Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V20/00—Scenes; Scene-specific elements
  • G06V20/60—Type of objects
  • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
  • G06V20/698—Matching; Classification
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/20—Image preprocessing
  • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/20—Image preprocessing
  • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
  • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
  • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

• the technical field of the invention is the characterization of microorganisms, in particular the characterization of yeasts or bacteria.
• microorganisms such as yeasts or bacteria, or their derivatives
• yeasts are widespread in various sectors, such as bakery, wine production, brewing or even the manufacture of dairy products.
• the application of yeasts or bacteria concerns many foods by means of probiotics, the latter being for example added to cereals or to animal feed.
• many industrial fields can use microorganisms. This is for example agriculture or horticulture, with the development of phytosanitary products or fertilizers more respectful of the environment, or the production of biofuels obtained from plants.
• Other applications relate to the field of pharmacy, medical diagnostics.
• the step of characterizing such microorganisms constitutes an essential link in the production chain.
• Microbiological controls are frequently used, on samples taken from culture media, in order to detect and enumerate living microorganisms. Cultivation on Petri dishes is still widely used, but has certain drawbacks, in particular the preparation, the duration of the analysis and the impossibility of detecting living and non-cultivable microorganisms.
• Document WO2018 / 215337 describes a device and a method making it possible to carry out a classification between living, dead or living but non-cultivable yeasts.
• Document EP3462381 describes a device and a method for detecting microorganisms in a food sample. The inventors propose an alternative method, so as to carry out an enumeration of living microorganisms, and developing, in a sample.
• a first object of the invention is a method for characterizing a sample, comprising microorganisms preferably placed in contact with a medium suitable for their development, the method comprising the following steps: a) illumination of the sample by a light source ; b) using an image sensor, obtaining images of the sample, each image representing the sample at a measurement instant; c) comparison of an image of the sample, at a measurement instant, with a reference image, representing the sample at a reference instant, so as to establish a comparison image, the comparison image being representative a variation of the sample between the measurement instant and the reference instant, the reference instant being chosen from among the measurement instants; d) repeating step c) for different measurement instants, so as to obtain different comparison images, each comparison image possibly being associated with a measurement instant; e) from the comparison images, defining at least one region of interest, each region of interest corresponding to a part of at least one comparison image in which a variation of the sample is detected; f) on several images of the sample, resulting from step b
• Step f) can include forming as many stacks of thumbnails as there are distinct regions of interest defined during step e). During step f), each stack of thumbnails is associated with a region of interest.
• thumbnailnail is understood to mean an image portion, of size smaller than the size of the image.
• Each image has pixels.
• Each thumbnail can have a number of pixels at least 10 times or at least 100 times less than the number of pixels in the image.
• the supervised artificial intelligence algorithm can be a neural network.
• the neural network can be a convolutional neural network.
• step e) comprises:
• the variation image can thus be formed by a combination of comparison images.
• Each pixel of the variation image is established from the value, at said pixel, of a comparison image.
• Step e) can include a determination of coordinates of interest, in the variation image, each coordinate of interest potentially corresponding to a microorganism. Each region of interest is then defined around each coordinate of interest. The shape and size of each region of interest are preferably predetermined.
• the process may comprise, following sub-step e-ii):
• step e-iii) comprises a definition of at least one region of interest around each coordinate of interest determined during sub-step e-ii).
• each thumbnail comprises a single region of interest defined during step e).
• the sample can be placed between the light source and the image sensor.
• no image formation optics are arranged between the sample and the image sensor.
• an optical system extends between the sample and the image sensor, the optical system defining an object plane and an image plane, the method being such that:
• the image sensor defines a detection plane, the detection plane being offset with respect to the image plane; - and / or the sample defines a sample plane, the sample plane being offset from the object plane.
• step b) comprises, at each measurement instant:
• the image of the sample can for example be obtained from the modulus and / or the phase and / or the real part and / or the imaginary part of the complex image obtained during sub-step b -ii).
