WO2002057997A1

WO2002057997A1 - Method and arrangement for segmenting white blood cells in a digital colour image

Info

Publication number: WO2002057997A1
Application number: PCT/SE2002/000050
Authority: WO
Inventors: Björn Nilsson
Original assignee: Cellavision Ab
Priority date: 2001-01-18
Filing date: 2002-01-17
Publication date: 2002-07-25
Also published as: WO2002057997A8

Abstract

The invention relates to a method for segmenting white blood cells in a series of images with at least one digital color image that has pixels with at least two color components, where red blood cells are also to be found in the series of images. The method comprises identifying a first and a second set of pixels in the series of images that depict red and while blood cells respectively using a first and a second rough measure; identifying first and second functions that give an approximate description of the relationship between the intensities of the two color components for pixels in the first and second set respectively; and determining whether a given pixel depicts white blood cells or not, on the basis of the given pixel's intensities of the two color components and the first and second functions.

Description

Method and arrangement for segmenting white blood cells in a digital colour image.

Technical Field

The present invention relates to a method for - segmenting objects in a first category in a series of images with at least one digital color image according to the preamble of claim 1. The invention also relates to an arrangement according to the preamble of claim 12, a computer program according to the preamble of claim 13 and a digital storage medium according to claim 14.

Cross reference to related applications: This application claims benefit from Swedish patent application no SE-0100142-9, filed January 18, 2001, and US provisional patent application no US-60/290316 , filed May 11, 2001. Background Art A method for analyzing peripheral blood and bone marrow is to carry out so-called differential counting of white blood cells, also called white blood corpuscles. This is normally carried out by viewing a stained blood smear on a slide through a microscope. White blood cells in the smear are identified and allocated to various categories. On the basis of the number of identified white blood cells in the various categories, an analysis can thereafter be carried out which can lead to a diagnosis being made concerning the organism from which the blood sample was obtained. Such identification and analysis can be carried out manually by an experienced analyst, who is able to distinguish between white blood cells in the various categories. However, a manual method is time-consuming and accordingly expensive. Consequently, a reliable automatic system for carrying out such an analysis is desirable.

In such a system, white blood cells are distinguished, or segmented, in digital microscope images, which comprise a large number of pixels. The distinguished white blood cells are thereafter categorized, for example by utilization of a neural network that has been "taught" to distinguish white blood cells of different categories.

In order for the subsequent classification to work well, it is important that the distinguishing or segment- ing method works well. By a well carried out segmentation of a given blood cell is meant that all pixels that are in an image and that depict the given blood cell are identified, but no other pixels. There can be several blood cells in an image and clustered groups of identi- fied pixels are processed as a segmented object. It is the identified pixels that are forwarded for categorization of the blood cell. If the segmentation is incorrect, that is if pixels that do not depict the given blood cell are forwarded, or when pixels that depict the given blood cell are not forwarded, the appearance of the cell will be distorted, which can result in an incorrect categorization. Too many incorrect categorizations lead to the system becoming less reliable. A well-known method for carrying out segmenting is the use of so-called thresholds. Using thresholds characteristics are compared, for example the gray-scale intensity, in a given pixel with a threshold value, in order to establish whether the pixel has such characteristics that are expected of the pixels that depict the sought object. If a dark type of sought objects is to be distinguished in a black and white image, for example, the gray-scale intensity of the given pixel can be compar- ed with a threshold value appropriate for the sought objects. If the gray-scale intensity of the given pixel is darker than the sought threshold value, this pixel is judged to depict a sought object, otherwise not.

When color images constitute input data for the seg- mentation method, the color information in the images can be utilized. A given pixel in an image has then usually an intensity component for each of the primary colors red (R) , green (G) and blue (B) . These components can be said to define a color vector. Based on this color vector, a so-called RGB projection can be calculated, as the scalar product of the color vector and a predetermined weighting vector.

