JP2008146278A - Cell contour extraction apparatus, cell contour extraction method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a cell contour extraction apparatus capable of extracting a cell contour from a cell staining image.
A cell contour extracting apparatus 10 includes an image acquiring unit 12 that acquires a stained image of a cell photographed by a microscope, a cell nucleus contour extracting unit 20 that detects a region of a cell nucleus from the stained image, and each pixel of the stained image. An edge detection unit 32 that detects a density gradient and edge with an adjacent pixel and an optimum path that goes around the cell nucleus are obtained by dynamic programming using the density gradient as a parameter, and the optimum path is extracted as a cell outline. A cell contour extraction unit 30, and the cell contour extraction unit performs a process of calculating a partial optimum path that connects pixel groups included in two pixel arrays arranged in different directions from a predetermined point in a cell nucleus region. Then, the pixel arrangement direction is changed repeatedly until the pixel arrangement direction goes around the predetermined point, and the optimum path is obtained based on the partial optimum path.
[Selection] Figure 1

Description

  The present invention relates to a technique for extracting an outline of an object included in an image, and more particularly, to a technique for extracting a cell outline from a stained image of a cell.

  Currently, stained images of cells are used for pathological diagnosis, and software for extracting cell morphology from image data is required.

  Conventionally, a technique for extracting a contour is known as an image processing technique for recognizing an object. One of the contour extraction methods is dynamic programming. The dynamic programming method is a method in which image data in a predetermined range around a contour model given as an initial value is sequentially evaluated by an evaluation function, and the contour model is gradually converged to an optimal solution (contour).

Patent Document 1 discloses an invention for extracting a contour by image processing using dynamic programming. In Patent Document 1, in order to solve the problem that a local minimum of an evaluation function is erroneously determined as an optimal solution, an image is processed by a Gaussian filter that can change the degree of blurring before applying dynamic programming. To do. The evaluation function is prevented from converging to the optimal solution of the local minimum by the greatly blurred image, and the optimum solution is obtained by sequentially reducing the degree of blurring. Patent Document 1 gives an example applied to the contour extraction of slugs and the extraction of a heart image from an ultrasound image of the heart, and shows that the contour can be appropriately extracted by the contour extraction technique.
JP-A-5-242247

  However, the invention described in Patent Document 1 cannot be applied to cell contour extraction.

  In the stained image of cells, not all cells in the image are stained uniformly. Since the staining intensity of the cell contour is strong and weak, and the staining intensity in the cell nucleus and cytoplasm is not constant, it is difficult to distinguish the cell contour from other parts. That is, the stained image of the cell has a larger noise than the image exemplified in Patent Document 1. For this reason, the cell contour cannot be extracted by the conventional method as described above.

  In addition, depending on the state of staining, the cell staining image may be captured as if the outline of the cell was partially cut. Since there are contours between cells in the cell image, the contour of the cut portion cannot be extracted unlike the case of detecting the contour of the object from the background image. However, it is desired to automatically extract a contour close to the true cell contour from such an image.

  In view of the above background, an object of the present invention is to provide a cell contour extraction apparatus capable of extracting a cell contour from a stained image of a cell.

  The cell contour extraction device of the present invention includes an image acquisition unit that acquires a stained image of a cell photographed by a microscope, a cell nucleus detection unit that detects a region of a cell nucleus from the stained image, and an adjacent pixel for each pixel of the stained image A concentration gradient calculating unit for obtaining a concentration gradient of the cell, and an optimum route that goes around the cell nucleus by dynamic programming using the concentration gradient as a parameter, and extracting the optimum route as a cell contour And the cell contour extraction unit calculates a partial optimal path that connects pixel groups respectively included in two pixel arrays arranged in a different direction from a predetermined point in the cell nucleus region, The direction of the pixel arrangement is changed repeatedly until the direction of the pixel arrangement makes a round around the predetermined point, and the optimum path is obtained based on the partial optimum path.

  In this way, a calculation process that uses the feature that a cell nucleus exists in a cell, that is, a partial optimization that connects pixel groups included in two pixel arrays arranged in a different direction from a predetermined point in the cell nucleus region. The cell contour can be appropriately extracted by dynamic programming in which the process of calculating the route is repeated.

  In the cell contour extraction device, the cell contour extraction unit obtains a plurality of candidate routes that are candidates for the optimal route based on the partial optimal route, and starts calculating the partial optimal route from the candidate routes. The difference between the distance from the predetermined point of the pixel that passes through the candidate path in the pixel array and the distance from the predetermined point of the pixel that passes through the candidate path in the pixel array that has been calculated is equal to or less than a predetermined threshold. A candidate route may be determined as the optimum route.

  By using the condition that the outline of the cell is closed as described above, a suitable optimum route can be obtained by obtaining a route in which the deviation of the partial optimum route at the start point and the end point of the calculation is equal to or less than a predetermined threshold. .

  In the cell contour extraction device, the cell contour extraction unit calculates a gap penalty based on a difference in distance from the predetermined point of pixels to be calculated that are included in different pixel arrays, and the gap penalty and the density The partial optimum route may be obtained based on the gradient.

