cavity selection part 24 selects glandular cavities from a sample image formed by photographing an object O. Paying attention to an arbitrary single glandular cavity among the selected glandular cavities, a distance transformation part 26 calculates a distance from a profile of the glandular cavity as to each of pixels around the glandular cavity. A coordinate transformation part 28 performs coordinate transformation so that the image around the glandular cavity is transformed into an image of the coordinate system consisting of an axis matching the profile of the glandular cavity and an axis corresponding to the distance from the profile. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image processing apparatus that processes an image obtained by imaging a sample, and more particularly to analysis of concentric structures contained in a biological tissue specimen.

  Until now, various attempts have been made to analyze biological specimens, particularly pathological specimens, by image processing.

  In order to analyze a pathological specimen, it is necessary to recognize a structure constituting the specimen, that is, a structure constituted by a cell, a cell nucleus, or a plurality of cells. It is possible to extract information effective for diagnosis support for the first time by accurately recognizing them and combining them as features.

  A structure composed of a plurality of cells includes the gland duct of the prostate. This is an organ for secreting prostate fluid, and has a structure in which cells constituting the wall surface of the cavity surround the cavity (space where no tissue exists). Normal gland ducts or relatively mild prostate cancer gland ducts are generally circular, but cancerous prostates often have distorted glandular duct shapes.

  Patent Document 1 discloses a technique for performing classification determination for diagnosis support by combining pathological specimens, particularly prostate specimens, and combining morphological characteristics of cell nuclei and morphological characteristics of cavities.

  Patent Literature 2 discloses a technique for recognizing a circular object. This is a concentric observation area that has the same center as the center of the circular object, and it recognizes the circular object by imaging and processing the luminance profile for one round in this observation area, and compares it with other circular objects. It gives the characteristic to do.

Further, Patent Document 3 discloses another technique for recognizing a circular object. In this method, after identifying the center and contour of a circular object, a feature amount that does not depend on the rotation of the object is obtained, thereby enabling recognition and comparison verification independent of the orientation (rotation angle) of the circular object.
JP 2001-59842 A JP-A-7-21381 JP 2002-32812 A

  Unlike an industrial product, a biological specimen cannot be expressed by a set of simple figures such as a circle or a straight line, and it is usually impossible that the same shape repeatedly appears. Normal gland ducts or relatively mild prostate cancer gland ducts are generally circular, but far from the perfect circle as envisaged in industrial tests. In particular, cancerous tissue loses structural order, and this tendency is remarkable.

  In the technique disclosed in Patent Document 1, cell nuclei and cavities contained in tissue are separately analyzed. However, when analyzing a gland duct that is a main organ of the prostate, a structure constituted by cell nuclei, cavities, and the like. Must be recognized. In this prior art, the gland duct is a structure in which cells surround the cavity, and each cell contains a cell nucleus, so there is a lack of consideration regarding the structure of the tissue, and the disclosed analysis method has high accuracy. Recognition of the ducts is not expected.

  The technique disclosed in Patent Document 2 is based on the premise that the object has a center and is approximately rotationally symmetric with respect to the center. That is, it is a technique based on an object such as a wheel, and is not applicable to recognition of a biological specimen.

  The technique disclosed in Patent Document 3 is based on the premise that the object has almost the same shape as if the object was cast. That is, the technique is based on an object such as a coin, and is not applicable to the recognition of a biological specimen.

  The present invention has been made in view of the above points, and an object of the present invention is to provide an image processing apparatus that can easily analyze the structure of a sample, in particular, the concentric structure of a biological tissue specimen.

One aspect of the image processing apparatus of the present invention is:
Contour setting means for setting the contour of the region of interest of the sample image;
Coordinate transformation is performed so that a part of the sample image including at least the region of interest set by the contour setting unit is an image of a coordinate system including an axis corresponding to the contour and an axis corresponding to the distance from the contour. Coordinate conversion means for
It is characterized by comprising

  ADVANTAGE OF THE INVENTION According to this invention, the image processing apparatus which can analyze easily the structure of a sample, especially the concentric structure of a biological tissue specimen can be provided.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  In this embodiment, in order to recognize a substantially concentric structure constituting at least a part of the sample, the sample image including the structure and the vicinity thereof is displayed on the horizontal axis of the outline of the object occupying the center of the structure. The image is converted to a plane with the distance as the vertical axis.

