FR3082943A1 - METHOD FOR COUNTING SMALL PARTICLES IN A SAMPLE - Google Patents

Abstract

The invention is a method of counting small particles in a sample. It consists of illuminating a sample and acquiring an image of it by an image sensor. A holographic reconstruction algorithm is applied to the acquired image, so as to obtain a stack of reconstructed images. Each reconstructed image is representative of a light wave of exposure to which the image sensor is exposed. Each reconstructed image is subject to segmentation, so as to isolate regions of interest corresponding to one or more particles. Using each region of interest, the method allows a count of the number of particles in the sample.

Description

Description

TECHNICAL AREA

The technical field of the invention is the observation of a sample, in particular a sample comprising small particles, by imaging without a lens, or by so-called defocused imaging.

PRIOR ART

Lensless imaging has developed in recent years, and its applications for the characterization of samples are increasing, for example in the field of medical diagnosis, cell culture, biological analysis or in the field of agribusiness. This technique makes it possible to obtain, at low cost, an image of a sample according to a high field of observation. The application of holographic reconstruction algorithms to images obtained according to a lensless configuration makes it possible to obtain exploitable images of samples, which constitutes a low-cost alternative to conventional microscopy, while benefiting from a field of extended observation.

Various documents describe the use of lensless imaging for the observation of samples containing particles. WO2016151248 or WO2017178723 describe the use of lensless imaging to characterize cells in a sample. The technique described in these documents is based on the acquisition of an image using an image sensor placed facing a sample. The application of a holographic reconstruction algorithm makes it possible to form profiles representative of the light wave to which the image sensor is exposed. From the profiles, the cells placed in the sample are characterized. The use of profiles makes it possible to be insensitive to the position of a cell to be characterized in the sample, and in particular to the distance between the cell and the sensor.

A difficulty associated with the application of holographic reconstruction algorithms is that it is necessary to know the distance between the image sensor and the sample. However, when the sample contains particles, the position of the latter, relative to the image sensor, can vary, in the thickness of the sample. Thus, the images obtained by holographic reconstruction can be clear for certain particles, for which the distance, relative to the image sensor, corresponds to the reconstruction distance taken into account. On the other hand, the images obtained by holographic reconstruction are blurred for the particles whose distance, relative to the image sensor, is different from the reconstruction distance taken into account. This property is exploited in digital focusing algorithms, the latter being intended to estimate a distance between a particle and an image sensor. More precisely, such algorithms exploit the fact that a particle is clear when the distance which separates it from the sensor corresponds to the reconstruction distance considered. The use of these algorithms makes it possible to go back to the distance between particles and a sensor, thus opening the way to a determination of the 3D position of particles, as well as to their enumeration. Such an algorithm is for example described in the publication Kemper B Application of a 3D tracking, LED illumination and multi-wavelength techniques for quantitative cell analysis in digital holographie microscopy, Proc. SPIE 7184, Three-Dimensional and Multidimensional Microscopy: Image Acquisition and Processing XVI, 71840R (March 4, 2009). Another particle tracking algorithm, in this case sperm, is described in W02014012031. In these publications, a position tracking algorithm is applied to each particle. However, when the number of particles is large, it is tedious to perform position tracking for all the particles, taken individually. Thus, when the size of the particles decreases, and their number increases, the implementation of digital focusing algorithms can prove to be difficult to carry out an accurate count of the particles present in the observed field. The inventors have developed a method allowing more precise counting of particles in a sample.

STATEMENT OF THE INVENTION

A first object of the invention is a method of counting particles in a sample, the sample comprising a medium in which the particles are immersed, the medium being held by a holding element, the medium preferably being stationary relative to the holding element, the method comprising the following steps:

a) illumination of the sample by a light source, the sample being arranged between the light source and an image sensor;

b) acquisition, using the image sensor, of an image of the sample, the sample being arranged between the light source and the image sensor, each image being representative of a light wave d exposure, to which the image sensor is exposed under the effect of the illumination of the sample;

c) from the image acquired during step b), application of a digital propagation operator to obtain a complex expression of the exposure light wave in different reconstruction planes, each reconstruction plan being associated at a reconstruction distance, relative to the image sensor;

d) formation of a stack of reconstructed images, each reconstructed image being obtained from the module and / or from the argument of a complex expression in a reconstruction plane, each reconstructed image being associated with a different reconstruction distance one from the other ;

e) segmentation of each reconstructed image, so as to isolate, on each reconstructed image, regions of interest, spaced apart from each other, each isolated region of interest corresponding to a particle or to a cluster of particles;

f) from the regions of interest isolated in step e), estimation of an amount of particles in the sample.

Thus, the method performs a combination of holographic reconstruction techniques and image segmentation, which is particularly suitable for samples containing a large number of particles. This allows in particular to avoid making a determination of the position of each particle, taken individually.

The holding element has the function of maintaining or confining the environment in which the particles are immersed.

The sample can be placed on a support to hold the sample between the image sensor and the light source.

In step c), the reconstruction distances can be obtained by determining a focusing distance, corresponding to a distance, relative to the image sensor, at which a maximum number of particles are located, each reconstruction distance being established by applying an offset from the focusing distance.