  In this way, by calculating the gap penalty based on the difference in distance between the two pixels to be calculated from the predetermined point, if the two pixels are separated, a strong gap penalty is set and the separated pixels It is possible to make it difficult for the route connecting the two to be selected as the partially optimal route. Thereby, a highly continuous route can be selected as a partially optimal route. Note that, in the conventional dynamic programming method described in Patent Document 1, a partial optimum path is selected from eight pixels adjacent to the pixel to be calculated, so that the optimum path is configured by the adjacent pixels. It had been. In the present invention, if the pixel arrays are adjacent to each other, it is possible to be selected as a partially optimal path even if the pixels are not adjacent to each other. Therefore, an optimal path constituted by pixels that are not adjacent can also be obtained. it can. It is desirable to use the center of gravity or center of the cell nucleus as the predetermined point in the cell nucleus.

  The cell contour extraction apparatus includes an edge detection unit that extracts a plurality of edges in the stained image based on the density gradient and assigns edge identifiers to pixels constituting each edge, and the cell contour extraction The unit may calculate the gap penalty based on a difference in distance from the predetermined point of pixels to be calculated included in different pixel arrays and the difference between the edge identifiers.

  In this way, by using the difference between the edge identifiers of a plurality of edges obtained in advance as a gap penalty, it is possible to appropriately obtain the optimum route by reflecting the edge detection result by the edge detection unit in the dynamic programming. Various methods including known methods can be adopted for edge detection by the edge detection unit.

  In the cell contour extracting apparatus, the cell contour extracting unit may exclude the region of the cell nucleus detected by the cell nucleus detecting unit from the search region of the optimum route.

  Thereby, the calculation processing amount by a cell outline extraction part can be reduced. Further, it is possible to prevent erroneous detection of a route passing through the cell nucleus region as a cell contour.

  The cell contour extraction device receives a position of a point indicated in an image displayed on the display unit and a display unit that displays the contour extracted by the cell contour extraction unit on the image of the cell. And an input unit, wherein the cell contour extraction unit may obtain an optimum route passing through the instructed point.

  In this way, by obtaining the optimum route that passes through the point indicated on the image, the obtained cell contour can be corrected.

  An outline extraction apparatus according to another aspect of the present invention includes an image acquisition unit that acquires an image, a density gradient calculation unit that calculates a density gradient with an adjacent pixel for each pixel of the image, and the density gradient based on the density gradient. An edge detection unit that extracts a plurality of edges in an image and assigns an edge identifier to pixels constituting each of the edges; an outline of an object included in the image; and a dynamic that uses the density gradient as a parameter A contour extraction unit that calculates the optimum route by a planning method, wherein the contour extraction unit calculates a gap penalty based on the difference between the edge identifiers, and determines an optimal route based on the gap penalty and the concentration gradient. Ask.

  As described above, by using the edge identifiers of a plurality of edges obtained in advance as gap penalties, the detection result detected by the edge detection unit can be reflected in the dynamic programming and the optimum route can be obtained appropriately.

  The cell contour extraction method of the present invention includes a step of acquiring a stained image of a cell photographed by a microscope, a step of detecting a region of a cell nucleus from the stained image, and a concentration gradient between adjacent pixels for each pixel of the stained image And calculating an optimal path that goes around the cell nucleus by dynamic programming using the concentration gradient as a parameter, and extracting the optimal path as a cell outline, Processing for calculating a partial optimum path connecting pixel groups respectively included in two pixel arrays arranged in a different direction from a predetermined point in the pixel array until the direction of the pixel array makes a round around the predetermined point And repetitively changing the direction to obtain the optimum route based on the partial optimum route.

  Thereby, similarly to the cell contour extraction apparatus of the present invention, the cell contour can be appropriately extracted by calculating the optimum path that goes around a predetermined point in the region of the cell nucleus by dynamic programming. Various configurations of the cell contour extracting apparatus of the present invention can be applied to the cell contour extracting method of the present invention.

  The program of the present invention is a program for extracting the outline of a cell from a stained image of a cell, the step of acquiring a stained image of a cell photographed by a microscope in a computer, and a region of a cell nucleus from the stained image. A step of detecting, a step of obtaining a density gradient with an adjacent pixel for each pixel of the stained image, and an optimum path that goes around the cell nucleus is obtained by dynamic programming using the density gradient as a parameter, and the optimum A step of extracting a path as a cell outline, a process of calculating a partial optimal path connecting pixel groups respectively included in two pixel arrays arranged in a different direction from a predetermined point in the region of the cell nucleus, The pixel array direction is changed repeatedly until the pixel array direction makes a round around the predetermined point, and the partial optimum process is performed. And a step of determining the optimal path based on.

  Thereby, similarly to the cell contour extraction apparatus of the present invention, the cell contour can be appropriately extracted by calculating the optimum path that goes around a predetermined point in the region of the cell nucleus by dynamic programming. Various configurations of the cell contour extraction apparatus of the present invention can be applied to the program of the present invention.

  The present invention relates to a region of a cell nucleus by a dynamic programming method that repeatedly performs a process of calculating a partial optimum path that connects pixel groups included in two pixel arrays arranged in different directions from a predetermined point in the region of the cell nucleus. By calculating the optimum path that goes around the predetermined point, it is possible to appropriately extract the cell contour from the stained image of the cell.

  Hereinafter, a cell outline extracting apparatus according to an embodiment of the present invention will be described with reference to the drawings.