  In this embodiment, the sample to be imaged is a prostate pathological specimen stained with hematoxylin and eosin as shown in FIG. In the figure, the substantially concentric structure is a cancerous gland duct, and the object occupying the center is a cavity inside the gland duct (gland duct lumen or simply glandular cavity). Hematoxylin and eosin staining is the most widely used method for staining pathological specimens and can be easily and inexpensively performed.

  As shown in FIG. 1A, the image processing apparatus 10 according to the present embodiment includes an illumination 12, an (objective and imaging) optical system 14, a camera 16, a memory 18, a binarization unit 20, and a particle analysis unit 22. , Gland cavity selection unit 24, distance conversion unit 26, coordinate conversion unit 28, frequency distribution generation unit 30, frequency peak position detection unit 32, and gland duct determination unit 34.

  That is, as shown in FIG. 1B, the illumination 12 is irradiated from the opposite side to the camera 16 to the imaging target O placed on the stage (not shown). Then, the transmitted light is imaged on an imaging surface of an imaging element (not shown) included in the camera 16 by the optical system 14. Here, the camera 16 is selected to be suitable for imaging and analysis of the target O, and an RGB camera is used in this embodiment. An image of the object O captured by the camera 16, for example, a pathological specimen image as shown in FIG. 2 is stored in the memory 18.

  Thereafter, the binarization unit 20 binarizes the image of the object O stored in the memory 18 according to the luminance, and extracts a high luminance region. Thereby, in the case of the pathological specimen image shown in FIG. 2, a binarized image as shown in FIG. 3 is obtained. In the transmission observation of the pathological specimen, since the illumination light is directly observed in a region such as a cavity where the tissue is not present, the luminance is remarkably higher than that in the region where the tissue is present. Therefore, if the luminance information is used, the cavity and the tissue can be easily separated.

  Further, if the particle analysis is performed after extracting the high luminance region and the selection is performed according to the size, position, and shape, the cavity region can be separated with higher accuracy. For example, a high-luminance region that is less than a specific size is regarded as noise and excluded from the target, and a high-luminance region adjacent to the image edge is regarded as a non-cavity background and excluded from the target. Therefore, the particle analysis unit 22 extracts a high-luminance region from the binarized image to perform particle analysis, and the glandular cavity selection unit 24 sets each extracted high-brightness connected region as a glandular cavity. Select as.

  Next, the coordinates of the images in the vicinity of the plurality of glandular cavities thus selected are each subjected to coordinate conversion by the procedure described below.

  First, as shown in FIG. 4A, paying attention to an arbitrary single gland duct, the distance conversion unit 26 calculates the distance from the glandular cavity for each pixel near the glandular cavity. However, the distance between the pixel and the glandular cavity here refers to the minimum distance among the Euclidean distances from the pixel to all the pixels constituting the glandular cavity. This algorithm is known as distance transformation in mathematical morphology (morphological operation), and “in a filter shape capable of efficient morphological operation” (Aoi et al., Information Processing Society of Japan, Vol. 41, No. 12). , Pp. 3344-3351).

  FIG. 5A shows a set of points (hereinafter referred to as equidistant lines) having the same distance from the glandular cavity (central oblique line area) 100 in order to visually explain the result of the distance conversion. is there. In FIG. 5A, equidistant lines 102 of distances “20”, “40”, “60”, “80”, “100” are presented, and the distances “20”, “40”, “60” are presented. , “80” equidistant line 102 is represented by a broken line, and “100” equidistant line 102 is represented by a solid line. In general, equidistant lines form a closed curve, and do not have intersections with objects and other equidistant lines that are distance criteria. Therefore, the arbitrary equidistant line 102 can be uniquely identified using the distance from the object (in this embodiment, the glandular cavity 100).

