Computer-implemented method for biological sample images

Abstract

Computer-implemented method for biological sample images The present invention relates to a computer-implemented method for automatically analyzing a digital image (103) of a biological sample (101). The method comprises a change (203) from a first color space to a second color space. Then, a fitting is performed taking into account multiple dimensions of the second color space to classify pixels.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for classifying pixels of a digital image of a biological sample. Background Art

[0002] Methods for automatically processing images of biological samples are well known. However, automatic classification of pixels (e.g., automatic classification based on pixel color) is usually unsatisfactory. For example, pixels located in areas that are actually not stained are often incorrectly classified as stained.

[0003] The document MEAS-YEDID V et al.: “Automatic color space selection for biological image segmentation”, Pattern Recognition, 2004, ICPR2004, Proceedings of the 17th International Cambridge Conference, Piscataway, NJ, USA, August 23-26, 2004, United Kingdom; IEEE, Los Alamitos, California, USA, Volume 3, August 23, 2004 (2004-08-23), pages 514-517, describes automatic color space selection for biological image segmentation. Summary of the invention

[0004] Therefore, an object of the present invention is to provide a method for automatic processing of images of biological samples, which method improves the accuracy of pixel classification.

[0005] According to a first aspect, the object of the invention is a method for classifying pixels of a digital image of a biological sample, the method comprising the following steps:

[0006] (a) receiving a first digital image representing a biological sample to obtain a value X1 associated with each pixel in the first digital image; and obtaining at least one other value X1 for each pixel from the first digital image and / or from another digital image i , i is an integer such that 2≤i≤n, n is an integer such that n≥2;

[0007] (b) Use matrix multiplication to change the space

[0008] I=A x X

[0009] To obtain a second digital image, X is a value including X1 to X n An n-row, 1-column matrix, I is a matrix that includes values ​​from I1 to I m is a matrix of m rows and 1 column, where m is an integer equal to or greater than 2, and I1 to I mis a value associated with each pixel in the second digital image, and A is a matrix of m rows and n columns;

[0010] (c) based on the values ​​of I1 and I2 of at least some pixels in the second digital image j (where j>1 and j≤m) to determine the classification condition; and

[0011] (d) classifying pixels in the second digital image that meet the classification condition into the first category;

[0012] Wherein, step (c) comprises:

[0013] determining a set of one or more distributions, the set comprising at least one distribution of the number of pixels in the second digital image as a function of I1, and the set comprising as a function of I j at least one distribution of the number of pixels in the second digital image that is a function of (where j>1 and j≤m),

[0014] fitting a distribution of the set to determine at least one fitting parameter, and

[0015] Determining a classification condition according to the at least one fitting parameter.

[0016] In the method according to the invention, for each pixel of the first digital image, several colors of the first digital image are combined in a common matrix X, or at least one color of the first digital image is combined with at least one color of another digital image in a common matrix X. Thus, the matrix integrates the relevant information in a mathematically easy-to-use tool in a first space. Then, the matrix X is transformed to obtain information in a second space that is easy to process information, and the matrix I is obtained according to this transformation. A frequency analysis is then performed to determine the classification conditions, which frequency analysis includes fitting one or more distributions about at least two coefficients of the matrix I (i.e., two dimensions in the second space). For the pixels of the second digital image, the classification conditions are checked, for example, these pixels are considered in turn to classify (or not classify) the pixels into the first category.

[0017] In the present invention, it is very meaningful to use information from multiple dimensions in the second space to determine the classification conditions. In fact, the intersection of such different information sources makes it possible to obtain particularly reliable classification conditions.

[0018] Even if the pixel value X in all embodiments of the present invention i and I j Instead of the actual color, you can also use the pixel value X i and I j The matrix X can be viewed as defining pixels in a first color space, and the matrix I can be viewed as defining pixels in a second color space.i With pixel value I j The spatial change between them can be called a change in color space.

[0019] When at least one X is obtained from another digital image i When (i ≥ 2), pixels from the first digital image and the other digital image are matched in such a way that there is overlap between structures or objects in the first digital image and structures or objects in the other digital image. The above method is also applicable when more than two images are used as input to the method. When the first digital image and the other digital image are used to determine the value of X, the second digital image is a combination of information from the first digital image and the other digital image.

[0020] In the method according to the invention, at least one distribution (as a function of at least one of the elements of the matrix I) is taken into account to determine the classification condition. The at least one distribution is included in a set of one or more distributions. One distribution of the set is a function f(I1) of I1. One distribution of the set is I j (m≥j>1), where the distribution may be the same as the distribution that is a function of I1 [and may be written as f(I1, I2)], or the distribution may be another distribution g(I j )(m≥j>1). The one or more distributions may be functions of other coefficients of the matrix I.

[0021] Using a distribution makes it possible to take into account information from the full range of color values. For example, one could use a one-dimensional distribution of I1 and I2 or I3 or any I j (m≥j>1); or I1 and I2 or I3 or any I j (m≥j>1) two-dimensional distribution.

[0022] The classification conditions preferably include the values ​​of the considered pixel in several dimensions of I and reference values ​​in said dimensions of I.

[0023] Preferably, the classification conditions include both I1 and I i The condition for the value of the considered pixel in . For example, the classification condition can be

[0024]

[0025] in and R are the reference values ​​obtained according to step (c), and X=(I1, I2) is the pixel under consideration.

[0026] The fit can generally determine approximately the color ranges corresponding to the color or stain of the pixel. The fit parameters quantify these color ranges. The classification criteria preferably reflect the separation between the color ranges.

[0027] In the context of this document, the biological sample is preferably a tissue section or a plurality of consecutive tissue sections (e.g., with a spacing of 2 to 5 μm). The biological sample may also be, for example, a biological tissue, a blood sample, one or several cells, etc. The method according to the invention may be used in histology, cytology, hematology, etc.

[0028] The determination of a set of one or more distributions, wherein the set includes at least one distribution of the number of pixels in the second digital image as a function of I1, wherein the set includes at least one distribution of the number of pixels in the second digital image as a function of I1. j At least one distribution of the number of pixels in the second digital image that is a function of (where m≥j>1).

[0029] The method according to the invention is particularly easy to implement in software. When executed by an electronic device such as a computer, the method is also particularly fast.

[0030] The method may comprise another further step, in particular any further step between steps (a) and (b) including, for example, image processing. A step of increasing or decreasing the resolution may be performed between steps (a) and (b). Steps (b) to (d) may then be performed for pixels having a pixel size different from that of the originally captured image.

[0031] Preferably, the steps of the method are performed in the following order: (a), (b), (c) and (d). However, it is possible, for example, to perform part of step (d) before performing part of step (c).

