Multichannel extended depth-of-field method for automated digital cytology

Abstract

The present invention provides a method (M) for generating a color-fidelity EDF image, i.e., an extended depth-of-field image, from a color volume (1), in which voxels of fixed indices 1 to L are 2D images acquired using a microscope at different focal depths in the z direction, said method comprising the steps of generating a grayscale volume (2), applying an invertible color-to-grayscale transform to said volume (1) (M10), applying a wavelet transform to the grayscale volume (2) to obtain a 3D wavelet coefficient matrix (3) (M20), selecting wavelet coefficients using a predefined coefficient selection rule, and generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM (M40). , applying (M50) an inverse wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D grayscale EDF image (4); generating a 2D color composite image CC; applying (M70) an inverse color-to-grayscale transform to the 2D grayscale EDF image (4) to obtain a 2D color EDF image (5); converting (M80) the 2D color composite image CC and the 2D color EDF image (5) to a color space including at least one chromaticity component and at least one intensity component; and concatenating (M90) the at least one chromaticity component of the 2D color composite image CC and the at least one intensity component of the 2D color EDF image (5) to obtain a color-fidelity extended depth-of-field image.

Description

[Technical Field]

[0001] The present invention relates to the field of image processing of biological sample images, in particular to the field of image analysis of urine samples in the context of automated digital cytology. [Background technology]

[0002] One limitation of conventional optical microscopy is the inability to obtain perfectly focused images of objects whose thickness along the optical axis exceeds the depth of focus. To overcome this problem, in digital cytology, a series of images are acquired while the object of interest is displaced along the optical axis (z-axis), resulting in a color volume containing the series of acquired images, which is therefore sometimes called an "image stack," "z-stack," "3D image," or "3D image stack." Naturally, different parts or details of the object of interest are in focus in different images of the color volume.

[0003] To facilitate the analysis and storage of the obtained volume, one commonly used strategy consists of fusing the acquired series of images into a single composite image. To do so, among all the images of the volume, i.e., among the stack of images, the image containing the most information can be selected. This "best focus" selection method requires very little computation. However, the "best focus" method results in a loss of information when the object of interest is thicker than the depth of field of the acquisition module, for example in the presence of cell clumps.

[0004] The extended depth of field (EDF) method is used to fuse the images of a z-stack into a single 2D image (the EDF image) in which all parts of the object of interest appear to be in focus, while avoiding the loss of information associated with the previously mentioned "best focus" method.

[0005] EDF methods are typically classified into three categories: spatial domain techniques, transform-based methods such as wavelet transform (WT)-based methods, and deep learning-based methods.

[0006] One spatial domain approach that is relatively easy to implement consists of selecting in-focus regions in each image of the z-stack and then fusing the selected regions to obtain one single 2D image in focus with all different parts of the object of interest.

[0007] Another example of a spatial domain technique is the smooth manifold extraction (SME) method described by A. Shihavuddin et al. in “Smooth 2D manifold extraction from 3D image stack” (2017). Summary of the Invention [Problem to be solved by the invention]

[0008] One common drawback of spatial domain techniques is the loss of information in the presence of thick or overlapping objects of interest, such as cell clumps, and therefore these techniques are not suitable for use with transmission and semi-transmission images, such as cytology images obtained with transmission microscopy.

[0009] Common wavelet transform-based methods include the discrete wavelet transform (DWT) algorithm and the stationary wavelet transform (SWT) algorithm, of which A.P. Bradley and P.C. Bamford show in “A One-pass Extended Depth of Field Algorithm Based on the Over-complete Discrete Wavelet Transform” (2004) that the latter performs better.

[0010] Typically, wavelet transform-based methods for computing an EDF image of a stack of images consist of applying a wavelet transform algorithm to all images in the stack to obtain a wavelet transform stack, and then applying a coefficient selection rule to select the most relevant wavelet coefficients in the wavelet transform stack. Typically, the coefficient selection rule is clearly defined for grayscale volumes, and some prior art techniques compute the EDF image of a stack by applying a wavelet transform to grayscale images, i.e., images containing only one channel. Such techniques may also be applied to color images, i.e., images containing more than one color channel. One possible approach for managing multi-channel images is to apply a wavelet transform, a coefficient selection strategy, and an inverse wavelet transform sequentially to each channel, resulting in one EDF image for each channel.

[0011] The EDF images thus obtained are then merged to obtain a single color EDF image. The main drawback of this method is the occurrence of color saturation and / or artifacts. To overcome this problem, B. Foster et al. (2004) propose a preprocessing step of color-to-grayscale conversion and a postprocessing step of color reassignment in “Extended Depth-of-Focus for multi-channel microscopy images: a complex wavelet approach.” However, this color reassignment step can lead to color inaccuracies. Furthermore, methods based on the complex wavelet transform (CWT), such as the method used by B. Foster et al., tend to produce ringing artifacts near the edges of objects of interest, thus limiting the recovery of image details.

[0012] Therefore, a major challenge for EDF methods associated with automated digital cytology is to accurately reproduce all details of the object of interest while preserving color fidelity.

[0013] High color fidelity is required in several fields, such as urine cytopathology, where color is a key feature for bladder cancer detection. Urine cytopathology is the method of choice for noninvasive bladder cancer detection and involves analyzing microscope slides containing urothelial cells to search for abnormalities. While this method provides satisfactory results for detecting advanced cancer, it is time-consuming, expensive, and ineffective for detecting early-stage bladder cancer. Early detection is a key challenge for improving patient survival while significantly reducing treatment costs, which are known to be higher for advanced-stage cancers. Automated solutions could aid in early-stage cancer management by shortening analysis time, thus accelerating the diagnostic process, and by implementing segmentation algorithms to improve the detection and quantification of abnormalities.

[0014] The present invention aims to overcome the problems associated with generating EDF images from color volumes in the field of image analysis of biological samples. [Means for solving the problem]

[0015] The present invention provides a computer-implemented method for generating a color-fidelity extended depth-of-field (EDF) image from a color volume of a biological sample having dimensions (N, M, L) and values ​​I(n, m, l), where l voxels (n, m, l) from 1 to L are images acquired at different focal depths in the z direction using a microscope or scanner, the method comprising the following steps: a) receiving a color volume and generating a grayscale volume by applying a reversible color-to-grayscale transform to the color volume; b) applying a wavelet transform to each set of N×M voxels of the grayscale volume with the same l index and (n,m) indices ranging from (1,1) to (N,M) to obtain a 3D wavelet coefficient matrix; where the value of each voxel with (n,m,l) index contains a set of wavelet coefficients, c) for each group of L voxels of the 3D wavelet coefficient matrix having the same (n, m) index and l index ranging from 1 to L, selecting one set of wavelet coefficients using a predefined coefficient selection rule; d) generating a 2D coefficient map CM and a 2D wavelet coefficient matrix WCM; where the value of each pixel with index (n,m) CM(n,m) is the l-index of the voxel of the 3D wavelet coefficient matrix containing the set of wavelet coefficients selected by the coefficient selection rule, and the value of each pixel with index (n,m) WCM(n,m) is the set of wavelet coefficients selected from the 3D wavelet coefficient matrix by the coefficient selection rule. e) applying an inverse wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D grayscale EDF image; f) generating a 2D color composite image CC; where the value of each pixel with index (n,m) CC(n,m) is the value of the voxel in the color volume with index l I(n,m,l) equal to the value CM(n,m) of the 2D coefficient map CM. g) applying the inverse of the reversible color-to-grayscale transformation to the 2D grayscale EDF image to obtain a 2D color EDF image; h) converting the 2D color composite image CC and the 2D color EDF image into a color space comprising at least one chromaticity component and at least one intensity component; i) concatenating at least one chromaticity component of the 2D color composite image CC and at least one intensity component of the 2D color EDF image to obtain a color-fidelity extended depth-of-field (EDF) image; The present invention relates to a method, comprising:

[0016] Advantageously, the method of the present invention enables accurate color reproduction and recovery of detail in EDF images obtained from color volumes. In particular, the method significantly reduces the color inaccuracies associated with prior art color reassignment strategies without compromising the level of detail.

[0017] Detail recovery is fundamental to achieve accurate segmentation, and in the context of automated digital cytology, accurate cytoplasmic and nuclear segmentation is important to (i) improve the detection of abnormal cells, (ii) facilitate cell counting, and (iii) calculate with high accuracy biomarkers that correlate with clinically relevant information, such as the nucleus / cytoplasm ratio.

[0018] In the method of the present invention, the set of wavelet coefficients includes at least four wavelet coefficients, and the subset of wavelet coefficients selected using the predefined coefficient selection rule may include at least one wavelet coefficient.

[0019] The number of wavelet coefficients in the 3D wavelet coefficient matrix depends on the decomposition level of the wavelet transform. In one embodiment, the wavelet transform includes multiple decomposition levels that are applied in succession. In one embodiment, the wavelet transform includes only one level of decomposition so that a color-fidelity extended depth-of-field (EDF) image with restored detail can be obtained with fewer computational steps. In this embodiment, the set of wavelet coefficients includes four wavelet coefficients, and the selected subset of wavelet coefficients includes at least one but not more than four coefficients.

[0020] In one embodiment, in the selecting step, a first subset of wavelet coefficients is used with a first predefined coefficient selection rule, and a second subset of wavelet coefficients is used with a second predefined coefficient selection rule. In this embodiment, the step of generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM includes: - generating a first 2D coefficient map; - generating a second 2D coefficient map; - combining the first 2D coefficient map and the second 2D coefficient map to obtain one 2D coefficient map CM; - generating a 2D wavelet coefficient matrix WCM; where the value of each pixel WCM(n,m) is the value of the voxel of the 3D wavelet coefficient matrix with (n,m) index and with l index equal to the value CM(n,m) of the resulting 2D coefficient map CM.

[0021] The values ​​of the voxels of the 3D wavelet coefficient matrix having an l index equal to the values ​​of the 2D coefficient map CM may contain the full set of wavelet coefficients.

[0022] In the first 2D coefficient map, the value of each pixel with an (n,m) index is the l index of a voxel of a 3D wavelet coefficient matrix that contains the subset of wavelet coefficients selected by the first coefficient selection rule.

[0023] In the second 2D coefficient map, the value of each pixel with an (n,m) index is the l index of a voxel in the 3D wavelet coefficient matrix that contains the subset of wavelet coefficients selected by the second coefficient selection rule.

[0024] In the 2D wavelet coefficient matrix WCM, a set of wavelet coefficients at a pixel with an (n, m) index is a set of coefficients in the 3D wavelet coefficient matrix that have the same (n, m) index and have the value of the 2D coefficient map as the l index. In this particular embodiment, the 2D coefficient map is obtained by combining the first coefficient map and the second coefficient map.

[0025] In this embodiment, in the step of generating a first 2D coefficient map, a first 2D wavelet coefficient matrix WCM is also generated. Similarly, in the step of generating a second 2D coefficient map, a second 2D wavelet coefficient matrix WCM is generated. However, in contrast to the first and second 2D coefficient maps, the first and second 2D wavelet coefficient matrices do not need to be combined. In fact, in this particular embodiment, the 2D wavelet coefficient matrix WCM is generated based on the first and second 2D coefficient maps.

[0026] This embodiment allows for selecting detailed information through a first coefficient selection rule and obtaining a denoising effect through a second coefficient selection rule. By combining the first 2D coefficient map and the second 2D coefficient map, one 2D coefficient map CM can be obtained, from which one wavelet coefficient matrix WCM is calculated. Advantageously, this combining step allows for obtaining a wavelet coefficient matrix WCM and a 2D coefficient map CM that have both the detailed information obtained through the first coefficient selection rule and the denoising effect obtained through the second coefficient selection rule.

[0027] According to one embodiment, the step of combining the 2D coefficient maps comprises: - filtering the first 2D coefficient map and the second 2D coefficient map, preferably using a median filter; - averaging the first and second 2D coefficient maps; - rounding the averaged 2D coefficient maps to obtain one 2D coefficient map CM; Includes:

[0028] This embodiment makes it possible to avoid an interpolation step. The value obtained in the rounded 2D coefficient map CM is used to select a wavelet coefficient from the 3D wavelet coefficient matrix having an l index equal to the obtained value. The wavelet coefficient matrix WCM is constructed based on the selected coefficients.

[0029] In one embodiment, the color-fidelity extended depth-of-field (EDF) images obtained with the present invention are converted to a different color space, preferably a color volume color space, to facilitate further steps of image analysis, such as segmentation, by selecting the color space in which the algorithms used in the further steps of image analysis perform best.

[0030] In one embodiment, the reversible color-to-grayscale transformation is principal component analysis (PCA), which has the advantage of being a standard and easy-to-implement technique for dimensionality reduction, and also has the advantage of being reversible.

[0031] According to one embodiment, the step of applying a wavelet transform applies a stationary wavelet transform (SWT). In the presence of overlapping or partially overlapping objects, the SWT is associated with better performance metrics than other methods, such as the CWT-EDF and the "best focus" method. Furthermore, in the presence of color images, the SWT-EDF achieves the best color fidelity compared to other wavelet-based methods followed by color reconstruction strategies.

[0032] In one embodiment, the color volume is a biomedical image, and the method further comprises segmenting a color-fidelity extended depth-of-field (EDF) image, where segmentation is applied to an image to improve performance when detail is restored and color fidelity is high.