  Further, as shown in FIG. 5B, when an arbitrary point P outside the object B is given, it is clear that there is only one equidistant line C including the point. Further, if a starting point S that is a reference for circulation is provided on the equidistant line C, the distance from the starting point S to the P obtained along the equidistant line C with respect to the equidistant line C including the point P. The position of the point P on the equidistant line C can be uniquely described by (hereinafter, “line distance”) or by normalizing the line distance by the circumference of the equidistant line C (hereinafter, “normalized line distance”). Note that the range of the normalized line distance is always (0, 1) and does not depend on the circumference of the equidistant line C. Therefore, an arbitrary point P outside the glandular cavity (object B) uses a set of the distance D from the glandular cavity (object B) and the normalized line distance L from the start point S in the equidistant line C of the distance D. (L, D), and this description is unique. That is, the point P on the image expressed as (X, Y) in the XY coordinate system can be expressed as coordinates (L, D) in the coordinate system based on the above description method, as shown in FIG. There is a one-to-one correspondence between the two.

  Based on the above principle, the coordinate conversion unit 28 converts the image near the glandular cavity into the LD coordinate system. For example, when coordinate conversion is performed on an image in the vicinity of the glandular cavity as shown in FIG. 4A, a coordinate conversion result image as shown in FIG. 4B is obtained. Here, the horizontal axis in the coordinate conversion result image represents the normalized line distance L, and the value increases from left to right. The vertical axis represents the distance D from the glandular cavity, and the value increases from top to bottom. In the coordinate conversion result image of FIG. 4B, the pixels having the same ordinate have the same distance from the glandular cavity.

  By the way, as shown in FIG. 6, the gland duct in the prostate has a concentric structure in which the gland cavity 100 is located at the center, the epithelium 104 around it, and the cell nucleus 106 around it. On the other hand, as shown in FIG. 4C, it is known that the blood vessel has no epithelium around the cavity, and the distance to the cell nucleus is very short or does not exhibit a concentric structure.

  Here, for example, a process for determining the degree of progression of prostate cancer based on the Gleason classification is considered. The Gleason classification is a technique widely used in the field of pathological diagnosis as a diagnostic criterion for determining the degree of progression of prostate cancer. Since the Gleason classification is based on the task of recognizing and analyzing the gland duct, it must be able to recognize the gland duct correctly, for example, without being confused with a blood vessel.

  Therefore, the frequency distribution generation unit 30 extracts dark purple pixels representing the cell nucleus 106 from the coordinate conversion result image shown in FIG. 4B, and the frequency distribution in the horizontal axis direction as shown in FIG. 7A. Get. Similarly, for blood vessels, the coordinate conversion result image is obtained as shown in FIG. 4D, and a frequency distribution in the horizontal axis direction as shown in FIG. 7B is obtained. The gland duct has a large peak of pixels representing the cell nucleus 106 near the distance “20” from the glandular cavity, whereas such a tendency is not observed in the blood vessel. Therefore, by detecting the frequency and peak position by the frequency peak position detection unit 32 and observing them by the gland duct determination unit 34, it is possible to distinguish between the gland duct and the blood vessel and recognize only the gland duct. If the frequency distribution is smoothed before the peak position is detected, the processing is more robust against noise.

  Although the above description is about the case where attention is paid to a single glandular cavity, generally, a specimen of a prostate includes a plurality of gland ducts. In this case, the glandular cavity selection unit 24 can recognize all the gland ducts included in the pathological specimen image by repeating the coordinate conversion and the gland duct recognition process while sequentially changing the target cavity.

  Note that the gland cavity selection unit 24 may manually select the gland duct of interest instead of automatically selecting the gland duct of interest.

  Further, the coordinate conversion result image shown in FIGS. 4B and 4D may be displayed on a display device (not shown) as necessary. In this display, in addition to a method for displaying each coordinate conversion result image alone, a method for displaying a plurality of coordinate conversion result images in an array, and a method for displaying an original image and a coordinate conversion result image in parallel. Is applicable.