(First embodiment)
FIG. 1 is a diagram illustrating a configuration of a cell contour extracting apparatus 10 according to the first embodiment. The cell contour extraction apparatus 10 includes an image acquisition unit 12 that acquires a stained image of a cell, an image data storage unit 18 that stores image data acquired by the image acquisition unit 12, and an image stored in the image data storage unit 18. A cell nucleus contour extracting unit 20 that extracts data cell contours by processing data and a cell contour extracting unit 30 that extracts cell contours are included.

The image acquisition unit 12 acquires, for example, image data obtained by photographing a tissue of a cell stained with hematoxylin and eosin (hereinafter referred to as “HE staining”) with an optical microscope.
FIG. 2 is a diagram illustrating an example of a cell image acquired by the image acquisition unit 12. In the present embodiment, as the image data, for example, data digitized in a format such as JPEG or GIF is acquired.

[Cell nucleus contour extractor]
The cell nucleus contour extraction unit 20 has a function of extracting the contour of the cell nucleus from the stained image of the cells. The HE-stained cell image is stained so that the difference between the cell nucleus and the cytoplasm can be discriminated. The cell nucleus contour extraction unit 20 extracts a cell nucleus region based on the difference in hue, saturation, and brightness between the cell nucleus and the cytoplasm.

  The cell nucleus contour extraction unit 20 includes a gray scale conversion unit 22, a connected component extraction unit 24, and a convex hull processing unit 26. The gray scale conversion unit 22 has a function of converting a stained image of cells to be processed into a gray scale. The gray scale conversion unit 22 increases the intensity of pixels in the cell nucleus region during conversion to gray scale.

The gray scale conversion unit 22 once converts the image data into an HSV model so that the pixels can be separated by the difference in color and brightness. The gray scale conversion unit 22 acquires typical cell nucleus color information from the input unit 14. Specifically, the stained image to be processed is displayed on the display unit 16 and the user is allowed to select one of the cell nuclei shown in the displayed image. When the user clicks the mouse, the input unit 14 inputs the hue, saturation, and brightness data of the pixel designated by the mouse to the grayscale conversion unit 22. The gray scale conversion unit 22 calculates the intensity (gray scale value) of each pixel of the stained image according to the following equation (1) using the color selected by the user as a reference.

  The connected component extraction unit 24 extracts connected components from the grayscale image by a threshold method. FIG. 3A is a diagram schematically showing a gray scale image of a cell nucleus. Actually, the cell nucleus N has a color having a concentration corresponding to the staining, but is omitted in FIG. FIG. 3B is a diagram illustrating the intensity of the pixel value on the BB line in FIG. The connected component extracting unit 24 detects a region having a predetermined threshold value or more as a connected component from the pixel value data as shown in FIG. Since cell nuclei may not be uniformly dyed as shown in FIG. 3B, the connected component extraction unit 24 may adjust the threshold so that the connected components can be appropriately extracted when detecting the cell nuclei. Good.

  The convex hull processing unit 26 performs a convex hull process on the connected component region detected by the connected component extracting unit 24 to determine a cell nucleus region and a cell nucleus contour. By the convex hull process, even when the cell nucleus is not uniformly dyed, the cell nucleus region and the cell nucleus contour can be detected appropriately.

  FIG. 4A and FIG. 4B are diagrams for explaining the convex hull. FIG. 4A shows an image before the convex hull processing, and FIG. 4B shows an image after the convex hull processing. A convex hull is a polygon that includes all points of interest and has a minimum area. The convex hull processing unit 26 draws a convex hull, counts the number of connected components as an area, and obtains a polygon that minimizes the number of connected components. By performing convex hull processing on the cell nucleus region shown in FIG. 4A, a cell nucleus region as shown in FIG. 4B is determined.

[Cell outline extraction unit]
The cell contour extraction unit 30 has a function of extracting a cell contour from a cell image. The cell contour extraction unit 30 includes an edge detection unit 32, a search region determination unit 34, a dynamic programming calculation unit 36, and a cell contour determination unit 38. Note that the cell contour extraction unit 30 performs processing on the grayscale image converted by the grayscale conversion unit 22.

  The edge detection unit 32 detects an edge from the cell image using a canny edge detector, and assigns an identifier to the obtained edge.

  Here, edge detection by the canny edge detector will be described. The canny edge detector first smoothes the cell image using a Gauss filter. This reduces the noise component contained in the cell image. Next, the canny edge detector performs density gradient detection using a Sobel filter. Thus, the canny edge detector detects the concentration gradient, and corresponds to the “concentration gradient detection unit” in the claims. The Sobel filter is a filter for detecting a vertical or horizontal density gradient (darkness difference) with respect to the pixel of interest.

FIG. 5A is a diagram showing an example of a Sobel filter for detecting a density gradient in the horizontal direction, and FIG. 5B is a diagram showing an example of a Sobel filter for detecting a density gradient in the vertical direction. The canny edge detector uses the density value g (i, j) of the pixel of interest using the cell value a and the pixel value f in the filters shown in FIGS. 5 (a) and 5 (b). It calculates | requires by the following formula | equation (2).

  The canny edge detector detects edges using values of density gradients in the vertical and horizontal directions. First, the canny edge detector calculates the density gradient direction and the magnitude of each pixel from the vertical density gradient and the horizontal density gradient. “Direction of density gradient” and “magnitude of density gradient” are obtained by vector synthesis of the density gradient in the vertical direction and the density gradient in the horizontal direction.

