Device for detecting particles by lensless imaging

Abstract

The invention relates to a device for detecting particles by lensless imaging, the device comprising: a memory (4) for receiving a plurality of z-stack images obtained from a lensless image of a biological sample; a selector (6) for determining a focus image on the basis of the z-stack image, in which each pixel in the focus image is associated on the one hand with the z-stack image and on the other hand with the intensity of the pixel in the z-stack image, the selector (6) being arranged to determine, for each z-stack image and a given pixel, the intensity of the pixel in the z-stack image; a reduction score is calculated based on the intensity of the given pixel and the intensity of the adjacent pixel, and the z-stack image with the highest reduction score is selected, so that the z-stack image of the given pixel is determined; a selector (10) for determining, for each pixel of a focus image, whether the pixel is a maximum in a local neighborhood in the focus image centered on the pixel or in a z-stack image associated with the pixel in the focus image, and if so, storing the pixel in a list of selected points, and if not, storing the pixel in a list of selected points. And the computer (12) is arranged to calculate a metric for each pixel in the list of selected loci based on the intensities of these pixels to produce an image of metrics for distinguishing particles associated with each pixel in the list of selected loci.

Description

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

[0002] Lensless imaging consists in placing the sample between a light source and an image sensor, without optical magnification between them. The image sensor thus collects an image of the light intensity transmitted by the sample, also called a hologram, formed by the interference pattern between the light waves emitted by the light source and transmitted by the sample, and the diffracted waves resulting from the sample diffracting the light waves emitted by the light source.

[0003] These interference patterns have been the subject of research in order to reconstruct images and thus exploit the information they contain about the sample being observed. As the technique has no optical elements and simple materials, it is considered particularly promising and has been the subject of numerous developments in the field of hematology.

[0004] Under platelet-rich plasma (PRP), red blood cells (RBC) can be counted because the number of RBCs is small enough not to mask other particles. Whole blood must be lysed before only platelet populations (PLTs) and white blood cells (WBCs) can be counted, but lysed RBCs create background noise, which creates additional difficulties.

[0005] Patents with publication numbers FR3034196 and FR3049348 teach a method for calculating statistics from detected cells (in particular z-curves) for counting and differentiating white blood cells. Unlike the prior art, these methods require a prerequisite: the use of an accurate method to reconstruct the image from the diffraction image, capable of preserving the local information of the punctual particles and without leaving any major artifacts. For example, patent (publication number: FR3049347) teaches such a reconstruction method.

[0006] To date, only the WBC-PLT method described in a patent (publication number: FR3082943) allows counting PLTs in PRP and lysed blood (using different settings), and additional work is needed to understand how to distinguish RBCs (in PRP) from other particles. The method is based on detecting particles in an image stack reconstructed around a mean plane; the detection planes are then merged to build a detection map. This map is used directly to count the particles. In the presence of several types of particles, it is appropriate to apply this general strategy to each type of particle. On each object (connected component) of the detection mask thus calculated, a metric is calculated to determine the nature of the object or its number. This last point constitutes an extremely unfavorable limitation: in the case of an image containing a large number of particles, many particle multiplets are present (statistically), and the method cannot solve this problem in the presence of multiple types of particles. Finally, attempts to use this method on lysed blood were unsuccessful.

[0007] Another approach has been developed, which is based on a deep neural network (DNN) model. The training is performed on the basis of augmented samples to obtain adequate input, taking into account the counts given by a reference machine. The performances are very good, but they are limited by the assumptions peculiar to this type of methods: having enough training data and a sufficiently reliable reference. Data acquisition is a long and expensive process, which is a limiting factor for the application of DNN-based techniques. In addition, the training is strongly influenced by the reference machine, which raises questions about the variability of the measurement machine. Moreover, since the training is performed on noisy images, it induces not only dissolving noise training for both counting and classification. Finally, this type of training allows counting but not classification of particles, since there is no reference to construct.

[0008] Therefore, no single solution fully meets the requirements for PLT counting in the context of lensless imaging of whole blood.