The difference between two successive reconstruction distances corresponds to a reconstruction step, the reconstruction step preferably being less than 20 μm.

The particles are preferably inscribed in a circle or in a sphere whose diameter is less than 20 μm or 10 μm or 5 μm.

The particles can in particular be thrombocytes, that is to say platelets of blood. During step e), the segmentation of each reconstructed image can be carried out by applying a thresholding, according to an intensity threshold, on each reconstructed image. Each reconstructed image comprising pixels, the value of the intensity threshold can be variable from one pixel to another. The value of the intensity threshold at one pixel can be established by:

defining a neighborhood around each pixel;

calculating an average intensity of pixels belonging to the neighborhood;

the threshold value, at each pixel, depending on the average intensity thus calculated.

The method may include a step of analyzing the regions of interest resulting from step e), so as to:

- determine a number of particles associated with each region of interest;

- or divide regions of interest encompassing several particles into different regions of interest comprising only one particle.

According to one embodiment, during step f), a region of interest whose extent is limited to one pixel is only taken into account if it appears in two adjacent reconstructed images.

The light source can be a laser source. Preferably, the light wave emitted by the light source has a spectral bandwidth less than 10 nm. Preferably, the light wave emitted by the light source has a spectral band centered on a wavelength less than 500 nm, and preferably between 400 nm and 500 nm.

A second object of the invention is a device for counting particles of a sample, comprising:

- a light source capable of emitting an incident light wave propagating towards the sample;

- an image sensor;

- a support, configured to hold the sample between the light source and the image sensor;

and a processor, configured to receive an image of the sample acquired by the image sensor and to implement steps c) to f) of a method according to the first object of the invention.

According to one embodiment, no image forming optics are disposed between the sample and the image sensor.

According to one embodiment, the device comprises an optical system defining a focusing configuration, according to which the image sensor is conjugated with a plane passing through the sample, the device being such that the image sensor or the sample are offset from the focus configuration, so that in step b), the image sensor acquires a defocused image of the sample.

Other advantages and characteristics will emerge more clearly from the description which follows of particular embodiments of the invention, given by way of nonlimiting examples, and represented in the figures listed below.

FIGURES

Figures IA and IB represent examples of devices allowing an implementation of the invention.

FIGS. 2A and 2B are reconstructed images, called module images, or amplitude image, of a sample comprising plates.

Figure 3 shows schematically the main steps of a method according to the invention.

FIGS. 4A, 4B and 4C show the influence of an adjustment of an image segmentation parameter on the identification of regions of interest, the latter being associated with particles.

FIGS. 5A and 5B, as well as 5C and 5D, illustrate the application of filtering of the watershed type on regions of interest. Figures 5A and 5C are figures before filtering. Figures 5B and 5D are figures obtained respectively after applying a filtering of Figures 5A and 5C.

FIG. 6 represents the number of particles counted by implementing the invention (ordinate axis) as a function of the number of particles counted according to a reference method (abscissa axis). In FIG. 6, the results of measurements have been compared respectively without taking into account the regions of monopixel interest, comprising only one pixel, and with taking into account certain regions of monopixel interest.

FIGS. 7A, 7B and 7C represent the number of particles counted by implementing the invention (ordinate axis) as a function of the number of particles counted according to a reference method (abscissa axis). In FIG. 7A, there is a laser diode emitting in purple / blue, as well as a light emitting diode emitting in blue. In FIG. 7B, different laser diodes emitting either in purple / blue or in red were used. In FIG. 7C, a laser diode emitting in purple / blue was used, as well as a light emitting diode emitting in red.

FIG. 8 is another example of a device suitable for implementing the invention. EXPLANATION OF PARTICULAR EMBODIMENTS

FIG. 1A represents an example of a device according to the invention. A light source 11 is configured to emit a light wave 12, called incident light wave, propagating in the direction of a sample 10, along a propagation axis Z. The light wave 12 is emitted according to a spectral illumination band Δλ.

Sample 10 is a sample that we wish to characterize. The sample comprises a 10 _m medium, for example a liquid medium, in which 10 _p particles are immersed. The 10p particles can be cells, or cell fragments, or microorganisms, or microalgae, or their fragments. The term particles can also denote solid particles, in particular microbeads, for example metallic microbeads, glass microbeads or organic microbeads, commonly used in biological protocols. They may also be insoluble droplets bathing in a liquid medium, for example lipid droplets in an oil-in-water emulsion. Thus, the term particle designates both endogenous particles, initially present in the sample examined, and exogenous particles, added to this sample before analysis. The invention is particularly intended for small particles. By small particle is meant a particle inscribed in a circle or in a sphere whose diameter is less than 20 μm, and preferably less than 10 μm or 5 μm. A targeted application is the counting of blood platelets (thrombocytes). More generally, the particles are small relative to the thickness of the sample, for example being at least 5 times or 10 times smaller than the thickness of the sample. The medium 10m can be a solid medium or a medium forming a gel. The 10m medium can include a body fluid, for example, and without limitation, blood plasma, urine, lymph or cerebrospinal fluid. The 10 _m medium is preferably transparent or translucent. The invention is particularly suitable for small particles.

