Method for characterizing a particle based on a hologram

Description

DOMAINE TECHNIQUE

[0001] The technical field of the invention is the characterization of particles, for example biological particles, from an image. ART ANTERIEUR

[0002] Holographic imaging has been undergoing significant developments for several years, particularly in the field of sample analysis. A large number of applications have been described for characterizing samples, particularly samples intended for biological or diagnostic applications. Many publications concern, for example, samples containing particles, the objective being to characterize the particles.

[0003] In a popular configuration, the image of the sample can be acquired using a lensless imaging configuration, with no imaging optics being placed between the sample and the image sensor. In the image acquired by the image sensor, each particle produces a diffraction pattern, the latter representing a signature of the particle. WO2008090330 describes such a configuration, when the source used is a light-emitting diode. In this document, it is shown that this imaging modality provides a high field of observation, and allows rapid identification of cells or bacteria using a simple device.

[0004] Document US7465560 describes a device for identifying bacterial colonies based on a diffraction pattern formed on an image sensor. In this document, the bacterial colony under examination is illuminated using a laser beam. This document shows that a diffraction pattern constitutes a signature of a bacterial colony. A diffraction pattern can be classified, for example, by using a decomposition according to Zernike polynomials. This is described, for example, in US8787633 or in WO2014184390.

[0005] The image acquired by the image sensor can be reconstructed, known as holographic reconstruction, to obtain a clearer image of the particles. Sophisticated holographic propagation algorithms, allowing a so-called reconstructed image of the sample to be obtained, are described in WO2016189257 or in WO2017162985.

[0006] The application of such algorithms can make it possible to obtain a complex image of the sample. This image can be segmented for particle counting purposes, as described in WO2018115734. By forming a plurality of complex images of the sample, it is possible to establish profiles parallel to the axis of propagation of the light, to characterize particles. This is for example described in WO2018215337.

[0007] A difficulty may arise when the sample contains a high concentration of particles. Indeed, on the image acquired by the image sensor, the diffraction patterns corresponding to each particle may overlap. As a result, characterization of the particles, based on their respective diffraction patterns, becomes more difficult to implement.

[0008] The inventors sought to improve the methods described above, by allowing a characterization of particles from diffraction patterns associated with each of them. The method is particularly suitable for a sample in which, due to the density of the particles, the diffraction patterns associated with each of them overlap. EXPOSE DE L'INVENTION

[0009] A first subject of the invention is a method for characterizing a particle present in a sample, the sample extending between an image sensor and a light source, the image sensor extending along a detection plane, the method comprising the following steps: a) illuminating the sample by the light source, the light source emitting an incident light wave propagating along a propagation axis; b) acquiring an image of the sample by the image sensor, the image sensor being exposed to an exposure light wave, the image comprising a plurality of elementary diffraction patterns, each elementary diffraction pattern corresponding to a particle; c) from the acquired image, reconstructing a complex image representative of a complex amplitude of the exposure light wave, in at least one reconstruction surface preferably passing through the sample, the reconstruction being carried out by implementing an iterative reconstruction algorithm, the algorithm comprising, during each iteration, an update of a phase of the exposure light wave in the detection plane or in the reconstruction surface;d) selecting a region of interest from the complex image, the selected region of interest corresponding to a particle of interest; e) forming an extracted complex image from the region of interest selected in step d); f) applying a propagation operator to the extracted complex image formed in step e), so as to obtain, in a propagation plane, an elementary diffraction pattern of the particle selected in step c), such that the elementary diffraction pattern obtained is isolated from elementary diffraction patterns of other particles present in the sample; g) characterizing the particle of interest as a function of the elementary diffraction pattern resulting from step f).

[0010] In step f), the elementary diffraction pattern can take the form of a complex image. We can then obtain a representation of the elementary pattern, by considering the modulus or the phase of the complex image, or even its real part or its imaginary part.

[0011] In step f), the propagation plane can be the detection plane. But it can also be a plane located at a distance from the sample less than or greater than the distance between the sample and the detection plane.

[0012] According to one embodiment, step c) comprises, at each iteration, an application of a holographic propagation operator to an image of the exposure light wave, in the detection plane, so as to obtain a complex expression of the exposure light wave, in at least one plane along which the sample extends, forming the complex image; step d) comprises, from the reconstructed complex image, obtaining an observation image, the particle of interest being selected on the observation image.

[0013] Step f) may comprise the application of a holographic propagation operator to the region of interest selected during step e), so as to obtain the elementary diffraction pattern, corresponding to the particle of interest, in the propagation plane, the part of the reconstructed complex image, located outside the region of interest not being propagated.

[0014] The method may include an estimation of a distance between the particle of interest and the detection plane, such that during step f), the propagation operator takes into account the distance thus estimated.

[0015] In step f), propagation is performed considering a predetermined propagation distance.

[0016] According to one embodiment, during step f), a complex elementary diffraction pattern is obtained, in the reconstruction plane, of the particle selected during step c). During step g), the characterization is then carried out from the phase or the modulus of the complex elementary diffraction pattern, or from its real part or its imaginary part.

[0017] Step g) may include the determination of an axial profile, representing a radial average of the elementary diffraction pattern.