[0009] The present invention improves this situation. To this end, a device for detecting microparticles by lensless imaging is proposed, comprising: a memory for receiving a plurality of z-stack images obtained from lensless images of biological samples; an initiator for determining a focus image based on the z-stack image, wherein, on the one hand, each pixel is associated with a z-stack image, and on the other hand, is associated with the intensity of the pixel in the z-stack image, and the initiator is used to determine the z-stack image of a given pixel by calculating a parsimony score (based on the intensity of the given pixel and the intensity of neighboring pixels) and selecting the z-stack image with the highest parsimony score for each z-stack image and a given pixel; a selector, which is configured to determine, for each pixel of the focus image, whether the pixel is a maximum value in a local neighborhood centered on the pixel in the focus image, or a maximum value in a z-stack image associated with the pixel in the focus image, and if so, to store the pixel in a selected site list, and a computer, which is configured to calculate a metric value for each pixel in the selected site list based on the intensities of these pixels, thereby generating an image of the metric values ​​in order to distinguish the microparticles associated with each pixel in the selected site list.

[0010] The device is particularly advantageous because it allows reliable enumeration of PLTs and can be embedded in a laboratory setting.

[0011] According to various embodiments, the present invention may have one or more of the following features:

[0012] - the computer is configured to implement a metric selected from a metric based on modeling the intensity of a neighborhood of each pixel in the selected list of sites in the z-stack image associated with the pixel in the focus image by a Gaussian by combining one or more of the covariance, height and eccentricity of the Gaussian, and a metric based on the average of the z-gradient of the intensities of the pixels in the neighborhood centered at each pixel in the selected list of sites, the gradient being evaluated for the z-stack image corresponding to each pixel in the selected list of sites in the focus image,

[0013] - the launcher was set to use a parsimonious score based on a pq-average, where the intensity is modulated by a Gaussian window centered at a given pixel,

[0014] - The enabler was set to use a pq mean with p equal to 1 / 2 and q equal to 3, a square neighborhood of 17 pixels, and a Gaussian with a standard deviation of 2.5,

[0015] - the selector is set to go through the list of selected sites and, when two pixels are in close proximity, retain the pixel with the highest correlation intensity in the focus image, and

[0016] -The apparatus further comprises a filter arranged to recalculate the focus image by recalculating the parsimony score for each given pixel by restricting the pixel neighborhood to pixels in the focus image whose associated z-stack image is the same as or immediately adjacent in z to the image associated with the given pixel.

[0017] The present invention also relates to a method for detecting particles by lensless imaging, comprising the following steps:

[0018] a) receiving a plurality of z-stack images obtained from a lens-free image of a biological sample,

[0019] b) determining a focus image, wherein each pixel is associated with a z-stack image on the one hand and with the intensity of the pixel in the z-stack image on the other hand, the determination for a given pixel comprising the following operations:

[0020] b1) calculating a parsimony score for each z-stack image and a given pixel, the parsimony score being derived from the intensity of the given pixel and the intensities of neighboring pixels, and

[0021] b2) Select the z-stack image with the highest parsimony score,

[0022] c) for each pixel of the focal image, determining whether the pixel is a maximum in a local neighborhood centered on the pixel in the focal image or a maximum in a z-stack image associated with the pixel in the focal image, and if so, storing the pixel in a selected site list, and

[0023] d) calculating a metric value for each pixel in the selected site list based on the intensities of those pixels so as to generate a metric value image that distinguishes the microparticles associated with each pixel in the selected site list.

[0024] According to various embodiments, the present invention may have one or more of the following features:

[0025] - in operation d), the metric is selected from a metric based on modeling the intensity of a neighborhood of each pixel in the selected list of sites in the z-stack image associated with the pixel in the focus image by a Gaussian by combining one or more of the covariance, height and eccentricity factors of the Gaussian, and a metric based on the average of the z-gradients of the pixel intensities of the neighborhood centered around each pixel in the selected list of sites, the gradient being evaluated for the z-stack image corresponding to each pixel in the selected list of sites in the focus image,

[0026] - Operation b1) uses a pq-mean based parsimony score where the intensity is modulated by a Gaussian window centered at a given pixel,

[0027] - Operation b1) uses the pq mean as the parsimony score, where p equals 1 / 2 and q equals 3, the square neighborhood is 17 pixels, and the standard deviation of the Gaussian equals 2.5,

[0028] - Operation c) comprises browsing the list of selected sites and, when two pixels are adjacent, retaining the pixel with the highest correlation strength in the focus image, and

[0029] - Operation b) also includes b3) recalculating the focus image by recalculating the parsimony score for each given pixel by restricting the pixel neighborhood to pixels in the focus image whose associated z-stack image is the same as or immediately adjacent in z to the image associated with the given pixel.