• the reconstruction plane is a plane along which the sample extends.
• the reconstruction plane is parallel to a detection plane along which the image sensor extends.
• the method may comprise, following step h), a step i) of counting regions of interest considered as comprising microorganisms developing in the sample.
• the method may include training of the supervised artificial intelligence algorithm, the training being carried out using calibration samples comprising developing microorganisms whose location is known.
• the measurement instants are between an initial instant and a final instant.
• the reference instant can correspond to the initial instant or to the final instant.
• the reference instant is an instant before or after each instant of measurement. It may for example be the instant immediately preceding or the instant immediately subsequent to each instant of measurement.
• a second object of the invention is a device for characterizing a sample, the sample comprising microorganisms, the device comprising:
• the device may include characteristics described in connection with the first subject of the invention.
• FIG. 1 is an example of a device allowing an implementation of the invention.
• FIG. 2 shows schematically the main steps of a method implemented by the invention.
• FIGS. 3A to 3L are images of a sample obtained at different successive instants, forming a series of images. These are images obtained by using a device as shown in FIG. 1.
• FIG. 4A shows an example of a region of interest comprising microorganisms growing in the sample.
• FIG. 4B shows intensity profiles of several images of the same series of images of the sample, in the region of interest shown in FIG. 4A.
• FIG. 4C shows intensity profiles of several comparison images, resulting from image comparisons of the series of images described in connection with FIG. 4B.
• FIG. 4D shows a horizontal profile of a variation image established for the series of images described in connection with FIG. 4B.
• each profile is determined along a dotted line drawn in Figure 4A.
• Fig. 5A is a variation image corresponding to the series of images shown in Figs. 3A to 3L.
• Figure 5B is the variation image shown in Figure 5A after detection of regions of interest.
• Figure 6 is a region of interest defined from image 5B.
• FIGS. 7A to 7H are thumbnails extracted from the images of the same series of images, each thumbnail corresponding to the region of interest shown in FIG. 6.
• FIGS. 8A to 8C are examples of stacks of thumbnails forming the input data of an algorithm of the convolutional neural network type.
• FIG. 9 is a diagram of an architecture of a convolutional neural network.
• Figures 10A to 10F show a stack of vignettes representative of the development of microorganisms.
• Figures 10G to 10L show a stack of thumbnails representative of a movement in the sample.
• FIG. 11 is a graph showing the evolution of the performance of the method as a function of the temporal extent of the series of images.
• FIGS. 12A to 12C are a comparison of detection of microorganisms respectively according to a reference method, according to a method based on an analysis of the morphology of the variation image, and according to a method according to the invention.
• FIGS. 12D to 12F are a comparison of detection of microorganisms respectively according to a reference method, according to a method based on an analysis of the morphology of the variation image, and according to a method according to the invention.
• FIGS. 12G to 121 are a comparison of detection of microorganisms respectively according to a reference method, according to a method based on an analysis of the morphology of the variation image, and according to a method according to the invention.
• FIG. 13A is a graph showing, for different samples, a number of microorganisms enumerated respectively with a method based on an analysis of the morphology of the variation image (y-axis) and a reference method (x-axis).
• FIG. 13B is a graph showing, for different samples, a number of microorganisms enumerated respectively with a method according to the invention (y-axis) and a reference method (x-axis).
• FIG. 14 is another example of a device allowing an implementation of the invention.
• FIG. 1 represents an example of a device according to the invention.
• a light source 11 is able to emit a light wave 12, called an incident light wave, propagating in the direction of a sample 10, along a propagation axis Z.
• the light wave is emitted according to a spectral illumination band Dl .
• Sample 10 is a sample that it is desired to characterize. It comprises in particular 10 fc microorganisms. Sample 10 may contain nutrients allowing the development of microorganisms.