The sample 10 is maintained by a holding element 15. The function of the holding element 15 is to maintain the medium 10 _m in which the particles 10 _p are arranged. The holding element 15 is preferably transparent or translucent, in particular between the light source and the sensor. The holding element 15 can be a transparent blade, for example a glass slide, or even a well plate, or an enclosure. In this example the holding element 15 is a fluid chamber. The thickness e of the sample 10, along the propagation axis Z, typically varies between 10 μm and 1 cm, and is preferably between 20 μm and 1 mm. The sample 10 is maintained on a support 10s at a distance d from an image sensor 20. The medium 10 _m in which the particles 10 _p are bathed is preferably immobile with respect to the holding element 15 or with respect to to the image sensor 20. The particles 10 _p can be stationary relative to the holding element 15 or relative to the image sensor 20.

The distance D between the light source 11 and the fluid chamber 15 is preferably greater than 1 cm. It is preferably between 2 and 30 cm. Advantageously, the light source 11, seen by the sample, is considered to be punctual. This means that its diameter (or its diagonal) is preferably less than a tenth, better a hundredth of the distance between the fluid chamber 15 and the light source 11.

The light source is preferably a point source, for example a laser light source or a light emitting diode. As described below, better results are obtained when the light source is a laser source, due to a narrower emission spectral band.

When the light source 11 is a light-emitting diode, it is preferably associated with a diaphragm 18, forming a spatial filter, as shown in FIG. 1B. The opening of the diaphragm is typically between 5 μιτι and 1 mm, preferably between 50 μιτι and 500 μιτι, for example 150 μιτι. The diaphragm can be replaced by an optical fiber, a first end of which is placed facing the light source 11 and a second end of which is placed opposite the sample 10. A diffuser 17 can be placed between the light-emitting diode and the diaphragm, as described in document US2017317125.

Preferably, the emission spectral band Δλ of the incident light wave 12 has a bandwidth less than 100 nm, and preferably less than 20 nm, or even 10 nm. By spectral bandwidth is meant a width at half height of said spectral band. As described below, the narrower the spectral band, the better the particle counting accuracy.

The sample 10 is placed between the light source 11 and an image sensor 20. The image sensor 20 is able to form an image I _o of the sample 10 according to a detection plane P _o . The acquired image is also designated by the term hologram. In the example shown, it is an image sensor comprising a pixel matrix, of CCD type or a CMOS. The pixel matrix forms the detection plane P _o . The detection plane P _o preferably extends perpendicular to the axis of propagation Z of the incident light wave 12. The detection plane extends along a radial plane XY, defined by two axes X and Y, the plane radial being perpendicular to the axis of propagation Z.

The distance d between the sample 10 and the pixel matrix of the image sensor 20 is preferably between 50 μιτι and 2 cm, preferably between 100 μιτι and 2 mm.

Note, in the embodiments shown in Figures IA and IB, the absence of magnification optics or image formation between the image sensor 20 and the sample 10. This does not prevent the presence possible focusing microlenses at each pixel of the image sensor 20, the latter having no function of enlarging the image acquired by the image sensor, their function being to optimize the collection efficiency of light by the pixels.

Under the effect of the incident light wave 12, the particles 10 _p present in the sample can generate a diffracted wave 13, capable of producing, at the level of the detection plane P _o , interference, in particular with a portion of the incident light wave 12 'transmitted by the sample. The transmitted light wave 12 'corresponds to a part of the incident light wave 12 not absorbed by the sample. Thus, the light wave 14, transmitted by the sample, and to which the image sensor 20 is exposed, designated by the term exposure wave, can comprise:

a component 13 resulting from the diffraction of the incident light wave 12 by each particle of the sample;

a component 12 ′ resulting from the transmission of the incident light wave 12 by the sample, part of the latter being able to be absorbed in the sample.

These components form interference in the detection plane P _o . Also, the image acquired by the image sensor includes interference patterns (or diffraction patterns), due to the different particles 10 _p of the sample.

A processor 21, for example a microprocessor, is able to process each image I _o acquired by the image sensor 20, according to the steps described below. In particular, the processor is a microprocessor connected to a programmable memory 22 in which is stored a sequence of instructions for performing the image processing and calculation operations described in this description. The processor can be coupled to a screen 24 allowing the display of images acquired by the image sensor 20 or calculated by the processor 21.

Due to the absence of image forming optics, an image I _o acquired by the image sensor 20, also called a hologram, does not allow a sufficiently precise representation of the observed sample to be obtained. The acquired image I _o may in particular include a large number of interference patterns, and may not be easily usable to identify the particles present in the sample and to count them.

A holographic propagation operator h can be applied to the image I _o acquired by the image sensor, so as to calculate a complex quantity representative of the exposure light wave 14. It is then possible to reconstruct a complex expression A of the exposure light wave 14 at any point of coordinates (x, y, z) of space, and in particular in a reconstruction plane P _z located at a distance \ z \ from the sensor image 20, called the reconstruction distance, this reconstruction plane being for example the plane P _w along which the sample extends. A first way to obtain the complex expression A (x, y, z) at a point (x, y, z) is to apply the expression: A (x, y, z) = I ₀ (x, y, z) * h (1), * designating the convolution product operator.