[0018] Step g) may include: a comparison of the elementary diffraction pattern, obtained during step f), with at least one diffraction pattern corresponding to a known particle; or a comparison of the elementary diffraction pattern, obtained during step f), with at least one diffraction pattern modeled by considering a particle whose refractive index and / or size are known; or a classification of the elementary diffraction pattern, obtained during step f), taking into account diffraction patterns obtained using known particles, or modeled diffraction patterns, during a learning phase.

[0019] According to one embodiment, step g) comprises a decomposition of the elementary diffraction pattern, according to a basis of Zernike polynomials, so as to obtain coordinates of the elementary diffraction pattern in the basis; the characterization of the particle of interest is carried out as a function of the coordinates of the elementary detection pattern.

[0020] In step g), the characterization can be carried out based on modeled diffraction patterns, each diffraction pattern corresponding to a particle for which a parameter is known, the parameter being a size and / or a refractive index, the characterization aiming to determine a value of the parameter for the analyzed particle of interest. The modeling can in particular be based on a hypothesis according to which the particle is spherical.

[0021] The method may include iterative adjustment of the particle parameter, each iteration including: i) taking into account a parameter or a set of parameters of the particle; ii) obtaining a modeled diffraction pattern according to the parameter or set of parameters taken into account; iii) determining a difference between the modeled diffraction pattern and the elementary diffraction pattern resulting from step f); iv) estimating a parameter, or a set of parameters, aimed at reducing the difference thus determined, the estimated parameter or set of parameters being taken into account during the following iteration.

[0022] According to one embodiment, during steps i) to iv), a set of parameters is taken into account, the set of parameters comprising a distance between the particle and the detection plane.

[0023] A second object of the invention is a device for observing a sample, the sample comprising particles, the device comprising: a light source, configured to illuminate the sample; a sample holder, configured to receive the sample; an image sensor, arranged such that when the sample is arranged on the holder, it extends between the image sensor and the light source, the image sensor being configured to acquire an image of the sample; a processor, configured to implement steps c) to f) or c) to g) of a method according to the first subject of the invention, from an image acquired by the image sensor.

[0024] According to one embodiment, no imaging optics are disposed between the image sensor and the processor.

[0025] According to one embodiment, the device comprises an optical system extending between the sample and the image sensor, the optical system defining an image plane and an object plane, the device comprising a means for adjusting the optical system, or the sample, or the image sensor, such that: the detection plane is offset from the image plane; and / or the object plane is offset from a sample plane, along which the sample extends.

[0026] Other advantages and characteristics will emerge more clearly from the following description of particular embodiments of the invention, given as non-limiting examples, and represented in the figures listed below. FIGURES

[0027] There figure 1 shows a first device allowing an implementation of the invention. The figure 2 shows the main steps of a process implementing the invention. figure 3A is an example of an image acquired by the sensor. The figure 3B is an image of the modulus of a reconstructed complex image, in a plane along which the sample extends, from the image acquired by the sensor. The figure 3C is an image of the phase of a reconstructed complex image, in a plane along which the sample extends, from the image acquired by the sensor. The figure 3D is an image of the modulus of a complex image extracted from a region of interest, corresponding to a particle of interest, of the reconstructed complex image. The figure 3E is an image of the phase of a complex image extracted from a region of interest, corresponding to a particle of interest, of the reconstructed complex image. The figure 3F shows an image of the modulus of a diffraction pattern propagated from the extracted complex image. The figure 4 shows an example of a representative radial average of the modulus of a diffraction pattern. The figure 5A illustrates test results performed on a sample containing blood particles. The figure 5B illustrates test results performed on a sample containing blood particles. The figure 5C illustrates test results performed on a sample containing blood particles. The figure 5D illustrates test results performed on a sample containing blood particles. The figure 6 shows another device allowing an implementation of the invention. EXPOSE DE MODES DE REALISATION PARTICULIERS

[0028] There figure 1 represents an example of a device according to the invention. A light source 11 is configured to emit a light wave 12, called an incident light wave, propagating towards a sample 10, along a propagation axis Z. The light wave is emitted according to an illumination spectral band Δλ.

[0029] Sample 10 is a sample that we wish to characterize. It includes in particular a medium 10m in which 10p particles are bathed. The medium 10m may be a liquid medium. It may include a bodily fluid, obtained for example from blood or urine or lymph or cerebrospinal fluid. It may also be a culture medium, comprising nutrients allowing the development of microorganisms or cells. By particle, we mean in particular, and in a non-exhaustive manner: a cell, whether a cultured cell or a body cell, for example a blood cell; a microorganism, for example a bacterium or a yeast or a microalgae; a solid particle, for example a microbead, the microbead being able to be functionalized, so as to promote grafting with an analyte; a particle forming an emulsion in the 10m medium, in particular a particle insoluble in the 10m medium, an example being a lipid droplet in an aqueous medium.

[0030] A 10 p particle can be solid or liquid. The invention is more particularly directed to particles whose size is less than 1 mm, or even 100 µm, or even 50 µm. By particle size, we mean a diameter or a diagonal.

[0031] By characterize, we mean in particular: to determine the nature of a particle, that is to say to classify this particle among one or more predetermined classes; to determine the state of a particle, among one or more predetermined states; to estimate a size of a particle, or its shape, or its volume or any other geometric parameter; to estimate an optical property of one or more particles, for example the refractive index or an optical transmission property, or an optical thickness (product of the thickness by the refractive index); the method may also include counting particles based on their characterization.