[0033] The present invention also provides a system for analyzing a biological sample, comprising: - receiving a color volume and generating a grayscale volume by applying a reversible color-to-grayscale transformation to the color volume; - applying a wavelet transform to each set of NxM voxels of the grayscale volume with the same l index and (n,m) indices ranging from (1,1) to (N,M) to obtain a 3D wavelet coefficient matrix; where the value of each voxel with (n,m,l) index contains a set of wavelet coefficients, - for each group of L voxels of the 3D wavelet coefficient matrix having the same (n,m) index and l index ranging from 1 to L, selecting one set of wavelet coefficients using a predefined coefficient selection rule; - Generate the 2D wavelet coefficient matrix WCM, where the value WCM(n,m) of each pixel with index (n,m) is a set of wavelet coefficients selected from the 3D wavelet coefficient matrix by the coefficient selection rule: - Generate a 2D coefficient map CM, where the value CM(n,m) of each pixel with (n,m) index is the l-index of the voxel in the 3D wavelet coefficient matrix containing the set of wavelet coefficients selected by the coefficient selection rule. - applying the inverse wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D grayscale EDF image; - Generate a 2D color composite image CC, where the value of each pixel with index (n,m) CC(n,m) is the value of the voxel in the color volume with index l I(n,m,l) equal to the value CM(n,m) of the 2D coefficient map CM. - applying a reversible inverse color-to-grayscale transformation to the 2D grayscale EDF image to obtain a 2D color EDF image; - converting the 2D color composite image CC and the 2D color EDF image into a color space comprising at least one chromaticity component and at least one intensity component; - concatenating at least one chromaticity component of the 2D color composite image CC and at least one intensity component of the 2D color EDF image to obtain a color-fidelity extended depth-of-field (EDF) image; The present invention relates to a system having a processor and a computing module configured to:

[0034] According to one embodiment, the system for analyzing biological samples according to the present invention is an automated digital cytology system for analyzing urine samples, in this embodiment the system further comprises an acquisition module configured to acquire a color volume, the color volume comprising at least two color images.

[0035] According to one embodiment, the system of the present invention is a bladder cancer detection system.

[0036] The invention also relates to a computer program product for analyzing biological samples, comprising instructions that cause a computer to carry out the steps of the method according to the invention when the program is executed by a computer.

[0037] The invention also relates to a computer-readable storage medium comprising instructions that, when the program is executed by a computer, cause the computer to carry out the steps of the method according to any one of the preceding embodiments.

[0038] definition For the purposes of the present invention, the following terms have the following meanings: - "Stack" refers to a volume containing at least two 2D images acquired using a microscope at different focal depths along the optical axis of the microscope.

[0039] - "EDF image" refers to a 2D image obtained from a stack of 2D images with the extended depth of field method, i.e. from a volume.

[0040] "Color space" refers to the mathematical representation of colors in a system of coordinate axes, each color being defined by a vector whose projections on the axes of the coordinate system are the different components of the color in said color space.

[0041] "Chrominance component" refers to the projection of a color vector in a color space on an axis that contains chrominance values ​​but not luminance values. A chrominance component may contain some or all of the chrominance information of a color vector.

[0042] "Intensity component" refers to the projection of a color vector in a color space on an axis that contains luminance values ​​and no chromaticity information. The intensity component may contain some or all of the luminance information of the color vector.

[0043] - "Chi-squared distance" refers to the distance between two normalized color histograms.

[0044] "Color fidelity" refers to a 2D image obtained from a color volume while accurately reproducing the colors of the volume, i.e., while minimizing the distance metric between the distribution of chromaticity components of the obtained 2D image and the distribution of chromaticity components of the original 3D stack. Preferably, the distance metric is the chi-squared distance (χ) between the color histogram of the 2D image and the color histogram of the color volume.

[0045] - "Processor" should not be construed as being limited to hardware capable of executing software, but refers in a general manner to a processing device that may include, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device (PLD). A processor may also encompass one or more graphics processing units (GPUs), whether utilized for computer graphics and image processing or other functions. Furthermore, instructions and / or data enabling the execution of the associated and / or resulting functions may be stored, for example, on an integrated circuit, any processor-readable medium such as a hard disk, an optical disk such as a CD (compact disc), a DVD (digital versatile disc), a RAM (random access memory), or a ROM (read-only memory). Instructions may be stored in hardware, software, firmware, or any combination thereof, among others.

[0046] The following detailed description will be better understood when read in conjunction with the drawings. For purposes of illustration, the present invention is shown in preferred embodiments. It should be understood, however, that the application is not limited to the precise color spaces, reversible transformations, objects of interest, and aspects shown in the drawings. Accordingly, when features recited in the appended claims are followed by reference signs, it should be understood that such signs are included merely to enhance comprehension of the claims and in no way limit the scope of the claims. [Brief explanation of the drawings]

[0047] [Figure 1] 1 is a flow chart that schematically represents a first non-limiting example of the method of the present invention. [Figure 2] Figure 2 shows non-limiting examples of images obtained at various steps of the method of the present invention: Figure 2a is an example of an image belonging to the initial 3D stack; Figure 2b shows a grayscale EDF image obtained from the color volume to which the image of Figure 2a belongs; Figure 2c shows a color EDF image obtained with a color assignment technique according to the prior art; and Figure 2d shows a color EDF image obtained with the color reconstruction technique according to the present invention. [Figure 3] 1 is a flow chart that schematically illustrates a second non-limiting example of the method of the present invention. [Figure 4] 1 is a flow chart that schematically illustrates the main steps of the method of the present invention. [Figure 5] FIG. 1 is a block diagram that schematically illustrates a particular mode of a system for generating a color-fidelity extended depth-of-field (EDF) image from a color volume, in accordance with the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0048] While various embodiments have been described and illustrated, the detailed description should not be construed as limiting, as those skilled in the art may make various modifications to the embodiments without departing from the true spirit and scope of the present disclosure, as defined by the appended claims.

[0049] This specification illustrates the principles of the present disclosure, and it will thus be understood that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the present disclosure and are included within its scope.

[0050] All examples and conditional language listed herein are intended for instructional purposes to assist the reader in understanding the concepts contributed by the inventors to further the principles and techniques of the present disclosure, and should not be construed as being limited to such specifically listed examples and conditions.

[0051] Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, such equivalents are intended to include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

[0052] The present invention relates to a computer-implemented method for generating an extended depth-of-field (EDF) image from a volume containing a stack of 2D microscopy images acquired at different focal depths in the z-direction. The EDF image obtained by this method is characterized by improved color fidelity and detail recovery. The volume in this method is a color volume.

[0053] According to the representation in FIG. 4, the method used in the present invention includes a receiving step for receiving Volume 1. According to one embodiment, Volume 1 is a stack of 2D color images acquired by moving a microscope slide on which an object of interest is placed along the z direction parallel to the optical axis of a bright-field microscope. Volume 1 has dimensions (N, M, L), and each voxel has a value I(n, m, l). A set of voxels in Volume 1 with a fixed l index ranging from 1 to L are 2D images acquired at different focal depths in the z direction using a microscope. The dimension L is equal to the number of acquired images. The l index of a voxel is the focal depth at which the first image containing said voxel was acquired. In the present invention, the dimension L of Volume 1 is 2 or more, and therefore, Volume 1 is a stack of 2D images including at least two 2D images acquired at different focal depths.