• the sample 10 may for example comprise ground food, the latter being intended for food, for example for animal nutrition. It is for example fat, meat, and vegetable fibers, in particular in the form of flour.
• the mixture can also include food supplements, for example vitamins.
• the mixture also contains microorganisms, in particular bacteria or yeasts, used as a food supplement, in the form of probiotics.
• An objective of the invention is to evaluate a quantity of microorganisms developing in the sample.
• the term quantity refers to a number or a concentration.
• microorganism is meant in particular a yeast, a bacterium, a spore, a fungus or a cell, whether it is a eukaryotic or prokaryotic cell, or a microalgae.
• Sample 10 can be solid. It can for example take the form of a powder, obtained by grinding food containing microorganisms.
• the sample can comprise culture medium, suitable for the development of microorganisms. It may be a culture medium which is liquid or takes the form of an agar.
• the sample 10 is, in this example, contained in a chamber 15.
• the chamber 15 may have a thickness e, along the axis of propagation, typically varying between 10 ⁇ m and 5 mm, and is preferably between 20 ⁇ m and 500 pm.
• the sample is maintained on a support 10s at a distance d from an image sensor 16.
• the concentration of microorganisms can vary between 500 per microliter and 5000 per microliter.
• the distance D between the light source 11 and the chamber 15 is preferably greater than 1 cm. It is preferably between 2 and 30 cm.
• the light source, seen by the sample is considered to be point. This means that its diameter (or its diagonal) is preferably less than a tenth, better still a hundredth of the distance between the fluidic chamber 15 and the light source.
• the light source is a light emitting diode. It is generally associated with diaphragm 18, or spatial filter.
• the aperture of the diaphragm is typically between 5 ⁇ m and 1 mm, preferably between 50 ⁇ m and 500 ⁇ m. In this example, the diaphragm is supplied by Thorlabs under the reference P150S and its diameter is 150 ⁇ m.
• 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 FIG. 1 also comprises a diffuser 17. , arranged between the light source 11 and 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 a.
• the scattering angle a varies between 10 ° and 80 °.
• the source light can 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.
• the emission spectral band D1 of the incident light wave 12 has a width of less than 100 nm.
• spectral bandwidth is meant a width at mid-height of said spectral band.
• the sample 10 is placed between the light source 11 and the image sensor 16 previously mentioned.
• the latter preferably extends parallel, or substantially parallel to the plane P 10 along which the sample extends.
• 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.
• the sample extends along an XY plane, perpendicular to the axis of propagation Z.
• the image sensor 16 is able to form an image I 0 of the sample 10 according to a detection plane P 0 .
• a detection plane P 0 is an image sensor comprising a matrix of pixels, of the CCD type or a CMOS.
• the detection plane P 0 preferably extends perpendicularly to the axis of propagation Z of the incident light wave 12.
• the distance d between the sample 10 and the matrix of pixels of the image sensor 16 is preferably between 50 ⁇ m and 2 cm, preferably between 100 ⁇ m and 2 mm.
• the microorganisms 10 fe present in the sample can generate a diffracted wave 13, capable of producing, at the level of the detection plane P 0 , interference, in particular with a part of the incident light wave 12 'transmitted by the sample. Furthermore, the sample can absorb part of the incident light wave 12.
• the light wave 14, transmitted by the sample, and to which the image sensor 16 is exposed designated by the term "wave exposure ", may include: a component 13 resulting from the diffraction of the incident light wave 12 by the microorganisms present in 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 interferences in the detection plane.
• the image acquired by the image sensor includes interference figures (or diffraction figures). This image forms a hologram, which is a signature of the content of the sample, and of its evolution over time.
• the processor is a microprocessor connected to a programmable memory 22 in which is stored a sequence of instructions for perform 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.