[0030] The invention also relates to a computer program comprising instructions for executing the method according to the invention, a data storage medium recording such a computer program, and a computer system comprising a processor coupled to a memory recording such a computer program.

[0031] Other characteristics and advantages of the invention will become more apparent on reading the following description, taken from examples given for illustrative and non-limiting purposes, taken from the accompanying drawings:

[0032] - Figure 1 A general diagram showing a device according to the invention,

[0033] - Figure 2 express Figure 1 Example of an operating cycle of the device,

[0034] - Figure 3 Indicated in Figure 2 An implementation example of the function implemented in the first operation,

[0035] - Figure 4 Indicated in Figure 2 An implementation example of the function implemented in the second operation,

[0036] - Figure 5 Indicated in Figure 2 An implementation example of the function implemented in the third operation of

[0037] - Figure 6 represents the lens-free image and the z-stack image derived from it,

[0038] - Figure 7 Indicates that Figure 3 The focus image obtained,

[0039] - Figure 8 Indicates that Figure 4 The focus image obtained,

[0040] - Fig. 9 Indicates that Figure 5 The focus image obtained,

[0041] - Fig.10 Indicates that from Figure 5 The image obtained is an image of the first metric value obtained, and

[0042] - Fig.11 Indicates that from Figure 5 An image of a second metric value obtained in the obtained image.

[0043] The drawings and the description below mostly contain elements of a specific nature, so they not only contribute to a better understanding of the invention but also contribute to the definition of the invention where appropriate.

[0044] This specification may refer to elements that may be protected by copyright. The right holder has no objection to anyone reproducing this patent document or its specification in the same manner as in an official document. Otherwise, he reserves all his rights.

[0045] Figure 1 A general diagram showing an apparatus 2 according to the invention. The apparatus 2 comprises a memory 4, an enabler 6, a filter 8, a selector 10 and a computer 12. As shown below, the filter 8 is optional.

[0046] The memory 4 receives data that are the subject of the functions and calculations implemented by the device 2. The device 2 mainly processes reconstructed images obtained from lensless images, the image focus being taken at different depths. Patent publication number FR3082944 describes an implementation example of such images. More specifically, the patent describes how to reconstruct multiple images from lensless images, which will be referred to hereinafter as "z-stack images" or "z-focus images", and which correspond to the focus of lensless imaging at a specific depth.

[0047] Figure 6 The acquisition images of lensless imaging are shown, as well as the module and phase images reconstructed from the acquisition images. As shown below, these are images that combine module and phase and are used as z-stack images in the context of the present invention. Alternatively, only the phase image or the module image can be used when the phase image or the module image carries the information of interest.

[0048] The difficulty in processing these images is that platelets constitute large and small elements within the pixel range, their number is very large, while the image background is very noisy due to the lysis of erythrocytes, so it is impossible to distinguish platelets from erythrocyte fragments with the naked eye. The data of a z-stack image include, 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 resolution of the images is 3840 × 2748 pixels and the intensities are modified: the z-stack images are processed in the form of |I-1| instead of |I|, because an image without particles after reconstruction must theoretically have a constant value at 1. Therefore, by processing in terms of |I-1|, the amplitude of the perturbations caused locally by diffraction can be taken into account.

[0049] The memory 4 may be any type of data storage suitable for receiving digital data: hard disk, hard disk flash, any form of flash drive, RAM, disk, local or cloud distributed storage, etc.

[0050] In the example described herein, the memory 4 receives all the data mentioned above and more generally all the data related to the device 2, i.e. the programs and software exemplifying the actuator 6, the filter 8, the selector 10 and the computer 12, their parameters, the input data received, the output data and the data stored in the buffer memory. The data calculated by the device may be stored on or on any type of memory similar to the memory 4. These data may be erased or retained after the device has performed its task.

[0051] The launcher 6, the filter 8, the selector 10 and the computer 12 directly or indirectly access the memory 4. They can be implemented in the form of appropriate computer code executed on one or more processors. The term "processor" refers to any processor suitable for the calculations described below. Such a processor can be implemented in any known manner, such as a microprocessor of a personal computer, laptop, tablet or smartphone, a dedicated chip of the field programmable gate array (FPGA) or system on chip (SoC) type, a computing resource in a grid or cloud, a graphics processor (GPU) cluster, a microcontroller or any other form suitable for providing the computing power required for the implementation described below. One or more of these elements can also be implemented in the form of a dedicated electronic circuit (such as an ASIC). Combinations of processors and electronic circuits can also be considered. Processors dedicated to machine learning can also be considered.