[0032] The method according to the invention may at any stage comprise selecting at least a portion of a first digital image, another digital image or a second digital image as "one or more portions of interest". Preferably, the method continues only for the selected one or more portions. Preferably, the selection is performed between steps (a) and (b). For example, in an embodiment of the invention, the following steps are included between steps (a) and (b) or before step (a): a biological practitioner or software selects one or more regions of interest on the first digital image. This facilitates the removal of apparently healthy areas as well as artifacts (folds, spots, etc.). Preferably, steps (b), (c) and (d) are then performed only on one or more regions of interest.

[0033] In an embodiment of the present invention, at least a portion of the first digital image or the second digital image is selected as "one or more portions of interest" based on the RGB color values ​​of the pixels between steps (a) and (b). For example, thresholding based on the color values ​​of the pixels in terms of blue, green or red may be included. This is particularly important for cancer, and a region of interest corresponding to the epithelium may be selected by thresholding the distribution of the blue color value.

[0034] Preferably, based on the values ​​of I1 and I j The classification condition is determined by the value of (m≥j≥1).

[0035] In the framework of this article, a digital image is a matrix of pixels. Some digital images may be color digital images in which each pixel is associated with the values ​​of three colors.

[0036] Preferably, each pixel is present in both the first digital image and the second digital image, but each pixel has a different value in the first digital image and the second digital image.

[0037] In the first picture, the three values ​​associated with each pixel are: a value for red, a value for green, and a value for blue. The first picture can be considered to be in a first color space (a red, green, and blue (RGB) color space). In the second picture, the three values ​​associated with each pixel are: a value for I1, a value for I2, and a value for I3. The second picture can be considered to be in a second color space. The second color space is not an RGB color space. The second color space can be referred to as an Ohta color space.

[0038] The color value I1 may be referred to as “intensity.” The color value I2 may be referred to as “red relative to blue.” The color value I3 may be referred to as “magenta relative to green.”

[0039] In the framework of this article, the classification condition may include several specific conditions. Preferably, the classification condition is met if all the specific conditions are met.

[0040] In the present framework, the distribution is preferably a histogram.

[0041] In an embodiment of the present invention, steps (a), (b), (c) and (d) are fully automated.

[0042] The method according to the invention is particularly suitable for "stained" biological samples, i.e. biological samples colored by any staining method or naturally colored. The staining method can be based on, for example, proteins, DNA, RNA, cell differentiation, etc. The staining method can involve coloring by dyes and / or can involve fluorescence and / or visible light, i.e. hematoxylin, eosin, saffron. The staining method can involve chromogenic compounds. The staining method can involve antibody / antigen reactions.

[0043] Any biological or histological, especially histochemical staining method can be used for staining. Preferably, staining is performed by a staining device. For example, staining can be blue (nuclei counterstained with hematoxylin) and brown (proteins from various sources detected by traditional peroxidase / DAB system immunoassay). For example, staining can allow detection of proliferating cells (detection of ki67 or cyclin A), apoptotic cells (cleaved caspase 3, cleaved cytokeratin 18, etc.), membrane antigens (growth factor receptors, adhesion receptors, death receptors, etc.), cytoplasmic proteins such as Bcl-2 family proteins, nuclear antigens (transcription factors, etc.), etc.

[0044] In an embodiment of the present invention, the biological sample is a tissue section obtained from a tissue block suspected of having cancer.

[0045] In step (a) any digital photography method may be used, for example using a camera or a scanner. The picture preferably represents the intact biological sample.

[0046] In an embodiment of the present invention, the first class includes pixels that are classified as representing a first type of structure or a first type of object (e.g., a tumor region). Pixels that are not classified as the first class are considered not to represent a first type of structure or a first type of object. For example, pixels of the first class may correspond to "labeled," "colored," or "stained" pixels, while pixels that do not belong to the first class are considered "uncolored," "unstained," or "background" pixels.

[0047] The first class of pixels may also correspond to pixels that are colored or dyed with a first color, whereas pixels not belonging to the first class may be uncolored, undyed or may be colored or dyed with another color, for example dyed with another color.

[0048] In an embodiment of the present invention, a method for segmenting the first digital image into a matrix region and a tumor region may also be performed, for example, in parallel with the method for classifying pixels according to the present invention. Such a segmentation method is, for example, the method described in WO2004084135. An embodiment of the present invention includes a process for classifying pixels according to the present invention and a step of determining a first class of pixels belonging to a tumor region.

[0049] After the method according to the invention, the first class of pixels may be used for further analysis. The further analysis may include allocating pixels (preferably, first class of pixels) as macro pixels. The macro pixels are preferably hexagonal. The allocation is preferably tiling. In an embodiment of the invention, the further analysis includes allocating first class of pixels belonging to a tumor region as macro pixels.

[0050] A histogram of the macropixels can then be determined. This histogram can be used, for example, for tumor classification or for heterogeneity studies of tumors.

[0051] In an embodiment of the present invention, step (c) includes determining a further classification condition, and step (d) includes classifying pixels of the second category in the second digital image that satisfy the further classification condition. For example, the classification condition may correspond to a first color value range in a color space of the second image, and the further classification condition may correspond to a second color value range in the color space of the second image.

[0052] In an embodiment of the present invention, the set of one or more distributions includes a function of I1 and I j (m≥j>1). The distribution preferably has at least two dimensions. The distribution can be fitted by a sum of m-dimensional Gaussians, for example.

[0053] In a first embodiment of the present invention, the set of one or more distributions includes a function of I1 and I j (m≥j>1). The set may, for example, include four distributions depending on I1 and I2 or depending on I1, I2 and I3.

[0054] In a second embodiment of the present invention, the set of one or more distributions includes a first distribution that is a function of I1 and a distribution that is a function of I j (m≥j>1). The first distribution and the second distribution may be one-dimensional.

[0055] In an embodiment of the present invention, the classification condition for a considered pixel is related to the value of the considered pixel in at least one of I1, I2 and I3.

[0056] This makes it possible to easily classify pixels according to their values ​​in the second color space.The "considered" pixel is the pixel that is currently being classified.

[0057] Preferably, the classification condition comprises a threshold of at least one of I1, I2 and I3 or a threshold of a combination of at least two of I1, I2 and I3. In an embodiment of the present invention, the classification condition is that the pixel must have an I1 value higher than a threshold, and the threshold has been determined based on the distribution of I1.

[0058] In an embodiment of the present invention, the classification condition is related to the value of the pixel considered in I1 and at least one I j (where m ≥ j > 1). In other words, the dimensions used for fitting and classification are the same.