The color information in an image can, as mentioned, be utilized for segmenting an object of a particular type from a digital color image. A weighting vector appropriate for this type of object and a threshold value appropriate for this type of object can then be used. A given pixel's RGB projection, derived using the weighting vec- tor, is compared with the threshold value in order to determine whether the pixel depicts an object of the sought type or not .

The way of carrying out segmenting described above, for analysis of a blood smear, works relatively well for correctly distinguishing a white blood cell or a red blood cell against the background in an image. If a red blood cell and a white blood cell are adjacent to each other or partially obscure each other, it can, however, be difficult to segment the white blood cell correctly. This is because the difference in color composition and intensity between red blood cells and the cytoplasm of some white blood cells is relatively small. The contrast in the transition between red and white blood cell can thus be weak. In the same way, it is of course difficult to segment a red blood cell, but, as mentioned above, it is primarily the white blood cells that are of interest. The normal procedure is to find a working segmentation condition, consisting of a weighting vector for an RGB projection and a threshold value, with which condition it is possible to distinguish a red and a white blood cell. Such a condition is, however, not particularly robust, but works only under almost constant conditions. In practice, the thickness of the smear can vary, both within one slide and between samples. In addition, the intensity and shade of the stain can vary. Furthermore, different laboratories use many different vari- ants of stains. A constant segmentation condition will therefore give incorrect results in many cases. Summary of the Invention

An object of the present invention is to solve the above-mentioned problems completely or partially.

This object is achieved by a method according to claim 1, an arrangement according to claim 12, a computer program according to claim 13 and a digital storage medium comprising the computer program according to claim 14.

More specifically, a first aspect of the invention relates to a method for segmenting objects in a first category in a series of images with at least one digital microscope color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the series of images . The method is characterized by the steps of identifying a first set of pixels in an image in the series of images using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category; identifying a first function that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set; identifying a second set of pixels in an image in the series of images using a second rough measure, so that the pixels in the second set predominantly depict objects in the first category; identifying a second func- tion that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the second set; and determining whether a given pixel in an image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and the first and second functions .

With such a method, the distinguishing is adapted to the appearance of the sample, so that effective segmentation can be carried out even if there are great variations in, for example, staining and layer thickness within or between images in the series of images . The segmentation is therefore reliable enough for correct classifi- cation to be carried out .

The first and the second function can preferably be essentially linear, which results in a simple method.

The step of determining whether the given pixel depicts an object in the first category or not preferably comprises the partial steps of identifying an element line for the given pixel, which in the intensity plane of the two color components passes through both the given pixel's intensities of both colors and an intersection between the first and the second function; allocating the given pixel an angular value based on the angle of the element line in relation to lines corresponding to said first and second functions; and determining whether the given pixel depicts an object in the first category or not based on the angular value. The use of such angular values has proved to be very reliable for segmentation of biological material .

According to a preferred embodiment, the given pixel's angular value assumes a first extreme value if the element line coincides with the first function's line or has an angle in relation to the second function's line which is greater than the angle between the first and the second function's lines; a second extreme value if the element line coincides with the second function's line or has an angle in relation to the first function's line which is greater than the angle between the first and the second function's lines; and a value between the first and the second extreme value if the element line has smaller angles in relation to the first and the second function's lines than the angle between the first and the second function's lines.

According to a preferred embodiment, the distance between the given pixel's intensities and the intersec- tion of the first and the second function in the plane of the two color components can also be used to determine whether the given pixel depicts an object in the first category or not. This makes the segmentation even more reliable. Objects in the first category are preferably white blood cells and objects in the second category are red blood cells. The method according to the invention has been found to be particularly suitable for segmentation of white blood cells in the presence of red blood cells.

The two color components are preferably red and blue. These primary color components are particularly suitable for segmentation of white blood cells in the presence of red blood cells, at least with the use of so-called MGG (May-Grύnwald-Giemsa) staining or so-called Wright staining.