  The canny edge detector then detects edges based on the direction and magnitude of the density gradient. FIG. 6A is a diagram for explaining processing for detecting an edge based on a density gradient. In FIG. 6A, pixels with diagonal lines indicate pixels whose density gradient is larger than the first threshold value, and pixels with vertical lines indicate that the density gradient is equal to or lower than the first threshold value and the second threshold value (however, , Pixels larger than the second threshold <the first threshold) are shown.

  The canny edge detector detects an edge by connecting pixels having a density gradient greater than the first threshold based on the direction of the density gradient of each pixel.

  Here, an example of starting edge detection from the pixel P1 will be described. Around the pixel P1, there are three pixels having a density gradient greater than the first threshold. If the direction of the density gradient of the pixel P1 is the right direction, the canny edge detector connects the pixel P1 and the pixel P2, as shown by the solid line arrow in FIG. Note that the direction of the density gradient obtained by calculation is not necessarily the eight directions of upper, lower, left and right, upper right, lower right, upper left, and lower left, but in the edge detection process, the direction of the density gradient is rounded to one of the above eight directions. Process.

  Next, similarly for the pixel P2, if there is a pixel P3 larger than the first threshold in the direction of the density gradient, the pixel P2 and the pixel P3 are connected. In this way, the edges are extended by sequentially connecting pixels having a density gradient greater than the first threshold, and the edges are detected.

  After obtaining an edge by connecting pixels having a density gradient greater than the first threshold value, the pixel at the edge of the edge is connected to a pixel having a density gradient greater than the second threshold value. Taking the pixel P4 as an example in FIG. 6A, as in the case described above, when there is a pixel P5 having a density gradient greater than the second threshold in the direction of the density gradient of the pixel P4 at the end, FIG. Extend the edge in that direction as indicated by the dotted line. Through the above processing, pixels can be connected as shown by a solid line and a dotted line in FIG. As a result, the edge can be detected as shown in FIG.

  The canny edge detector gives an identifier to the detected edge. As shown in FIG. 6B, the edges E1 and E2 are identified and the edge to which each pixel belongs is stored.

  The dynamic programming calculation unit 36 extracts cell contours by calculation based on dynamic programming. The dynamic programming calculation unit 36 applies the dynamic programming in a region having a predetermined radius centered on the center of gravity of the cell nucleus (hereinafter, this region is referred to as a “search region”) using the concentration gradient as a parameter to obtain an optimum route. Ask for. In the present embodiment, dynamic programming is applied on the assumption that the angular component of the cell contour viewed from the center of gravity of the cell nucleus is monotonously displaced.

  The search area determination unit 34 has a function of determining an area having a predetermined radius centered on the center of gravity of the cell nucleus as a search area for dynamic programming. The search area determination unit 34 has a function of excluding a cell nucleus area from a search area having a predetermined radius. Since there is no cell outline in the cell nucleus area, excluding the cell nucleus area from the search area in advance reduces the computational burden for extracting the cell outline, and the path that passes through the cell nucleus area Can be extracted by mistake.

[Dynamic programming]
FIG. 7 is a diagram illustrating an example in which the contour of a cell having a cell nucleus A is calculated by dynamic programming. In FIG. 7, a circular region R centered on the center of gravity G of the nucleus A is a search region. The diameter of the circular region R is set in advance such that the cell to be searched is included according to the magnification of the cell image.

  The dynamic programming calculation unit 36 calculates an optimum path that goes around the cell nucleus A based on the density gradient of each pixel group (for example, pixels on the lines t1, t2, and t3) aligned in the radial direction of the circular region R. Ask.

  FIG. 8 is a diagram for explaining the route search by the dynamic programming calculation unit 36. FIG. 8 is a schematic diagram in which the search region R shown in FIG. 7 is developed in the circumferential direction. The vertical axis indicates the distance from the cell nucleus center G, and the horizontal axis indicates the angle. Lines t1 to t3 illustrated in FIG. 7 correspond to the lines t1 to t3 illustrated in FIG. In FIG. 8, a large number of fine lines that look like wrinkles indicate edges. Each pixel of the image shown in FIG. 8 has information on the density gradient value and the identifier of the edge to which the pixel belongs.

  FIG. 9A is a diagram showing the density gradient value of each pixel in the developed view shown in FIG. 8, and FIG. 9B is a diagram showing the identifier of the edge of each pixel. In FIG. 9B, “0” is described as an edge identifier for pixels that are not edges.

  The dynamic programming calculation unit 36 obtains a partial optimum path that connects pixel groups arranged in different directions from the center of gravity G as the first step of calculating the optimum path.

  FIG. 9C is a diagram for explaining a method for obtaining a partial optimum route. FIG. 9C shows an example in which a partial optimum route to a pixel at coordinates (r, t) in the t column is obtained. For example, the t-1 column corresponds to the pixel arrangement on the line t2 shown in FIGS. 7 and 8, and the t column corresponds to the pixel arrangement on the line t3.

As shown by an arrow in FIG. 9C, the dynamic programming calculation unit 36 compares the coordinates (r, t) and the coordinates (r + k, t−1) in the t−1 column, respectively. Calculate the optimal route. In FIG. 9C, the arrows do not extend from the upper two rows. This is because the upper two rows are areas of the cell nucleus A and are excluded from the search area by the search area determination unit 34. The dynamic programming calculation unit 36 obtains a dynamic programming score (hereinafter referred to as “DP score”) in order to obtain a partially optimal route. The following formula (3) is a formula for obtaining the DP score d (r, t) at the coordinates (r, t).