[0059] Preferably, the classification condition includes a threshold value of I1 and a threshold value of at least one of I2 and I3 or a threshold value of a combination of I1, I2 and I3. In an embodiment of the present invention, the classification condition is that the value of I1 of the pixel must be higher than a first threshold value and the value of I2 of the pixel must be higher than a second threshold value, the first threshold value has been determined based on the distribution of I1, and the first threshold value has been determined based on the distribution of I2.

[0060] In an embodiment of the present invention, empirical calibration parameters are used in determining classification conditions, which have been determined by showing a biological practitioner calibration biological samples prepared in the same manner as the biological samples.

[0061] There may be multiple empirical calibration parameters. Preferably there are two empirical calibration parameters. More preferably, one calibration parameter corresponds to I1 and one calibration parameter corresponds to I j , where I j It is the color value used to determine the classification condition. The determination of the empirical calibration parameters can be much earlier than the use of the claimed method. Preferably, in the preparation of the calibration biological sample and the preparation of the biological sample analyzed by the claimed method, the preparation conditions including, for example, staining tools and staining dyes are the same. "Calibration biological sample" can also be referred to as "simulated biological sample". The use of empirical calibration parameters makes it possible to take into account the experience of biological practitioners as to which pixels they consider to be "colored" and "uncolored". Therefore, the classification method according to the present invention can be calibrated. For example, if the classification condition involves a threshold, the use of empirical calibration parameters can affect the value of the threshold. Preferably, each color value used in step (c) has a calibration parameter. More preferably, there is a calibration parameter for I1 and a calibration parameter for I2 or I3.

[0062] For example, for stained tissue sections, using empirical calibration parameters allows taking into account the experience of histologists about which pixels they consider to be "stained" and "unstained." In this case, "preparation" preferably includes staining.

[0063] If empirical calibration parameters are used in determining the classification condition, these are preferably taken into account in determining the classification condition as a function of the at least one fitting parameter.

[0064] In an embodiment of the present invention, one or more Gaussian distributions are used to fit one or more distributions in the collection.

[0065] The at least one fitting parameter preferably comprises one or more means of one or more Gaussian distributions. The at least one fitting parameter preferably comprises one or more means of one or more Gaussian distributions and one or more standard deviations of the Gaussian distributions. Determining the classification condition based on the at least one fitting parameter preferably involves a weighted sum of the means and the standard deviations. If an empirical calibration parameter is used when determining the classification condition, the empirical calibration parameter is preferably taken into account in at least one weight of the weighted sum.

[0066] In an embodiment of the present invention, a Gaussian mixture model is used for fitting, which can adapt particularly well to the changes of the histogram.

[0067] In an embodiment of the present invention, the maximum likelihood method is used for fitting.

[0068] Another method such as the mean square method can also be used for fitting. The inventors found that using the maximum likelihood method provides better results than other methods because the parameter estimation is based on a sub-sampled pixel set. This method is particularly suitable for wide virtual slides (WSI).

[0069] For example, a sum of two-dimensional Gaussian functions may be used for fitting.

[0070] In an embodiment of the present invention, the method includes classifying pixels of the second category that do not satisfy the classification condition in the second digital image.

[0071] For example, these pixels can be regarded as undyed, or can be regarded as stained with another color, or can be regarded as having another color. In an embodiment of the present invention, for example, based on further classification conditions, pixels that do not meet the classification conditions are classified into multiple classes. Therefore, the pixels can be classified into a first class (first color or first stained color), a second class (second color or second stained color) and a third class (third color or no staining).

[0072] In an embodiment of the invention, the method comprises determining the matrix A based on principal component analysis, independent component analysis or factorial analysis of the matrix X. In practice, it makes sense to focus on the dimensions representing the relevant information in order to perform one or more fits on the coefficients of the matrix I that concentrates the most relevant information. Another possibility is that the matrix A is known before the method.

[0073] In an embodiment of the present invention referred to as the Ohta embodiment, n=3, m=3, X1 is the red value R in the first digital image, X2 is the green value G in the first digital image, X3 is the blue value B in the first digital image, and

[0074]

[0075] In the Ohta embodiment of the present invention, the equations for the change of the color space in step (b) are based on principal component analysis, as demonstrated by Ohta in his publication. Therefore, the separation between the colors in the color space of the second digital image is much stronger, which can be referred to as the "second color space" or "Ohta color space". Therefore, the method according to the present invention is particularly suitable for distinguishing pixels of different colors, which increases the classification accuracy of the method.

[0076] In addition, since the classification condition is based on the intensity I1 of the pixel in the second color space and at least another color value I2, I3 of the pixel. The use of I1 is particularly suitable for distinguishing signals from backgrounds. The use of I2 or I3 is particularly suitable for distinguishing colors. Therefore, the use of I1 and I2 or I3 brings some supplementary information. Therefore, the use of the combination of I1 and I2 or I3 greatly improves the accuracy of the method.

[0077] The inventors have shown that the combination of I1 with I2 and / or I3 in Ohta space is particularly effective for distinguishing, for example, brown and blue, and that this combination can also be very effective for distinguishing other colors.

[0078] In the first picture, the three values ​​associated with each pixel are: a value for red, a value for green, and a value for blue. The first picture can be considered to be in a first color space, which is a red-green-blue (RGB) color space. In the second picture, the three values ​​associated with each pixel are: a value for I1, a value for I2, and a value for I3. The second picture can be considered to be in a second color space. The second color space is not an RGB color space. The second color space can be referred to as an Ohta color space.

[0079] The color value I1 may be referred to as “intensity.” The color value I2 may be referred to as “red relative to blue.” The color value I3 may be referred to as “magenta relative to green.”

[0080] In an embodiment of the present invention, the first digital image shows at least a portion of a biological sample stained according to a first staining technique. The portion of the biological sample is preferably a first tissue slice. The first digital image is preferably an RGB image. For example, if the first digital image is a fluorescent image or a polarized image, the first digital image may be a grayscale level.

[0081] In the framework of this document, preferably, the first dyeing technique differs from the second dyeing technique in the chemistry used for coloring.

[0082] In an embodiment of the invention, the other digital image shows at least a portion of the biological sample stained according to the second staining technique. In this case, the method according to the invention makes it possible to combine the information provided by several staining techniques to obtain a common condition or criterion for classifying pixels. The portion of the biological sample visible on the other digital image can be the second tissue slice, or the first tissue slice after staining the unstained first tissue slice according to the second staining technique.

[0083] In an embodiment of the present invention, the first digital image is obtained from an optical image, a fluorescent image or a polarization image. Within the framework of the present invention, any other imaging technology may also be used. Another digital image is one that can be obtained from any one of the optical image, the fluorescent image or the polarization image. "Obtained from..." means that the digital image may be an optical image, a fluorescent image, a polarization image, a combination of at least two of these images, or an image containing information from at least one of these images. "Optical image" preferably refers to an image under visible light, more preferably a colored image.