The first rough measure preferably comprises identi- fying and excluding pixels that depict background by the use of thresholds. This provides a simple and usually adequately reliable method of roughly identifying a set of pixels that depict red blood cells.

The first rough measure can further comprise identi- fying and excluding areas of pixels, which areas can be assumed to contain white blood cells, which assumption is based on the intensity of the green color component. This gives a "purer" set of pixels that depict red blood cells . The second rough measure preferably comprises the identification of at least one pixel cluster with low intensity of the green color component. This is a simple way to identify pixels that depict nuclei of white blood cells . According to one embodiment, the second rough measure can further comprise the partial steps of selecting an area associated with the pixel cluster, which area has a size that is larger than the expected size of a white blood cell; selecting at least one pixel in the area, which pixel has an appearance that differs from the expected appearance of pixels that depict white blood cells, said at least one pixel constituting an initial quantity; determining object similarity values for pixels in the area; and identifying the second set of pixels based on the initial quantity and the object similarity values. This rough measure provides more pixels from the cytoplasm and thus a better selection. A second aspect of the invention relates to an arrangement for segmenting objects in a first category in a series of images with at least one digital color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the series of images. The arrangement is characterized by means for identifying a first set of pixels in an image in the series of images using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category; means for identifying a first function that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set; means for identifying a second set of pixels in an image in the se- ries of images using a second rough measure, so that the pixels in the second set predominantly depict objects in the first category; means for identifying a second function that gives an approximate description of the rela- tionship between the respective intensities of the two color components for pixels in the second set; and means for determining whether a given pixel in an image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and the first and second functions.

The arrangement has corresponding advantages to those of the method, and can be varied in a similar way. A third aspect of the invention relates to a computer program for segmenting objects in a first category in a series of images with at least one digital color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category are also being found in the series of images. The computer program is characterized by instructions corresponding to the steps of identifying a first set of pixels in an image in the series of images using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category; identifying a first function that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set; identifying a second set of pixels in an image in the series of images using a second rough measure, so that the pixels in the second set predominantly depict objects in the first category; identifying a second function that gives an approximate de- scription of the relationship between the respective intensities of the two color components for pixels in the second set; and determining whether a given pixel in an image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and the first and second functions.

A fourth aspect of the invention relates to a digital storage medium containing such a program. The computer program and thus the storage medium have corresponding advantages to those of the method, and can be varied in a similar way. Brief Description of the Drawings

Fig. 1 shows an example of a microscope image in which the present invention can be used.

Fig. 2 shows a diagram in which the pixels in the image in Fig. 1 have been plotted based on their respective intensities of red and blue.

Fig. 3 shows a red-blue plane in which two approxi- mating functions have been identified according to an embodiment of the invention.

Fig. 4 shows an example where angular values have been determined for the pixels shown in the image in Fig. 1. Figs 5a and 5b show an arrangement according to an embodiment of the invention.

Fig. 6 shows a flow chart for a method according to an embodiment of the invention. Description of Preferred Embodiments

Fig. 1 shows an example of a microscope image in which the present invention can be used. In the image there is a set of blood cells of varying types in a stained blood smear (stained in accordance with the MGG method) . There are four white blood cells VI - V4, one of which (V3) probably indicates a diseased state. In the image there are also a large number of red blood cells, Rl, R2, etc. In addition, there are in the image a number of thrombocytes T, which appear as small dark spots. The blood cells appear against a background B. The image shown in Fig. 1 is a black and white version of a color image. As is known by those skilled in the art, in such a color image white blood cells appear as slightly more blue than the background and the red blood cells. The contrast between the cytoplasm of the white blood cells and the background therefore appears clearer in the color image on which a method according to the invention is used than in the black and white image reproduced here . It is, as mentioned, particularly difficult to segment certain categories of white blood cells when their cytoplasm is adjacent to red blood cells. This is the case with the white blood cell VI, which is adjacent to the red blood cell Rl . The present invention relates to a reliable and effective way of carrying out such a segmentation.

