Method and device for detecting, classifying and quantifying single-cell organisms

Abstract

The invention relates to a device (1) for detecting, classifying and quantifying single-cell organisms, which uses hyperspectral technology to detect, classify and quantify the presence of single-cell organisms. The device comprises at least one frame (8) for attaching components; at least one hyperspectral image acquisition component (9); at least one standardised lighting system (10); and at least one sample-holding carrier (11) with controlled horizontal movement.

Description

[0001] METHOD AND DEVICE FOR DETECTION, CLASSIFICATION AND QUANTIFICATION OF UNICELLULAR ORGANISMS

[0002] DESCRIPTIVE MEMORY

[0003] The present invention relates to a device for detecting, classifying and quantifying unicellular organisms (1) and a method for detecting, classifying and quantifying unicellular organisms (2), intended to detect, classify and quantify with precision and in real time the presence of pathogenic and non-pathogenic unicellular organisms present on the surface of a sponge moistened with peptone water or similar, which has been rubbed on a surface, work material or biological tissue under study.The method applies to a compact size device that allows the scanning and spectral acquisition of a plurality of samples simultaneously in real time, and for each object of study it executes a preprocessing of the spectral data that allows in the next step to start a process of stochastic spectroscopy and active learning that establishes with high precision and in real time indicators of detection, classification and quantification of pathogenic and non-pathogenic unicellular biological material present on the surface of the sponge. The method (2) and device for detection, classification and quantification of unicellular organisms (1) allows to shorten the detection and quantification times compared to standardized analytical techniques and the use of kits that in the best of cases report presence or absence after 24 - 48 hours.As will be explained later, it also offers advantages over hyperspectral imaging (HSI) techniques based on the detection of pathogens using a technique that combines conventional microscopy and illumination and which, at the process level, discriminates hyperspectral identities from spectral signatures at the pixel level in the Fourier domain.

[0004] The technology uses a methodology that in the previous stage includes the development of a data set of each pathogenic and non-pathogenic unicellular biological material that seeks to be detected, in this objective a standardized laboratory methodology is developed, which for each biological material considers the preparation of a plurality of smears that contain known concentrations of the isolated unicellular organism such as Listeria monocytogenes, Staphylococcus aureus, Salmonella spp, coliforms, Escherichia coli or fungi to mention a few, for each isolate a set of samples is obtained by means of a sponge smear that represents the presence, absence and concentration of the target isolate.The surface of each sponge that contains the spectral information of the isolate of each unicellular organism is subjected to scanning and registration of the hyperspectral image, thus forming a data set of digital images that contain the specific hyperspectral data of the pathogenic and non-pathogenic unicellular biological material.

[0005] 1

[0006] SUBSTITUTE SHEET (RULE 26) presence, absence, and concentration have been recorded by smear on the sampling sponge. The specific hyperspectral data contained in the data set is then used to contrast, using stochastic spectroscopy, the hyperspectral data recorded in new samples obtained by sponge smear. It considers the processing, segmentation, and analysis of each image and, when deemed necessary, is integrated with active learning processes controlled in the data processing unit. This process allows the detection, classification, and quantification of the presence of pathogenic and non-pathogenic unicellular organisms present in a smear obtained from a surface, material, or biological tissue to be reported in real time and with high precision.

[0007] The device (1) develops application in the field of food safety, such as meat, vegetable and fruit processing processes, however, it also has application in other areas such as pharmaceutical production where the processes control contamination or presence of pathogenic and non-pathogenic unicellular organisms. The device and the methodology allow, for example, to detect, classify and quantify in real time the presence of Listeria monocytogenes, Staphylococcus aureus, Salmonella spp, coliforms, Escherichia coli or fungi on a biological tissue, material or work surface.These are small organisms shaped like cocci, bacilli, spirals, or filamentous organisms whose presence generates different levels of risk to food safety, process safety, or cross-contamination, among others. They are subject to continuous monitoring on work surfaces, inputs, work materials, raw materials, products in transit, and finished products.