[0054] The volume 1 according to the present invention can be represented in any color space. In one embodiment, the initial color volume, i.e., the stack of 2D images, is an RGB volume.

[0055] The present invention will be better understood when read in conjunction with the drawings. For purposes of illustration, Method M is shown in a preferred embodiment. It should be understood, however, that the application is not limited to this particular embodiment, and that the drawings are not intended to limit the scope of the claims to the embodiment depicted therein. Accordingly, when a reference sign follows a feature recited in the appended claims, it should be understood that such sign is included merely to enhance comprehension of the claims and in no way limits the scope of the claims.

[0056] Furthermore, those skilled in the art will appreciate that any flowcharts, flow diagrams, etc., whether or not a computer or processor is explicitly shown, represent various processes that may be substantially represented on a computer-readable medium as so executed by such a computer or processor.

[0057] As shown in FIG. 1, the method M described herein may include a step of generating a grayscale volume 2 from a color volume 1 by applying a reversible color-to-grayscale transformation M10 to the color volume 1.

[0058] In FIG. 1, the color volume 1 is represented schematically as a cube.

[0059] The method M optionally includes a step of pre-processing the received color volume 1 before generating the grayscale volume 2. The pre-processing step may in particular include a transformation of the color volume 1, for example restoration to an appropriate restoration standard.

[0060] In an advantageous mode, the pre-processing step may include a standardization phase, which may increase the efficiency of downstream processing of the volume 1. Such standardization may be particularly useful when several color volumes 1 originating from different sources (possibly including different imaging systems) are processed with the method of the present disclosure.

[0061] The color to grayscale transformation needs to be a reversible transformation in order to convert the grayscale EDF image 4 obtained after applying the inverse wavelet transform M50 back to a color image.

[0062] In one embodiment, the reversible color-to-grayscale transformation is a principal component analysis (PCA) transformation, and in this particular embodiment, its inverse transformation is an inverse PCA that allows for the reconstruction of the original color variables from the principal components.

[0063] Since the reversible color to grayscale conversion causes a loss of color information, a color reconstruction strategy must be applied afterwards. The steps that make it possible to obtain the color reconstruction according to the method are detailed below.

[0064] The method M of the present invention includes a step M20 of applying a wavelet transform to the grayscale volume 2. By applying the wavelet transform, a 3D wavelet coefficient matrix 3 having dimensions (N, M, L) can be obtained, where the value of each voxel with an index (n, m, l) contains a set of wavelet coefficients. Said set of coefficients contains at least four wavelet coefficients. The term "3D" refers to a three-dimensional matrix. The wavelet coefficient matrix 3 is a 3D matrix because it has dimensions (N, M, L).

[0065] In the present invention, the color volume 1, the grayscale volume 2, and the 3D wavelet coefficient matrix 3 are three-dimensional matrices with dimensions (N, M, L), while the wavelet coefficient matrix WCM, the coefficient map CM, the grayscale EDF image 6, the color composite image CC, the color EDF image 7, and the color-fidelity extended depth-of-field (EDF) image are 2D, i.e., two-dimensional, matrices with dimensions (N, M).

[0066] The step M20 of applying a wavelet transform to the grayscale volume 2 is performed in parallel for each set of N×M voxels of the grayscale volume 2 having the same l index and (n,m) indices ranging from (1,1) to (N,M). Thus, in said step M20, for the grayscale volume 2 having dimensions (N,M,L), L wavelet transforms are applied.

[0067] The wavelet transform applied in step M20 is preferably the stationary wavelet transform (SWT), since SWT-EDF achieves the best color fidelity among wavelet-based methods with color reconstruction. This is particularly advantageous if, after the concatenation step M90, the image obtained by this method M is further segmented. In practice, SWT-EDF outperforms both the CWT-EDF method and the "best focus" method according to the common segmentation evaluation metric Intersection over Union (IoU).

[0068] The wavelet transform generates a decomposition of a signal as wavelet coefficients, which include approximation and detail coefficients. In the context of images, the wavelet transform is known as decomposing the original image into sub-images of wavelet coefficients. This decomposition is reversible, and the original image can be reconstructed from the wavelet coefficients by applying an inverse wavelet transform. In particular, the wavelet transform of an image is calculated by applying a first decomposition to each row of pixels in the image, and then applying a second decomposition to each column of the results of the first decomposition. A one-level wavelet decomposition of an image is the result of the first and second decompositions, i.e., a row decomposition and a column decomposition. A one-level wavelet decomposition of an image generates four matrices of wavelet coefficients (sub-images). The pixel values ​​of the sub-images contain wavelet coefficients. However, several levels of subsequent decomposition can be performed. At each decomposition level, four sub-images are obtained: three "detail" sub-images containing details of the original image and a fourth "approximation" sub-image containing an approximation of the original image. The pixel values ​​of the sub-images contain wavelet coefficients. In particular, the pixel values ​​of the three "detail" sub-images include horizontal detail coefficients in the first sub-image, vertical detail coefficients in the second sub-image, and diagonal detail coefficients in the third sub-image. The horizontal detail coefficients, vertical detail coefficients, and diagonal detail coefficients are functions of the horizontal detail, vertical detail, and diagonal detail of the original image, respectively. The pixel values ​​of the fourth approximation sub-image include approximation coefficients. The approximation coefficients are functions of the approximation of the original image. In practice, at each decomposition level, the image is filtered with a high-pass filter to generate detail coefficients and a low-pass filter to generate approximation coefficients. At each decomposition level, further such filtering is performed on the image obtained in the previous decomposition step. After one or more decomposition levels, the original image can be reconstructed by applying an inverse wavelet transform to the decomposition.

[0069] In the method M according to the present invention, the wavelet transform applied in step M20 includes one level of decomposition, so that each image of the grayscale volume 2 is decomposed into four sub-images containing wavelet coefficients. An image of the grayscale volume 2 means a set of N×M voxels of the grayscale volume 2 having the same l-index, so that the l-th image of the grayscale volume 2 is a set of voxels with a fixed l-index and (n,m)-indexes ranging from (1,1) to (N,M).

[0070] The method M also comprises a selecting step M30 configured to select, for each group of L voxels of the 3D wavelet coefficient matrix 3 having the same (n,m) index and an l index ranging from 1 to L, a set of wavelet coefficients using a predefined coefficient selection rule, said set of wavelet coefficients comprising at least one detail or approximation coefficient from the sub-image defined above.

[0071] In one embodiment, the entire set of wavelet coefficients is used in further steps of the present invention.

[0072] In one embodiment, a subset of wavelet coefficients is selected from the entire set of wavelet coefficients, and said subset is further used to calculate a 2D coefficient map CM. In this particular embodiment, several 2D coefficient maps can be calculated, in particular, each 2D coefficient map is calculated based on a different subset of wavelet coefficients, and then several 2D coefficient maps are combined to obtain one 2D coefficient map.