The function of the propagation operator h is to describe the propagation of light between the image sensor 20 and a point of coordinates (x, y, z), located at a distance | z | of the image sensor. It is then possible to determine the module M (x, y, z) and / or the phase φ (x, y, z) the light wave 14, at the reconstruction distance | z |, with:

M (x, y, z) = abs [A (x, y, z) J;

<p (x, y, z) = arg [A (x, y, z)].

The operators abs and arg respectively designate the module and the argument.

The propagation operator is for example the Fresnel-Helmholtz function, such that: h (x, y, z) = - e ^1Zn xexp (in ~ ^{x + y} ) (2) with i ² = —1.

- ⁷ d / .z λζ

The complex expression A of the light wave 14, at any point of coordinates (x, y, z) of space, is such that: A (x, y, z) = M (x, y, z) e ^I ' ^{p <} ' ^x ' ^y ' ^z \

In the remainder of this description, the coordinates (x, y) designate a radial position in a radial plane XY parallel to the detection plane P _o . The z coordinate designates a coordinate along the propagation axis Z.

The complex expression A is a complex quantity whose argument and modulus are respectively representative of the phase and the intensity of the light wave 14 of exposure detected by the image sensor 20. The convolution product of the image I _o by the propagation operator h makes it possible to obtain a complex image A _z representing a spatial distribution of the complex expression A in a reconstruction plane P _z , extending at a distance | z | of the detection plane P _o , parallel to the latter. In this example, the detection plane P _o has the equation z = 0. The complex image A _z corresponds to a complex image of the sample in the reconstruction plane P _z . The image A _z is defined at radial coordinates (x, y), so that A _z (x, y) = A (x, y, z). Image A _z also represents a two-dimensional spatial distribution of the complex expression of the exposure wave 14. Such a process, designated by the term holographic reconstruction, makes it possible in particular to reconstruct an image of the module or of the phase of the exposure light wave 14 in the reconstruction plan. For this, we can form images M _z and <p _z representing respectively the module or the phase of the complex image A _z , with M _z = mod (A _z ) and φ _ζ = arg (A _z ). In the following text, an image obtained from the module or from the phase of a complex image is designated by the term reconstructed image.

The application of a holographic reconstruction operator, from an acquired image, according to expression (1), can be accompanied by the formation of noise affecting the reconstructed images. In order to limit the appearance of this noise, the application of the holographic reconstruction operator is carried out according to iterative holographic reconstruction algorithms. Such reconstruction algorithms are for example described:

in document WO2016189257, in particular in steps 100 to 150 described in the latter;

or in document WO2017162985, and more precisely according to steps 110 to 160 described in the latter.

The images represented in FIGS. 2A and 2B were obtained by applying a holographic reconstruction algorithm to an image acquired by an image sensor 20. The sample observed was a sample of blood plasma rich in platelets (PRP - acronym of PlateletRich Plasma), obtained by centrifugation of a blood plasma. The sample 10 was placed in a fluid chamber 15 of thickness e equal to 100 μm, placed at a distance of 1 mm from an image sensor of the IDS UI-1492LE type. To obtain FIGS. 2A and 2B, a holographic reconstruction algorithm described in WO2017162985 was applied to an image I _o acquired by the image sensor, by considering respectively two reconstruction distances z _x and z ₂ . The reconstruction distances respectively define two reconstruction planes P _± and P ₂ , extending in the sample, parallel to the detection plane P _o . We thus obtained two complex images A _± and ri ₂ representative of a complex expression of the light wave of exposure 14 in each reconstruction plane P _± and P ₂ considered. FIGS. 2A and 2B correspond respectively to the modulus of each complex image obtained. These are images of the module and M ₂ , respectively representative of the module of the light wave of exposure 14 at distances z _x and z ₂ relative to the detection plane P _o .

The difference z ₂ - z _x between the two reconstruction distances is small, preferably less than or equal to 20 pm, and preferably less than or equal to 10 pm. A part of the plates appearing in each of these figures are surrounded by a black dotted circle in the two figures. Despite the slight offset between the reconstruction distances z _x and z ₂ , it can be seen that certain plates are visible in FIG. 2B and not in FIG. 2A, the latter being circumvented by a white dotted circle in FIG. 2B. Also, in two consecutive figures, respectively reconstructed by considering close reconstruction distances, certain particles suddenly appear or disappear. In addition, when plates are visible in the two consecutive figures, they appear slightly contrasted.

When the size of the particles is larger, the use of reconstructed images makes it possible to perform a three-dimensional localization of the particles, by implementing a digital focusing algorithm, known in the field of digital holography. According to such an algorithm, a plurality of images reconstructed according to different reconstruction distances are considered, these images forming a stack of images. For each particle identified, an evolution of a sharpness indicator is established on the different reconstructed images. The coordinate, along the Z axis, of the particle considered corresponds to the coordinate of the reconstructed image for which the sharpness indicator is the highest.