[0032] The sample 10 is, in this example, contained in a fluidic chamber 15. The fluidic chamber 15 is for example a Countess or Cellvision, Gene Frame ®< type fluidic chamber with a thickness e=100 µm. The thickness e of the sample 10, along the propagation axis, typically varies between 10 µm and 1 cm, and is preferably between 20 µm and 500 µm. The sample extends along a plane P 10 , called the sample plane, perpendicular to the propagation axis Z. The sample plane is defined by the X and Y axes shown on the figure 1 The sample is held on a support for 10s at a distance d from an image sensor 16. The particle concentration can vary between 50 per microliter and 100,000 per microliter.

[0033] The distance D between the light source 11 and the fluidic 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 point-like. This means that its diameter (or its diagonal) is preferably less than one tenth, better one hundredth of the distance between the fluidic chamber 15 and the light source. On the figure 1 , the light source is a light-emitting diode. It is generally associated with a diaphragm 18, or spatial filter. The aperture of the diaphragm is typically between 5 µm and 1 mm, preferably between 50 µm and 500 µm. In this example, the diaphragm is supplied by Thorlabs under the reference P150S and its diameter is 150 µm. 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. The device shown in the figure 1 also comprises a diffuser 17, arranged between the light source 11 and the diaphragm 18. The use of such a diffuser makes it possible to overcome constraints of centering the light source 11 relative to the opening of the diaphragm 18. The function of such a diffuser is to distribute the light beam, produced by an elementary light source 11 according to a cone of angle α. The use of such a diffuser is described in WO2016078946. Preferably, the diffusion angle α varies between 10° and 80°. Alternatively, the light source may be a laser source, such as a laser diode. In this case, it is not useful to associate a spatial filter or a diffuser with it. When the light source is a laser source, it is advantageously used as described in application WO2017109428.

[0034] Preferably, the emission spectral band Δλ of the incident light wave 12 has a width of less than 100 nm, or even less than 50 nm. Spectral bandwidth means a width at half-height of said spectral band.

[0035] The sample 10 is arranged between the light source 11 and the image sensor 16 previously mentioned. The image sensor 16 defines a detection plane P 0 , preferably extending parallel, or substantially parallel to the plane P 10 along which the sample extends. The term substantially parallel means that the two elements may not be strictly parallel, an angular tolerance of a few degrees, less than 20° or 10° being allowed.

[0036] The image sensor 16 is capable of forming an image I 0 of the sample 10 according to the detection plane P 0 . In the example shown, it is an image sensor 16 comprising a pixel matrix, of the CCD or CMOS type. The detection plane P 0 preferably extends perpendicular to the propagation axis Z of the incident light wave 12. The distance d between the sample 10 and the pixel matrix of the image sensor 16 is preferably between 50 µm and 2 cm, preferably between 100 µm and 2 mm.

[0037] In this embodiment, we note the absence of magnification or image formation optics between the image sensor 16 and the sample 10. This does not prevent the possible presence of focusing microlenses at each pixel of the image sensor 16, the latter not having a function of magnifying the image acquired by the image sensor, their function being to optimize the detection efficiency.

[0038] As mentioned in the applications cited in the prior art, 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 detection plane P 0 , interference, in particular with a part 12' of the incident light wave 12 transmitted by the sample. Furthermore, the sample can absorb a part of the incident light wave 12. Thus, the light wave 14, transmitted by the sample, and to which the image sensor 16 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, a part of the latter being able to be absorbed in the sample.

[0039] These components form interferences in the detection plane. Also, the image acquired by the image sensor includes interference patterns (or diffraction patterns), each interference pattern being able to be associated with a 10 p particle of the sample.

[0040] A processor 20, for example a microprocessor, is capable of processing each image I0 acquired by the image sensor 16. In particular, the processor is a microprocessor connected to a programmable memory 22 in which is stored a sequence of instructions for carrying out 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 16 or calculated by the processor 20.

[0041] When the interference patterns are sufficiently far apart, each interference pattern can be used to characterize the particle with which it is associated. This is described in documents US8787633 or WO2014184390 cited in the prior art. However, when the quantity of particles present in the sample is large, the interference patterns overlap, which no longer allows for correct characterization using the methods of the prior art.

[0042] The inventors have designed a method that is particularly suitable for the situation in which the elementary diffraction patterns of each particle overlap. The main steps of the method are described in figure 2 These steps are described in connection with the figures 3A à 3F These figures were obtained using a blood sample purified with Dextran. Dextran causes red blood cells to aggregate, which leads to their sedimentation. The supernatant plasma, depleted in red blood cells compared to the initial blood, is recovered. It is then diluted in PBS (Phosphate Buffer Saline) buffer at physiological pH. The experimental conditions are as follows: light source: 405 nm laser source; image sensor: IDS UI-1924LE CMOS sensor, comprising 3840 by 2748 pixels, the size of each pixel being 1.67 µm x 1.67 µm, providing a useful detection area of ​​approximately 30 mm 2<; distance between the sample and the image sensor: between 1 mm and 1.5 mm.