Fig. 2 shows a diagram in which all the pixels in the image in Fig. 1 have been plotted based on their re- spective intensities of red and blue. A pixel's red intensity is put on the vertical axis and the blue intensity on the horizontal axis. In the diagram the plotted pixels can be likened to two elongated "clouds" 11, 13, which meet in an inverted V shape. It appears that the first cloud 11, which has greater variation in red intensity, contains almost exclusively pixels that depict white blood cells and background. The second cloud 13, which is more concentrated, contains pixels that depict red blood cells and background. The pixels in both clouds that depict the background are concentrated in a small area around a background central point 15. A white cell central point 16 can also be determined for the pixels that depict white blood cells, and a corresponding red cell central point 17 can be determined for the pixels that depict red blood cells. It can be said that the pixels of the red and white blood cells respectively deviate from the background central point in two different directions. This is because the red and the white blood cells are affected in different ways by the staining.

According to an embodiment of the present invention, a first function, a white-function 12 in the red-blue plane, is determined, which function approximately describes the relationship between the intensities of red and blue in pixels that depict white blood cells.

Similarly, a corresponding second function, a red-function 14 for the pixels that depict red blood cells, to be determined. White blood cells are then said to constitute objects in a first category, and red blood cells objects in a second category. On the basis of these red- and white-functions, it can thereafter be determined with great certainty, whether a given pixel depicts a red or a white blood cell.

These red- and white-functions can be determined from smaller sets of pixels that depict red and white blood cells respectively. These sets can be identified using rough measures, so that a majority of the pixels in a set depict objects of the required category.

IDENTIFICATION OF PIXELS THAT DEPICT RED BLOOD CELLS

Unlike white blood cells, it can be assumed that red blood cells will be found in each image in a series of images comprising microscope images of a blood smear. The most suitable way of finding a set of pixels that depict red blood cells in an image is to exclude the pixels that can be assumed to depict the background. Of the remaining amount of pixels in an image, a clear majority will then depict red blood cells. Individual small thrombocytes and possibly some white blood cells can be found, but only a small part of the remaining pixels depict such objects. The pixels that depict the background can suitably be excluded by the use of thresholds . It is then assumed that the pixels that are lighter than a certain threshold value depict the background. It can, as will be described later, be expedient also to record the characteristics of the pixels that are excluded by the use of thresholds. It can, for example, be advantageous to calculate the background central point 15, as will be described below.

The pixels that are not excluded thus depict principally red blood cells and constitute a roughly selected set of pixels according to an embodiment of the invention.

It is also possible to exclude pixels that can depict white blood cells or thrombocytes, in order to further refine the selected set. Thrombocytes can usually be removed by the use of thresholds, as they are darker in the green component than red blood cells.

Nuclei in white blood cells have also usually a markedly low or dark green component and can thereby usually be found. Thereafter, it is possible to exclude an area around such nuclei, in order to remove the whole white blood cell . The size of the area is then preferably of the order of magnitude that white blood cells tend to be at the level of magnification used, or larger. Identification of a set of red blood cells does not normally need to be carried out for each image in a series of images . DETERMINATION OF APPROXIMATE FUNCTION FOR RED BLOOD CELLS

Based on the roughly identified set of pixels that depict red blood cells, an approximate function is there- after to be determined, which function describes the relationship between the red and the blue intensity of pixels in the set . This is named the red cell function 14. The function can be essentially linear, that is of the form R = k_r-B+m_r, where R and B describe the red and the blue intensity respectively. The parameters k_r and m_r are then to be determined. For the set of pixels that de- pict red blood cells, this can be carried out by ordinary linear regression, which is an operation well-known to those skilled in the art.

Alternatively, the function R = k_r-B+m_r can be determined as the function that passes through both the back- ground central point 15 and the red cell central point 17.