The variable d (r + k, t−1) of the first term in the max function on the right side of Expression (3) is the value of each pixel in the t−1 column immediately before the DP score d (r, t) to be obtained. Shows DP score. The variable of the second term in the max function indicates a gap penalty (gap_penalty). When the pixel at the coordinate (r, t) and the pixel at the coordinate (r + k, t−1) to be calculated in the t−1 column have the same edge identifier, the gap penalty is the same as the pixel at the coordinate (r, t) and the coordinate ( If any of the pixels of r + k, t−1) does not have an edge identifier, a weak gap penalty G ₁ is set, and if the edge identifiers are different, a strong gap penalty G ₂ (G ₂ <G ₁ ) is set. . As a result, when the edge identifiers are different, a strong gap penalty is applied, so that it is difficult to select the partial optimum route. Whether or not the edge identifiers are the same is determined using the data shown in FIG.

  The coefficient k multiplied by the gap penalty in the second term is the value of the calculation target pixel (r, t) and the calculation target pixel (coordinate (r + k, t−1)) in the t−1 column in the radial direction. It indicates whether the pixel is shifted (difference in distance from the center of gravity G). As a result, the gap penalty increases as the difference between the two pixels to be calculated increases, so that it is difficult to select the partial optimal path.

  The dynamic programming calculation unit 36 stores the pixels in the t−1 column that give the DP score of the pixel at the coordinates (r, t) as a partially optimal path to the coordinates (r, t). Here, the pixel of the t−1 column that gives the DP score of the pixel at the coordinate (r, t) is the pixel having the largest numerical value in the max function, and the pixel shift (k) in the gap penalty (gap_penalty). This is the pixel with the smallest value multiplied by.

  The dynamic programming calculation unit 36 performs the above calculation for all the pixels in the t column, and obtains a partial optimum path for each pixel. When the calculation for the t column is completed, the dynamic programming calculation unit 36 sequentially calculates t + 1 column, t + 2 column,... The dynamic programming calculation unit 36 calculates a partial optimum path while gradually changing the selection direction of the pixel column to be calculated, and performs calculation until the pixel column to be selected makes a round and returns to the pixel column from which the calculation is started. Do. That is, calculation is performed for pixel rows in all directions around the center of gravity G. When the calculation for all the pixels in the search region is completed, the dynamic programming calculation unit 36 sequentially follows the partially optimal route, thereby forming a plurality of routes. This is the optimum route candidate route (see FIG. 9D).

  The cell contour determination unit 38 has a function of determining a cell contour based on the DP score calculated by the dynamic programming calculation unit 36. After the DP score calculation by the dynamic programming calculation unit 36 is performed to the end (line t1 in FIG. 7), the cell contour determination unit 38 sorts the plurality of candidate routes in descending order according to the DP score at the final point. . The cell contour determination unit 38 looks at the sorted candidate routes in descending order of the DP score, and obtains a route that satisfies the boundary condition. A route that satisfies the boundary condition is a route in which the value of the vertical axis (shift in the radial direction) at the start point and end point of the calculation is within a certain range. With this configuration, an appropriate route can be extracted as the cell contour by utilizing the characteristic that the cell region is surrounded by the closed contour.

  FIG. 10 is a diagram illustrating a route search result. As shown by the dotted line in FIG. 10, there are a plurality of candidate routes, but ultimately, the route indicated by the solid line that satisfies the boundary condition is selected as the optimum route.

  FIGS. 11-13 is a figure which shows operation | movement of the cell outline extraction apparatus 10 of this Embodiment. FIG. 11 is a diagram illustrating the overall operation of the cell contour extraction apparatus 10, FIG. 12 is a diagram illustrating the operation of cell nucleus contour extraction, and FIG. 13 is a diagram illustrating the operation of cell contour extraction.

  As illustrated in FIG. 11, the cell contour extraction device 10 acquires a stained image of a cell (S10). Next, the cell contour extraction apparatus 10 extracts the cell nucleus contour by the cell nucleus contour extraction unit 20 (S12), and extracts the cell contour by the cell contour extraction unit 30 (S14). The cell contour extracting apparatus 10 displays the extracted cell contour on the display unit 16 (S16).

  The cell nucleus contour extraction process will be described in detail with reference to FIG. The cell contour extraction device 10 displays the acquired stained image on the display unit 16 (S20), and accepts selection of a cell nucleus from the user (S22). When the user clicks the mouse on the displayed image, the input unit 14 of the cell contour extracting apparatus 10 transmits information on the hue, brightness, and saturation of the pixel indicated by the mouse to the grayscale conversion unit 22. . The gray scale conversion unit 22 converts the cell image into a gray scale image using the hue, brightness, and saturation information received from the input unit 14 (S24).

  Next, the cell contour extraction apparatus 10 detects a connected component from the grayscale image of the cell using a predetermined threshold (S26), and performs a convex hull process on the detected connected component region to thereby extract a region of the cell nucleus. Determine (S28).

