Particle detection device using lensless imaging

Abstract

The present invention relates to an apparatus for detecting particles using lensless imaging. The apparatus comprises a memory (4) configured to receive a plurality of z-stack images obtained from lensless imaging of a biological sample, and a picker (6) configured to determine, based on the z-stack images, a focused image in which each pixel is associated with the z-stack image on the one hand and with the intensity of the pixel in the z-stack image on the other hand. The picker is configured to determine the z-stack image for a given pixel by calculating, for each z-stack image and for a given pixel, a parsimony score based on the intensity of the given pixel and the intensities of neighboring pixels, and selecting the z-stack image with the highest parsimony score. The apparatus further comprises a selector (10) and a computer (12).

Description

[Technical Field]

[0001] The present invention relates to the field of imaging of samples, in particular biological samples, by lensless imaging. [Background technology]

[0002] Lensless imaging consists in placing a sample between a light source and an image sensor, without any optical magnification between them. The image sensor collects an image (also called a hologram) of the light intensity transmitted by the sample. This image is formed by the interference pattern between, on the one hand, the light waves emitted by the light source and transmitted by the sample, and, on the other hand, the diffracted waves resulting from the diffraction by the sample of the light waves emitted by the light source.

[0003] These interference patterns are the subject of research to reconstruct images that exploit the information contained in the observed sample. This technique is considered particularly promising due to the simplicity of the materials involved, as no optics are required, and is the subject of numerous developments in the field of hematology.

[0004] Counting red blood cells (hereinafter "RBCs") is possible in the context of platelet-rich plasma (hereinafter "PRP") administration because the number of RBCs is low enough so as not to mask other particles. In the case of whole blood, they must be lysed, and only the platelet population (hereinafter "PLTs") and white blood cells (hereinafter "WBCs") can be counted, with the added difficulty of having a noisy background due to the lysed RBCs.

[0005] Patent Publications FR3034196 and FR3049348 teach the calculation of statistics from detected cells, particularly z-profiles, to count and differentiate WBCs. Unlike the state of the art, these methods require a prerequisite: image reconstruction from diffraction images using an accurate method that can preserve local information of pseudopoint particles and does not leave significant artifacts. For example, Patent Publication FR3049347 teaches such a reconstruction method.

[0006] To date, only the WBC-PLT method described in Patent No. FR3082943 can count PLTs on PRP and in lysed blood (using different settings), and additional work is needed to learn how to distinguish RBCs (in PRP) from other particles. This method is based on detecting particles in a stack of images reconstructed around a mean plane, and then merging the detection planes to establish a detection map. This map is then used directly to count particles. When multiple particle types are present, it is appropriate to adapt this general strategy to each particle type. For each object (connected component) in the thus calculated detection mask, metrics are calculated to determine the object's characteristics or their number. This last point is a significant limitation: in images containing a large number of particles, many particle multiplets are (statistically) present, and this method cannot address this issue when multiple particle types are present. Ultimately, attempts to use this method on lysed blood have been unsuccessful.

[0007] Other methods have been developed that are based on deep neural network (DNN) models. Training was performed on an extended sample basis to ensure sufficient input, taking into account the counts provided by a reference machine. While performance was quite good, the method is limited by the assumptions inherent in this type of method, namely, the lack of sufficient training data with a sufficiently reliable reference. Data acquisition is a lengthy and expensive process, which is a limiting factor for the application of DNN-based techniques. Furthermore, training is strongly influenced by the reference machine, which creates issues related to the variability of the measurement machine. Furthermore, training is performed on noisy images, which induces training noise in the dissolution as well as counting and classification. Finally, since there is no constructible reference, this training allows for particle counting but not classification. Summary of the Invention [Problem to be solved by the invention]

[0008] Therefore, in the context of lensless imaging in whole blood, no technique provides complete satisfaction for PLT counting. [Means for solving the problem]

[0009] The present invention improves this situation by proposing an apparatus for detecting particles using lensless imaging, comprising: a memory configured to receive a plurality of z-stack images obtained from lensless imaging of a biological sample; an initiator configured to determine, based on the z-stack images, a focused image in which each pixel is associated with the z-stack image on the one hand and with the intensity of this pixel in the z-stack image on the other hand, the initiator configured to determine the z-stack image for a given pixel by calculating, for each z-stack image and for a given pixel, a parsimony score based on the intensity of the given pixel and the intensities of neighboring pixels, and selecting the z-stack image with the highest parsimony score; a selector configured to determine, for each pixel of the focused image, whether this pixel is maximum in the focused image or in a local neighborhood centered on this pixel in the z-stack images associated with this pixel in the focused image, and if so, to save this pixel in a list of selected regions; and a computer configured to calculate, for each pixel in the list of selected regions, a metric value based on the intensities of these pixels, and to generate an image of the metric value that allows particles associated with each pixel in the list of selected regions to be distinguished from one another.