[0052] The initiator 6, the filter 8, the selector 10 and the computer 12 are presented separately here because they perform different functions. This modular description is intended to provide a better understanding of the functional blocks implemented by the device 2. It goes without saying that two or more of these elements can still be combined together as long as the functional relationship remains comparable.

[0053] Figure 2 Shows Figure 1 An example of an operating cycle of the device in order to better understand the respective functions of the launcher 6, the filter 8, the selector 10 and the computer 12.

[0054] The loop starts at operation 200, where the initiator 6 executes the Focus() function. The Focus() function accesses a z-stack image from the same lens-free image and returns an image called a focus image, in which each pixel is associated with a z-stack image index on the one hand and the intensity of the pixel in the z-stack image on the other hand. The purpose of the Focus() function is to determine the best focus depth for each pixel in order to find the most relevant plane to infer the properties of each pixel. This function will Figure 3 Further described in. Figure 7 An example of an obtained focus image is shown.

[0055] The focus image is then reprocessed in optional operation 210, where filter 8 executes the Refine() function. As shown below, the Focus() function uses a parsimony score and a Gaussian window, which allows a high-quality focused image to be obtained while retaining a certain regularity, without which the pixels of the same cell are not necessarily in the same focus. However, in platelet detection, this regularity comes at a cost. In fact, due to the large number of cells present in the cavity, the distance between two cells may be very small, and it is easy for cells to be located in adjacent windows and interfere with the parsimony score. The Gaussian window reduces this phenomenon, but if two cells are several pixels apart, the cell with higher "intensity" will take precedence over the other, while the cell with lower "intensity" may have non-optimal focus. The Refine() function aims to solve this problem by improving the focused image from the focal plane found in operation 200, and only considering pixels in the same or very close planes in the parsimony score for each pixel. This function will be achieved by Figure 4 For further description. Figure 8 An example of an improved focus image is shown.

[0056] The selector 10 then uses the focus image (possibly refined) to select pixels that are likely to form plaque sites. To do this, the selector 10 executes the Selec() function in operation 220. This function facilitates pixel detection rather than through connected components so that multiple cases can be optimally handled as simply as possible. It is important not to "forget" any site and to treat the target to be searched as a unique pixel so that the operation of distinguishing between target pixels (platelets) and background pixels is reduced to the final operation. This function will be in Figure 5 Further described in. Fig. 9 An example of the obtained site images is shown.

[0057] Finally, in operation 230, the computer 12 uses the list of sites determined by the selector 10 and executes the Class() function, in which a metric is calculated to generate an image that allows platelet pixels to be distinguished from background pixels. Thus, this operation also allows the PLTs to be counted and their concentration to be estimated. Fig.10 and 11 The classification images obtained for two different metrics are shown.

[0058] Figure 3 An implementation example of the Focus() function executed by Launcher 6 is shown. As described above, the purpose of this function is to determine for each pixel from the z-stack image the z focal plane that best suits that pixel. To this end, the applicant has discovered that a parsimony score applied to a neighborhood of points for each pixel can achieve unparalleled results. The parsimony score has the advantage of being applicable to pixel objects because it seeks to maximize the level of a single pixel at the expense of all other pixels. There are several parsimony scores, and the Focus() function described here uses a modified version of the pq-mean metric (or pq mean). The pq mean has good theoretical properties while being fairly simple to calculate, which makes it easy to embed the pq mean into a machine. Therefore, in the example described here, the metric used has the following formula: where X represents the neighborhood containing all intensities around a given pixel, G is the Gaussian window applied to the neighborhood pixel intensities, * represents the convolutional product, and p and q are parameters of the parsimony score, 0<p≤1<q.

[0059] In the examples described here, the applicant has determined that p=1 / 2 and q=3 give the best results. However, other combinations also provide good results, such as p=1 / 3 and q=4 or p=1 / 2 and q=4. In the case of pixel search, in this paper, the dimension of the neighborhood is determined by a Gaussian. Thus, starting from a standard deviation of 2.5, for a Gaussian covering 3 sigma - representing a traditional approach - a neighborhood of 17 pixels is obtained (3 times 2.5 rounded to 8, times 2, plus the center pixel). Other Gaussian window sizes can be retained.