[0084] In an embodiment of the present invention, X2 is obtained from another digital image, and at least one X is obtained from other digital images. i (i>2). For example, X2, X3, and X4 may be based on another digital image, and X5, X6, and X7 may be based on yet another digital image.

[0085] In an embodiment of the present invention, X1 is obtained by combining the first digital image with other digital images, and at least one X is obtained by combining another digital image with other digital images. i (2≤i≤n). For example, information from the first digital image and the other digital images may be combined to provide X1, X2, and X3, and information from another digital image and the other digital images may be combined to provide X4, X5, and X6.

[0086] The classification conditions determined at step (c) can then be reused for other biological samples.

[0087] The invention also relates to a data processing device comprising means for performing the method according to any embodiment of the invention.

[0088] The present invention also relates to a computer program comprising instructions, which, when executed by a data processing device, cause the data processing device to perform a method according to any embodiment of the present invention. The data processing device may be a computer, a tablet, a smart phone, a controller, a processor, and may be included in any type of device.

[0089] The invention also relates to a computer-readable medium comprising instructions which, when executed by a data processing device, cause the data processing device to perform a method according to any embodiment of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] In order to better understand the present invention, it will now be described by way of example with reference to the accompanying drawings, in which:

[0091] - Figure 1a A flow chart illustrating a first example of a method for classifying pixels according to an embodiment of the present invention is shown;

[0092] - Figure 1b A flow chart illustrating a second example of a method for classifying pixels according to an embodiment of the present invention is shown;

[0093] - Figure 2 is a flowchart illustrating an exemplary embodiment of determining a classification condition according to the present invention;

[0094] - Figure 3 is a flow chart illustrating an exemplary embodiment of determining at least one empirical calibration parameter according to the present invention;

[0095] - Figure 4 is a flow chart showing an exemplary embodiment of the steps of determining a classification condition and classifying a pixel into a first class according to the present invention;

[0096] - Figure 5 A first digital image in a first example of a first embodiment of the present invention is shown;

[0097] - Figure 6a yes Figure 5 an enlarged view of a portion of the first digital image;

[0098] - Figure 6b Shown with Figure 6a The portion of the first digital image shown in FIG. 1 corresponds to the portion of the second digital image;

[0099] - Figure 7 shows a first histogram in a first example of a first embodiment of the present invention;

[0100] - Figure 8 shows a second histogram in the first example of the first embodiment of the present invention;

[0101] - Figure 9a A picture submitted to an organizational scholar in a first example of a first embodiment of the present invention is shown;

[0102] - Figure 9b Shown with Figure 9aThe same image, where different types of dots indicate whether the histologist considers the location indicated by the dot as stained or unstained;

[0103] - Fig.10a The boundary between “stained” and “unstained” areas obtained without using empirical calibration parameters is shown;

[0104] - Fig.10b shows the boundary between "stained" and "unstained" regions obtained using the empirical calibration parameters described herein;

[0105] - Fig.11 In a first example of the first embodiment of the present invention, an image obtained by classification according to the present invention is shown;

[0106] - Fig.12 shows a binary mask in a first example of a first embodiment of the present invention;

[0107] - Fig.13 Shown in accordance with Fig.11 Apply the classification shown on the image Fig.12 The result of the binary mask of

[0108] - Fig.14 shows the use of hexagonal macropixels in Fig.13 The tiling on top;

[0109] - Fig.15 Shows the Fig.14 The result of the calculation performed on the macro pixel;

[0110] - Fig.16 A histogram providing the number of macropixels as a function of the transmission factor is shown;

[0111] - Fig.17 shows a first digital image in a second example of the first embodiment of the present invention;

[0112] - Fig.18a yes Fig.17 an enlarged view of a portion of the first digital image;

[0113] - Fig.18b Shown with Fig.18a The portion of the first digital image shown in FIG. 1 corresponds to the portion of the second digital image;

[0114] - Fig.19 shows a first histogram in a second example of the first embodiment of the present invention;

[0115] - Fig. 20 shows a second histogram in a second example of the first embodiment of the present invention;

[0116] - Fig.21a A picture submitted to an organizational scholar in a second example of the first embodiment of the present invention is shown;

[0117] - Figure 21b Shown with Fig.21a The same image, where different types of dots indicate whether the histologist should consider the location indicated by the dot as stained or unstained;

[0118] - Fig.22a The boundary between “stained” and “unstained” areas obtained without using empirical calibration parameters is shown;

[0119] - Figure 22b shows the boundary between "stained" and "unstained" regions obtained using the empirical calibration parameters described herein;

[0120] - Fig.23 shows an image obtained by classification according to the present invention in a second example of the first embodiment of the present invention;

[0121] - Fig.24 shows a binary mask in a second example of the first embodiment of the present invention;

[0122] - Fig.25 Shown according to Fig.23 Apply the classification shown on the image Fig.24 The result of the binary mask of

[0123] - Fig.26 shows the use of hexagonal macropixels in Fig.25 The tiling on top;

[0124] - Fig. 27 Shows the Fig.26 The result of the calculation performed on the macro pixel;

[0125] - Fig.28 A histogram providing the number of macropixels as a function of the transmission factor is shown;

[0126] - Fig.29 A first digital image in an example of a second embodiment of the present invention is shown;

[0127] - Fig.30a yes Fig.29 an enlarged view of a portion of the first digital image;

[0128] - Fig.30b Shown with Fig.30a The portion of the first digital image shown in corresponds to a portion of the second digital image;

[0129] - Fig.31a shows a two-dimensional distribution in an example of a second embodiment of the present invention;

[0130] - Fig.31b Shows the Fig.31a Four Gaussian curves are fitted to the two-dimensional distribution of

[0131] - Fig.32 Shows the use of Fig.31b The results of the analysis of the four ellipses;

[0132] - Fig.33a yes Fig.29 A magnified image of a given area on a

[0133] - Figure 33b is with Fig.33a Same Fig.32 Magnified view of the area above;

[0134] - Fig.34 shows the use of hexagonal macropixels in Fig.32 The tiling on top;

[0135] - Fig.35 Macropixels with colors corresponding to their collagen amounts are shown;

[0136] - Fig.36 Macropixels with colors corresponding to their cytoplasmic amount are shown;

[0137] - Fig.37 The macropixel distribution as a function of its collagen amount is shown;

[0138] - Fig.38 The distribution of macropixels as a function of their cytoplasmic amount is shown;

[0139] - Fig.39 A first digital image in an example of the present invention is shown; and

[0140] - Fig.40 Shown with Fig.39 The portion of the first digital image shown in FIG. DETAILED DESCRIPTION

[0141] The present invention will be described with respect to specific embodiments and with reference to some drawings, but the present invention is not limited thereto. The drawings described are only schematic and non-limiting. In the drawings, for illustrative purposes, the size of some elements may be exaggerated and not drawn to scale.