[0010] This device is particularly advantageous because it can perform reliable counting of PLTs and can be integrated into laboratory equipment.

[0011] According to various embodiments, the invention may have one or more of the following features. The computer is configured to implement a metric selected from: a metric based on Gaussian modeling of the neighboring intensities of each pixel in the list of selected sites in the z-stack image associated with this pixel in the focus image by combining one or more of the covariance, height, and eccentricity coefficients of the Gaussian (Gaussian distribution); and a metric based on averaging the z-gradient of the intensities of neighboring pixels centered around each pixel in the list of selected sites, the gradient evaluated for the z-stack image corresponding to each pixel in the list of selected sites in the focus image (FocImg[]). The initiator is configured to use a pq-means based parsimony score whose intensity is modulated by a Gaussian window centered on a given pixel. The initiator is configured to use pq averaging with a Gaussian with p equal to 1 / 2, q equal to 3, a square neighborhood of 17 pixels, and a standard deviation equal to 2.5 The selector is configured to browse through the list of selected regions and, if two pixels are in close proximity, to keep the one whose associated intensity in the focus image is most significant. The apparatus further comprises a filter configured to recalculate the focus image by recalculating the parsimony score for each given pixel by limiting the neighboring pixels to those pixels whose associated z-stack image in the focus image is the same as that associated with the given pixel, or to those pixels that are immediately neighboring in z.

[0012] The present invention relates to a method for particle detection using lensless imaging, the method comprising: a) receiving a plurality of z-stack images obtained from lensless imaging of a biological sample; b) determining a focus image associated with each pixel on the one hand with a z-stack image and on the other hand with the intensity of this pixel in this z-stack image, said determination comprising, for a given pixel: b1) for each of the z-stack images, and for a given pixel, by calculating a parsimony score from the intensity of the given pixel and the intensities of neighboring pixels; and b2) The action taken by selecting the z-stack image with the highest parsimony score; c) for each pixel in the focus image, determining whether this pixel is the largest in the focus image or in a local neighborhood centered on this pixel in the z-stack image associated with this pixel in the focus image, and if so, saving this pixel in a list of selected regions; d) calculating a metric value for each pixel in the list of selected sites based on the intensity of those pixels, enabling an image of the metric values to be generated that allows particles associated with each pixel in the list of selected sites to be distinguished from one another.

[0013] According to various embodiments, the invention may have one or more of the following features. In operation d), the metric is selected from a metric based on Gaussian modeling of the intensities of the neighborhood of each pixel in the list of selected sites in the z-stack image associated with this pixel in the focused image by combining one or more of the Gaussian's covariance, height, and eccentricity coefficients, and a metric based on an average of the z-gradients of the intensities of neighboring pixels centered around each pixel in the list of selected sites, the gradient evaluated for the z-stack image corresponding to each pixel in the list of selected sites in the focused image. · Operation b1) uses a parsimony score based on pq-means whose intensity is modulated by a Gaussian window centered on a given pixel. Action b1) uses as the parsimony score the pq average with p equal to 1 / 2, q equal to 3, a square neighborhood of 17 pixels, and a Gaussian with standard deviation equal to 2.5. Operation c) involves browsing the list of selected regions and, if two pixels are immediately neighbors, keeping the one whose associated intensity in the focused image is most significant. Operation b) further includes b3) recalculating the focus image by recalculating the parsimony score for each given pixel by limiting the neighboring pixels to pixels whose associated z-stack image in the focus image is the same as that associated with the given pixel, or immediate neighboring pixels in z.