  Next, the cell contour extraction process will be described in detail with reference to FIG. The cell contour extraction apparatus 10 smoothes the grayscale image using a Gauss filter (S30), and applies a Sobel filter to the smoothed image to obtain a density gradient with adjacent pixels (S32). Specifically, the vertical and horizontal Sobel filters are applied to obtain the density gradient in the vertical direction and the density gradient in the horizontal direction, and the density gradient direction and magnitude are obtained by vector synthesis of each density gradient. . Next, the cell contour extraction apparatus 10 detects an edge by using the concentration gradient value by the method described with reference to FIGS. 6A and 6B (S34). The cell contour extracting apparatus 10 assigns an identifier to the detected edge (S34).

  Next, the cell contour extracting apparatus 10 determines a search area (S36). The search area determination unit 34 pays attention to the cell nucleus in the image, and determines a circular area centered on the center G of the cell nucleus as the search area. Further, the search area determination unit 34 sets the concentration gradient value of the cell nucleus area to 0 to exclude it from the search area, and limits the search area to an area where no cell nucleus exists (S38).

  Subsequently, the cell contour extracting apparatus 10 sequentially calculates partial optimum paths between the respective pixel columns arranged in different directions from the center of gravity G using dynamic programming (S40), and traces the partial optimum paths. The optimum route is determined based on the DP score and the boundary condition at the final point among the plurality of candidate routes generated in step S42. The cell contour extraction apparatus 10 determines whether or not the contour extraction processing has been completed for all cell nuclei detected in the image (S44), and if the processing has been completed for all cell nuclei (YES in S44), The cell extraction process is terminated. If there is a cell nucleus that has not been subjected to contour extraction processing, the process proceeds to step S36 for determining a search region for obtaining a cell contour including the cell nucleus, and the cell contour is obtained by the same processing as described above.

FIG. 14 is a diagram illustrating a result of extracting the cell contour L by the cell contour extracting apparatus 10. The cell contour extraction apparatus 10 displays the extraction result of the contour L as shown in FIG.
Heretofore, the configuration and operation of the cell contour extraction apparatus 10 of the present embodiment have been described.

  In this embodiment, since the optimal path that goes around a predetermined point in the cell nucleus region is calculated by dynamic programming using the feature that the cell nucleus exists in the cell, the contour of the cell is calculated. Can be appropriately extracted.

  In the present embodiment, the dynamic programming calculation unit 36 assumes that the angular component of the cell outline viewed from the center of gravity G of the cell nucleus is monotonously displaced, and advances the calculation in one direction, thereby reducing the calculation process. . In the conventional method, since any pixel in the eight directions around the pixel to be calculated is obtained as a partial optimal path, the calculation may not converge due to the optimal path going to the right or the left. In some cases, the calculation cost until convergence increases. According to the present embodiment, the calculation is easily performed because the calculation is performed in a range of 360 degrees while changing the angle in one direction around the center of gravity G of the cell nucleus.

  In the present embodiment, a partial optimum path is obtained by calculating pixels included in two pixel columns arranged in different directions, and the fact that the pixels are adjacent is not a condition for the optimal path. There is a possibility that a path connecting pixels is required as an optimal path. Thereby, for example, an appropriate optimum route can be obtained even when a part of the contour is cut.

  In the present embodiment, a strong gap penalty is set when the edge identifiers of the edges detected by the edge detection unit 32 are different, and a weak gap penalty is set when the edge identifiers are the same. It is easy to select a path connecting pixels belonging to the partial optimum path. Using the result of edge detection by the edge detection unit 32, a partially optimal route can be obtained appropriately.

  In the present embodiment, the dynamic programming calculation unit 36 multiplies the gap penalty by a coefficient k indicating how many pixels in the radial direction of the two pixels to be calculated are shifted, so that the radial shift is large. In this case, a gap penalty acts strongly, and it becomes difficult to select a partially optimal route. As a result, a route with high continuity is selected as the partially optimal route.

  In the present embodiment, the calculation has only to be started from one of the pixel columns arranged in a predetermined direction from the center of gravity G of the cell nucleus, so that the user does not need to set the starting point for searching for the optimum route.

(Second Embodiment)
Next, a cell contour extracting apparatus according to the second embodiment of the present invention will be described. The basic configuration of the cell contour extracting apparatus of the second embodiment is the same as that of the cell contour extracting apparatus 10 of the first embodiment (see FIG. 1). The cell contour extraction apparatus according to the second embodiment displays the cell nucleus contour and the cell contour obtained by the apparatus on the display unit 16, and corrects the cell nucleus contour and the cell contour based on an instruction input from the input unit 14. It has the function to do.

  The cell contour extracting apparatus according to the second embodiment displays the cell nuclei detected by the cell nucleus contour extracting unit 20 on the stained image, and receives information on the correct region of the cell nuclei from the input unit 14. As an input interface, for example, when there is a portion included in the cell nucleus region, an interface is prepared such that the mouse is right-clicked, and a portion to be excluded from the cell nucleus region is left-clicked. The cell nucleus contour extraction unit 20 performs processing for including or excluding a predetermined region including the clicked position in the cell nucleus region. At this time, in order to improve the processing speed, a region including the cell nucleus to be corrected may be cut out as a region of interest (ROI).