However, the inventors have found that when the number of particles is high, it is not optimal to implement a digital focusing algorithm, on each particle, taken individually, as carried out in the prior art. Indeed, this is too consuming in computing resources: memory and / or computing time. An additional difficulty appears when the particles are small, typically less than 20 pm or 10 pm: the particles then appear only on certain reconstructed images, close to each other, and their contrast is low, which does not allow not an optimal implementation of digital focus. However, a digital focusing algorithm can however be implemented, so as to estimate an average focusing distance z ^, the latter corresponding to a distance, relative to the image sensor 20, at which a majority of the particles are located. 10 _p where a large number of particles are found.

The inventors have implemented a method, combining holographic image reconstruction and segmentation of reconstructed images, the main steps of which are described below, in connection with FIG. 3, in order to obtain a reliable count of small particles. , such as platelets, in a sample.

Step 100: Illumination of the sample 10 and acquisition of an image / ₀ of the sample 10 by the image sensor 20, this image forming a hologram. An advantage of the configuration without lens, shown in FIGS. 1A or 1B is the wide field observed, making it possible to simultaneously address a high volume of sample. The field observed depends on the size of the image sensor, being slightly smaller than the detection surface of the latter, due to the spacing between the pixels of the image sensor and the sample. The field observed is generally greater than 10 mm ² , and is typically between 10 mm ² and 50 mm ² , which is significantly higher than with a conventional microscope. In applications related to particle counting, this makes it possible to perform a counting based on a small number of acquired images.

Tests were carried out by considering several sources of light, in different spectral bands, and of different temporal coherences. These tests are described below in connection with FIGS. 6A to 6C.

Step 110: Obtaining a stack of complex images.

This step includes a reconstruction of several complex images by performing digital reconstructions at different reconstruction distances from the detection plane P _o . We then obtain a stack of complex images Aj. Each complex image Aj of the image stack is reconstructed at a reconstruction distance z _; , with z _{/ = 1} <z _; <z _{/ = /} , where:

Zj _{= 1} is a minimum distance from the image sensor, along the Z axis, at which a complex image Zl _{/ = 1} is reconstructed;

Zj ₌ j is a maximum distance from the image sensor, along the Z axis, at which a complex image Aj ₌ j is reconstructed.

Each index j is a natural integer representing a reconstruction distance, with 1 <j <],] being the number of complex images Aj in the image stack.

Preferably, the reconstruction distances z _; frame the sample. If the sample 10 extends, along the propagation axis Z, between two distances z _min and z _max , relative to the image sensor 20, with z _min <z _max , the minimum reconstruction distances z _{/ = 1} and maximum Zj ₌ j are such that z _{/ = 1} <z _min <z _max <Zj ₌ j. The inventors have found that if the first reconstruction distance z _{/ = 1} or the last reconstruction distance Zj ₌ j are too far from the platelets, the risk of false detection is increased. It is therefore preferable to have an a priori as to the probable distances of the platelets from the detection plane P _o .

The reconstruction distances z _; can be established by determining a focusing distance z?, at which a major part of the wafers are found, or a large number of wafers, the focusing distance being obtained by a digital focusing algorithm. The reconstruction distances z _; can then be determined by operating a progressive shift from the focusing distance Zf.

The respective reconstruction distances z ₇ , z _{; +1} of two reconstructed images Aj, A _z . _{+ i} adjacent can be separated by a distance comparable to the particle size. When the latter are plates, the difference between two adjacent reconstruction distances is preferably between 5 pm and 15 pm, for example 10 pm. Such a difference corresponds to a reconstruction step. By adjacent reconstruction distance is meant a reconstruction distance the indices j of which are consecutive.

Each complex image Aj of the stack of images is obtained from the image I _o acquired during step 100, according to the different propagation distances z _{7 = 1} ... z ₇ .... Zj ₌ j . Preferably, each complex image Aj is reconstructed by applying a reconstruction algorithm limiting the reconstruction noise, as previously described. One can for example use an algorithm described in WO2017162985. By starting from, we mean from the image I _o itself or from an image obtained by processing the image I _o , for example a normalization or taking into account the square root of the image .

At the end of step 110, there is a stack of complex images Aj, each complex image being representative of a complex expression of the exposure light wave 14, in a reconstruction plane Pj located at a distance z ₇ from the detection plane P _o .

Step 120: Obtaining a stack of reconstructed images.

The images Aj from step 110 are complex images, in the sense that they correspond to a spatial distribution of complex quantities, in a reconstruction plane Pj associated with the reconstruction distance z _; . From each complex image Aj, a reconstructed image Ij can be obtained representing a spatial distribution of a scalar quantity in the reconstruction plane Pj associated with the reconstruction distance z _; . The reconstructed image Ij can be obtained from:

an image of the module Mj of each complex image Aj, with Ij (x, y) = M ₇ (x, y) = mod (ri, (x, y));

and / or an image of the phase <pj of each complex image Aj, with <p ₇ (x, y) = arg (Aj (x, y));

and / or an image of the real part Rej of each complex image ri ,, with ffe ₇ (x, y) = Re (Aj (x, yyi, Re designating the operator real part;

and / or an image of the imaginary part Irrij of each complex image ri ,, with Imj (x, y) = Im (Aj (x, y)), Im designating the operator imaginary part.