  When the coordinate conversion result image is recorded on a recording medium by a recording device (not shown), the correspondence between the original image and the coordinate conversion result image is also recorded. Alternatively, if images before and after coordinate conversion are recorded as a set, it is convenient for reuse.

  As described above, according to the present embodiment, an image of a pathological specimen is taken, the glandular cavity of the prostate is recognized, the pixels around the glandular cavity are coordinate-converted, and the gland duct is recognized separately from the blood vessel. be able to.

  If this method is used, even if the object to be recognized is not a simple geometric combination, it is possible to easily analyze the state of the neighborhood by the above coordinate conversion. Since they have the same ordinate, it is possible to analyze the vicinity of the object using simple processing such as projection and frequency distribution. This is particularly suitable for the analysis of a biological specimen whose object is often indefinite shape.

  As mentioned above, although this invention was demonstrated based on one Embodiment, this invention is not limited to one Embodiment mentioned above, Of course, a various deformation | transformation and application are possible within the range of the summary of this invention. It is.

  For example, in the above-described embodiment, a biological specimen (pathological specimen) is used as the imaging target O, but the present invention can also be applied to other specimens.

(Appendix)
The invention having the following configuration can be extracted from the specific embodiment.

(1) contour setting means for setting the contour of the region of interest of the sample image;
Coordinate transformation is performed so that at least a part of the sample image including the region of interest set by the contour setting unit is an image of a coordinate system including an axis corresponding to the contour and an axis corresponding to the distance from the contour. Coordinate conversion means for
An image processing apparatus comprising:

  Here, the contour setting means corresponds to the illumination 12, the optical system 14, the camera 16, the memory 18, the binarization unit 20, the particle analysis unit 22, and the glandular cavity selection unit 24 in FIG. The coordinate conversion means corresponds to the distance conversion unit 26 and the coordinate conversion unit 28 in FIG.

  According to the image processing apparatus described in (1), it is possible to provide an image processing apparatus that can easily analyze the structure of the sample.

  (2) The image processing apparatus according to (1), wherein a range of an axis corresponding to the contour is normalized.

  According to the image processing apparatus described in (2), it is possible to perform processing independent of the size of the contour of the attention area.

  (3) The image processing apparatus according to (1) or (2), wherein the sample is a biological specimen.

  According to the image processing apparatus described in (3), the structure of the biological specimen can be analyzed.

  (4) The image processing apparatus according to (3), wherein the biological specimen is a pathological specimen.

  According to the image processing apparatus described in (4), it is possible to analyze the structure of the pathological specimen.

  (5) The image processing apparatus according to (4), wherein the pathological specimen is a prostate specimen.

  According to the image processing apparatus described in (5), the structure of the prostate can be analyzed.

  (6) The image processing device according to (5), wherein the region of interest is a lumen of a gland duct of the prostate.

  According to the image processing apparatus described in (6), it can be used for diagnosis of prostate cancer.

  (7) Any one of (1) to (6), further comprising a classification unit that classifies a plurality of regions included in the sample image based on the image coordinate-transformed by the coordinate transformation unit. An image processing apparatus according to 1.

  Here, the classification means corresponds to the frequency distribution generation unit 30, the frequency peak position detection unit 32, and the gland duct determination unit 34 in FIG.

  According to the image processing apparatus described in (7), the structures can be automatically classified.

  (8) The image processing apparatus according to (7), wherein the classification unit classifies the gland ducts and blood vessels of the prostate.

  According to the image processing apparatus described in (8), since the prostate gland can be classified, it can be used for diagnosis of prostate cancer.

  (9) The image processing apparatus according to any one of (1) to (8), further comprising display means for displaying an image coordinate-converted by the coordinate conversion means.

  Here, the display means corresponds to a display device (not shown).

  According to the image processing apparatus described in (9), the user can easily grasp the structure of the sample from the displayed image.

  (10) The image processing apparatus according to (9), wherein the display unit displays a plurality of coordinate-transformed images side by side.