[0073] The coefficient selection rule applied in step M30 can be any coefficient selection rule known to those skilled in the art. In particular, the coefficient selection rule is adapted to a grayscale image or a volume. For multi-channel images, applying the coefficient selection rule to all color channels and merging the results may generate false colors, so it is necessary to first convert the color image to a grayscale image, then calculate a 2D grayscale EDF image, and finally perform color reconstruction.

[0074] In the step M30 of selecting wavelet coefficients, one or more coefficient selection rules may be applied.

[0075] The method M of the present invention may further comprise a step M40 of generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM, where the value WCM(n,m) of each pixel with an (n,m) index is a set of wavelet coefficients selected from the 3D wavelet coefficient matrix 3 by a coefficient selection rule, and the value CM(n,m) of each pixel with an (n,m) index is the l index of a voxel in the 3D wavelet coefficient matrix 3 that contains the set of wavelet coefficients selected by the coefficient selection rule. For example, the value of pixel CM(2,3) includes an n index equal to 2, an m index equal to 3, and an l index of a voxel in 3D wavelet coefficient matrix 3 selected from all voxels with l ranging from 1 to L.

[0076] According to one embodiment, in the selecting step M30, a first subset of wavelet coefficients is selected using a first predefined coefficient selection rule and a second subset of wavelet coefficients is selected using a second predefined coefficient selection rule. In this embodiment, the step M40 of generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM comprises a number of steps M41, M42, M43 and M44.

[0077] Step M41 involves generating a first 2D wavelet coefficient matrix and a first 2D coefficient map from a first subset of wavelet coefficients. In this embodiment, step M41 is followed by step M42, which involves generating a second 2D wavelet coefficient matrix and a second 2D coefficient map from a second subset of wavelet coefficients. Finally, in step M43, the first 2D coefficient map and the second 2D coefficient map are combined to obtain a final 2D coefficient map CM. Based on the final 2D coefficient map CM, a 2D wavelet coefficient matrix WCM is generated in step M44. The value WCM(n,m) of each pixel in the 2D wavelet coefficient matrix WCM is a set of coefficients of a 3D wavelet coefficient matrix with an (n,m) index and an l index equal to the value CM(n,m) of the final 2D coefficient map CM.

[0078] The step M43 of combining the first and second 2D coefficient maps to obtain one 2D coefficient map CM may comprise several steps. In particular, it may include: - first filtering the first 2D coefficient map and the second 2D coefficient map, preferably using a median filter; - then average the first and second 2D coefficient maps, - finally, rounding the averaged 2D coefficient maps to obtain one 2D coefficient map CM; may include:

[0079] In one embodiment, the median filter is a 3x3 median filter.

[0080] In one embodiment, where first and second coefficient selection rules are used to select first and second subsets of wavelet coefficients, respectively, two 2D wavelet coefficient matrices and two 2D coefficient maps are generated. In the first 2D wavelet coefficient matrix, the value of each pixel with an (n,m) index is a first subset of wavelet coefficients selected from the 3D wavelet coefficient matrix 3 by the first coefficient selection rule. In the second 2D wavelet coefficient matrix, the value of each pixel with an (n,m) index is a second subset of wavelet coefficients selected from the 3D wavelet coefficient matrix 3 by the second coefficient selection rule. In the first 2D coefficient map, the value of each pixel with an (n,m) index is the l index of a voxel in the 3D wavelet coefficient matrix 3 that includes the first subset of wavelet coefficients selected by the first coefficient selection rule. In the second 2D coefficient map, the value of each pixel with an (n,m) index is the l index of a voxel in the 3D wavelet coefficient matrix 3 that contains a second set of wavelet coefficients selected by the second coefficient selection rule.

[0081] According to one particular embodiment, a first selection rule is applied to select the most relevant coefficients from the subset of coefficients comprising detail coefficients, and a second selection rule is applied to select the most relevant coefficients from the subset of coefficients comprising approximation coefficients. The detail coefficients comprise horizontal detail coefficients, vertical detail coefficients, and diagonal detail coefficients. In this particular embodiment, the first coefficient selection rule is: - calculating for each voxel of the 3D wavelet coefficient matrix with (n,m,l) indices the sum of the absolute values ​​of the horizontal detail coefficients, vertical detail coefficients and diagonal detail coefficients with (n,m,l) indices; - selecting, among all voxels of the 3D wavelet coefficient matrix with (n,m) indices and l indices ranging from 1 to L, the voxel containing the highest sum of the absolute values ​​of the horizontal detail coefficients, the vertical detail coefficients and the diagonal detail coefficients; Includes:

[0082] In this embodiment, the second coefficient selection rule is: - for each voxel of the 3D wavelet coefficient matrix with (n,m,l) indices, calculating the Laplacian variance of the approximation coefficients with (n,m,l) indices; - selecting, among all voxels of the 3D wavelet coefficient matrix with (n,m) index and l index ranging from 1 to L, the voxel containing the highest absolute value of the Laplacian variance of the approximation coefficients; Includes:

[0083] In this embodiment, the first 2D wavelet coefficient matrix contains information about the image details, while the second 2D wavelet coefficient matrix is ​​a smoothed representation of the image due to the denoising effect obtained from the low-pass filter of the wavelet transform.

[0084] According to one embodiment, the method comprises a step M50 of applying the inverse wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D greyscale EDF image 4.

[0085] In one embodiment, the method includes a step M60 of generating a 2D color composite image CC, where the value CC(n,m) of each pixel with (n,m) index is the value I(n,m,l) of the voxel of the color volume 1 with l index equal to the value CM(n,m) of the 2D coefficient map CM.

[0086] In one embodiment, the method also includes a step M70 of applying the inverse of the reversible color-to-grayscale transformation to the 2D grayscale EDF image 4 to obtain a 2D color EDF image 5.

[0087] In one embodiment, the method also includes a step M80 of transforming the 2D color composite image CC and the 2D color EDF image 5 into a color space that includes at least one chromaticity component and at least one intensity component.

[0088] The term color space refers to the mathematical representation of colors in a coordinate system where each color is defined by a vector, and the projections of the vectors on the axes of the coordinate system are the various components of the color in said color space.

[0089] Most color spaces use three components to represent color. Traditionally, biomedical images are acquired by devices such as cameras and scanners whose sensors have three channels: red (R), green (G), and blue (B), thus producing images in the RGB color space. In the RGB color space, the R, G, and B components are highly correlated, and furthermore, luminance information is not available. Therefore, for some applications, conversion from the RGB color space to another color space with a different coordinate axis system is appropriate, as in step M80. Each coordinate axis of the color space contains color information, which may be a luma value, a luminance, a function of luminance, or alternatively, chrominance (a combination of chroma and luma values) or a function of chrominance. Preferably, in the color space, each coordinate axis contains either chromaticity information, i.e., chrominance or a function of chrominance, or intensity information, i.e., a luma value or a function of luminance or luminance, and none of the coordinate axes contains a combination of the aforementioned chromaticity and intensity information.