[0008] General practice for pathogen detection, for example, in the food industry, establishes the implementation of a quality assurance program with a sampling and pathogen detection plan using standardized laboratory analytical techniques. These include obtaining samples from each material or surface where a risk of contamination has been identified, or from the product itself—raw materials, products in transit, or finished products—which are then subjected to analytical processes involving microorganism isolation, cultivation, identification, and quantification. For this purpose, in addition to standardized analytical techniques, various solutions for detecting single-cell microorganisms are also identified in the form of microbiological analysis kits. Their response time, like the laboratory analytical route, is typically within 24 hours at best.These techniques use isothermal PCR-type analytical methods or alternative methods that utilize culture broth. Depending on the pathogen, these methods will require 24 to 48 hours of pre-enrichment, and depending on the content of the smear, may require a larger amount of growth broth. Many of these solutions establish a reporting system that defines the presence / absence of the pathogen without providing qualitative or quantitative indicators. As indicated in Pl.

[0009] 2

[0010] SUBSTITUTE SHEET (RULE 26) US20220120674, the state of the art identifies solutions based on analytical chemistry that include PCR, immunoassay, basic microscopy, infrared, Fourier transform (FTIR), and Raman Laser Spectroscopy (RLS). These are methodologies that require executing different analytical steps in the laboratory that pose limitations in terms of response time, cost and labor. The polymerase chain reaction (PCR) is a high-performance technique but requires sample preparation to eliminate inhibitors from the sample that can result in a false negative and add reagents necessary to execute the technique.

[0011] The state of the art also identifies solutions for pathogen detection using hyperspectral imaging (HSI) based on the detection of pathogens by recording an image using a technique that combines conventional optical microscopy and spectroscopy to analyze the image at each pixel or in the X and Y planes of an image circumscribed to an area of ​​interest. In this case, the system used to record the area of ​​interest may involve two approaches: a) a non-static approach in which only one point or line of the sample is scanned at a time or b) a static system, in which the entire area is scanned at once. In the static system, different registration techniques are identified in the image, where each wavelength in the spectral range of interest is scanned at the same time or filters are placed on the sensor to detect the specific wavelength of interest.Therefore, the optical system needs to direct the light exiting the sample towards the detection system. The spectroscopic component of HSI can include spectral ranges from ultraviolet (UV) to infrared (IR), but the most common are visible (VIS) and near-infrared (NIR). CN113065403 is a solution that uses machine learning and performs hyperspectral image analysis as a cell classification method based on the following steps: 1) obtain a three-dimensional cell hyperspectral image S * M * N, where S represents different wavenumbers, each wavenumber corresponds to a two-dimensional image with the size of M * N; 2) preprocess and segment the hyperspectral image to obtain cell image blocks; 3) machine learning to classify cell image patches and obtain cell classification results in a tissue.This solution performs preprocessing of hyperspectral images of cells, including at least region of interest (ROI) extraction and noise filtering, binarization processing, and image morphology processing, and the preprocessing order can be adjusted, including, in some cases, a filter can be used to reduce the noise factor and improve image quality.

[0012] In the field of hyperspectral image processing, a huge variety of techniques are identified in the state of the art, for example, noise reduction techniques, baseline correction and normalization, and feature extraction techniques,

[0013] 3

[0014] SUBSTITUTE SHEET (RULE 26) such as dimensionality reduction methods, including principal component analysis (PCA). For the classification problem, techniques range from the most classical, where the best results seem to have been obtained with super vector machines (SVM), to the most recent, which use active learning and deep learning techniques. Furthermore, since spectral data can show significant variations due to factors such as instrument calibration, sample preparation, or measurement conditions, sophisticated transfer learning techniques have been used, in which models are pre-trained on large spectral data sets and tuned to specific tasks by exploiting prior knowledge. For example, WO2013093913A1 describes a method for the spectroscopic detection and identification of microorganisms in a culture.This method combines the acquisition of spectral images with a learning algorithm in order to accelerate the analysis of a bacterial culture and WO02022254332 which refers to a method and device suitable for the acquisition of hyperspectral images of bacterial colonies, yeast, molding and inorganic particles in a Petri dish.