[0043] Etape 100 : image acquisition I 0 by the image sensor. The image I0 has different elementary diffraction patterns superimposed on each other. The figure 3A represents such an image. Obviously, the methods for characterizing diffraction patterns presented in the prior art documents are not applicable, due to the excessively high density of particles in the sample.

[0044] Etape 110 : reconstruction of an image A 10 complex in the sample plane. From the image I 0 obtained in step 100, a holographic reconstruction algorithm is used to obtain a complex image A10 , preferably in a reconstruction plane P 10 , along which the sample extends. The complex image A 10 corresponds to a complex expression of the exposure light wave 14 in the reconstruction plane considered. This step may in particular implement an iterative algorithm as described in WO2017162985, and more precisely in steps 110 to 150 described in this application.

[0045] During this iterative algorithm, at each iteration, a propagation of the acquired image, or of an image resulting from a previous iteration, is carried out in the reconstruction plane, in this case the plane of the sample. This produces a complex image propagated in the plane of the sample. A noise level in the reconstructed complex image is then determined, then a complex image is adjusted, in the plane of the sample, to reduce the noise level in the complex image propagated in the plane of the sample. The adjustment is carried out by minimizing a criterion quantifying a noise level in the complex image propagated in the plane of the sample. The adjustment consists in particular of adjusting the phase of the exposure light wave, in the plane of the sample. The complex image estimated in the detection plane, during each iteration, is used, during the following iteration, by being propagated in the plane of the sample.The inventors observed that the implementation of such an iterative algorithm led to particularly efficient results.

[0046] Alternatively, one can implement an iterative algorithm as described in WO2016189257. Such an algorithm assumes that an image of the sample is formed in several spectral bands.

[0047] According to such algorithms, the phase of the exposure light wave 14 is progressively adjusted in the detection plane. In WO2016189257, the phase adjustment is performed by iterations by averaging, during each iteration, the phase of light waves reconstructed in the sample plane, according to different spectral bands. In WO2017162985, the phase adjustment is performed by iteration so as to minimize, during each iteration, the reconstruction noise of a complex image reconstructed in the sample plane.

[0048] Following the implementation of the holographic reconstruction algorithm, an image can be obtained I 10, called the observation image of the sample, from the reconstructed complex image I 10. The observation image I 10 is an image allowing to visually represent the reconstructed complex image A 10 . It can be: of an image of the reconstructed complex image module: I 10 = mod ( A 10), mod denoting the modulus operator; or of an image of the phase of the reconstructed complex image: I 10 = arg ( A 10), arg denoting the argument operator; or of an image of the real part of the reconstructed complex image: I 10 = Re ( A 10), Re denoting the real part operator; or of an image of the imaginary part of the reconstructed complex image: I 10 = Im ( A 10), Im designating the imaginary part operator; or a combination of the previously cited images.

[0049] Generally speaking, an observation image is an image formed from scalar quantities obtained from a complex image, the latter being formed from complex quantities.

[0050] THE figures 3B et 3C respectively represent an observation image I 10 corresponding to the modulus or phase of a part of a complex image A 10 reconstructed, in the plane of the sample, from the hologram represented on the figure 3A . The observation image I 10 of the figure 3B , obtained by the module of the reconstructed complex image, allows to separate each particle from one another. It allows to select a particle of interest 10 p,i that one wishes to characterize. Etape 120 : selection of a region of interest ROI i , corresponding to the particle of interest 10 p,i .

[0051] An interest in the observation image I 10 is that it is obtained from a complex image A 10 reconstructed in a reconstruction plane P 10 passing through the sample 10. The spatial resolution of the observation image I 10 allows easy separation of 10p particles from each other. From the observation image I 10 , a user can select a particle. The selection can also be done by a microprocessor, by applying an image segmentation algorithm, so as to isolate regions of interest from the background of the image. The segmentation can be done by thresholding, for example Otsu thresholding or Bradley thresholding. The region of interest ROI i then corresponds to a particle, or to a group of particles close to each other, forming the same trace on the reconstructed complex image, the latter case corresponding in particular to the formation of a bacterial colony.

[0052] At the end of step 120, we have a region of interest ROI i of the observation image I 10 , which corresponds to a particle of interest 10 p,i that we wish to characterize. The region of interest ROI i can also be applied to the reconstructed complex image A 10 , so as to extract, from this image, only the region of interest corresponding to the particle of interest 10 p,i .

[0053] Preferably, the region of interest ROI i is centered on the particle of interest 10 p,i , or on the group of particles considered. Etape 130

[0054] Extraction of a part of the complex image located in the region of interest ROI i selected in step 120, so as to obtain an extracted complex image A 10,i. The part of the image thus extracted forms a thumbnail. The figures 3D et 3E respectively show the module mod ( A 10 ,i ) and the phase arg ( A 10 ,i ) of the extracted complex image A 10, i In these figures, the extracted complex image includes the region of interest ROI i of the reconstructed complex image A 10 . The pixels arranged around the region of interest are representative of an average background level obtained in the reconstructed complex image A 10 . Alternatively, the pixels of the extracted complex image A 10, i arranged around the region of interest ROI i can be set to an arbitrary value, for example a gray level equal to 1. Etape 140

[0055] Propagation, in a propagation plane, of the extracted complex image A 10,i . The propagation plane may notably be the detection plane, which is considered below. However, this is not necessary. The propagation plane, in which the extracted complex image is propagated, may be different from the detection plane. The region of interest ROI i is extracted from a complex image A 10 obtained using a holographic reconstruction algorithm involving a progressive adjustment of the phase of the exposure light wave. It can therefore be assumed that in the extracted region of interest, the phase value is correctly estimated. Also, the exposure light wave 14 is correctly described in the extracted region of interest ROI i . The region of interest ROI i can then be propagated by simply using a holographic propagation operator h, known in the field of holography, for example a Fresnel operator, such as: h x y z = 1 jλz e j 2 π z λ exp jπ x 2 + y 2 λz x, y designate coordinates in the detection plane P 0 ; z represents a coordinate along the Z propagation axis.