[0090] In contrast to RGB, YUV, YES, YT1T2, CIE L * a * b * , YcbCr, CIE L * U * V * Color spaces such as RGB, HSV, etc., contain at least one coordinate axis that contains only intensity information and no chromaticity information, and at least one axis that contains only chromaticity information and no intensity information. In other words, each of these color spaces contains at least two axes along which intensity and chromaticity information are separated.

[0091] More specifically, YUV, YES, YT1T2, CIE L * a * b * , YcbCr, CIE L * U * V *contains one luminance component and two chrominance components.

[0092] Conversely, the color space HSV contains two intensity components, which are hue and saturation, and one chromaticity component, which is luminance / value.

[0093] In the above example, the color space of the converting step M80 contains three components. However, in alternative embodiments, color spaces with four or more dimensions may be used. These color spaces are typically used in print and graphic design.

[0094] According to one embodiment, the converting step M80 is configured to convert the 2D color composite image CC and the 2D color EDF image 5 into the YUV color space, where the luma component Y is the intensity component and U and V are the chromaticity components.

[0095] According to another embodiment, the color space of the converting step M80 is the following color space: CIE L * a * b * , YcbCr, CIE L * U * V * , one of the HSVs.

[0096] In one embodiment, the method further comprises a step M90 of concatenating at least one chromaticity component of the 2D color composite image CC and at least one intensity component of the 2D color EDF image 5 to obtain a color-fidelity extended depth-of-field (EDF) image.

[0097] Thus, in the above example (step M80 of converting the 2D color composite image CC and the 2D color EDF image 5 to YUV color space), the concatenating step M90 may include concatenating the U component of the 2D color composite image CC, the V component of the 2D color composite image CC, and the Y component of the 2D color EDF image 5.

[0098] Since the chromaticity information of the 2D color composite image CC is obtained from the initial color volume 1 in step M60 of generating the 2D color composite image CC, the 2D color composite image CC has improved color fidelity compared to volume 1. Since the detail information of the 2D color EDF image 5 is maximized by the coefficient selection rule, the 2D color EDF image 5 has improved recovery of details from color volume 1. By concatenating the intensity component of the 2D color EDF image 5 with the chromaticity component of the 2D color composite image CC, it is advantageously possible to combine recovery of details from color volume 1 with high color fidelity to color volume 1. As a result, method M makes it possible to obtain a “color-faithful” extended depth-of-field (EDF) image in which details from the initial volume 1 are recovered.

[0099] 1, the color space has one intensity component and two chromaticity components, and in the concatenation step M90, one intensity component of the color EDF image 7 and two chromaticity components of the color composite image CC are concatenated. Thus, in this particular embodiment, the color-fidelity extended depth-of-field (EDF) image has one intensity component and two chromaticity components.

[0100] The color-fidelity extended depth-of-field (EDF) image obtained in the concatenation step M90 may be further converted to any color space. Optionally, it is converted to the color space of the initial color volume 1, which may be the RGB color space. In another embodiment of the invention, the color-fidelity extended depth-of-field (EDF) image obtained is converted to a color space different from the color space of the initial volume 1.

[0101] As mentioned above, the color space of the converting step M80 may be a four- or more-dimensional color space. In this case, after the concatenating step M90, the color-fidelity extended depth-of-field (EDF) image may be converted to a color space with a smaller number of components. This embodiment is particularly advantageous because four or more components of such a color space usually contain redundant information, in the sense that they can be derived from the first three components.

[0102] According to one embodiment, the color-fidelity extended depth-of-field (EDF) image obtained in the concatenation step M90 is further segmented. Preferably, the segmentation step is performed using a U-net segmentation model.

[0103] A non-limiting example of steps M10 to M90 of the method according to the invention is shown in Figure 1. In Figure 1, the selection step M30 is not shown. Various modifications of this step can be implemented in steps M10 to M70 without departing from the result of obtaining a 2D wavelet coefficient matrix WCM, a 2D coefficient map CM, a 2D color composite image CC, and a 2D color EDF image 5.

[0104] The method M of the present invention makes it possible to obtain a 2D image with improved color fidelity and detail recovery from a color volume 1 by concatenating in color space the intensity component of the 2D color EDF image 5 and the chromaticity component of the color composite image 2D color composite image CC.

[0105] One advantage of the present invention will be better understood when read in conjunction with Figure 2. In the example shown in Figure 2, the objects of interest are cells 6, some of which form cell clumps 7. In particular, Figures 2a, 2b, and 2d represent non-limiting examples of images obtained before and after performing various steps of the method M of the present invention. Figures 2c and 2d show, respectively, a color image obtained from the image of Figure 2b after a color reassignment strategy according to the prior art and a color image obtained from the image of Figure 2b after a color reassignment strategy according to the present invention.

[0106] More specifically, Figure 2a is an example of an image belonging to a first color volume 1, which contains an in-focus object 8 and an out-of-focus object 9. In a color volume 1 of a biological sample, one color is usually dominant. In the image of Figure 2a, the dominant color of all objects is blue, except for element 10, which has a red dominant color. In the color volume 1 to which the image of Figure 2a belongs, blue is the dominant color, and red is a rare color. All objects in the images shown in Figures 2b-2d are in focus. Figure 2b is an example of a 2D color EDF image 5 obtained as the output of step M70, which applies the inverse of the reversible color-to-grayscale transformation to the 2D grayscale EDF image 4. In particular, the image of Figure 2b is obtained after inverse PCA. The red color of element 10 is lost after inverse PCA. Figure 2c shows a 2D color EDF image 5 obtained from the image of Figure 2b, to which a prior art color assignment technique is applied. The red color of element 10 is partially restored according to the prior art color assignment strategy. 2d shows a 2D color EDF image 5 obtained with the method M of the present invention. In particular, the image of FIG. 2d is obtained after converting the 2D color EDF image 5 and the 2D color composite image CC into the YUV color space and then concatenating the luma component Y of the 2D color EDF image 5 with the chroma components U, V of the 2D color composite image CC. In contrast to the image of FIG. 2c, in the image of FIG. 2d, the red color of element 10 is fully restored, and element 10 has the same color as in the image of FIG. 2a.

[0107] The color fidelity obtained by the prior art and the present method is compared using the chi-squared distance (χ). In particular, the chi-squared distance (χ) between the RGB-normalized color histogram of each image shown in Figure 2 and the color histogram of the image shown in Figure 2a is calculated. The chi-squared distance (χ) between the image shown in Figure 2d and the image shown in Figure 2a is smaller than the chi-squared distance (χ) between the image shown in Figure 2c and the image shown in Figure 2a. These results indicate that the image shown in Figure 2d obtained by the present method M advantageously has better color fidelity than the image shown in Figure 2c obtained by the prior art method. In particular, the present invention ensures accurate reproduction of elements with rare colors, i.e., elements whose chrominance components are not the dominant chrominance components in the original color volume 1. The main problems observed in the 2D color EDF image 5 are a lack of color accuracy and the presence of artifacts. The present method achieves the best color fidelity compared to other wavelet-based methods and subsequent color reconstruction strategies. For example, the example in FIG. 2 shows that elements with rare colors are better reproduced with the present method M than with prior art color reconstruction methods.