Next, correction of the cell contour by the cell contour extraction apparatus according to the second embodiment will be described.
FIG. 15A is a diagram illustrating an image of the cell contour L1 obtained by the cell contour extraction apparatus according to the second embodiment. From FIG. 15A, it can be visually confirmed that the displayed cell outline L1 is deviated from the true cell outline. When the user designates a point C that is a part of the true cell contour, the cell contour extraction device recalculates the optimum route with the passage through the point C as a constraint.

  Specifically, the dynamic programming calculation unit 36 calculates a partial optimum path starting from the pixel at the point C, and performs calculation until it matches the previously obtained cell contour. In this case, since passing through the point C is given as a constraint, it is not necessary to obtain the partial optimum path for all the pixels included in the pixel array as in the case of obtaining the partial optimum path for the first time. The optimum route is obtained by sequentially connecting the partially optimum routes starting from C.

  When the partially optimal route starting from the point C matches the previously obtained route, the dynamic programming calculation unit 36 stops the calculation of the partially optimized route and replaces the optimum route before the correction instruction with the new optimum route. A path connecting the optimal path determined in step (2) is obtained as a new cell outline. Thereby, as shown in FIG.15 (b), the cell outline L2 which passes along the point C is extracted.

  FIG. 16 is a diagram illustrating an operation of the cell contour extracting apparatus according to the second embodiment. The cell contour extraction apparatus acquires a stained image of the cell (S50), and extracts a cell nucleus contour from the stained image (S52). The processing up to this point is the same as that of the cell contour extraction device of the first embodiment.

  The cell contour extraction apparatus according to the second embodiment displays the extracted cell nucleus contour on the display unit 16 (S54), and accepts a cell nucleus contour correction instruction. If there is a correction instruction (YES in S56), the cell contour extracting apparatus corrects the cell nucleus contour in accordance with the correction instruction (S58). If there is no correction instruction (NO in S56), the process proceeds to the step of extracting the cell outline without correcting the cell nucleus outline.

  Subsequently, the cell contour extracting apparatus according to the second embodiment extracts a cell contour as in the first embodiment (S60). The cell nucleus contour extraction apparatus displays the extracted contour on the display unit 16 (S62), and accepts the designation of the contour from the user. When the designation of the contour is accepted (YES in S64), the cell contour extraction device recalculates the contour passing through the designated point (S60). When the designation of the contour is not accepted (NO in S64), the cell contour extraction device ends the contour extraction process.

  The cell contour extraction apparatus according to the second embodiment can manually correct the cell nucleus contour and the cell contour by the user, so that contour extraction with higher accuracy can be performed.

  As mentioned above, although the cell outline extraction apparatus and the cell outline extraction method of the present invention have been described in detail with reference to the embodiments, the present invention is not limited to the above-described embodiments.

  In the above-described embodiment, the cell contour extraction device and the cell contour extraction method have been described. However, a program for realizing the functions of the above-described cell contour extraction device is also included in the scope of the present invention.

  In the above-described embodiment, the example in which the Sobel filter is used to obtain the density gradient of the pixel has been described. However, the filter for obtaining the density gradient is not limited to the Sobel filter. For example, a Laplacian filter or a Prewitt filter may be used.

  In the above-described embodiment, the example in which different gap penalty values are used according to the difference between the edge identifiers of the two pixels to be calculated has been described. However, a constant value may be used as the gap penalty. In this case, since it is not necessary to perform edge detection, the cell contour extraction apparatus can be realized with a simpler configuration than the embodiment.

  In the above-described embodiment, an example of searching for an optimum route in a circular search region when calculating a cell contour by dynamic programming has been described, but the search region is not limited to a circular shape. For example, based on the shape of surrounding cells, the shape of the cell may be predicted, and a search area where the search can be performed efficiently may be set.

  In the above-described embodiment, an example of extracting the outline of a cell has been described. However, the present invention can also be used for extracting the outline of an object other than a cell. If the contour to be extracted is not a contour surrounding a closed region and the method of calculating around a predetermined point cannot be adopted, dynamic programming with a change in edge identifier as a gap penalty is applied. Even if it only does, there exists an effect which can extract an outline appropriately.

  As described above, the present invention is useful as a cell contour extraction device or the like that extracts a cell contour from a cell staining image, and can be applied to, for example, a pathological diagnosis device or the like.

The figure which shows the structure of the cell outline extraction apparatus of 1st Embodiment. Figure showing an example of a stained cell image (A) The figure which shows the gray scale image of a cell typically (b) The figure which shows the intensity | strength of the pixel value of a BB direction (A) The figure which shows the image of the cell nucleus before performing a convex hull process (b) The figure which shows the image of the cell nucleus after performing a convex hull process (A) The figure which shows the example of the Sobel filter which calculates | requires the density gradient of a horizontal direction (b) The figure which shows the example of the Sobel filter which calculates | requires the density gradient of a vertical direction (A) The figure for demonstrating the edge detection by a canny edge detector (b) The figure which shows the example of the edge detected by the canny edge detector Illustration for explaining dynamic programming Illustration for explaining dynamic programming (A) The figure which shows the density gradient value of each pixel in the expanded view shown in FIG. 8 (b) The figure which shows the edge identifier of each pixel in the expanded view shown in FIG. 8 (c) It is for demonstrating calculation of a partial optimal path | route. Figure (d) Diagram showing an example of a partially optimal route Diagram showing multiple candidate routes and optimal routes that are candidates for the optimal route The figure which shows the flow of operation | movement of the cell outline extraction apparatus of 1st Embodiment. The figure which shows the flow of operation | movement of the cell nucleus outline extraction by the cell outline extraction apparatus of 1st Embodiment. The figure which shows the flow of the operation | movement of the cell outline extraction by the cell outline extraction apparatus of 1st Embodiment. The figure which shows the cell outline extraction result by the cell outline extraction apparatus of 1st Embodiment (A) The figure which shows the cell outline extraction result by the cell outline extraction apparatus of 2nd Embodiment (b) The figure which shows the corrected cell outline The figure which shows the flow of operation | movement of the cell outline extraction apparatus of 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Cell outline extraction apparatus 12 Image acquisition part 14 Input part 16 Display part 18 Image data storage part 20 Cell nucleus outline extraction part 22 Gray scale conversion part 24 Connected component extraction part 26 Convex hull processing part 30 Cell outline extraction part 32 Edge detection part 34 Search area determination unit 36 Dynamic programming calculation unit 38 Cell contour determination unit