[0142] Furthermore, terms such as first, second, third, etc. in the description and claims are used to distinguish similar elements and not necessarily to describe a sequential or chronological order. Where appropriate, embodiments of the invention may operate in other sequences than described or illustrated herein.

[0143] Additionally, while various embodiments are referred to as "preferred," this should be interpreted as exemplary ways in which the invention may be implemented and should not be construed as limiting the scope of the invention.

[0144] The term "comprising" used in the claims should not be interpreted as being limited to the means listed thereafter, and the term does not exclude other elements or steps. It should be interpreted as specifying the presence of the mentioned features, integers, steps or components, but not excluding the presence or addition of one or more additional features, integers, steps or components or combinations thereof. Thus, the scope of the expression "a device comprising A and B" should not be limited to a device consisting of only components A and B, but means that, with respect to the present invention, only components A and B are listed in the device, and further, the claims should be interpreted as including equivalents of those components.

[0145] In the drawings, the same reference numbers may be used to refer to the same or similar elements.

[0146] Figure 1a A flow chart illustrating an example of a method 100 for classifying pixels according to an embodiment of the present invention is shown, wherein no further digital image is used. Preferably, a biological sample 101 is stained 201. The result is a stained biological sample 102. A picture is then made of at least a portion of the biological sample 101 or at least a portion of the stained biological sample 102 to provide 202 a first digital image 103 representing at least a portion of the stained or unstained biological sample 101 to an apparatus comprising data processing means.

[0147] The first digital image 103 comprises pixels. The first digital image 103 is in a first space having n dimensions (n≥2), and each pixel has pixels referred to as X1 to X2. n In the embodiment of the present invention where only the first digital image 103 is used to determine X, the values ​​X2 to X n is obtained 202 b together with the value X1 at step 202 , since these values ​​are included in the first digital image 103 .

[0148] The first space is preferably a color space, and more specifically an RGB color space. Thus, the three color values ​​associated with each pixel of the first digital image 103 are: a value R for red (= X1), a value G for green (= X2), and a value B for blue (= X3).

[0149] According to the method 100, a spatial (or color space) change 203 is then performed, starting for example from the RGB values ​​of the pixels of the first digital image 103. The color space change 203 comprises determining the second digital image 104 in a second color space by performing the following matrix multiplication for each pixel.

[0150] I=A x X

[0151] The matrix A is the same for all pixels. The components I1 to I m is the pixel value in the second digital image 104, where m≥2.

[0152] In the Ohta embodiment of the present invention, the color values ​​of the pixels in the second digital image 104 are I1, I2 and I3. I1, I2 and I3 are determined by the following formula:

[0153]

[0154] Then the classification condition 105 is determined 204. The determination 204 is made based on:

[0155] the values ​​of I1 for at least some pixels of the second digital image 104, and

[0156] At least one I of at least some pixels of the second digital image 104 j (m≥j>1).

[0157] In an Ohta embodiment of the present invention, the determination 204 is made based on:

[0158] the values ​​of I1 for at least some pixels of the second digital image 104, and

[0159] The values ​​of I2 and / or the values ​​of I3 of at least some pixels of the second digital image 104 .

[0160] The value of I1 is particularly useful for distinguishing pixels of interest. The value of I2 or I3 is particularly useful for distinguishing colors or stains. The combination of I1 and I2 or I3 allows accurate determination of colored or stained pixels of interest.

[0161] Preferably, the classification condition 105 is determined to represent a condition that is satisfied only by colored or stained pixels and not satisfied by non-colored pixels.

[0162] Then, the pixels in the second digital image 104 that satisfy the classification condition 105 are classified into the first class 106. In other words, these pixels are selected as pixels belonging to the first class 106.

[0163] In an embodiment of the present invention, a third digital image is formed using pixels of the first class 106, and the third digital image is a mask, preferably a binary image or a grayscale image. The first color in the binary image corresponds to pixels of the first class 106, and the second color in the binary image corresponds to pixels not of the first class 106.

[0164] The present invention also relates to a method for biological image processing, comprising the classification method 100 and at least one of the following methods:

[0165] Determine at least one empirical calibration parameter 113 (see, for example Figure 3 );

[0166] A method for segmenting the first digital image into a stroma region and a tumor region;

[0167] • assigning pixels (preferably pixels of the first class 106) as macropixels; and

[0168] Further analysis is performed based on the first category of 106 pixels.

[0169] In an embodiment of the invention, the method of segmenting the first digital image into stroma regions and tumor regions determines a binary mask that can be superimposed on the third binary image and selects only pixels from the tumor region and pixels of the first class.

[0170] Preferably, only those selected pixels are assigned as macropixels and / or used for any further analysis.

[0171] This further analysis preferably involves statistically analyzing a function of the number of macropixels having a given characteristic.

[0172] Figure 1b A flow chart illustrating an example of a method 100 for classifying pixels according to an embodiment of the present invention is shown, wherein another digital image 103b is used. Preferably, a first tissue slice 101 of a biological sample is stained 201 according to a first staining technique. The result is a first stained tissue slice 102 of the biological sample. Then, at least a portion of the first tissue is imaged to provide 202 a first digital image 103 representing at least a portion of the first tissue slice to an apparatus comprising a data processing device. The first digital image 103 comprises pixels. Each pixel has at least one value referred to as X1 in the first digital image 103.

[0173] Then or simultaneously, a second tissue section of the biological sample 101 is stained 201b according to a second staining technique. The result is a second stained tissue section 102b of the biological sample. Then, a picture is made of at least a portion of the second tissue section to provide a further digital image 103b representing at least a portion of the second tissue section to an apparatus comprising data processing means. The further digital image 103b comprises values ​​associated with its pixels. When the apparatus receives the further digital image 103b, there is at least one value X associated with each pixel of the further digital image 103b obtained 202b by the apparatus. i (i>1). Preferably, each pixel of the further digital image 103b corresponds to a single pixel in the first digital image 103. For each pixel, there is a matrix X=(X1, ..., X n ) T , which combines one or more values ​​obtained from the first digital image 103 and one or more values ​​obtained from the further digital image 103b.

[0174] The method then uses the matrix X to compare with the reference Figure 1a Steps 203 to 205 are continued in a similar manner as described.