[0014] The invention also relates to a computer program comprising instructions for carrying out the method according to the invention, a data storage medium on which such a computer program is recorded, and a computer system comprising a processor connected to a memory, the memory having recorded thereon such a computer program. [Brief explanation of the drawings]

[0015] Other characteristics and advantages of the invention will become more apparent from the following description, with reference to examples given for illustrative and non-limiting purposes, and with reference to the drawings, in which: [Figure 1] 1 shows a schematic diagram of an apparatus according to the present invention. [Figure 2] 2 shows an example of an operating loop of the device of FIG. 1. [Figure 3] 3 shows an example of the implementation of the function implemented in the first operation of FIG. 2. [Figure 4] 3 shows an example of the implementation of the function implemented in the second operation of FIG. 2. [Figure 5] 10 shows an example of the implementation of the function implemented in the third operation of FIG. 2. [Figure 6] 1 shows a lensless image and a z-stack image derived therefrom. [Figure 7] This shows the focused image obtained by the function in FIG. [Figure 8] The focus image obtained by the function in FIG. [Figure 9] This shows the focused image obtained by the function in FIG. [Figure 10] 6 represents an image of a first metric obtained from an image obtained by the function of FIG. 5. [Figure 11]6 represents an image of a second metric obtained from an image obtained by the function of FIG. 5.

[0016] The drawings and description below contain, for the most part, elements of a specific nature, and therefore they not only serve to better understand the invention, but also contribute, where appropriate, to its definition. DETAILED DESCRIPTION OF THE INVENTION

[0017] This description may contain material that may be protected by copyright. The Rights Holder has no objection to anyone reproducing this patent document or its description as it appears on the official files. The Rights Holder fully reserves its rights to the remainder.

[0018] Figure 1 shows a schematic diagram of an apparatus 2 according to the invention. The apparatus 2 comprises a memory 4, an initiator 6, a filter 8, a selector 10 and a computer 12. As will be seen below, the filter 8 is optional.

[0019] Memory 4 receives data that are the subject of functions and calculations performed by device 2. Device 2 primarily processes reconstructed images obtained from lensless images and focused at various depths. Patent published under FR3082944 describes an example implementation of this type of image. More specifically, the patent describes how a plurality of images, hereinafter referred to as "z-stack images" or "z-focus images," are reconstructed from the lensless images, corresponding to the focusing of the lensless imaging at a particular depth.

[0020] Figure 6 shows acquired images from lensless imaging and modulus and phase images reconstructed from the acquired images. As will be shown below, these are combined modulus and phase images used as z-stack images in the context of the present invention. Alternatively, only the phase image or only the modulus image can be used if it contains the information of interest.

[0021] The difficulty in processing these images is that platelets constitute elements with sizes in the pixel range, are very numerous, and are present in an image with a very noisy background due to RBC (red blood cell) lysis. Therefore, it is impossible to distinguish PLTs (platelets) on the one hand from RBC debris on the other hand with the naked eye. The data of a z-stack image comprise, on the one hand, pixels identified by their coordinates and, on the other hand, intensity values associated with these pixels. In the example described here, the images have a resolution of 3840 × 2748 pixels, and their intensities are modified. That is, the z-stack images are processed in the form |I-1| instead of |I|, since a particle-free image after reconstruction should theoretically have a constant value of 1. Thus, by working with |I-1|, the amplitude of the disturbances locally induced by diffraction is taken into account.

[0022] The memory 4 can be any type of data storage suitable for receiving digital data, such as a hard disk, a hard disk flash memory, any form of flash memory drive, RAM, a magnetic disk, local or cloud distributed storage, etc.

[0023] In the example described here, the memory 4 receives all the data mentioned above, and more generally all data relating to the device 2, i.e. the programs and software instantiating the initiator 6, the filter 8, the selector 10 and the computer 12, their parameters, the received input data, the output data and the data stored in a buffer memory. The data calculated by the device can be stored in any type of memory similar to the memory 4 or in the memory 4. These data can be erased after the device has performed or maintained its task.

[0024] The initiator 6, the filter 8, the selector 10 and the computer 12 have direct or indirect access to the memory 4. They may be implemented in the form of suitable computer code executed by one or more processors. The term "processor" means any processor adapted for the calculations described below. Such a processor may be implemented in any known manner: a microprocessor for a personal computer, laptop, tablet or smartphone, a dedicated chip of FPGA or SoC type, a computing resource on a grid or cloud, a cluster of graphics processors (GPUs), a microcontroller or any other form suitable for providing the computing power required for the implementations described below. One or more of these elements may also be implemented in the form of specialized electronic circuits, for example, an ASIC. A combination of a processor and an electronic circuit is also conceivable. A processor dedicated to machine learning is also conceivable.