Claims (9)

An image acquisition unit for acquiring a stained image of a cell photographed by a microscope;
A cell nucleus detection unit for detecting a region of a cell nucleus from the stained image;
A density gradient calculation unit for obtaining a density gradient with an adjacent pixel for each pixel of the stained image;
A cell contour extraction unit that obtains an optimal route that goes around the cell nucleus by dynamic programming using the concentration gradient as a parameter, and extracts the optimal route as a cell contour;
With
The cell contour extraction unit performs a process of calculating a partial optimum path connecting pixel groups included in two pixel arrays arranged in different directions from a predetermined point in the cell nucleus region, and the direction of the pixel array is A cell contour extraction device that repeatedly performs the pixel arrangement by changing the direction of the pixel arrangement until it makes a round around the predetermined point, and obtains the optimum route based on the partial optimum route.   The cell contour extraction unit obtains a plurality of candidate routes that are candidates for the optimum route based on the partial optimum route, and the candidate route is selected from the candidate routes in the pixel array in which the calculation of the partial optimum route is started. A candidate path whose difference between the distance from the predetermined point of the passing pixel and the distance from the predetermined point of the pixel passing through the candidate path in the pixel array for which the calculation has been completed is equal to or less than a predetermined threshold is determined as the optimal path. The cell contour extraction device according to claim 1, wherein the cell contour extraction device is determined.   The cell contour extraction unit calculates a gap penalty based on a difference in distance from the predetermined point of pixels to be calculated included in different pixel arrays, and the partial contour based on the gap penalty and the density gradient The cell contour extraction device according to claim 1, wherein an optimum route is obtained. A plurality of edges in the stained image is extracted based on the density gradient, and includes an edge detection unit that assigns edge identifiers to pixels constituting each edge,
4. The cell according to claim 3, wherein the cell contour extraction unit calculates the gap penalty based on a difference in distance from the predetermined point of pixels to be calculated included in different pixel arrays and the difference between the edge identifiers. 5. Contour extraction device.   The cell contour extraction device according to claim 1, wherein the cell contour extraction unit excludes a region of the cell nucleus detected by the cell nucleus detection unit from the search region of the optimum route. A display unit for displaying the contour extracted by the cell contour extraction unit on the cell image;
An input unit for receiving position information of an indicated point in the image displayed on the display unit;
With
The cell contour extraction device according to claim 1, wherein the cell contour extraction unit obtains an optimum path through the instructed point. An image acquisition unit for acquiring images;
A density gradient calculation unit for obtaining a density gradient with an adjacent pixel for each pixel of the image;
Based on the density gradient, an edge detection unit that extracts a plurality of edges in the image and assigns edge identifiers to pixels constituting each edge;
A contour extraction unit that calculates a contour of an object included in the image by an optimum route search by dynamic programming using the density gradient as a parameter;
With
The contour extracting unit calculates a gap penalty based on the difference between the edge identifiers, and obtains an optimum path based on the gap penalty and the density gradient. Obtaining a stained image of the cells taken by a microscope;
Detecting a region of a cell nucleus from the stained image;
Obtaining a density gradient with adjacent pixels for each pixel of the stained image;
An optimum path that goes around the cell nucleus is obtained by dynamic programming using the concentration gradient as a parameter, and the optimum path is extracted as a cell outline, from a predetermined point in the region of the cell nucleus A process of calculating a partial optimum path that connects pixel groups included in two pixel arrays arranged in different directions is performed by changing the direction of the pixel array until the direction of the pixel array makes a round around the predetermined point. Repeatedly performing the step of determining the optimal route based on the partially optimal route;
A cell contour extraction method comprising: A program for extracting the outline of a cell from a stained image of the cell,
Obtaining a stained image of the cells taken by a microscope;
Detecting a region of a cell nucleus from the stained image;
Obtaining a density gradient with adjacent pixels for each pixel of the stained image;
An optimum path that goes around the cell nucleus is obtained by dynamic programming using the concentration gradient as a parameter, and the optimum path is extracted as a cell outline, from a predetermined point in the region of the cell nucleus A process of calculating a partial optimum path that connects pixel groups included in two pixel arrays arranged in different directions is performed by changing the direction of the pixel array until the direction of the pixel array makes a round around the predetermined point. Repeatedly performing the step of determining the optimal route based on the partially optimal route;
A program that executes

Images