In general, a reconstructed image Ij is an image of scalar quantities obtained from a complex reconstructed image Aj. It can in particular be established from the module and / or from the phase and / or from the real part and / or from the imaginary part of a complex image Aj calculated during step 110.

At the end of step 120, there is a stack of reconstructed images I ^ Jj ... Ij, each reconstructed image Ij being associated with a reconstruction distance Zj.

Step 130: Segmentation of each reconstructed image Ij.

FIGS. 2A and 2B are two images reconstructed by considering the module of complex images. We are talking about module images. On these images, we observe that each plate corresponds to dark pixels, forming a local contrast. During step 130, each reconstructed image Ij is subject to segmentation, so as to delimit regions of interest ROItj. The double index i, j denotes a region of interest i in a reconstructed image Ij. By image segmentation is meant a partition of the image so as to group the pixels, for example as a function of their intensity.

The image segmentation results in a segmented image in which the regions of interest ROIi j, spaced from one another, are delimited. Each region of interest can correspond to a particle, or to a cluster of particles. The segmentation makes it possible to define at least two classes of pixels: a class of pixels representing the background of the image and at least one class of pixels belonging to a region of interest ROItj. The segmentation can be carried out according to a value of a predetermined intensity threshold. The threshold value can be adjusted on each image, for example according to an Otsu thresholding, consisting in determining a value of an intensity threshold from the histogram of the image, the threshold allowing an optimal separation of the pixels according to the two classes previously described.

Other segmentation methods are known to those skilled in the art. We can for example cite the method described in Bradley D. Adaptive thresholding using the integral image. This method is based on a definition of a neighborhood of each pixel, in which an average intensity (or gray level) is calculated. It is then observed whether the intensity of the pixel considered detaches from a factor applied to the average intensity calculated in the vicinity. The neighborhood of each pixel is determined by considering a window extending on either side of the pixel, the size of which can be configured. The parameters of this segmentation process are:

the factor which is applied to the average intensity calculated in the window, so as to obtain the threshold;

the size of the window.

The inventors used a window size of 25 x 25 pixels. Figures 4A, 4B, 4C show the result of a segmentation, the threshold values being respectively greater than or equal to 1.1 times the average level, 1.07 times the average level and 1.05 times the average level.

In FIGS. 4A to 4C, the areas circled with a dark line correspond to plates, while the dark points correspond to regions of ROI interest resulting from the segmentation. The value of the factor requires adjustment, in order to compromise between false negatives and false positives. In FIGS. 4B and 4C, it is observed that the threshold values of 1.07 times or 1.05 times the average level lead to a high number of false positives. These appear in the form of dark dots that are not circled.

At the end of step 130, there are, on each reconstructed image Ij, regions of interest ROIij with each region of interest being associated with one or more plates in the image.

Step 140 Enumeration.

During step 140, the regions of interest ROItj resulting from the segmentation of the reconstructed images Ij are counted. According to one embodiment, a composite image can be constituted, comprising the different regions of interest ROItj resulting from the segmentation of the reconstructed images Ij. These regions of interest are then merged into one image. Alternatively, the regions of interest are counted on each reconstructed image independently of one another, not taking into account the regions of interest overlapping in the radial plane XY, that is to say - say delimiting the same region in the radial XY plane.

Certain regions of interest may comprise several plates arranged too close to one another so that the image segmentation can separate them. In order to take account of such a possibility, the ROItj regions of interest can be processed to identify regions of interest comprising several plates. The treatment can be carried out according to a morphological criterion, in order to assign a number of platelets per region of interest according to the shape and / or the extent of each region of interest. We can for example refer to reference shapes, corresponding to regions of interest comprising a predetermined number of plates.

The treatment can also be carried out by applying a so-called watershed filter, usually designated by the term anglosaxon watershed. During the application of such a filter, known to those skilled in the art, the image is assimilated to a topographic relief, in which a rise in water is simulated. The application of the filter makes it possible to separate regions of interest separated by pixels whose intensity is lower than the level considered.

FIGS. 5A to 5D are experimental examples, carried out on samples comprising platelets, showing the effectiveness of filtering of the watershed type. Figures 5A and 5B correspond to the same scenario before and after the implementation of the filter. It is observed that the filter allows a separation of a region of interest a (FIG. 5A) into two distinct regions of interest b etc (FIG. 5B). Figures 5C and 5D correspond to the same scenario before and after the implementation of the filter. It is observed that the filter allows separation of a region of interest a (FIG. 5C) into three distinct regions of interest b, c and d (FIG. 5D).