• the inventors are based on the fact that under the effect of the development of microorganisms in the sample, forming colonies, the hologram formed on the image sensor changes. This evolution can be interpreted with the naked eye or by means of simple image processing algorithms when the sample is not very dense. However, when the number of microorganisms is high, the interpretation of the images is more difficult and can lead to errors in the counting of the microorganisms or in their location.
• the inventors have designed a method for detecting and counting microorganisms developing in a sample. By develop is meant to multiply, so as to form clusters or colonies. The method is based on the observation of the development of these clusters or of these colonies over time.
• Step 100 Arrangement of the sample on the support.
• the sample is placed on the support 10s, in the field of observation of the image sensor. It is then illuminated by the light source 11.
• Step 110 Obtain a series of images of the sample
• images I oi are acquired successively at different measurement instants t j , extending over an acquisition time range.
• the index i is a strictly positive integer designating the rank of each image in the series of images.
• the acquisition time range can be between 1 h and 20 h, or even more.
• the images I i can be acquired according to a determined acquisition frequency, for example every hour.
• An image I oi acquired by the image sensor 16, also called a hologram, does not make it possible to obtain a sufficiently precise representation of the observed sample. This is due to the lack of magnification optics between the sample and the image sensor.
• a holographic propagation operator h can be applied to each image I oi acquired by the image sensor, so as to calculate a quantity representative of the exposure light wave 14. It is then possible to reconstruct a complex expression of the light wave 14 at any coordinate point (x, y, z) in space, and in particular in a reconstruction plane P z located at a distance
• Ai (x, y, z) / 0, i (x, y, z) * h * denoting the operator product of convolution.
• the function of the propagation operator ha is to describe the propagation of light between the image sensor 16 and a point of coordinates (x, y, z), located at a distance ⁇ z ⁇ from the image sensor. It is then possible to determine the modulus j (x, y, z) and / or the phase f ⁇ c, g, z ) the light wave 14, at a point in the coordinate space (x, y, z ), with :
• ⁇ Pi (x, y, z) arg [Ai (x, y, z) ⁇ .
• the coordinates (x, y) denote a radial position in a radial plane XY perpendicular to the axis of propagation Z.
• the coordinate z denotes a coordinate along the axis of propagation Z.
• the complex expression A t is a complex quantity whose argument and modulus are respectively representative of the phase and intensity of the exposure light wave 14 detected by the image sensor 16 at the measurement instant t j .
• the product of convolution of the image I oi by the propagation operator h makes it possible to obtain a complex image Ai representing a spatial distribution of the complex expression of A (x, y, z) in the reconstruction plane considered.
• the latter is the plane P 10 along which the sample extends.
• the set of images l t successively obtained forms a series of images of the sample, each image obtained being representative of the sample at a measurement instant t.
• Each image l is associated with a measurement instant t, which corresponds to the acquisition instant of the image / 0 (.
• Each image l represents the sample at the instant t j .
• Obtaining a complex image can be accompanied by significant reconstruction noise, usually designated by the term “twin image”. Iterative algorithms have been described, making it possible to obtain, by holographic reconstruction, a complex image A t while minimizing the reconstruction noise.
• Such holographic reconstruction algorithms are for example described in document WO2017162985 (steps 100 to 170, shown diagrammatically in FIG. 2A of WO2017162985) or in document WO2016189257 (steps 100 to 500, diagrammatically in FIG. 4 of WO2016189257).
• FIGS. 3A to 3L represent images / d obtained by a device such as represented in FIG. 1, during an acquisition time period of 11 hours, from an initial instant t t (FIG. 3A). These images were obtained by holographic reconstruction from I oi images acquired every hour. Each acquired image I oi has undergone a holographic reconstruction, as described in WO2017162985, so as to obtain a reconstructed complex image A it in the plane of the sample. Images / j were thus obtained, respectively representative of the sample at each measurement instant t it by determining the modulus of the complex image.
• the sample comprises granules of animal nutrition, crushed, comprising Saccharomyces cerevisiae yeasts in a culture medium of YPD (Yeast Peptone Dextrose) type.