[0175] In the framework of the present invention, the first digital image 103 and / or the further digital image 103b may be obtained from an optical image (eg an RGB image), a fluorescence image or a polarization image. In practice, any type of image is suitable.

[0176] Figure 1a The method considers only the first digital image 103 as input for digital processing. Figure 1b The method of the invention considers a first digital image 103 and a further digital image 103b as inputs for digital processing. Within the framework of the invention, at least one further digital image can also be used to obtain at least one component of X. For example, the first digital image 103 and the other digital image can be combined to obtain X1, while the further digital image 103b and the other digital image can be combined to obtain X2. Such an image combination can actually be performed by principal component analysis to extract the most relevant information directly from the image.

[0177] Figure 2 is a flow chart illustrating an exemplary embodiment of determining 204 a classification condition 105 according to the present invention.

[0178] According to this exemplary embodiment, a set 111 of one or more distributions is determined 211. The set 111 includes at least one distribution of the number of pixels in the second digital image 104 as a function of I1. The set 111 includes at least one distribution of the number of pixels in the second digital image 104 as a function of I1. jAt least one distribution of the number of pixels in the second digital image 104 as a function of (m≥j>1) (eg, I2 or I3).

[0179] For example, the set 111 may include a first histogram representing the number of pixels in the second digital image as a function of I1 and a second histogram representing the number of pixels in the second digital image as a function of I2.

[0180] Then, in Figure 2 In the illustrated embodiment, one or more distributions of the set 111 are fitted 212 using parametric functions to extract one or more parameters 112 of these parametric functions.

[0181] Then, in Figure 2 In the illustrated embodiment, the determination 204 also takes into account 213 at least one empirical calibration parameter 113 in addition to the one or more parameters 112 to calculate the classification condition 105 .

[0182] In an embodiment of the present invention, steps 211 and 212 are performed, but step 213 is not performed. In other words, the empirical calibration parameters 113 are not used, but the fitting 212 of the distribution is used to determine the classification condition 105.

[0183] Figure 3 is a flow chart illustrating an exemplary embodiment of determining 300 at least one empirical calibration parameter 113 according to the present invention.

[0184] According to this exemplary embodiment, the calibration biological sample 301 is stained or prepared 401 in the same manner as the biological sample 102 analyzed by the classification method 100 is stained or prepared. The calibration biological sample becomes the stained or prepared calibration biological sample 302. Then an RGB picture of the stained or prepared calibration biological sample 302 is taken 402, and points are placed on some pixels of the picture. Then, the picture with the points 303 is shown 403 to a biological practitioner (such as a doctor for interpreting biological samples). The biological practitioner gives an indication of the staining. The indication of the staining can, for example, express which points the biological practitioner thinks are stained / unstained, or stained with a first color / stained with a second color / … / unstained. The result 304 can have the format of a table of RGB values ​​and indicate the staining or color. Then, based on the RGB values ​​and the corresponding staining or color indication, the empirical calibration parameters 113 are determined 404. The practitioner's point is intersected with two masks (or binary images), one mask for the labeled image (or stained / colored image) and the other mask for the unlabeled image. When a labeled point is associated with its mask (or binary image), the point is considered "true", otherwise it is considered "false". Scores are calculated to determine the first calibration parameter and / or the second calibration parameter.

[0185] Figure 4 1 is a flow chart showing an exemplary embodiment of the steps of determining 204 the classification condition 105 and classifying the pixel 205 into the first class 106 according to the present invention. Figure 4 In the description of , the index J is an integer strictly greater than 1 and equal to or less than m, where m is the number of elements in the matrix I.

[0186] Those skilled in the art will understand that Figure 4 Some of the steps shown in parallel may be performed sequentially, and vice versa.

[0187] According to this exemplary embodiment, the determination 211 of the set 111 of one or more distributions includes the determination 501 of a first histogram 601 that is an I1 histogram and the determination 502 of a first histogram 601 that is an I1 histogram. j Determination 502 of a second histogram 602 of histograms The set 111 comprises the first histogram 601 and the second histogram 602 .

[0188] Then, in step 212, a Gaussian mixture model having a first I1 Gaussian curve and a second I1 Gaussian curve is used to fit the first histogram 601. The first I1 Gaussian curve preferably corresponds to the stained / colored signal of the I1 scale, and the second I1 Gaussian curve corresponds to the background of the I1 scale. Typically, the first I1 Gaussian curve reaches a much higher value than the second I1 Gaussian curve. The first I1 Gaussian curve is written as:

[0189]

[0190] Here, μ1 and σ1 603 are some fitting parameters in the fitting parameters 112 .

[0191] In step 212, using a first I j Gaussian Curve and Second I j The second histogram is fitted using a Gaussian mixture model of the sum of Gaussian curves. j The Gaussian curve preferably corresponds to I j Scaled dyed / colored signal, II I j The Gaussian curve corresponds to I j Scale background. Usually, the first j The Gaussian curve reaches a value higher than the second I j The Gaussian curve has much higher values. j The Gaussian curve is written as:

[0192]

[0193] Among them, μ j and σ j are some fitting parameters in the fitting parameters 112 .

[0194] In step 213, the first calibration parameter ITC1 605 is considered 505 to determine the first threshold IT1 as IT1 = μ1 + ITC1 * σ1, and the second calibration parameter ITC j , to determine the second threshold as IT j =μ j +ITC j *σ j .

[0195] Classification conditions 105 include:

[0196] First specific condition 607: I1>IT1, and

[0197] Second specific condition 608: I j >IT j .

[0198] The classification step 205 includes:

[0199] · comparing 507 the pixels of the second digital image 104 with a first threshold IT1 and selecting 609 only the pixels where I1>IT1; and

[0200] Compare the selected pixel with the second threshold IT j Compare 508, only I j >IT jThe pixels of are classified in the first class 106 .

[0201] Figure 4 is an exemplary method of the first embodiment of the present invention. In the first embodiment of the present invention, the set of one or more distributions includes: providing a first distribution of the number of pixels in the second digital image as a function of the I1 value of the pixel; and providing I1 as a function of the pixel. j A second distribution of the number of pixels in the second digital image as a function of the value of (j>1).Examples 1 and 2 illustrate a first embodiment of the present invention.

[0202] According to a second embodiment of the present invention, the set of one or more distributions includes providing a function of the I1 value of a pixel and a function of the I1 value of a pixel. j The distribution of the number of pixels in the second digital image as a function of (2≤j≤m) values. The multidimensional distribution is preferably fitted using a sum of at least two or three Gaussians. A Gaussian mixture model can be used for fitting.