  According to the image processing apparatus described in (10), it is possible to compare a plurality of coordinate-converted images.

  (11) The image processing apparatus according to (9), wherein the display unit displays the sample image and the coordinate-converted image side by side.

  According to the image processing apparatus described in (11), images before and after coordinate conversion can be compared.

  (12) The image according to any one of (1) to (11), further comprising recording means for recording the sample image and the image coordinate-transformed by the coordinate transformation means as a set. Processing equipment.

  Here, the recording means corresponds to a recording device (not shown).

  According to the image processing apparatus described in (12), the images before and after the coordinate conversion are not separated, which is convenient for reuse.

(A) is a figure which shows the structure of the image processing apparatus which concerns on one Embodiment of this invention, (B) is a figure which shows the cross section of the dye | stained tissue sample. It is a figure which shows the image of the pathological specimen of the prostate dye | stained with hematoxylin and eosin. It is a figure which shows the binarized image which binarized the pathological sample image of FIG. (A) is a diagram showing an image of a gland duct and its vicinity, (B) is a diagram showing a coordinate conversion result image of the image of (A), (C) is a diagram showing an image of a blood vessel and its vicinity, (D) is a diagram showing a coordinate conversion result image of the image of (C). (A) is a diagram for visually explaining the distance transformation result, (B) is a diagram for explaining the unique designation of the position based on the equidistant gland, and (C) is a concept of the coordinate transformation result. It is a figure which shows LD coordinate system for demonstrating. It is a schematic diagram for demonstrating the concentric structure of a gland duct. (A) is a figure which shows the frequency distribution of the pixel calculated | required from the coordinate transformation result image of FIG.4 (B), (B) is the frequency distribution of the pixel calculated | required from the coordinate transformation result image of FIG.4 (D). FIG.

Explanation of symbols

    DESCRIPTION OF SYMBOLS 10 ... Image processing apparatus, 12 ... Illumination, 14 ... Optical system, 16 ... Camera, 18 ... Memory, 20 ... Binarization part, 22 ... Particle analysis part, 24 ... Glandular cavity selection part, 26 ... Distance conversion part, 28 DESCRIPTION OF SYMBOLS ... Coordinate conversion part, 30 ... Frequency distribution generation part, 32 ... Frequency peak position detection part, 34 ... Gland duct determination part, 100 ... Glandular cavity, 102 ... Equidistant line, 104 ... Epithelium, 106 ... Cell nucleus, O ... Object.

Claims (12)

Contour setting means for setting the contour of the region of interest of the sample image;
Coordinate transformation is performed so that at least a part of the sample image including the region of interest set by the contour setting unit is an image of a coordinate system including an axis corresponding to the contour and an axis corresponding to the distance from the contour. Coordinate conversion means for
An image processing apparatus comprising:   The image processing apparatus according to claim 1, wherein a range of an axis corresponding to the contour is normalized.   The image processing apparatus according to claim 1, wherein the sample is a biological specimen.   The image processing apparatus according to claim 3, wherein the biological specimen is a pathological specimen.   The image processing apparatus according to claim 4, wherein the pathological specimen is a prostate specimen.   6. The image processing apparatus according to claim 5, wherein the region of interest is a lumen of a gland duct of the prostate.   7. The image processing according to claim 1, further comprising a classification unit that classifies a plurality of regions included in the sample image based on the image coordinate-transformed by the coordinate transformation unit. apparatus.   The image processing apparatus according to claim 7, wherein the classification unit classifies the gland ducts and blood vessels of the prostate.   9. The image processing apparatus according to claim 1, further comprising display means for displaying an image coordinate-converted by the coordinate conversion means.   The image processing apparatus according to claim 9, wherein the display unit displays a plurality of coordinate-converted images side by side.   The image processing apparatus according to claim 9, wherein the display unit displays the sample image and the coordinate-converted image side by side.   12. The image processing apparatus according to claim 1, further comprising recording means for recording the sample image and the image coordinate-transformed by the coordinate transformation means as a set.

Images