[0108] FIG. 3 illustrates a particular embodiment of the method of the present invention.

[0109] More precisely, in the example shown in Figure 3, the color space of the transforming step M80 has three components, so the result of the transforming step M80 can be represented diagrammatically as a pile of three prisms, each prism representing one component.

[0110] As mentioned above, in a concatenation step M90, at least one intensity component of the color EDF image 7 and at least one chromaticity component of the color composite image CC are concatenated.

[0111] In the example of FIG. 3, the intensity component is represented by the prisms above and in the middle of the pile, and the chromaticity component is represented by the prism below the pile.

[0112] For example, the top prism may represent a hue (H) component, the middle prism may represent a saturation (S) component, and the bottom prism may represent a value (V) component. Thus, in this example, the concatenating step M90 may include concatenating the H component of the 2D color composite image CC, the V component of the 2D color EDF image 5, and the S component of the 2D color EDF image 5.

[0113] In the example shown in FIG. 3, a color-fidelity extended depth-of-field (EDF) image has two intensity components and one chrominance component.

[0114] The invention also relates to a system 11 for analyzing a color volume 1 acquired from a biological sample.

[0115] The system 11 will now be described with reference to FIG.

[0116] Said system 11 comprises at least one input 12 adapted to receive a color volume 1. The color volume may be stored in one or more local or remote databases 13. The latter may take the form of storage resources available from any kind of suitable storage means, in particular an EEPROM (Electrically Erasable Programmable Read Only Memory) such as a RAM or a flash memory, possibly in an SSD (Solid State Disk).

[0117] The system 11 includes a processor 14, preferably configured to perform the steps of the method M according to any one of the embodiments described above.

[0118] While the system 11 described herein is versatile and includes several functions that can be performed alternatively or in any cumulative manner, other implementations within the scope of this disclosure include systems having only a portion of the functionality of the present invention.

[0119] Each of systems 11 is advantageously a device or physical portion of a device designed, configured, and / or adapted to perform the aforementioned functions and provide the aforementioned effects or results. In alternative implementations, any of systems 11 is embodied as a set of devices or physical portions of devices, whether grouped on the same machine or on different, possibly remote, machines. System 11 could, for example, be distributed across a cloud infrastructure and have functionality available to users as cloud-based services, or have remote functionality accessible through an API.

[0120] In the following, modules should be understood as functional entities rather than material, physically distinct components. As a result, they can be grouped together in the same tangible, concrete component or distributed across several such components. Also, each of these modules may itself be shared between at least two physical components. Furthermore, modules may be implemented in hardware, software, firmware, or any mixture thereof. They are preferably embodied within at least one processor of system 11.

[0121] According to one embodiment, the system 11 further includes a visualization module for displaying the volume 1, the grayscale volume 2, the 2D color composite image CC, the 2D grayscale EDF image 4, the 2D color EDF image 5, and the color-fidelity extended depth-of-field (EDF) image.

[0122] The visualization module may further display the 3D wavelet coefficient matrix 3, the 2D wavelet coefficient matrix WCM, and the 2D coefficient map CM.

[0123] The visualization module may be connected directly to the GPU. According to a variant, the visualization module is external to the system 11 and is connected to the system 11 by cable or wirelessly for transmitting the display signals. In this case, the system 11 comprises an interface for transmission or connection adapted to transmit the display signals to external display means, such as for example an LCD or plasma screen or a video projector.

[0124] The system 11 shown in Figure 5 interacts with a user interface 15, through which a user can input and retrieve information. The user interface 15 includes any suitable means for inputting or retrieving data, information, or instructions, in particular visual, tactile, and / or audio capabilities, which may include any or some of the following means well known to those skilled in the art: a screen, a keyboard, a trackball, a touchpad, a touchscreen, a loudspeaker, a voice recognition system.

[0125] In one embodiment, the biological sample is a urine sample. According to this embodiment, the system 11 may include an automated digital cytology system configured to acquire a volume 1 including at least two 2D color images.

[0126] The present invention further relates to a computer program product for generating a color-fidelity extended depth-of-field (EDF) image from a volume 1 obtained from a microscope, the computer program product comprising instructions that cause the computer to perform the steps of the method according to any one of the preceding embodiments when the program is executed by a computer.

[0127] A computer program product for performing the above-described methods can be written as a computer program, code segments, instructions, or any combination thereof, for individually or collectively instructing or configuring a processor or computer to operate as a machine or special-purpose computer that performs the operations performed by the hardware components. In one example, the computer program product includes machine code that is executed directly by the processor or computer, such as machine code generated by a compiler. In another example, the computer program product includes higher-level code that is executed by the processor or computer using an interpreter. A programmer skilled in the art can easily write instructions or software based on the block diagrams and flowcharts shown in the figures and corresponding descriptions herein that disclose algorithms for performing the operations of the above-described methods.

[0128] The invention further relates to a computer-readable storage medium comprising instructions that, when the program is executed by a computer, cause the computer to carry out the steps of the method according to any one of the preceding embodiments.

[0129] According to one embodiment, the computer readable storage medium is a non-transitory computer readable storage medium.

[0130] A computer program implementing the method of the present embodiment can generally be distributed to users on a distribution computer-readable storage medium, such as, but not limited to, an SD card, an external storage device, a microchip, a flash memory device, a portable hard drive, and a software website. From the distribution medium, the computer program can be copied to a hard disk or similar intermediate storage medium. The computer program can be executed by loading computer instructions from the distribution medium or intermediate storage medium into the computer's execution memory and configuring the computer to operate according to the method of the present invention. All of these operations are familiar to those skilled in the art of computer systems.

[0131] The instructions or software and any associated data, data files, and data structures for controlling a processor or computer to implement the hardware components and perform the aforementioned methods are recorded, stored, or fixed on one or more non-transitory computer-readable storage media. Examples of non-transitory computer-readable storage media include read-only memory (ROM), random-access memory (RAM), flash memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state disk, and any device known to one of ordinary skill in the art that can store the instructions or software and any associated data, data files, and data structures in a non-transitory manner and provide the instructions or software and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the instructions. In one example, the instructions or software and any associated data, data files, and data structures are distributed across a network of connected computer systems such that the instructions and software and any associated data, data files, and data structures are stored, accessed, and executed by processors or computers in a distributed fashion. [Explanation of symbols]

[0132] 1 Color Volume 2 Grayscale Volume 3 3D wavelet coefficient matrix 4 2D grayscale EDF images 5 2D color EDF images 6 cells 7 Cell clumps 8 Objects in focus 9. Out of focus objects 10 Red Elements 11 System 12 Input 13 Database 14 processors 15 User Interface 16 Output