The background central point is calculated as the mean values of the blue and red intensities respectively for the set of pixels that depict the background. The red cell central point is calculated as the mean values of the blue and red intensities respectively for the set of pixels that depict red blood cells. The central points thus correspond to mean values. Alternatively, points in the red-blue plane that are based on median values or mo- dal values can be used. Both these concepts are well- known to those skilled in the art.

Irrespective of whether the above-mentioned function is determined by linear regression or by utilizing the above-mentioned central points, it is advantageous in as- sociation with this to determine a measure of the uncertainty in the estimation of the function. This provides a measure of reliability for the whole operation. Such a measure of uncertainty can be based on the standard error of the estimation for the red cell central point or on the standard errors for the respective regression parameters .

IDENTIFICATION OF PIXELS THAT DEPICT WHITE BLOOD CELLS Rough identification of pixels that depict white blood cells and determination of an approximate function that is associated with these pixels can preferably be carried out for each white blood cell that appears in a series of images. This is preferable as white blood cells as a group show large variations.

A first rough method of identifying pixels that depict white blood cells is to utilize the previously mentioned low green component that is found in the white blood cell's nucleus. Pixels with a green intensity below a particular threshold value are thus included in the set of pixels that are considered to depict white blood cells. This provides principally pixels that depict the nucleus of a white blood cell .

A second more precise method for identifying pixels that depict white blood cells is to utilize an active contour model which is so arranged that it starts from an area outside the white blood cell . This can be carried out as follows.

Firstly, as in the first method, coherent areas of pixels in the digital image are identified that have a green component below a particular threshold value, which coherent areas are of a particular minimum size. These can be assumed to depict the nucleus of a white blood cell. Thereafter, an area is selected in the digital image which has the nucleus at its center and which is sufficiently large to include with great certainty the whole white blood cell. The pixels in the selected area are used for further processing.

Thereafter object similarity values are determined for the pixels in the selected area with regard to red and blue intensity. The object similarity value of a pixel is a measure of how similar this pixel is to the expected pixels in the white blood cell.

The determination of this object similarity value can be carried out based on the function 14 already determined for red blood cells and the central point 15 determined for background pixels. Each given pixel in the selected quantity has a corresponding position in the diagram in Fig. 2. The given pixel can be allocated an object similarity value that is a (not necessarily linear) function of the angle between a line that is drawn through this position and the background central point 15 and the line of the red cell function 14. This is carried out in such a way that small angles give small object - similarity values, and large angles give large object - similarity values.

Thereafter an initial quantity is selected from the pixels in the area. The initial quantity is selected from the pixels that have the lowest object similarity values, so that the initial quantity with great certainty does not include pixels in the white blood cell. Pixels in the initial quantity are allocated a first object likelihood value, for example zero. The object likelihood value is a measure of how probable it is that a pixel, put in context, depicts a sought object. The initial quantity is thereafter allowed to increase by including pixels that are adjacent to the initial quantity. When a new pixel is included in the quantity, it is allocated an object likelihood value which is a function (for example the sum) of its object similarity value and the object likelihood value of an adjacent, already included, pixel. The object likelihood values thus become higher the later a pixel is included.

This is carried out preferably until all the pixels in the image have been allocated object likelihood val- ues. The object likelihood value of a given pixel then depends on its distance from the initial quantity before this has started to grow, on its object similarity value and on object similarity values of any pixels that are located between the given pixel and the initial quantity before it has started to grow.

It is also possible to carry out the above-mentioned determination of object likelihood values approximately, by applying a modified distance transform. The distance transform will then be modified in such a way that it de- termines an object likelihood value for a given pixel, which object likelihood value depends on the object - similarity value of the given object, the Euclidean distance to the initial quantity and object similarity val- ues of pixels that are located between the given object and the initial quantity.