[0056] The propagation may include a convolution of the extracted complex image, including the region of interest ROI i , by the propagation operator h, so that: A 0 , i x y z = A 10 , i x y z ∗ h Or : A 0, i is the elementary diffraction pattern obtained in the detection plane P 0 ; A 10, i is the extracted complex image, including the region of interest ROI i , of the reconstructed complex image A 10 .

[0057] According to this step, only the region of interest ROI i of the complex image A10 is propagated in the propagation plane (in this example the propagation plane), so as to obtain an elementary diffraction pattern corresponding to the particle of interest isolated from the other particles.

[0058] This step allows us to obtain an image representing the elementary diffraction pattern A 0, i corresponding to the particle of interest 10 p,i without this figure being superimposed on the diffraction figures corresponding to the other 10 p particles of the sample, in particular the particles arranged in the vicinity of the particle of interest 10 p,i . An elementary diffraction figure is obtained, which allows a more reliable characterization of the particle of interest 10 p,i . The elementary diffraction figure A 0, iis a complex image, that is to say formed of complex amplitudes, from which we can extract the modulus, or the phase, or the real part, or the imaginary part, so as to obtain an observation diffraction pattern I 0, i .

[0059] There figure 3F is an observational diffraction pattern I 0, i obtained from the modulus of an elementary diffraction pattern A 0, i , the latter being obtained by propagation of the extracted complex image A 10, i represented on the figures 3D et 3E , in the propagation plane, which in this example is the detection plane.

[0060] The particles are not necessarily coplanar. Also, according to one embodiment, during step 110, several observation images are formed I 10, zto each observation image being associated a reconstruction distance z with respect to the detection plane P 0 . Each observation image I 10, z results from a complex image A 10, z obtained by implementing an iterative holographic reconstruction algorithm from the image I 0 formed in the detection plane P 0 . The difference between two successive reconstruction distances z can be between a few µm and a few tens of µm. For each particle of interest 10 p,i we can determine a distance z i relative to the detection plane. The distance z i can be determined by digital focusing, consisting of determining the distance at which the trace of the particle of interest 10 p,i , on the different observation images I 10, z , is the sharpest. According to this embodiment, the region of interest ROI i corresponding to each particle of interest 10 p,i is extracted from the observation image I 10, z i trained at a distance z i associated with the particle.

[0061] In step 140, the propagation of the region of interest ROI i is then carried out taking into account, in the propagation operator, the propagation distance z i associated with the particle of interest 10 p,i .

[0062] Alternatively, whatever the distance z i associated with the particle of interest 10 p,i , the propagation of the extracted complex image is carried out by considering a constant and predetermined propagation distance. Thus, depending on the distance z i associated with them, the particles of interest 10 p,i are propagated according to different propagation planes.

[0063] The invention thus allows, for each particle of interest 10 p,i, a formation of an elementary diffraction patternA 0, i taking into account the distance z i between the particle of interest and the detection plane P 0 . Etape 150. Caractérisation

[0064] During this step, the particle of interest 10 p,i is characterized using the elementary diffraction pattern A 0,i resulting from step 140.

[0065] The characterization of the particle of interest 10 p,i is carried out from diffraction patterns corresponding to known particles, the latter being either modeled or acquired experimentally.

[0066] The characterization may consist of a classification of the particle of interest 10 p,i among predetermined particle classes. The classification may involve a learning phase, performed using one or more learning samples, comprising known particles.

[0067] Characterization can use deep learning techniques, usually referred to as "deep learning".

[0068] According to one embodiment, a classification is implemented assuming a reduction in the dimensionality of the elementary diffraction pattern. A 0, i .

[0069] A first solution is to perform a radial average of the elementary diffraction pattern, considering for example the modulus or the phase of the diffraction pattern. This solution is particularly suitable for spherical objects. Indeed, each diffraction pattern has a certain symmetry around a center of the diffraction pattern. It is possible to present the diffraction pattern according to a radial average, consisting of performing an average of the value of the pixels located at the same distance from the center of the diffraction pattern. The figure 4 represents such a radial average (curve a), obtained by considering the modulus of the elementary diffraction pattern. The radial average takes the form of an intensity profile developed in the detection plane, transverse to the elementary diffraction pattern, and passing through the center of the latter. On the figure 4 , the brace shows the part of the radial average with sufficient signal-to-noise ratio and spatial sampling to be used to characterize the particle of interest. Using the radial average allows the information present in the elementary diffraction pattern to be converted into a function whose variable r represents the distance from the center of the diffraction pattern. The variable r corresponds to the abscissa of the figure 4 .