[0203] In two or more dimensions, the Gaussians obtained from the fit can be represented by ellipses. For example, there can be a first ellipse corresponding to stained / colored pixels, a second ellipse corresponding to unstained / uncolored pixels, and a third ellipse corresponding to the background. The ellipses can be classified by their surfaces: ellipses with larger surfaces have lower priority, ellipses with smaller surfaces have higher priority. Empirical calibration parameters can be taken into account by multiplying or increasing or decreasing the length of the major and minor axes of the ellipse. Example 3 illustrates a second embodiment of the invention.

[0204] Example 1

[0205] Figures 5 to 16 A first example of the first embodiment of the present invention is shown. In this example, staining of tissue sections is performed using BIM.

[0206] Figure 5 A first digital image 103 (FIG. 1) is shown. Figure 6a yes Figure 5 An enlarged view of a portion of the first digital image 103 . Figure 6b Shown with Figure 6a The portion of the first digital image 103 shown in corresponds to a portion of the second digital image 104 (FIG. 1).

[0207] Figure 7 A first histogram 601 having a fit as the sum of a first I1 Gaussian curve 701 and a second I1 Gaussian curve 702 is shown ( Figure 4 ). Figure 8 A second histogram 602 having a fit as the sum of a first I2 Gaussian curve 703 and a second I2 Gaussian curve 704 is shown ( Figure 4 ).

[0208] Figure 9a A picture 303 with dots 705 is shown submitted 403 to an organizational scholar ( Figure 3 ). Figure 9b A picture 303 with dots is shown, where the type of dot indicates whether the histologist considers the location indicated by the dot to be stained (light grey dot 706) or unstained (dark grey dot 707). The dots that remain white correspond to pixels that the histologist cannot determine whether they are stained or not.

[0209] Fig.10a The boundary 809a between the "stained" and "unstained" regions obtained without using the empirical calibration parameters is shown. Fig.10b The boundary 809b between "stained" and "unstained" regions obtained using the empirical calibration parameters as described herein is shown. It is particularly evident in boxes 708a, 708b, 709a, 709b that the use of the empirical calibration parameters improves the classification.

[0210] Fig.11 The image resulting from classification 205 is shown: red pixels are pixels belonging to the first class 106, green pixels are pixels not belonging to the first class 106. The method applied in this example 1 involves selecting a region of interest, and black pixels are those pixels which are not considered to be part of the region of interest.

[0211] Fig.12 A binary mask is shown, where white pixels correspond to tumor areas.

[0212] Fig.13 Shown in Fig.11 The classification 205 shown is applied to the image obtained Fig.12 The result of the binary mask.

[0213] Fig.14 shows the use of hexagonal macropixels in Fig.13 Tile on top.

[0214] Fig.15 Shows the Fig.14 The color indicates the value obtained by the calculation.

[0215] Fig.16 A histogram providing the number of macropixels as a function of the transmission factor is shown.

[0216] Example 2

[0217] Figures 17 to 28A second example of the first embodiment of the present invention is shown. In this example, staining of tissue sections is performed using P-ERK.

[0218] Fig.17 A first digital image 103 (FIG. 1) is shown. Fig.18a yes Fig.17 An enlarged view of a portion of the first digital image 103 . Fig.18b Shown with Fig.18a The portion of the first digital image 103 shown in corresponds to a portion of the second digital image 104 (FIG. 1).

[0219] Fig.19 A first histogram 601 having a fit as the sum of a first I1 Gaussian curve 701 and a second I1 Gaussian curve 702 is shown ( Figure 4 ). Fig. 20 A second histogram 602 having a fit as the sum of a first I2 Gaussian curve 703 and a second I2 Gaussian curve 704 is shown ( Figure 4 ).

[0220] Fig.21a A picture 303 with dots 705 is shown submitted 403 to an organizational scholar ( Figure 3 ). Figure 21b A picture 303 with dots is shown, where the type of dot indicates whether the histologist considers the location indicated by the dot to be stained (light grey dot 706) or unstained (dark grey dot 707). The dots that remain white correspond to pixels that the biologist cannot determine whether they are stained or not.

[0221] Fig.22a The boundary 809a between the "stained" and "unstained" regions obtained without using the empirical calibration parameters is shown. Figure 22b The boundary 809b between "stained" and "unstained" regions obtained using the empirical calibration parameters as described herein is shown. It is particularly evident in boxes 708a, 708b, 709a, 709b that the use of the empirical calibration parameters improves the classification.

[0222] Fig.23 The resulting image from classification 205 is shown: red pixels are pixels belonging to the first class 106, green pixels are pixels not belonging to the first class 106. The method applied in this example 2 involves selecting a region of interest, and black pixels are those pixels which are not considered to be part of the region of interest.

[0223] Fig.24 A binary mask is shown, where white pixels correspond to tumor areas.

[0224] Fig.25 Shown in Fig.23 The classification 205 shown is applied to the image obtained Fig.24 The result of the binary mask.

[0225] Fig.26 shows the use of hexagonal macropixels in Fig.25 Tile on top.

[0226] Fig. 27 Shows the Fig.26 The color indicates the value obtained by the calculation.

[0227] Fig.28 A histogram providing the number of macropixels as a function of the transmission factor is shown.

[0228] Example 3

[0229] Figures 29 to 38 An example of the second embodiment of the present invention is shown. In this example, staining of a tissue section is performed using three colors RGB3, where red corresponds to collagen fibers, cyan corresponds to cytoplasm, and purple corresponds to cell nuclei.

[0230] Fig.29 A first digital image 103 (FIG. 1) is shown. Fig.30a yes Fig.29 An enlarged view of a portion of the first digital image 103 . Fig.30b Shown with Fig.30a The portion of the first digital image 103 shown in corresponds to a portion of the second digital image 104 (FIG. 1).

[0231] Fig.31a A two-dimensional distribution as a function of I1 and I2 is shown. The two-dimensional distribution forms a set of one or more distributions. Each point 710 represents the value of a pixel in the second color space. Fig.31b Shows the Fig.31a Four Gaussian curves are fitted to the two-dimensional distribution of the pixel. The first ellipse 711 corresponds to red in the RGB space. The second ellipse 712 corresponds to blue in the RGB space. The third ellipse 713 corresponds to green in the RGB space. The fourth ellipse 714 corresponds to the background pixel.

[0232] Fig.32 The results of the analysis of four ellipses with red (first ellipse - nucleus), blue (second ellipse - cytoplasm), green (third ellipse - collagen) and black for background pixels are shown. This result corresponds to Fig.29 The same area.

[0233] Fig.33a yes Fig.29 A magnified image of a given area in , Figure 33b is with Fig.32 A magnified view of the same area.