[0025] The initiator 6, filter 8, selector 10 and computer 12 are shown separately here because they each perform different functions. This modular description is intended to provide a better understanding of the functional blocks implemented by the device 2. Of course, two or more of these elements can be grouped together as long as the functional relationships are equivalent.

[0026] FIG. 2 shows an example of an operating loop of the apparatus of FIG. 1, allowing a better understanding of the individual functions of the initiator 6, filter 8, selector 10 and computer 12.

[0027] The loop begins at operation 200, where the initiator 6 executes the Focus() function. The Focus() function accesses the z-stack images from the same lensless image and returns an image, called the focus image, in which each pixel is associated with the z-stack image index on the one hand and the intensity of this pixel in the z-stack image on the other hand. The purpose of the Focus() function is to identify, for each pixel, what the best focus depth is and find the most relevant plane to infer the nature of each pixel. This function is further explained with reference to FIG. 3. FIG. 7 shows an example of an acquired focus image.

[0028] The focused image is then reprocessed in optional operation 210, and Filter 8 executes the Refine() function. As shown below, the Focus() function uses a parsimony score combined with a Gaussian window, which allows for a high-quality focused image while maintaining a certain regularity; otherwise, pixels of the same cell would not necessarily be in the same focus. However, in the context of platelet detection, this regularity comes at a price. Indeed, due to the large number of cells present in the chamber, the distance between two cells can be extremely small, and a cell may easily reside within a neighboring window, interfering with the parsimony score. The Gaussian window mitigates this phenomenon, but if two cells are several pixels apart, the cell with the higher "intensity" will dominate the other, and the cell with the lower "intensity" will likely have a suboptimal focus. The Refine() function aims to solve this problem by narrowing the focused image from the focal plane found in operation 200 and considering only pixels in the same or very close plane for each pixel's parsimony score. This function is further explained with reference to Figure 4. FIG. 8 shows an example of a refined focus image.

[0029] The selector 10 then uses the focus image (and possibly the refinement image) to select pixels that are likely to form platelet regions. To do this, the selector 10 executes the Selec() function in operation 220. This function prioritizes pixel detection and handles the multiplet case as simply and best as possible, regardless of connected components. The important thing is not to "forget" any regions, but to consider the objects being searched for as unique pixels, delegating the operation of distinguishing between object pixels (platelets) and background pixels to the final operation. This function is further explained with reference to Figure 5. Figure 9 shows an example of an image of a captured region.

[0030] Finally, in operation 230, the computer 12 uses the list of regions determined by the selector 10 to run the Class() function, and metrics are calculated to generate an image that can distinguish between platelet pixels and background pixels. This operation therefore allows PLTs to be counted and their concentration to be estimated. Figures 10 and 11 show the resulting classification images for two different metrics.

[0031] FIG. 3 shows an example implementation of the Focus() function executed by the initiator 6. As mentioned above, the purpose of this function is to determine, for each pixel in a z-stack image, the z focal plane that best suits that pixel. To do this, applicants have discovered that a parsimony score applied to each pixel's point neighborhood produces unparalleled results. Parsimony scores have the advantage of being adaptive to pixel objects because they attempt to maximize the level of a single pixel at the expense of all other pixels. Several parsimony scores exist, and the Focus() function described here uses a modified version of the pq-means metric (or pq-means). pq-means is fairly simple to calculate, yet has good theoretical properties and can be easily implemented on a machine. Thus, in the example described here, the metric used has the following formula:

number

[0032] In the example described here, the applicant has found that p=½ and q=3 give the best results. However, other combinations also provide good results, for example, p=½ and q=4, or p=½ and q=4. For pixel searches, the neighborhood size is determined here from a Gaussian. Thus, starting from a standard deviation of 2.5, a Gaussian covering 3 sigma (σ) (representing the traditional approach) results in a neighborhood of 17 pixels (3 × 2.5 rounded to 8, plus 2 times the center pixel). Other Gaussian window dimensions can be maintained.

[0033] Thus, in a first operation 300, a first current pixel cp is initialized with the coordinates of the first pixel cp0 in the z-stack image, and then in operation 310, the Wind() function reads all neighbors within a square window of 17 pixels centered on pixel cp and records their intensities in variable W. Variable W receives the neighbors of all z-stack images, so that all z-stack images are processed at once. Alternatively, the z-stack images can be processed serially or in parallel, with focus selection occurring after all of these calculations.