[0070] Another solution is a projection of an observation image resulting from the elementary diffraction patternA 0, i , according to a basis of Zernike polynomials. In this case, the elementary diffraction pattern is decomposed according to a predetermined number of Zernike polynomials Z n m n being an integer corresponding to the order of the polynomial, m being an integer such that m < n , and such that n . - m is even.

[0071] Zernike polynomials are known to those skilled in the art, and their use has been described in US8787633 or in WO2014184390. For example, from the elementary diffraction pattern, it is possible to obtain A 0, i , two observation images representing respectively the modulus and the phase of the elementary diffraction pattern. Each image is decomposed according to the first 15 Zernike polynomials Z 0 0 , Z 1 − 1 , Z 1 1 , Z 2 − 2 , Z 2 0 , Z 2 2 , Z 3 − 3 , Z 3 − 1 , Z 3 1 , Z 4 − 4 , Z 4 − 2 , Z 4 0 , Z 4 2 , Z 4 4 .

[0072] The decomposition allows to obtain coordinates in a basis defined by the polynomials considered. When the decomposition is carried out under 15 polynomials, we obtain a vector V i of dimension (15, 1). The classification of the elementary diffraction pattern A 0, i is performed using the vector resulting from the decomposition. When performing a decomposition of the modulus and phase of the elementary diffraction pattern A 0, i , we obtain a vector V i of dimension (30, 1).

[0073] To the vector resulting from the decomposition in the polynomial basis, a term representing a dimension or, more generally, a geometric parameter of the particle of interest can be added. By geometric parameter, we mean a perimeter, a radius or diameter, an area, an eccentricity, an estimate of the volume from the area. The geometric parameter of the particle can be obtained by considering the complex image A 10 reconstructed in the sample plane, during step 120, or an observation image I 10 obtained from the reconstructed complex image. In the example shown in connection with the figures 3A à 3F , we estimated a dimension of the particle of interest from the image of the module of the extracted complex image A 10, i represented on the figure 3D .

[0074] We can also add, to the vector resulting from the decomposition in the polynomial basis, an optical characteristic, for example an absorbance level, a refractive index value or a diffraction power value.

[0075] The elementary diffraction pattern A 0, i, or its projection, whether a radial average or a polynomial basis, can be compared to diffraction patterns obtained experimentally using known calibration particles. It can also be compared to diffraction patterns obtained by numerical modeling. According to such an embodiment, a database is created comprising modeled diffraction patterns. To do this, several parameter values ​​of a particle are taken into account. Then, a diffraction pattern obtained at the detection plane is modeled by numerical modeling. Such modeling can in particular be based on Mie theory. Mie theory is an elastic scattering model making it possible to obtain a solution to Maxwell's equations describing a light wave diffracted by a spherical particle illuminated by a monochromatic incident light wave of wavelength λ.Besides its spherical shape, the particle is characterized by a refractive index. n , the latter being able in particular to be a complex refractive index: n = Re n + jIm n ,

[0076] With j 2<=-1. Re And Im are respectively the real part and imaginary part operators. The particle is also characterized by its size (radius or diameter).

[0077] When using diffraction patterns modeled by Mie theory, it is preferable that the particles analyzed are spherical, or made spherical by the addition of a spherizing agent.

[0078] According to a variant, a progressive adjustment of the parameters of the modeled particle is carried out, in particular the refractive index and the size, so that the modeled diffraction pattern gradually approaches the elementary diffraction pattern corresponding to the particle of interest. The progressive adjustment can be done by iteration, by determining, at each iteration, a difference between the elementary diffraction pattern and a modeled diffraction pattern. The difference can be progressively reduced by implementing a gradient descent type algorithm. This makes it possible to carry out a progressive adjustment of the modeling parameters so as to progressively reduce the difference between the modeled diffraction pattern and the elementary diffraction pattern. The modeling parameters can be: the size (radius or diameter), the refractive index, and possibly the distance between the particle and the detection plane.The more the distance between the particle and the detection plane is known, the more precisely the elementary diffraction pattern is reconstructed.

[0079] On the figure 4 , a curve (curve b) is shown, corresponding to a radial average of the modulus of an elementary diffraction pattern simulated by Mie theory, for a particle with a radius of 2.59 µm and a refractive index of 1.43. This illustrates the possibility of characterizing a particle of interest, from modeled elementary diffraction patterns.

[0080] Furthermore, being able to model an elementary diffraction pattern allows different particles to be modeled, with different parameters, in particular size and refractive index. This also allows for interpolations between elementary diffraction patterns modeled between two different values ​​of the same parameter.

[0081] The classification of an elementary diffraction pattern obtained for a particle of interest can be performed by an SVM (Support Vector Machine) type algorithm, after a learning phase carried out on the basis of modeled diffraction patterns. Other classification algorithms, notably supervised ones, for example by PCA (Principal Component Analysis), can also be applied.

[0082] The use of modeled diffraction patterns can also be considered during characterization by a decomposition according to a polynomial basis.