[0234] Fig.34 shows the use of hexagonal macropixels in Fig.32 Tile on top.

[0235] Fig.35 Macropixels are shown with a color corresponding to their collagen amount.

[0236] Fig.36 Macropixels are shown with colors corresponding to their cytoplasmic amount.

[0237] Fig.37 The distribution of macropixels as a function of their collagen amount is shown.

[0238] Fig.38 The distribution of macropixels as a function of their cytoplasmic amount is shown.

[0239] Example 4

[0240] Fig.39 A first digital image 103 (FIG. 1) is shown showing whole cells immunostained with an anti-luciferase antibody 48 hours after transfection with a luciferase plasmid.

[0241] Fig.40 Shown with Fig.39 The portion of the first digital image 103 shown in FIG. 1 corresponds to the second digital image 104 ( FIG. 1 ).

[0242] A possible embodiment of the present invention is a method (100) for classifying pixels of a digital image of a biological sample (101), the method comprising the following steps:

[0243] (a) receiving (202) a first digital image (103) representing a biological sample (101), each pixel of the first digital image (103) being associated with a red value R, a green value G and a blue value B;

[0244] (b) The color space is changed using the following equation (203)

[0245]

[0246] to obtain a second digital image (104), I1, I2 and I3 being color values ​​associated with each pixel of the second digital image (104);

[0247] (c) determining a classification condition (105) based on the value of I1 and the values ​​of I2 and / or I3 of at least some pixels in the second digital image (104); and

[0248] (d) Classifying (205) pixels in the second digital image (104) that satisfy the classification condition (105) into the first class (106).

[0249] In this method, step (c) preferably comprises:

[0250] determining (211) a set (111) of one or more distributions, the set (111) comprising: at least one distribution of the number of pixels in the second digital image (104) as a function of I1, and at least one distribution of the number of pixels in the second digital image (104) as a function of I1; j at least one distribution of the number of pixels in the second digital image (104) as a function of (j=2 or 3),

[0251] fitting (212) one or more distributions of the set (111) to determine at least one fitting parameter (112), and

[0252] Determining a classification condition (105) based on at least one fitting parameter (112).

[0253] In other words, the present invention relates to a computer-implemented method for automatically analyzing a digital image 103 of a biological sample 101. The method comprises a change 203 from a first color space (e.g. RGB) to a second color space (e.g. Ohta). Then, fitting is performed to classify pixels taking into account multiple dimensions of the second color space. For example, a color intensity value and a color value of at least one of "red relative to blue" and "magenta relative to green" are used to select stained or colored pixels.

[0254] Although the invention has been described above with respect to certain embodiments, it will be readily appreciated that other embodiments are possible.

Claims

1. A method (100) for classifying pixels of a digital image of a biological sample (101), the method comprising the following steps: (a) receiving (202) a first digital image (103) representing the biological sample (101) to obtain a value X1 associated with each pixel of the first digital image (103), and obtaining (202b) at least one other value X1 for each pixel from the first digital image (103) and / or from another digital image (103b); i , where i is an integer such that 2≤i≤n, and n is an integer such that n≥2; (b) Use matrix multiplication to change the space (203) I=A x X To obtain a second digital image (104), X is a value including X1 to X n A matrix of n rows and 1 columns, I is a matrix including values ​​I1 to I m A matrix of m rows and 1 columns, where m is an integer equal to or greater than 2, and I1 to I m is a value associated with each pixel of the second digital image (104), and A is a matrix with m rows and n columns; (c) based on the values ​​of I1 and I2 of at least some pixels in the second digital image (104); j The classification condition (105) is determined (204) by the value of, where j>1 and j≤m; and (d) classifying (205) pixels of the second digital image (104) that satisfy the classification condition (105) into the first class (106); Wherein, step (c) comprises: determining (211) a set (111) of one or more distributions, the set (111) comprising at least one distribution of the number of pixels in the second digital image (104) as a function of I1, the set (111) comprising as a function of I1 j at least one distribution of the number of pixels in the second digital image (104) as a function of , wherein j>1 and j≤m, fitting (212) one or more distributions of the set (111) to determine at least one fitting parameter (112), and Determining the classification condition (105) according to the at least one fitting parameter (112); Wherein, the biological sample (101) is a stained, polarized or fluorescent biological sample, and the value X1 to the value X n is the color value in the color space.

2. The method according to claim 1, wherein: The set (111) of one or more distributions includes as a function of I1 and as a function of I j The distribution of a function of , where j>1 and j≤m.

3. The method according to claim 1, wherein: The set (111) of one or more distributions includes a first distribution as a function of I1 and a second distribution as a function of I j The second distribution of a function of , where j>1 and j≤m.

4. The method according to any one of claims 1 to 3, wherein: The classification condition for the pixel under consideration is related to the value of I1 of the pixel under consideration and at least one I j It is related to the value of , where j>1 and j≤m.

5. The method according to any one of claims 1 to 3, wherein: In the determination (204) of the classification condition (105), empirical calibration parameters (113) are used, which have been determined by presenting (403) to a biological practitioner a calibration biological sample (302) prepared in the same manner as the biological sample (101).

6. The method according to any one of claims 1-3, comprising determining the matrix A based on principal component analysis, independent component analysis or factorial analysis of the matrix X.

7. The method according to any one of claims 1 to 3, wherein: n=3, m=3, X1 is the red value R in the first digital image (103), X2 is the green value G in the first digital image (103), and X3 is the blue value B in the first digital image (103), wherein 8. The method according to any one of claims 1 to 3, wherein: The first digital image (103) shows at least a portion of the biological sample (101) stained according to a first staining technique.

9. The method according to claim 8, wherein: The further digital image (103b) shows at least a portion of the biological sample (101) stained according to a second staining technique.

10. The method according to any one of claims 1 to 3, wherein: The first digital image (103) is acquired from an optical image, a fluorescence image or a polarization image.

11. The method according to any one of claims 1 to 3, wherein: X2 is obtained from the other digital image (103b), and at least one X i is obtained from other digital images, where i>2.

12. The method according to any one of claims 1 to 3, wherein: X1 is obtained from the combination of the first digital image (103) and other digital images, and at least one X i is obtained from the combination of the another digital image (103b) and the other digital images, wherein 2≤i≤n.

13. A data processing device comprising means for performing the method according to any one of claims 1 to 12.

14. A computer program product comprising a computer program, the computer program comprising instructions which, when executed by a data processing device, cause the data processing device to perform the method according to any one of claims 1 to 12.

15. A computer-readable medium comprising instructions which, when executed by a data processing device, cause the data processing device to perform the method according to any one of claims 1 to 12.

Images