• the experimental parameters were as follows: light source: RGB dial LED Created MC-E Color; image sensor: UI-1492LE-M IDS monochrome CMOS sensor - 3840 x 2748 pixels; distance image sensor - sample: 1 mm; distance light source - sample: 5 cm.
• the microorganisms take the form of spots which gradually darken.
• the development of microorganisms is reflected, in the images of the series of images, in a progressive darkening.
• FIGS. 3A to 3L there is materialized, by a black frame, a region of interest of the image corresponding to the same colony of microorganisms. It is observed that the colony darkens under the effect of the multiplication of microorganisms. Darkening is attributed to increasing scattering and attenuation of light by microorganisms. This feature is used in step 140.
• Step 120 comparison of images.
• the objective of this step is to establish comparison images, each comparison image corresponding to a comparison between two images of the series of images of the sample.
• the comparison can take the form of a subtraction or a ratio.
• a comparison image I CO mp, i is calculated.
• Each comparison image is associated with an instant of measurement t j .
• Each comparison image I comP i represents a variation of the sample between the reference instant t re ⁇ i and the measurement instant t j .
• the comparison takes the form of a subtraction.
• the comparison can take the form of a ratio
• Step 130 Forming a variation image From each comparison image I CO mp, i> determining a variation var I picture, the latter being representative of a sample variation range during the time of acquisition.
• Each image of the sample is defined according to pixels of coordinates (x, y), in the detection plane formed by the image sensor. The same applies to each comparison image homp.i
• an image of variation I var is calculated.
• the value of the variation image I var (x, y) at each pixel (x, y) corresponds to the value of the comparison image I CO mp, i (X y) reflecting maximum darkening among the different images comparison.
• one selects, for each pixel (x, y), the comparison image I comP i whose value I comP i (x, y), for the pixel considered, translates a maximum difference of two images and the / re /,; under the effect of the development of microorganisms.
• the reference image / re, i is associated with a reference instant t re ⁇ i prior to the measurement instant t j associated with the image I u and the image reference is subtracted from each image in sample / j , which corresponds to expressions (1) and (2), for each pixel (x, y):
• the reference image / re / represents the sample at a reference instant t re fi after the instant of measurement of the image / j , and that the reference image is subtracted from each image / j , for each pixel (x, y), the development of microorganisms leads to a brightening of the reference image.
• the reference image / re / represents the sample at a reference instant t re fi after the instant of measurement of the image / j , and that the reference image is subtracted from each image / j , for each pixel (x, y)
• the variation image is established by using, for each pixel (x, y), the comparison image whose value corresponds to an extremum, for the pixel, among the set of comparison images / comp j .
• the extremum reflects an increase in diffusion under the effect of the development of microorganisms.
• the comparison image I CO mp, i is calculated, the extremum is either a maximum or a minimum.
• the variation image I var can therefore be formed from different comparison images I comP i .
• the reference image / re / is the initial image I t of the series of images.
• the variation image is such that:
• variation image is defined for the whole of the series of images resulting from step 110.
• Figure 4A shows a region of interest of one of the images shown in Figures 3A-3L, centered on growing microorganisms.
• region of interest is meant a part of an image of the sample comprising a microorganism or a colony of microorganisms.
• Intensity profiles were performed on the same region of interest, considering different images successively obtained, along the line shown in dotted lines in FIG. 4A.
• the profiles correspond to images obtained respectively 0 hour (see figure 3A - reference), 3 hours (see figure 3D - - reference / 4 ), 7 hours (see figure 3H - reference I 8 ) and 9 hours (see FIG. 3J - - reference I 10 ) after the initial instant t t .
• the initial instant corresponds to FIG. 3A.
• the profiles are shown in Figure 4B. It is observed that the presence of each microorganism results in lower gray levels.