Based on the object likelihood values that pixels in the selected area are allocated, the pixels are selected that will subsequently be used as the basis for the white cell function.

DETERMINATION OF APPROXIMATE FUNCTION FOR WHITE BLOOD CELLS

Based on the roughly identified set of pixels that depict white blood cells, an approximate function is thereafter to be determined, which function describes the relationship between the red and the blue intensity of pixels in the set. This is named the white-function 12. The function can also be essentially linear, that is of the form R = k_v-B+m_v, where R and B describe the red and the blue intensity respectively for pixels that depict white blood cells. The parameters k_v and m_v are then to be determined. For the set of pixels that depict white blood cells this can preferably be carried out by ordi- nary linear regression, which is an operation well-known to those skilled in the art.

Alternatively, the white-function R = k_v-B+m_v can be determined as the function that passes through both the background central point 15 and the white cell central point 16.

The white cell central point 16 is calculated in a corresponding way to the red cell central point 17. Irrespective of whether the above-mentioned white- function is determined by linear regression or by utilizing the above-mentioned central points, it is advantageous in association with this to determine a measure of the uncertainty in the estimation of the function. Such a measure can be based on the standard error of the estimation for the white cell central point 16 or on the - standard errors for the respective regression parameters. USE OF RED AND WHITE CELL FUNCTIONS Fig. 3 shows a red-blue plane in which two approximating functions, a red cell function 14 and a white cell function 12, have been identified in accordance with an embodiment of the invention.

According to the invention, it is to be determined whether a pixel in a digital microscope image depicts an object in a first category, that is in this case white blood cells, or not. This is carried out on the basis of the pixel's respective intensities of the two color components (in this case red and blue) and first and second functions (in this case red-function and white-function respectively) .

Fig. 3 shows as an example four different pixels, 30, 32, 33, 34, which have been plotted in the diagram in accordance with their respective red and blue inten- sities. With reference to Fig. 2, it can be assumed that the pixels 33 and 34 with great probability depict a white blood cell and a red blood cell respectively. This cannot be said with equally great certainty for the pix- els 30 and 32. The pixels 30 and 32 possibly depict a white cell and a red cell respectively.

Angular values are preferably determined for the pixels in a digital image. A corresponding element line 31 has been drawn for the pixel 30. The element line 31 for the pixel 30 is such that in the intensity plane of the two color components (red, blue) it passes through both the pixel's 30 intensities of both colors and an intersection 15 between the red-function and the white- function.

An angular value for a pixel 30 is based on the angles (β and a respectively) between the pixel's element line 31 and the red-function 14 and the white-function 12 respectively. Based on the amount of this angular value, it can be determined whether the pixel depicts an object in the first category or not.

The angular value can preferably be normalized. A given pixel's angular value then assumes a first extreme angular value, for example 0, if the pixel's element line coincides with the white-function. The same also preferably applies if the element line's angle β in relation to the red-function is larger than the angle between the red- and white-functions . The pixel 33 has thus the angular value 0. A given pixel's angular value assumes a second extreme angular value, for example 1, if the pixel's element line coincides with the red-function. The same also preferably applies if the element line's angle a in rela- tion to the white-function is larger than the angle between the red- and white-functions. The pixel 34 has thus the angular value 1.

If a pixel's element line has smaller angles (β and oi respectively) in relation to the lines of the red-function 14 and the white-function 12 than the angle between the functions, the angular value assumes a value between the first extreme value and the second extreme value, that is between 0 and 1. Preferably this depends upon the ratio between both angles a and β. The pixel 30 could therefore be given the angular value 0.3, while the pixel 32 could be given the angular value 0.9.

The distance x to the intersection 15 of the red- function 14 and the white-function 12 in the two color components' plane, can also be used to determine whether the given pixel 30 depicts an object in the first category or not .