[0060] Thus, in a first operation 300, the first current pixel cp is initialized with the first pixel cp0 in the z-stack image coordinates, and then in operation 310, the Wind() function retrieves all the neighbors in a square window of 17 pixels centered on the pixel cp and records their intensities in a variable W. The variable W receives the neighbors of all z-stack images so that all z-stack images can be processed at once. Alternatively, the z-stack images can be processed sequentially or in parallel and then the focus can be selected after all these calculations.

[0061] Then, in operation 320, the parsimony score is calculated by the Pars() function. The result is stored in the array CP[] which thus receives, for each focal depth z corresponding to a z-stack image, on the one hand the parsimony score of the pixel cp in the z-stack image considered and, on the other hand, the intensity of the pixel cp in the z-stack image considered.

[0062] The launcher 6 then executes the Max() function in operation 330. The Max() function traverses the array CP[] and returns the pair (z; I) with the highest parsimony score (where I is the intensity of the associated z-stack image stored in the array CP[]. In practice, the z-stack image with the highest parsimony score is the image in which the pixel has the best focus. Alternatively, the z corresponding to the maximum intensity can be determined by interpolating a curve of values ​​I, and, if necessary, the intensity values ​​can be interpolated in this plane for the rest of the calculations. In the latter case, this new plane can be introduced into the z-stack image data, where only the neighborhood of the point under consideration can be retained, which neighborhood is described below.

[0063] Then, the triple (cp; z; I) is stored in the array FocImg[] in operation 340. As can be seen from the figure, the array FocImg[] contains all the information of the focus image and is constructed step by step. Next, the array FocImg[] will be able to specify the focus image.

[0064] Finally, the next pixel is retrieved in operation 350 and the Focus() function continues with operation 310 unless all pixels have been processed, in which case it ends in operation 399 .

[0065] In the above, the Focus() function processes the pixels sequentially. It goes without saying that a parallel version of this processing is possible. Therefore, the focus image FocImg[] has an 8 pixel thick frame in which no z is identified since all measurements are zero. This has the advantage of limiting the risk of edge effects. At the end of the processing by device 2, a 35 pixel thick frame is removed during the counting. However, this is still negligible compared to the size of the images considered (10MP). The applicant has also found that the pq-mean is not the only parsimony score that gives satisfactory results. Therefore, combining the Gini measure (or Gini index) with a Gaussian window is a variant that is part of the scope of the invention.

[0066] Figure 4 An implementation example of an optionally executed Refine() function implemented by the filter 8 is shown. In fact, this function is very close to the Focus() function. In addition, similar or identical operations have reference numerals that vary only in the hundreds digit, and only the differences are described.

[0067] The only difference between the Focus() function and the Refine() function is the definition of the neighborhood window. Therefore, in operation 410, the filter 8 executes the Wind2() function, which defines the window W in a different way. In fact, since the goal is to consider only the neighborhood that is in the same focal plane or close (for example, one plane away) as the pixel cp, the Wind2() function defines the window W in a similar way to the Wind() function, but, at each pixel, it retrieves the z value of the pixel in the focus image FocImg[] and discards the pixel if the z value is different from the z of the z-stack image of the current pixel cp. Therefore, for each z-stack image, the parsimony score of each pixel is improved using only the pixels that are likely to be relevant.

[0068] The Refine() function can be repeated until a modification threshold of the parsimony score is reached. Alternatively, this function can be repeated a fixed number of times. The applicant has found that performing 4 times provides the best ratio between improvement and computational cost. The main benefit provided by the Refine() function is the improvement of the bias in the linear regression measured by device 2. In fact, tests conducted by the applicant on an existing database showed that using the Refine() function, the bias (the ordinate at the origin of the linear regression) changed from 35 gigaplatelets per liter to 24 gigaplatelets per liter, and ideally the bias is 0. In the case of embedded use, the Refine() function can still be omitted because its computational cost is very high. The Gaussian window is of less concern in the Refine() function and can be fixed or gradually relaxed at each iteration.

[0069] Figure 5 An example implementation of the Selct() function performed by the selector 10 is shown. Now that the focus image has been determined, the goal is to identify the pixels corresponding to potential platelet sites. This is equivalent to performing a local maximum search.