• Comparison images were formed by subtracting the initial image ( Figure 3A) so as to establish comparison images I CO mp, i, hompA > I comp, 8 > homp, io > as defined according to the expression ( 1).
• the profiles of each comparison image are shown in Figure 4C.
• the profile of the comparison image I CO mp, i is equal to 0.
• an image of variation l var according to expression (5) has been formed.
• the profile of the variation image I var is shown in FIG. 4D.
• the variation image l var is essentially formed from the values of the pixels of the comparison images I CO mp, 8 and
• FIG. 5A shows the image of variation I var described in the previous paragraph. Dark traces are observed, some of which correspond to colonies of developing microorganisms.
• the objective of the following steps is to determine, from the traces detected on the image of variation I var , those which correspond to a developing microorganism.
• Step 140 Determination and localization of regions of interest from the variation image
• the objective of this step is to detect regions of interest containing respectively at least one trace detected on the variation image.
• step 140 comprises a binarization of the variation image I var , on the basis of a predetermined threshold.
• Figure 5B shows the binarized l var variation image.
• each dark spot of the variation image likely to correspond to a microorganism or a colony of microorganisms, appears in the form of a light trace, the coordinates of which can be determined, for example the coordinates of the centroid .
• coordinates of interest x 7 , y ;
• Each coordinate of interest can correspond to a microorganism (or to a colony of microorganisms), or to a displacement of a particle in the sample 10.
• each region of interest ROI j Around each coordinate of interest (x ; -, y), we define a region of interest ROI j .
• the size of each region of interest is preferably predefined.
• each region of interest is a square with sides 65 pixels.
• the definition of the dimensions of the regions of interest ROI j can be carried out beforehand, for example on the basis of experimental tests.
• Each region of interest can be defined so that its center corresponds respectively to each coordinate of interest (x, y ; ).
• FIG. 6 represents an example of a region of interest defined around a coordinate of interest, the latter being the centroid of a white trace of the variation image after binarization, the trace being surrounded by a dotted circle on Figure 5B.
• Step 150 Extraction of thumbnails.
• a vignette V j is defined from each image of the sample / j , resulting from step 110.
• Each vignette V t corresponds respectively to a part of the image I l in each region d 'ROI j interest defined during step 140.
• a stack of vignettes V is obtained, each vignette corresponding to an extraction of region d 'ROI j interest in the sample image / d .
• the vignettes Vi formed from different images I u in the same region of interest ROI j , form a stack of vignettes V j .
• FIGS. 7A to 7H are examples of vignettes V tj associated with the same region of interest ROI j respectively extracted from different images. There is a variation in the appearance of the thumbnails between Figure 7A and Figure 7H. Thus, each stack of vignettes V j . is representative of a local variation of the sample, in the same region of interest ROI j .
• Step 160 Classification of each stack of vignettes V j .
• each stack of vignettes is used as input data IN of a convolutional neural network CNN, whose output layer OUT makes it possible to determine whether the stack of vignettes corresponds to the development of a microorganism. or not.
• Figures 8A, 8B and 8C are examples of thumbnail stacks.
• FIG. 9 schematically shows an example of different layers of a convolutional neural network.
• a convolutional neural network is known to those skilled in the art. It is for example described in Karpathy "Large-scale video classification with convolutional neural networks", 2014 IEEE Conference on Computer Vision and Pattern Recognition.
• the convolutional neural network comprises: an input layer IN: this layer is formed by a stack of vignettes V j ; several convolutional layers CONVi ... CONV k .
• Each convolution layer compotes a first level, formed of images obtained by convolution of at least one image of the previous layer by a filter. In this example, the size of the filter is 3 x 3.
• the parameters of each filter are established during a learning phase described below.
• the images forming the first level are then subjected to transformations, which may include:
• ⁇ “pooling”, this involves replacing the values of a group of pixels by a single value, for example the average, or the maximum value, or the minimum value of the group considered.