[0034] Then, in operation 320, the parsimony scores are calculated by the Pars() function, the results of which are stored in an array CP[] which receives, for each focal depth z corresponding to the z-stack image, on the one hand the parsimony scores of the pixels cp in the considered z-stack image and, on the other hand, the intensities of the pixels cp in the considered z-stack image.

[0035] The initiator 6 then executes the Max() function in operation 330. The Max() function looks through the array CP[] and returns the couple (z; I) (I is the intensity stored in the array CP[] of the associated z-stack image) with the highest parsimony score. In fact, the z-stack image with the highest parsimony score is the image in which this pixel has the best focus. Alternatively, it is also possible to determine the z corresponding to the maximum intensity by interpolating the curve of values I and, if necessary, interpolating intensity values within this plane for the remainder of the calculations. In the latter case, this new plane can be introduced into the z-stack image data, keeping only the neighborhood of the considered point, which will be described below.

[0036] Then, in operation 340, the triplet (cp;z;I) is stored in the array FocImg[]. The array FocImg[] contains all the information of the focus image and is built up step by step. In the following, the array FocImg[] can point to the focus image.

[0037] Finally, in operation 350, the next pixel is read, and if not all pixels have been processed, the Focus( ) function resumes operation 310. In that case, in operation 399, it ends.

[0038] In the above, the Focus() function processes pixels sequentially. Obviously, a parallelized version of this process is possible. Thus, the focus image FocImg[] has an 8-pixel-thick frame where z is not identified, since all measurements are zero. This has the advantage of limiting the risk of edge effects. At the end of processing by device 2, a 35-pixel-thick frame is removed during counting. Nevertheless, this is negligible compared to the size of the image considered (10 MP). The applicant also recognizes that the pq-average is not the only parsimony score that provides satisfactory results. Thus, the Gini metric (or Gini index) combined with a Gaussian window is a variant that is part of the scope of the present invention.

[0039] Figure 4 shows an example implementation of the optional Refine() function implemented by filter 8. In fact, this function is very close to the Focus() function, and similar or identical operations have reference numbers that only change in the hundreds digit, and only the differences are described.

[0040] The only difference between the Focus() and Refine() functions is the definition of the neighborhood window. Thus, in operation 410, the filter 8 executes the Wind2() function, which separately defines the window W. Indeed, since the goal is to consider only neighbors that are in the same focal plane as the pixel cp or are close (e.g., only one plane away), the Wind2() function defines the window W in a similar way to the Wind() function, but at each pixel, it reads the value z of this pixel in the focus image FocImg[] and discards this pixel if this z differs from the z of the z-stack image of the current pixel cp. In this way, in each z-stack image, the parsimony score for each pixel is refined using only potentially relevant pixels.

[0041] The Refine() function can be repeated until a threshold for parsimony score correction is reached. Alternatively, the function can be repeated a fixed number of times. The applicant has found that four runs provide the best ratio between improvement and computational cost. The main benefit provided by the Refine() function is the improvement of bias in the linear regression of measurements by device 2. Indeed, tests of existing databases performed by the applicant have shown that using the Refine() function, the bias (ordinate of the origin of the linear regression) goes from 35 giga platelets per liter to 24 giga platelets per liter, which would ideally be zero. In embedded use situations, the Refine() function can still be omitted due to its prohibitive computational cost. The Gaussian window is not critical for the Refine() function and can be fixed or gradually relaxed with each iteration.

[0042] Figure 5 shows an example implementation of the Selct() function executed by the selector 10. Once the focus image is determined, the goal is to identify among the pixels those that correspond to potential platelet sites, which means performing a search for local maxima.

[0043] Thus, in a first operation 500, a first current pixel cp is initialized using the coordinates of the first pixel cp0 in the z-stack image, and two operations 510 and 520 are initiated in parallel. In operation 510, a MaxLoc() function determines a Boolean value indicating whether the current pixel cp constitutes a local maximum in the z-stack image, the depth of which is associated with the current pixel cp in the focus image FocImg[]. This is done by comparing the intensity I of the current pixel cp with the intensities of all connected pixels. Operation 520 is very close, except that the neighborhood is acquired in the focus image FocImg[] itself, not the z-stack image of the focus of the current pixel cp. Then, in operation 530, the value m1 from operation 510 and the value m2 from operation 520 are compared using a logical OR. Thus, this test is positive if one of the two values m1 and m2 is equal to 1. In fact, two neighboring pixels do not necessarily have the same focus, and it is possible to have a pixel that is a local maximum on the focused image but not on the image of that focal plane.