Fig. 4 shows an example where angular values have been determined for the pixels in the image in Fig. 1. Pale pixels have there a low angular value. As the Figure shows, the contrast between the cytoplasm of the white blood cells and the adjacent red blood cells is now considerably stronger than in Fig. 1. The angular values in Fig. 4 can therefore be utilized to carry out better seg- mentation of the white blood cells in Fig. 1. For example, the angular values of the pixels can be used as determinants in the threshold method. Fig. 5a shows an application of an arrangement according to the invention. The arrangement is incorporated in a system with a microscope 51 and a computer system 52. The microscope 51 provides digital color images to the computer system 52, which carries out segmentation of objects of interest in the digital images. The computer system 52 and the microscope 51 can be integrated. Parts of the segmentation method can also be carried out distributed over a network. Fig. 5b shows an arrangement ac- cording to an embodiment of the invention. The arrangement incorporates a number of functional modules 53 - 57.

In a first module (ID1) 53, a first set of pixels in a digital image is identified using a first rough measure, so that the pixels in the first set predominantly depict objects in a particular category, for example red blood cells. The first module 53 consists of means for carrying out such an identification.

In a second module (FCN1) 54, a first essentially linear function is identified, which function gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set . The second module 54 consists of means for carrying out such an identification.

In a third module 55 (ID2) , a second set of pixels in a digital image is identified using a second rough measure, so that the pixels in the second set predominantly depict objects in a particular category, for exam- pie white blood cells. The third module 55 consists of means for carrying out such an identification.

In a fourth module (FCN2) 56, a second essentially linear function is identified, which function gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the second set. The fourth module 56 consists of means for carrying out such an identification.

In a fifth module (SEGM.) 57, it is determined whether pixels in an image depict an object in the first category or not, on the basis of their respective intensities of the two color components and said first and second functions. The fifth module 57 consists of means for making such a determination and outputs objects that have been segmented out of a digital image for further analysis .

All the modules described above can be realized by means of computer programs in the computer system 52. Such a computer program can be stored separately on a digital storage medium.

It is, however, also possible to implement some of the modules as circuit solutions, for example in the form of ASIC or FPGA circuits (ASIC = Application Specific Integrated Circuit; FPGA = Field Programmable Gate Array) .

Fig. 6 shows a flow chart for a method 60 according to an embodiment of the invention. The method is used for distinguishing/segmenting objects in a first category in a series of images with at least one digital color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the se- ries of images .

In a first step 61, a first set of pixels in an image in a series of images is identified using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category. By a measure is meant here and in the following a rule that is applied to a quantity of input data, that is in this case pixels in an image. The rule gives for a pixel the result either that the pixel depicts an object in the second category or that the pixel does not do so . In a second step 62, a first essentially linear function is identified, which function gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set . In a third step 63, a second set of pixels in an image in a series of images is identified using a second rough measure, so that the pixels in the second set predominantly depict objects in the first category.

In a fourth step 64, a second essentially linear function is identified, which function gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the second set . In a fifth step 65, it is determined whether a given pixel in an image in a series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color compo- nents and said first and second functions.

The series of images that is considered above contains preferably several images, but can also contain only one image. It is not necessary to carry out all the parts of the method on one and the same image. The first and the second step can be carried out on a first image, the third and fourth on a second image in the series of images, and the fifth step on a third image.

The invention is not limited to the embodiments described above, but can be varied within the scope of the appended claims.

Claims

1. A method for segmenting objects in a first category in a series of images with at least one digital mi- croscope color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the series of images, c h a r a c t e r i z e d by the steps of - identifying (61) a first set of pixels in a color image in the series of images using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category;

- identifying (62) a first function, which function gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the first set;

- identifying (63) a second set of pixels in a color image in the series of images using a second rough mea- sure, so that the pixels in the second set predominantly depict objects in the first category;

- identifying (64) a second function, which function gives an approximate description of the relationship between the respective intensities of the two color compo- nents for pixels in the second set; and

- determining (65) whether a given pixel in a color image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and said first and second functions.