• a “max pooling” is applied, which corresponds to replacing the values of groups of pixels from 2 by 2 by the maximum value in the group; a linear rectification, usually designated "RELU", making it possible in particular to remove certain values from the images of the first level. For example, negative values can be removed by replacing them with the value 0.
• each component of the vector VECT is a characteristic (or “feature”) of the stack of vignettes V j forming the input IN of the network.
• the VECT vector is used as the input vector of an NN neural network of the “fully connected” type, a term commonly used by those skilled in the art.
• Each of the components of the VECT vector forms a node feeding the neural network NN.
• the neural network NN comprises an output layer OUT directly connected to the vector VECT.
• the output layer has y n nodes. Each node corresponds to a class.
• a value of each node of the output layer is determined by applying an activation function to a linear combination of the values of the nodes of the input layer.
• the output layer OUT comprises two classes: the stack of labels corresponds to a developing microorganism; the stack of stickers does not correspond to a developing microorganism.
• each node y n is such that: where x m is the value of each node of the previous layer (terms of the vector VECT); b m is a bias associated with each node of the preceding layer x m f n is an activation function associated with the node of rank n of the layer considered; w mn is a weighting term for the node of rank m of the preceding layer and the node of rank n of the layer considered.
• each activation function f n is determined by those skilled in the art. It may for example be an activation function f n is a function of hyperbolic or sigmoid tangent type.
• the OUT layer comprises values making it possible to confirm or deny that the presence of microorganisms developing in the region of interest ROI j associated with the stack of vignettes V j .
• the network may include a final END layer, usually designated by the term “Softmax”, comprising as many components as there are classes, each component representing a probability of belonging to each class.
• Step 170 counting
• step 160 the microorganisms developing in the sample are counted, the latter being confirmed during step 160.
• the convolutional neural network has previously been the subject of training (step 90), so as to determine the parameters of the convolutional filters as well as the parameters linked to each node, that is to say the terms x m , b m , f n and w mn defined in connection with expression (7).
• the training is carried out using series of images of known samples.
• the inventors trained using 7000 thumbnail stacks of known class. 80% of the images were used to perform the actual training, while the remaining 20% were used to validate the algorithm.
• FIGS. 10A to 10F correspond to a stack of labels corresponding to the development of a microorganism.
• FIGS. 10G to 10L correspond to a stack of labels corresponding to movement in the sample, in a part of the sample not comprising a microorganism.
• the use of a supervised artificial intelligence algorithm makes it possible to discriminate, among the local variations detected by analyzing the variation image, those which actually correspond to the growth of microorganisms and those which correspond only to a movement effect and not to the development of microorganisms.
• FIG. 11 represents an evolution of the precision of the result supplied by the convolutional neural network (ordinate axis), as a function of the time period during which images are acquired (abscissa axis - unit: hour).
• the precision of the result is 87.8%, that is, 87.8% of the results are correct.
• the time period is 12 hours, the accuracy reaches 98%. Precision is the rate of correctly classified thumbnail stacks.
• the samples contained Saccharomyces cerevisiae yeasts distributed in a food matrix such as animal feed pellets. The values resulting from the application of the method described above were then compared with the reference values.
• step 130 an algorithm based on the formation of a variation image, as described in step 130, was also used.
• the counting of the developing microorganisms is then carried out by filtering and morphological processing operations. of the variation image: Gaussian filtering - Otsu filtering (known to those skilled in the art) - morphological processing (dilation / erosion) - cleaning of regions of interest considered to be aberrant.
• Gaussian filtering - Otsu filtering known to those skilled in the art
• morphological processing diilation / erosion
• FIGS. 12A, 12B and 12C correspond respectively to the application, on a sample, of the manual selection (reference method), of the algorithm based on a morphological analysis, and of the method described in connection with steps 100 to 170. It can be seen that the latter (FIG. 12C) is consistent with the reference method (FIG. 12A). In each image, the microorganisms detected are represented by a white point.