[0044] If the test in operation 530 is positive, the pixel cp is added to the array of selected regions SS[] in operation 540, and a test in operation 550 determines whether there remains a current pixel to process. If so, the Select() function resumes operations 510 and 520 for the new current pixel. Otherwise, the Sng1() function is executed in operation 560. The Sng1() function is used to take into account the fact that, among the selected pixels, it is quite common for two pixels to be on the same particle. Consequently, once the list SS[] is stopped, the Sng1() function scans it to identify pairs of directly adjacent selected pixels and keeps the one with the greatest intensity in the focus image FocImg[]. Finally, the Select() function ends in operation 599.

[0045] Above, the Select() function processes pixels sequentially. Of course, a parallel version of this process is also possible. Furthermore, to provide a visually meaningful representation, the output SS[] list can be equated to an image of the region where the selected region has a value of 1 and the other pixels have a value of zero. Again, the image of the region and the SS[] list refer to the same thing and can be used interchangeably or together. Alternatively, operation 560 can be omitted.

[0046] Once the image of the site is determined, a metric must be applied that can classify pixels as particles or background. The applicant has performed extensive research and identified two families of metrics that can distinguish particles.

[0047] The first family is of Gaussian character and includes the following two metrics: The first metric: for each site, a 3x3 dimensional (alternatively, a 5x5, 7x7, or other window can be used to detect particles of different sizes) neighborhood centered on the site of intensities in the z-stack images associated with the site in the focus image FocImg[] is modeled by a Gaussian, the inverse of whose standard deviation constitutes the metric. A second metric, where for each site, a 3x3 neighborhood of intensities in the z-stack images associated with the site in the focus image FocImg[] is modeled by a Gaussian, and its height constitutes the metric. Alternatively, the metric can be based on the variance divided by the height.

[0048] Figures 10 and 11 show how these two metrics can separate PLTs, WBCs and background using additional separation zones.

[0049] The second family is based on z-gradients. In this version, for each pixel in a 3x3 dimensional neighborhood centered on each feature, a z-gradient is calculated using the intensities of these pixels in the individual z-stack images. This gradient is evaluated around the z-depth associated with the feature in the focus image FocImg[]. The metric is then defined as the average of the calculated gradients.

[0050] For example, other metric families are contemplated. In particular, metrics can be calculated for neighborhoods other than 3x3 neighborhoods, and the metrics calculated for these neighborhoods may be of properties other than Gaussian modeling. The neighborhoods selected in this way can be projected into various bases, such as, but not limited to, the frequency base of the Fourier transform, a base formed by a set of orthogonal polynomials, a decomposition of the neighborhood into wavelets, etc.

[0051] The device and method according to the present invention can also be used for other types of samples for which characterization is desired. These samples may in particular include a medium in which the particles are immersed. The medium may be a liquid medium. It may include body fluids obtained, for example, from blood, urine, lymph, cerebrospinal fluid. It may also be a culture medium containing nutrients that allow the growth of microorganisms or cells. The term "particle" means, without limitation: Cells (cultured cells or somatic cells, e.g., blood cells) Microorganisms, such as bacteria, yeasts, and microalgae Solid particles, such as microbeads, which can be functionalized to facilitate binding to the analyte. Particles that form emulsions in a medium, especially particles that are insoluble in the medium, for example, lipid droplets in an aqueous medium.

Claims

1. A device for detecting particles using lensless imaging, A memory (4) configured to receive multiple z-stack images obtained from a lensless image of a biological sample, An initiator (6) configured to determine a focus image (FocImg[]) for each pixel based on the aforementioned z-stack image, which is associated with the z-stack image on the one hand and with the intensity of the pixel in the z-stack image on the other hand, and to determine a z-stack image for a predetermined pixel by calculating a saving score for each of the z-stack images and for a predetermined pixel based on the intensity of the predetermined pixel and the intensity of neighboring pixels, and by selecting the z-stack image with the highest saving score, A selector (10) is configured to determine, for each pixel in the focus image (FocImg[]), whether that pixel is the largest within the focus image (FocImg[]) or within the local neighborhood centered on that pixel in the z-stack image associated with that pixel in the focus image (FocImg[]), and if so, save that pixel to a list of selected regions (SS[]), An apparatus comprising: a computer (12) configured to calculate a metric value for each pixel in a list of selected regions based on the intensity of these pixels, and to generate an image of the metric values ​​that distinguishes the particles associated with each pixel in the list of selected regions from one another.