2. A method according to claim 1, in which said first and second functions are essentially linear.

3. A method according to claim 2, in which the step of determining whether said given pixel depicts an object in the first category or not comprises the partial steps of

- identifying an element line for the given pixel which in the intensity plane of the two color components passes through both the given pixel's intensities of both colors and an intersection between the first and the second function;

- allocating the given pixel an angular value based on the angle of said element line in relation to lines corresponding to said first and second functions;

- determining whether the given pixel depicts an object in the first category or not based on the angular value .

4. A method according to claim 3, in which the given pixel's angular value assumes a first extreme value if the element line coincides with the first function's line or has an angle in relation to the second function's line which is greater than the angle between the first and the second function's lines; assumes a second extreme value if the element line coincides with the second function's line or has an angle in relation to the first function's line which is greater than the angle between the first and the second function's lines; and assumes a value between the first and the second extreme value if the element line has smaller angles in relation to the first and the second function's lines than the angle between the first and the second function's lines.

5. A method according to claim 4, in which the distance between the given pixel's intensities and the intersection of the first and the second function in the plane of the two color components is also used to deter- mine whether the given pixel depicts an object in the first category or not.

6. A method according to any one of the preceding claims, in which said objects in the first category are white blood cells and said objects in the second category are red blood cells.

7. A method according to any one of the preceding claims, in which said two color components are red and blue.

8. A method according to any one of claims 6 or 7, in which said first rough measure comprises identifying and excluding pixels that depict background by the use of thresholds.

9. A method according to claim 8, in which said first rough measure further comprises identifying and ex- eluding areas of pixels, which areas can be assumed to contain white blood cells, which assumption is based on the intensity of the green color component.

10. A method according to any one of claims 6-9, in which said second rough measure comprises the identification of at least one pixel cluster with low intensity of the green color component .

11. A method according to claim 10, in which said second rough measure further comprises the partial steps of

- selecting an area associated with said pixel cluster, which area has a size that is larger than the ex- pected size of a white blood cell;

- selecting at least one pixel in the area, which pixel has an appearance that differs from the expected appearance of pixels that depict white blood cells, said at least one pixel constituting an initial quantity; - determining object similarity values for pixels in the area; and

- identifying said second set of pixels based on the initial quantity and the object similarity values.

12. An arrangement for segmenting objects in a first category in a series of images with at least one digital color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the series of images, c h a r a c t e r i z e d by

- means (53) for identifying a first set of pixels in a color image in the series of images using a first rough measure, so that pixels in the first set predominantly depict objects in the second category;

- means (54) for identifying a first function that gives an approximate description of the relationship be- tween the respective intensities of the two color components for pixels in the first set;

- means (55) for identifying a second set of pixels in a color image in the series of images using a second rough measure, so that the pixels in the second set pre- dominantly depict objects in the first category;

- means (56) for identifying a second function that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the second set; and - means (57) for determining whether a given pixel in a color image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and said first and second functions.

13. A computer program for segmenting objects in a first category in a series of images with at least one digital color image that depicts biological material and comprises a plurality of pixels that have at least two color components, objects in a second category also being found in the series of images, chara c t e r i z ed by instructions corresponding to the steps of

- identifying a first set of pixels in a color image in the series of images using a first rough measure, so that the pixels in the first set predominantly depict objects in the second category;

- identifying a first function that gives an approximate description of the relationship between the re- spective intensities of the two color components for pixels in the first set;

- identifying a second set of pixels in a color image in the series of images using a second rough measure, so that the pixels in the second set predominantly depict objects in the first category;

- identifying a second function that gives an approximate description of the relationship between the respective intensities of the two color components for pixels in the second set; and - determining whether a given pixel in a color image in the series of images depicts an object in the first category or not, on the basis of the given pixel's respective intensities of the two color components and said first and second functions.

14. A digital storage medium, comprising a computer program according to claim 13.