• Figures 12D, 12E and 12F correspond respectively to the application, on another part of the sample, of manual selection (reference method), of the algorithm based on a morphological analysis and of the method described in connection with the steps 100 to 170. It can be seen that the latter (FIG. 12F) is more consistent with the reference method (FIG. 12D).
• FIGS. 12G, 12H and 121 correspond respectively to the application, to another part of the sample, of manual selection (reference method), of the algorithm based on a morphological analysis and of the method described in connection with the steps 100 to 170. It can be seen that the latter (FIG. 121) is consistent with the reference method (FIG. 12G).
• the inventors carried out counts on 104 different samples. These samples included Saccharomyces cerevisiae yeasts arranged in three different media, each medium comprising animal nutrition granules.
• Baseline counts were determined by scanning the sample under a microscope. The reference counts were compared with the morphological analysis algorithm and the algorithm according to steps 100 to 170 previously described. Each algorithm was implemented using 11 images acquired during a time period of 10 hours, with an acquisition rate of 1 image per hour.
• FIG. 13A shows, for each sample, the reference count (x-axis) and the count performed with the morphological analysis algorithm (y-axis).
• the average of the absolute differences between the reference values and the values given by the morphological analysis algorithm is equal to 12.9%.
• FIG. 13B shows, for each sample, the reference count (x-axis) as a function of the count performed with the algorithm implementing the convolutional neural network (y-axis).
• the average of the absolute differences between the reference values and the values given by the algorithm based on neural networks is equal to 9.9%.
• each type of food matrix is respectively identified by a triangle, round or square type symbol.
• an image formation optic is disposed between the sample and the image sensor.
• the device comprises an optical system 19, defining an object plane P 0 b j and an image plane Pi m .
• the image sensor 16 is then arranged in a so-called defocused configuration, according to which the sample extends along a plane offset from the object plane, and / or the image sensor extends along a plane offset from the object plane. at the image level.
• 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 combined with a plane Pio along which s 'expands the sample.
• the offset d is preferably less than 500 ⁇ m, or even less than 200 ⁇ m. It is preferably greater than 10 ⁇ m or 20 ⁇ m.
• the object plane P 0 b j coincides with a plane Pio along which the sample extends and the image plane Pi m is offset with respect to the detection plane Po according to an offset d .
• the method described in connection with steps 100 to 170 is applicable to images acquired according to such a configuration.
• a lensless imaging configuration is preferred, due to the larger field of view that it provides.
• the detection plane coincides with the image plane and the sample plane coincides with the object plane. Each image is thus acquired according to a focused configuration.
• the invention can be implemented, without limitation, in the field of food processing, or in the control of industrial processes, in the control of the environment, or in the field of microbiology or clinical diagnosis. in biology, or in environmental control, or in the field of food processing or industrial process control.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• Multimedia (AREA)
• General Physics & Mathematics (AREA)
• General Health & Medical Sciences (AREA)
• Evolutionary Computation (AREA)
• Health & Medical Sciences (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Life Sciences & Earth Sciences (AREA)
• Biomedical Technology (AREA)
• Artificial Intelligence (AREA)
• Molecular Biology (AREA)
• Computing Systems (AREA)
• Databases & Information Systems (AREA)
• Medical Informatics (AREA)
• Software Systems (AREA)
• Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
• Apparatus Associated With Microorganisms And Enzymes (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=70154730

Family Applications (1)

Country Status (3)

Families Citing this family (1)

Family Cites Families (4)

• 2020
  • 2020-02-03 FR FR2001039A patent/FR3106897B1/fr active Active
• 2021
  • 2021-02-01 EP EP21701817.5A patent/EP4100867A1/de active Pending
  • 2021-02-01 WO PCT/EP2021/052287 patent/WO2021156192A1/fr not_active Ceased

Also Published As

Similar Documents

Legal Events