2. Computer (12) - A metric based on Gaussian modeling of the intensity of the neighborhoods of each pixel in a list of selected regions in the z-stack image associated with this pixel in the focus image by combining one or more of the Gaussian covariance, height, and eccentricity coefficient, and The apparatus according to claim 1, configured to implement a metric selected from: a metric based on the average of the z-gradients of the intensity of neighboring pixels centered on each pixel in a list of selected regions, wherein the gradient is evaluated with respect to the z-stack image corresponding to each pixel in the list of selected regions in the focus image (FocImg[]).

3. The apparatus according to claim 1, wherein the initiator (6) is configured to use a pq-average savings score whose intensity is modulated by a Gaussian window centered on a predetermined pixel.

4. The apparatus according to claim 3, wherein the initiator (6) is configured to use a pq mean with a Gaussian having p equal to 1 / 2, q equal to 3, a square neighborhood of 17 pixels, and a standard deviation equal to 2.

5.

5. The apparatus according to any one of claims 1 to 4, wherein the selector (10) is configured to browse a list of selected regions and, if two pixels are in immediate neighboring positions, to maintain the one with the most significant relevant intensity in the focus image (FocImg[]).

6. The apparatus according to any one of claims 1 to 4, further comprising a filter (8) configured to recalculate the focus image (FocImg[]) by recalculating a saving score for each predetermined pixel by limiting the neighboring pixels to pixels in z that are identical to the associated z-stack image in the focus image (FocImg[]) or to pixels in the immediate vicinity of z.

7. A particle detection method using lensless imaging, a) Operation to receive multiple z-stack images obtained from lensless images of biological samples, b) An operation in which each pixel is associated with a z-stack image on the one hand, and a focus image (FocImg[]) associated with the intensity of that pixel in the z-stack image on the other hand, wherein for a given pixel, b1) For each z-stack image, and for each predetermined pixel, calculate a saving score from the intensity of the predetermined pixel and the intensity of neighboring pixels, and, b2) The operation performed is to select the z-stack image with the highest saving score, c) For each pixel in the focus image (FocImg[]), determine whether this pixel is the largest within the focus image (FocImg[]) or within its local neighborhood in the z-stack image associated with this pixel in the focus image (FocImg[]), and if so, save this pixel to the list of selected regions (SS[]). d) A method comprising the operation of calculating a metric value for each pixel in a list of selected regions based on the intensity of these pixels, thereby generating an image of the metric values ​​that allows the particles associated with each pixel in the list of selected regions to be distinguished from one another.

8. In operation d), the measurement is - A metric based on Gaussian modeling of the intensity of the neighborhoods of each pixel in a list of selected regions in the z-stack image associated with this pixel in the focus image by combining one or more of the Gaussian covariance, height, and eccentricity coefficient, and The method according to claim 7, wherein the method is selected from: a metric based on the average of the z-gradients of the intensity of neighboring pixels centered on each pixel in a list of selected regions, wherein the gradient is evaluated with respect to the z-stack image corresponding to each pixel in the list of selected regions in the focus image (FocImg[]).

9. The method according to claim 7, wherein operation b1) uses a pq-average savings score whose intensity is modulated by a Gaussian window centered on a predetermined pixel.

10. The method according to claim 9, wherein operation b1) uses a pq mean with p equal to 1 / 2, q equal to 3, a square neighborhood of 17 pixels, and a standard deviation equal to 2.5 as the saving score.

11. The method according to any one of claims 7 to 10, wherein operation c) includes browsing a list of selected regions and, if two pixels are in immediate neighboring positions, retaining the one with the most significant relevant intensity in the FocImg[].

12. The method according to any one of claims 7 to 10, wherein operation b) further comprises b3) recalculating the focus image (FocImg[]) by recalculating the saving score for each predetermined pixel by limiting the neighbor pixels to pixels in z that are identical to the associated z-stack image in the focus image (FocImg[]) or to immediately neighboring pixels in z.

13. A computer program, when implemented by a computer, that includes instructions for performing the method according to any one of claims 7 to 10.

14. A data storage medium on which the computer program described in claim 13 is recorded.

Images