[0070] Thus, in a first operation 500, the first current pixel cp is initialized with the first pixel cp0 in the z-stack image coordinates, and then two operations 510 and 520 are started in parallel. In operation 510, the MaxLoc() function determines a Boolean value indicating whether the current pixel cp constitutes a local maximum in the z-stack image whose depth 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 similar, except that the neighborhood is performed on the focus image FocImg[] itself, and no longer on the z-stack image of the focus of the current pixel cp. Then, in operation 530, the value m1 in operation 510 and the value m2 in operation 520 are compared with a logical OR. Thus, if one of the two values ​​m1 and m2 is equal to 1, the test is positive. In practice, two adjacent pixels do not necessarily have the same focus, and there may be pixels that are local maxima on the focus image, but not on the image of its focus plane.

[0071] If the test of operation 530 is positive, the pixel cp is added to the array SS[] at the selected position in operation 540, and then a test is performed in operation 550 to determine whether there are any current pixels that need to be processed. If so, the Select() function continues to perform 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 in the selected pixels, two pixels are often located on the same particle. Therefore, once the list SS[] stops, the Sng1() function scans it to identify pairs of directly adjacent and selected pixels and retains the pixel with the largest intensity in the focus image FocImg[]. Finally, the Select() function ends in operation 599.

[0072] In the above, the Select() function processes pixels sequentially. It goes without saying that a parallel version of this process is possible. In addition, in order to provide a visually meaningful representation, the output SS[] list can be similar to a site image, where the value of the selected site is 1 and the value of other pixels is zero. Here again, the site image and the SS[] list can be used interchangeably or in conjunction because they specify the same thing. Alternatively, operation 560 can be omitted.

[0073] Once the site image is determined, it remains to apply metrics that allow the classification of pixels as microparticles or background. Applicants have conducted extensive research and identified two families of metrics that can distinguish between microparticles.

[0074] The first series has a Gaussian property and includes two measures:

[0075] - In the first metric, for each site, the intensity in the z-stack images associated with the site in the focal image FocImg[] is modeled by a Gaussian, and the inverse of its standard deviation constitutes the metric,

[0076] - A second metric where for each locus, a 3×3 neighborhood of intensities in the z-stack image associated with the locus in the focal image FocImg[] is modeled by a Gaussian whose height constitutes the metric. Alternatively, the metric can also be based on the variance divided by the height.

[0077] Fig.10 and 11 It shows how these two indicators separate PLTs, white blood cells (WBCs), and background, and increase the separation area.

[0078] The second series is based on the z-gradient. In this version, for each pixel of a neighborhood of size 3×3 centered at each site, the z-gradient is calculated using the intensity of these pixels in the corresponding z-stack image, which is evaluated around the z-depth associated with the site in the focal image FocImg[]. The metric is then defined as the average of the calculated gradients.

[0079] For example, other families of metrics can be considered. In particular, metrics can be computed on neighborhoods other than the 3×3 neighborhood, and metrics computed on these neighborhoods can have properties different from those of Gaussian modeling. Thus, the selected neighborhood can be projected onto different bases, such as, but not limited to: a frequency basis of the Fourier transform, a basis formed by a series of orthogonal polynomials, a decomposition of the neighborhood into wavelets.

[0080] The device and method of the invention can be used for other types of samples that need to be characterized. These samples may include in particular a medium in which the microparticles are immersed. The medium may be a liquid medium. It may include a body fluid, such as obtained from blood or urine or lymph or cerebrospinal fluid. It may also be a culture medium, including nutrients that allow the development of microorganisms or cells. The term "microparticle" refers in particular, but not limited to:

[0081] - cells, whether cultured cells or somatic cells, such as blood cells;

[0082] - microorganisms, such as bacteria or yeast or microalgae;

[0083] - solid particles, such as microbeads, which can be functionalized to facilitate the grafting of the analyte; or even

[0084] - Microparticles that form an emulsion in a medium, in particular microparticles that are insoluble in the medium, such as lipid droplets in an aqueous medium.