During step 140, a delicate point is the processing of regions of monopixel interest ROI ^ comprising only one pixel. In fact, it is unlikely that each monopixel region of interest corresponds to a brochure. A first possibility is not to take into account the regions of monopixel interest when counting. The risk is then to increase the number of false negatives, that is to say to obtain an incomplete count. In order to reduce the risk of occurrence of a false negative, a ROI ^ monopixel region of interest is considered to be representative of a plate if it appears on two adjacent reconstructed images of the image stack, that is to say - say on two reconstructed images β β ₊₁ or and Ij.

FIG. 6 represents, on different samples, the number of platelets N determined by implementing the invention (ordinate axis) as a function of the number of platelets determined by implementing a reference method (abscissa axis). The reference method is a laboratory automaton, of the HORIBA P60 type. Curve I corresponds to an implementation without taking into account the regions of monopixel interest. Curve II corresponds to an implementation taking into account the regions of monopixel interest appearing on two adjacent reconstructed images. On each curve, we have represented an equation of a straight line obtained by linear regression. It is observed that taking into account the regions of monopixel interest, subject to an appearance on two successive reconstructed images, makes it possible to approximate the slope of the correlation curve of 1, and to obtain a close correlation coefficient R ² from 1.

The inventors have examined the influence of the incident illumination wave 12 on the counting performance, and in particular the spectral band Δλ as well as the spectral bandwidth, the latter conditioning the temporal coherence of the wave.

We first used two light sources:

a laser diode emitting in blue / violet, and whose emission spectral band is centered on a wavelength of 405 nm (curves I and II of FIG. 7A);

a light emitting diode, and whose emission spectral band is centered on a wavelength of 450 nm and a bandwidth of 20 nm (curve III of FIG. 7A).

The laser diode was supplied at two different intensities: 120 mA (curve I) and 90 mA (curve II), these supply currents giving a bandwidth of 4.6 nm and 8.9 nm respectively. WO2017109428 describes a method for determining the optimum supply intensity of a laser diode, which is particularly suitable for the application of holographic observation of samples containing particles.

FIG. 7A represents the number N of platelets determined by implementing the invention (ordinate axis) as a function of the number of platelets determined by implementing a reference method (abscissa axis). For each test, a regression line has been indicated, the equation of which is given in FIG. 7A. It is observed that the most favorable configuration, that is to say that for which the slope, as well as the correlation coefficient R ² , come closest to 1, is the curve I, obtained with a laser source whose emission spectral band is centered on a wavelength close to 400 nm and whose spectral bandwidth is the smallest of the three configurations tested.

We then used two light sources:

the laser diode emitting in blue / violet, and whose emission spectral band is centered on a wavelength of 405 nm, and of spectral width 4.6 nm, which corresponds to a coherence length of 15 μιτι (curve I of Figure 7B);

a laser diode emitting in the red, and whose emission spectral band is centered on a wavelength of 653 nm (curves II and III of FIG. 7B).

The laser diode emitting in the red was supplied at two different intensities: 12 mA (curve II) and 11 mA (curve III), these supply currents giving a bandwidth of 5 nm and 8.9 nm respectively, which corresponds to a coherence length of 104 μιτι and 23 μιτι.

FIG. 7B represents the number N of platelets determined by implementing the invention (ordinate axis) as a function of the number of platelets determined by implementing a reference method (abscissa axis). A regression line, the equation of which is given in the figure, was indicated for each test, as well as a correlation coefficient R ² . We observe that the most favorable configuration, that is to say the one for which the slope comes closest to 1, is the curve I, obtained with a laser source whose emission spectral band is centered on a length. wave near 400 nm. Furthermore, the comparison of curves II and III shows that it is preferable that the spectral band is as fine as possible, so as to maximize the coherence length.

We then used three light sources:

the laser diode emitting in violet, previously described, whose emission spectral band is centered on a wavelength of 405 nm, and of spectral width 4.6 nm, which corresponds to a coherence length of 15 pm (curve I of Figure 7C); the laser diode emitting in the red, and whose emission spectral band is centered on a wavelength of 653 nm, and of spectral width 8.9 nm, which corresponds to a coherence length of 23 pm (curve II of Figure 7C);

a light emitting diode emitting in the red, the emission spectral band of which is centered on a wavelength of 640 nm, and of spectral width 20 nm (curve III of FIG. 7C).

FIG. 7C represents the number N of platelets determined by implementing the invention (ordinate axis) as a function of the number of platelets determined by implementing a reference method (abscissa axis). We observe that the most favorable configuration, that is to say the one for which the slope comes closest to 1, is the curve I, obtained with a laser source whose emission spectral band is centered on a length. wave near 400 nm. By using a shorter wavelength, it has been found that the resolution of the hologram, that is to say of the image acquired by the image sensor, is improved. This is the reason why values of correlation coefficient R ² are obtained which are closer to 1 when the wavelength is close to 400 nm. Furthermore, the comparison of curves II and III shows that it is preferable that the spectral band is as fine as possible.