[0083] The inventors implemented a method as previously described to characterize blood particles, in this case red blood cells. Each sample consisted of blood diluted 1 / 600 in a spherizing reagent. The red blood cell concentration, before dilution, was 2.41 10 6 red blood cells per mm 3 . The average globular volume was 92 µm 3 . The sample extended over a distance range of between 1050 and 1200 µm from the image sensor. Steps 100 to 150, previously described, were applied. 1000 particles of interest were distinguished, each of them corresponding to a red blood cell. A distance was estimated using complex images reconstructed at different distances z i of each particle of interest relative to the detection plane. An elementary diffraction pattern A 0, iwas formed for each particle of interest, taking into account the distance associated with it.

[0084] A radial profile was then determined forming a radial average of each elementary diffraction figure, such as that described in connection with the figure 4 An adjustment algorithm was implemented to progressively determine the parameters of a spherical particle whose radial profile, formed from the diffraction pattern modeled in the detection plane, approaches the radial profile formed from the elementary diffraction pattern.

[0085] The inventors established a relationship between the refractive index at 405 nm and the hemoglobin concentration in a particle. When modeling diffraction patterns, the radius, the refractive index (dependent on the hemoglobin concentration), and the distance from the detection plane were considered.

[0086] There figure 5A represents a distribution of the hemoglobin content of particles (y-axis) as a function of the distance from the detection plane (x-axis). The hemoglobin content is expressed in g / dL.

[0087] There figure 5B represents a distribution of the particle radius (y-axis) as a function of the distance from the detection plane (x-axis). The particle radius is expressed in µm.

[0088] There figure 5C represents a histogram of the number of particles (y-axis) as a function of the distance from the detection plane (x-axis). The distance is expressed in µm.

[0089] The method described above was implemented on different types of blood particles (red blood cells, white blood cells, platelets) placed in a sample of blood purified in Dextran. As previously indicated, Dextran has the effect of causing an aggregation of the red blood cells, which leads to their sedimentation. The supernatant plasma, depleted in red blood cells compared to the initial blood, is recovered. It is then diluted to 1 / 10 in PBS buffer. During this test, the concentration of white blood cells, red blood cells and platelets was respectively 5.1 10 3 per mm, 0.02 10 6 per mm, 198.10 3 per mm. For different particles of interest, elementary diffraction patterns were formed by implementing steps 100 to 150 described above. From each elementary diffraction pattern, a size and a refractive index were estimated, the latter depending on the hemoglobin level.In the case of white blood cells and platelets that do not contain hemoglobin, this concentration has no significance, but it is related to the optical index. The . figure 5D positions each characterized particle of interest in a reference frame whose abscissa axis corresponds to the radius (in µm) and the ordinate axis represents the hemoglobin level. This makes it possible to identify red blood cells (set a), white blood cells (set b) and platelets (set c).

[0090] In another test, the inventors implemented the method as previously described on different samples, comprising different types of white blood cells (leukocytes), divided into four classes: class 1: lymphocytes; class 2: monocytes; class 3: neutrophils; class 4: eosinophils.

[0091] Purified blood samples were prepared and particles corresponding to white blood cells were characterized. In these samples, the red blood cells were lysed, and the blood was diluted to 1 / 10 (dilution in a Whitediff lysis reagent - Horiba commercial kit). The characterization was carried out by implementing steps 100 to 150 described above, each isolated elementary diffraction pattern being obtained by decomposing the modulus and phase images according to Zernike polynomials, up to order n = 4. For each decomposition, 30 coordinates were obtained (15 for the modulus, 15 for the phase). The diameter of each white blood cell was also determined, the latter being estimated from the complex image reconstructed in the plane of the sample, and more particularly from the modulus image.For each class, a database of diffraction patterns obtained experimentally on purified samples (a single type of white blood cell - or a class - largely preponderant in the sample >80%) was available. This database was used for learning the classification. The classification was carried out on the basis of an SVM type algorithm.

[0092] The following table represents the classification results. The classes associated with each row and column correspond to the true classes and the classes determined by measurement, respectively. The classification performance was deemed satisfactory. Table 1 1 2 3 4 1 93% 3% 6% <1% 2 2% 77% 10% 3% 3 4% 14% 80% 8% 4 1% 6% 4% 89%

[0093] There figure 6 schematizes another device allowing implementation of the invention. Unlike the device shown in the figure 1 , the device of the figure 6 comprises an optical system 19 for imaging. The optical system 19 defines an image plane and an object plane. The optical system may be a lens or an objective. When acquiring the image of the sample, the image sensor is arranged in a defocused configuration. The detection plane is offset from the image plane and / or the plane along which the sample extends is offset from the object plane. The offset is generally small, preferably less than 1 mm, and typically in a range of 50 µm - 500 µm.

[0094] The invention may be implemented on different types of samples, for example in the field of diagnosis, biology, or environmental control, or in different industrial sectors, for example agri-food or process control.