Claims

1. A device for detecting particles by lensless imaging, comprising: A memory (4) for receiving a plurality of z-stack images obtained from a lensless image of a biological sample; an initiator (6) for determining a focus image (FocImg[]) based on the z-stack image, each pixel in the focus image (FocImg[]) being associated with a z-stack image on the one hand and with the intensity of the pixel in the z-stack image on the other hand, the initiator (6) being arranged to determine, for each z-stack image and a given pixel, the z-stack image of the given pixel by calculating 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 (10), the selector (10) It is configured to determine, for each pixel of the focus image (FocImg[]), whether the pixel is a maximum in a local neighborhood centered on the pixel in the focus image (FocImg[]), or whether it is a maximum in the z-stack image associated with the pixel in the focus image (FocImg[]), and if so, store the pixel in a selected site list (SS[]), and the computer (12) is configured to calculate a metric value for each pixel in the selected site list based on the intensity of these pixels to allow generation of a metric value image, thereby allowing differentiation of particles associated with each pixel in the selected site list.

2. An apparatus according to claim 1, wherein the computer (12) is configured to implement a metric selected from a metric based on modeling the intensity of the neighborhood of each pixel in the selected site list in the z-stack image associated with the pixel in the focus image by a Gaussian combined with one or more of the covariance, height and eccentricity of the Gaussian, and a metric based on the average value of the z-gradient of the intensity of the pixels in the neighborhood centered on each pixel in the selected site list, the gradient being evaluated for the z-stack image corresponding to each pixel in the selected site list in the focus image (FocImg[]).

3. An apparatus according to claim 1 or 2, wherein the enabler (6) is arranged to use a parsimony score based on a pq-average, wherein the intensity is modulated by a Gaussian window centred on the given pixel.

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

5.

5. An apparatus according to any one of the preceding claims, wherein the selector (10) is arranged to browse the list of selected sites and, when two pixels are adjacent, retain the pixel with the highest correlation strength in the focus image (FocImg[]).

6. The apparatus according to any of the preceding claims further comprises a filter (8) which is arranged to recalculate the focus image (FocImg[]) by recalculating the simplicity score of each given pixel in such a way as to restrict the pixel neighborhood to pixels in the focus image (FocImg[]) whose associated z-stack image is the same as or immediately adjacent in z to the image associated with the given pixel.

7. A method for detecting particles by lensless imaging, comprising the following operations: a) receiving a plurality of z-stack images obtained from a lens-free image of a biological sample, b) determining a focus image (FocImg[]), wherein each pixel is associated with, on the one hand, a z-stack image and, on the other hand, with the intensity of said pixel in said z-stack image, the determination comprising, for a given pixel, the following operations: b1) calculating, for each z-stack image and the given pixel, a parsimony score of the intensity of the given pixel and the intensities of neighboring pixels, and b2) selecting the z-stack image with the highest parsimony score, c) for each pixel of the focus image (FocImg[]), determining whether the pixel is a maximum in a local neighborhood centered on the pixel in the focus image (FocImg[]) or a maximum in the z-stack image associated with the pixel in the focus image (FocImg[]), and if so, storing the pixel in a selected site list (SS[]), and d) calculating a metric value for each pixel in the selected site list based on the intensity of these pixels to allow the generation of a metric value image allowing the microparticles associated with each pixel in the selected site list to be distinguished.

8. A method according to claim 7, wherein in operation d), the metric is selected from a metric based on modeling the intensity of the neighborhood of each pixel in the selected site list in the z-stack image associated with the pixel in the focus image by a Gaussian combined with one or more of the covariance, height and eccentricity of the Gaussian, and a metric based on the average value of the z-gradient of the intensity of the pixels in the neighborhood centered on each pixel in the selected site list, the gradient being evaluated for the z-stack image corresponding to each pixel in the selected site list in the focus image (FocImg[]).

9. The method according to claim 7 or 8, wherein operation b1) uses a pq-mean based parsimony score, wherein the intensity is modulated by a Gaussian window centered on the given pixel.

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

5.

11. The method according to any one of claims 7 to 10, wherein operation c) comprises browsing the selected site list and retaining the pixel with the highest correlation strength in the focus image (FocImg[]) when two pixels are adjacent.

12. A method according to any one of claims 7 to 11, wherein operation b) also includes operation b3) recalculating the focus image (FocImg[]) by recalculating the simplicity score of each given pixel by restricting the pixel neighborhood to pixels in the focus image (FocImg[]) whose related z-stack image is the same as or immediately adjacent in z to the image associated with the given pixel.

13. A computer program comprising instructions for performing the method according to one of claims 7 to 12 when implemented by a computer.

14. A data storage medium having recorded thereon the computer program according to claim 13.

Images