The invention has been described by implementing a lensless imaging configuration, according to which no imaging optics are interposed between the sample 10 and the image sensor 20. According to a variant, illustrated on FIG. 8, the device comprises an optical system 16, defining an object plane P _O bj and an image plane P _im . The image sensor 20 is then arranged in a so-called defocused configuration, according to which the sample extends in a plane offset from the object plane, and / or the image sensor extends in a plane offset from the plane at the image level. By defocused configuration is meant a configuration comprising an offset of the sample and / or of the image sensor with respect to a focusing configuration, according to which the detection plane Po is conjugated with a plane P _i0 according to which extends the sample. The offset δ is preferably less than 500 μm, or even less than 200 μm. It is preferably greater than 10 pm or 20 pm. The steps 100 to 140 previously described can be implemented in such a configuration. In the example shown in FIG. 8, the object plane P _O bj is coincident with a plane P _i0 along which the sample 10 extends and the image plane P _im is offset with respect to the detection plane P _o .

Although described in connection with a platelet count, the invention can be implemented for counting other particles, preferably of dimension less than 20 μm, or 10 μm or even 5 μm. The invention can be implemented, without limitation, in the field of health, environmental control, food or control of industrial processes.

Claims (16)

1. Method for counting particles in a sample, the sample comprising a medium (10 m) in which the particles (10 _p ) are immersed, the medium being held by a holding element (15), the medium being immobile with respect to the holding element (15), the method comprising the following steps: a) illumination of the sample (10) by a light source (11), the sample being arranged between the light source (11) and an image sensor (20); b) acquisition, using the image sensor (20), of an image (/ ₀ ) of the sample (10), the sample being placed between the light source (11) and the sensor d image (20), each image being representative of a light wave (14) of exposure, to which the image sensor (20) is exposed under the effect of the illumination of the sample; c) from the image acquired (/ ₀ ) during step b), application of a digital propagation operator (h) to obtain a complex expression (Aj) of the exposure light wave (14 ) in different reconstruction planes (P _; ), each reconstruction plane (z _; ) being associated with a reconstruction distance (z _; ), relative to the image sensor (20); d) formation of a stack of reconstructed images, (/ _x ... Ij), each reconstructed image (/) being obtained from the module and / or the argument of a complex expression (Aj) in a reconstruction plane (P _; ), each reconstructed image being associated with a reconstruction distance (z _; ) different from each other; e) segmentation of each reconstructed image, so as to isolate, on each reconstructed image, regions of interest (ROIij), spaced apart from each other, each isolated region of interest corresponding to a particle or a cluster of particles; f) from the regions of interest (ROItj) isolated in step e), estimation of a quantity of particles in the sample. 2. Method according to claim 1, in which in step c), the reconstruction distances are obtained by determining a focusing distance (Zf), corresponding to a distance, relative to the image sensor, at which a number maximum of particles are located, each reconstruction distance (z _; ) being established by applying an offset with respect to the focusing distance. 3. Method according to any one of the preceding claims, in which the difference between two successive reconstruction distances corresponds to a reconstruction step, the reconstruction step being less than 20 μm. 4. Method according to any one of the preceding claims, in which the particles are inscribed in a circle or in a sphere whose diameter is less than 20 µm or 10 µm or 5 µm. 5. Method according to any one of the preceding claims, in which the particles are thrombocytes. 6. Method according to any one of the preceding claims, in which during the step e), the segmentation of each reconstructed image is carried out by applying a thresholding, according to an intensity threshold, on each reconstructed image. 7. The method of claim 6, wherein each reconstructed image comprising pixels, the value of the intensity threshold is variable from one pixel to another. 8. The method as claimed in claim 7, in which the value of the intensity threshold at one pixel is established by: defining a neighborhood around each pixel; calculating an average intensity of pixels belonging to the neighborhood; the threshold value, at each pixel, depending on the average intensity thus calculated. 9. Method according to any one of the preceding claims, comprising a step of analyzing the regions of interest resulting from step e), so as to: - determine a number of particles associated with each region of interest; - or divide regions of interest encompassing several particles into different regions of interest comprising only one particle. 10. Method according to any one of the preceding claims, in which during the step f), a region of interest whose extent is limited to one pixel is only taken into account if it appears in two adjacent reconstructed images. 11. Method according to any one of the preceding claims, in which the light source is a laser source. 12. Method according to any one of the preceding claims, in which the light wave emitted by the light source has a spectral bandwidth less than 10 nm. 13. Method according to any one of the preceding claims, in which the light wave emitted by the light source has a spectral band centered on a wavelength less than 500 nm, and preferably between 400 nm and 500 nm. 14. Device for counting particles of a sample, comprising: - a light source (11) capable of emitting an incident light wave (12) propagating towards the sample (10); - an image sensor (20); - a support (10s), configured to hold the sample (10) between the light source (11) and the image sensor (20); a processor (21), configured to receive an image (/ ₀ ) of the sample acquired by the image sensor (20) and to implement steps c) to f) of a method according to any one of claims 1 to 13. 15. Device according to claim 14, in which no image-forming optics are arranged between the sample (10) and the image sensor (20). 16. Device according to claim 14, comprising an optical system (16) defining a focusing configuration, according to which the image sensor (20) is conjugated with a plane passing through the sample, the device being such that the image sensor (20) or the sample (10) are offset from the focus configuration, so that during step b), the image sensor (20) acquires an image defocused sample (10).