Claims

1. A method for characterizing a particle (10p) present in a sample (10), the sample lying between an image sensor (16) and a light source (11), the image sensor lying in a detection plane (P0), the method comprising the following steps: a) illuminating the sample with the light source, the light source emitting an incident light wave (12) that propagates along a propagation axis (Z); b) acquiring an image of the sample (I0) with the image sensor (20), the image comprising a plurality of elementary diffraction patterns, each elementary diffraction pattern corresponding to one particle; c) on the basis of the acquired image (I0), reconstructing a complex image (A10) representative of a complex amplitude of the exposure light wave on at least one reconstruction surface (P10) passing through the sample, this reconstruction being achieved by implementing an iterative reconstruction algorithm, the algorithm comprising, in each iteration, updating a phase of the exposure light wave (14) in the detection plane (P0) or on the reconstruction surface (P10); d) selecting a region of interest (ROIi) of the complex image, the selected region of interest corresponding to a particle of interest (10p,i); e) forming a complex image (A10,i) extracted from the region of interest (ROIi) selected in step d); f) applying a propagation operator (h) to the extracted complex image (A10,i) formed in step e), so as to obtain, in at least one propagation plane, an elementary diffraction pattern (A0,i) of the particle selected in step c), such that the elementary diffraction pattern obtained is isolated from elementary diffraction patterns of other particles present in the sample; g) characterizing the particle of interest (10p,i) depending on the elementary diffraction pattern (A0,i) resulting from step f).

2. The method as claimed in claim 1, wherein, in step f), the propagation plane is the detection plane.

3. The method as claimed in either one of the preceding claims, wherein: • step c) comprises, in each iteration, applying a holographic propagation operator to an image of the exposure light wave (14), in the detection plane, so as to obtain a complex expression for the exposure light wave, in at least one plane in which the sample lies, forming the complex image (A10); • step d) comprises, on the basis of the reconstructed complex image (A10), obtaining an observation image (I10), the particle of interest (10p,i) being selected in the observation image.

4. The method as claimed in any one of the preceding claims, wherein step f) comprises applying a holographic propagation operator to the region of interest (ROIi) selected in step e), so as to obtain the elementary diffraction pattern (A0,i), corresponding to the particle of interest, in the propagation plane, the portion, of the reconstructed complex image (A10), located outside the region of interest (ROIi) not being propagated.

5. The method as claimed in any one of the preceding claims, comprising an estimation of a distance between the particle of interest and the detection plane, such that, in step f), the propagation operator takes into account the distance thus estimated.

6. The method as claimed in any one of claims 1 to 4, wherein, in step f), the propagation is carried out in light of a predetermined propagation distance.

7. The method as claimed in any one of the preceding claims, wherein, in step f), a complex elementary diffraction pattern is obtained, in the reconstruction plane, of the particle selected in step c).

8. The method as claimed in claim 7, wherein, in step g), the characterization is carried out on the basis of the phase or of the modulus of the complex elementary diffraction pattern, or of its real part or of its imaginary part.

9. The method as claimed in any one of the preceding claims, wherein step g) comprises determining an axial profile, representing a radial average of the elementary diffraction pattern.

10. The method as claimed in any one of the preceding claims, wherein step g) comprises: • comparing the elementary diffraction pattern (A0,i), obtained in step f), with at least one diffraction pattern corresponding to a known particle; • or comparing the elementary diffraction pattern (A0,i), obtained in step f), with at least one diffraction pattern modeled in light of a particle the refractive index and / or size of which are known; • or classifying the elementary diffraction pattern (A0,i), obtained in step f), in light of diffraction patterns obtained using known particles, or diffraction patterns modeled in a training phase.

11. The method as claimed in either one of the preceding claims, wherein: • step g) comprises decomposing the elementary diffraction pattern (A0,i) into a basis of Zernike polynomials, so as to obtain coordinates of the elementary diffraction pattern in the basis; • the particle of interest being characterized depending on the coordinates of the elementary detection pattern.

12. The method as claimed in any one of the preceding claims, wherein, in step g), the characterization is carried out depending on modeled diffraction patterns, each diffraction pattern corresponding to one particle a parameter of which is known, the parameter being a size and / or a refractive index, the characterization aiming to determine a value of the parameter for the analyzed particle of interest.

13. The method as claimed in claim 12, comprising iteratively adjusting the parameter of the particle, each iteration comprising: i) taking into account a parameter or a set of parameters of the particle; ii) obtaining a modeled diffraction pattern dependent on the parameter or the set of parameters taken into account; iii) determining a difference between the modeled diffraction pattern and the elementary diffraction pattern resulting from step f); iv) estimating a parameter, or a set of parameters, aiming to decrease the difference thus determined, the parameter or set of parameters estimated being taken into account in the following iteration.

14. The method as claimed in claim 13, wherein, in steps i) to iv), a set of parameters is taken into account, the set of parameters comprising a distance between the particle and the detection plane.

15. A device for observing a sample (10), the sample containing particles, the device comprising: • a light source (11), configured to illuminate the sample; • a sample holder (10s), configured to receive the sample; • an image sensor (16), placed such that when the sample is placed on the holder, it lies between the image sensor and the light source, the image sensor being configured to acquire an image of the sample; • a processor (20), configured to implement steps c) to f) or c) to g) of a method as claimed in any one of the preceding claims, on the basis of an image acquired by the image sensor.

16. The device as claimed in claim 15, wherein no image-forming optic is placed between the image sensor and the sample.

17. The device as claimed in claim 15, comprising an optical system (19) lying between the sample and the image sensor, the optical system defining an image plane and an object plane, the device comprising means for adjusting the optical system, or the sample, or the image sensor, such that: • the detection plane is offset from the image plane; • and / or the object plane is offset from a sample plane in which the sample lies.