Claims

1. 1. A computer-implemented method (M) for generating a color-fidelity extended depth-of-field (EDF) image from a color volume (1) of a biological sample having dimensions (N, M, L) and values ​​I(n, m, l), where voxels (n, m, l) for fixed l from 1 to L are images acquired at different focal depths in the z-direction using a microscope, the method comprising the following steps: a) receiving a color volume (1) and generating a grayscale volume (2) by applying an invertible color-to-grayscale transform (M10) to the color volume (1); b) applying a wavelet transform (M20) to each set of N×M voxels of the grayscale volume (2) with the same l index and (n,m) indices ranging from (1,1) to (N,M) to obtain a 3D wavelet coefficient matrix (3), The value of each voxel with (n, m, l) index contains a set of wavelet coefficients, Steps and c) for each group of L voxels of the 3D wavelet coefficient matrix (3) having the same (n, m) index and l index ranging from 1 to L, selecting one set of wavelet coefficients using a predefined coefficient selection rule (M30); d) generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM (M40), The value WCM(n,m) of each pixel with (n,m) index is a set of wavelet coefficients selected from the 3D wavelet coefficient matrix (3) by the coefficient selection rule; The value C M (n, m) of each pixel with an (n, m) index is the l index of a voxel in the 3D wavelet coefficient matrix (3) that contains the set of wavelet coefficients selected by the coefficient selection rule. Steps and e) applying (M50) the inverse of the wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D grayscale EDF image (4); f) generating a 2D color composite image CC (M60), The value CC(n,m) of each pixel with index (n,m) is the value I(n,m,l) of the voxel of the color volume (1) with index l equal to the value CM(n,m) of said 2D coefficient map CM; Steps and g) applying (M70) the inverse of the reversible color to grayscale transformation to the 2D grayscale EDF image (4) to obtain a 2D color EDF image (5); h) converting the 2D color composite image CC and the 2D color EDF image (5) into a color space containing at least one chromaticity component and at least one intensity component (M80); i) concatenating (M90) at least one chromaticity component of the 2D color composite image CC and at least one intensity component of the 2D color EDF image (5) to obtain the color-fidelity extended depth-of-field (EDF) image; A method comprising:

2. In the selecting step (M30), a first subset of wavelet coefficients is selected using a first predefined coefficient selection rule and a second subset of wavelet coefficients is selected using a second predefined coefficient selection rule, and the step of generating (M40) the 2D wavelet coefficient matrix WCM and the 2D coefficient map CM comprises: generating a first 2D coefficient map from said first subset of wavelet coefficients (M41); generating a second 2D coefficient map from said second subset of wavelet coefficients (M42); a step (M43) of combining said first 2D coefficient map and said second 2D coefficient map to obtain one 2D coefficient map CM; generating a 2D wavelet coefficient matrix WCM (M44), the value of each pixel WCM(n,m) is the value of the voxel of the 3D wavelet coefficient matrix having an (n,m) index and having an l index equal to the value CM(n,m) of the obtained 2D coefficient map CM; 10. The method of claim 1, comprising:

3. The step (M43) of combining the first 2D coefficient map and the second 2D coefficient map comprises: filtering the first 2D coefficient map and the second 2D coefficient map; averaging the first and second 2D coefficient maps; rounding the averaged 2D coefficient map to obtain a single 2D coefficient map C; The method of claim 2 , comprising:

4. The method described in claim 3, wherein a median filter is used for filtering.

5. The method according to any one of claims 1 to 4, wherein the set of wavelet coefficients comprises at least four wavelet coefficients, and the subset of wavelet coefficients comprises at least one wavelet coefficient.

6. The method according to any one of claims 1 to 5, wherein the color-fidelity extended depth-of-field (EDF) image is transformed into a different color space.

7. A method according to any one of claims 1 to 6, wherein the color-fidelity extended depth-of-field (EDF) image is converted to the color space of the color volume (1).

8. The method according to any one of claims 1 to 7, wherein the reversible color to grayscale transformation is a principal component analysis (PCA).

9. The method according to any of claims 1 to 8, wherein the wavelet transform applied in the applying step (M20) is a stationary wavelet transform (SWT).

10. The method according to any of claims 1 to 9, wherein the color volume (1) is a biomedical image, and the method further comprises the step of segmenting the color-fidelity extended depth-of-field (EDF) image.

11. A system (11) for analyzing a biological sample, comprising: at least one input (12) adapted to receive a color volume (1); At least one processor (14), generating a grayscale volume (2) by applying (M10) a reversible color-to-grayscale transformation to the color volume (1); applying (M20) a wavelet transform to each set of N×M voxels of the grayscale volume (2) having the same l index and (n,m) indices ranging from (1,1) to (N,M) to obtain a 3D wavelet coefficient matrix (3); where the value of each voxel with (n, m, l) index contains a set of wavelet coefficients, For each group of L voxels of the 3D wavelet coefficient matrix (3) having the same (n, m) index and an l index ranging from 1 to L, select (M30) one set of wavelet coefficients using a predefined coefficient selection rule; generating a 2D wavelet coefficient matrix WCM and a 2D coefficient map CM (M40); where the value WCM(n,m) of each pixel with an (n,m) index is the set of wavelet coefficients selected from the 3D wavelet coefficient matrix (3) by the coefficient selection rule, and the value CM(n,m) of each pixel with an (n,m) index is the l index of a voxel in the 3D wavelet coefficient matrix (3) that contains the set of wavelet coefficients selected by the coefficient selection rule; applying (M50) the inverse of the wavelet transform to the 2D wavelet coefficient matrix WCM to obtain a 2D grayscale EDF image (4); generating a 2D color composite image CC (M60); where the value CC(n,m) of each pixel with index (n,m) is the value I(n,m,l) of the voxel of the color volume (l) with index l equal to the value CM(n,m) of the 2D coefficient map CM; applying (M70) the inverse of the reversible color to grayscale transformation to the 2D grayscale EDF image (4) to obtain a 2D color EDF image (5); Transforming (M80) the 2D color composite image CC and the 2D color EDF image (5) into a color space including at least one chromaticity component and at least one intensity component; concatenating (M90) the at least one chromaticity component of the 2D color composite image CC and the at least one intensity component of the 2D color EDF image (5) to obtain a color-fidelity extended depth-of-field (EDF) image; a processor configured to at least one output (16) adapted to provide said color-fidelity extended depth-of-field (EDF) image; A system (11) comprising:

12. The system (11) of claim 11, wherein the biological sample is a urine sample and the system includes an automated digital cytology system configured to acquire a color volume (1), the color volume (1) including at least two 2D color images.

13. A bladder cancer detection device comprising the system of claim 11 or claim 12.

14. A computer program for analyzing a biological sample, comprising instructions that cause the computer to carry out the steps of the method of any one of claims 1 to 10 when the computer program is run by the computer.

15. A non-transitory computer-readable storage medium comprising instructions that, when the program is executed by a computer, cause the computer to perform the steps of the method according to any one of claims 1 to 10.

Images