[0015] In general, conventional strategies designed to discriminate a hyperspectral image require supervised data (at the pixel level) and the generation of models trained with this information. The objective is to discriminate hyperspectral identities from spectral signatures in the Fourier domain at the pixel level. Fundamentally, all the methods presented have one common denominator: they are based on a feature-based discrimination hypothesis. There is no doubt that this feature-based approach has proven to be a very powerful strategy in many areas of active learning and pattern recognition. However, in the challenging context of hyperspectral image analysis and classification, its adoption presents several limitations. In this regard, we can mention the following:

[0016] Existence of spectral signatures

[0017] It is widely accepted that there is a specific link between the object to be detected and certain spectral characteristics, which usually require pixel-by-pixel extraction and processing. This applies to conventional techniques, such as the well-known spectral angle mapper (SAM) and its derivatives such as matched filtering and unmixing. However, difficulties arise when the sensor does not capture the spectral signature of the phenomenon studied. In this case, it is usual to affirm that the phenomenon is not active in the spectra recorded by that sensor. For example, in the case of the study of molecular-scale phenomena, it is common to use the FTIR (Fourier Transform Infrared) spectroscopic technique, which records waves in the infrared range, where characteristics can be distinguished.

[0018] 4

[0019] SUBSTITUTE SHEET (RULE 26) specific spectral characteristics related to bonds and other related vibrational phenomena.

[0020] Spectral signature extraction

[0021] Given the above association, in the implementation or deployment of classification or regression methods, a key requirement is to identify, extract, and analyze the spectral signature linked to the object to be detected. In other words, the focus is on discovering discriminatory features in the Fourier domain. In carefully controlled laboratory environments, this approach might be acceptable. However, in field conditions or in remote sensing applications, it is virtually impossible to control even the most basic parameters (acquisition noise, atmospheric conditions, incident light, circulating dust, ambient and sample humidity, instrument calibration, among many others). Robust feature extraction then becomes a challenge.

[0022] Partial and noisy labeling

[0023] To build effective inference models, data must be labeled for supervised training (classification and regression in machine learning). In the context of hyperspectral imaging, this is a significant limitation, as a single sensor capture can comprise millions of spectral measurements that would require pixel-based (high-resolution) labeling. Therefore, in practice, one must deal with either missing labels (unsupervised or partially supervised) or noisy labels (robust learning).

[0024] The device of the present invention defines a compact device that can be transported in a suitcase and enabled to operate with just the connection to a power supply source. It differs from current solutions due to its compact design characteristics, high precision and rapid detection of unicellular organisms present on the surface of a sponge that has been moistened with peptone water or other similar material and that has been rubbed on the surface of biological tissue, work surfaces and materials as appropriate.The device for detecting, classifying and quantifying unicellular organisms (1) is made up of at least one component holding frame (8); at least one hyperspectral image acquisition component (9); at least one standardized lighting system (10); at least one object carrying carriage with controlled horizontal movement (11) for obtaining the sample on the surface of a sampling set (12); it also has a method for detecting, classifying and quantifying unicellular organisms (2) that in terms of image processing and stochastic spectroscopy has: a) software for.

[0025] 5

[0026] SUBSTITUTE SHEET (RULE 26) acquisition and classification of hyperspectral image data; b) implementation of training, detection, classification, and quantification algorithms; and c) a dataset of the spectral signature of each target family or species of unicellular organism; and d) a digital platform for visualizing the results.

[0027] In terms of image processing, the device of the present invention utilizes spectroscopy, a scientific technique used to study and analyze the properties of light or electromagnetic emission and reflection. It provides valuable information about the composition, structure, and physical properties of various objects, from celestial bodies such as stars and galaxies to microscopic laboratory samples. By examining spectral patterns, spectroscopy provides a wealth of information about their chemical composition, temperature, density, velocity, and other characteristics. It has applications in a wide range of disciplines, including astronomy, chemistry, physics, environmental sciences, and materials science, and plays a pivotal role in advancing our understanding of the universe and the world around us.Imaging spectroscopy represents an important step in increasing the complexity of the wide range of existing remote sensing, field, and laboratory spectroscopic techniques, as it multiplies the amount of available data by incorporating the spatial dimension. Specifically in the field of reflectance spectroscopy, with the advent of multispectral and hyperspectral imaging, the main challenge is the handling of large amounts of data and the subsequent inference from them, which includes data quality assurance, to address weak and / or noisy measurements, high data dimensionality (the curse of dimensionality), and, most especially, the high cost of annotating and labeling them to make them suitable for supervised methods.

[0028] In the field of image processing, the device develops stochastic imaging spectroscopy processes that offer a radically different perspective for hyperspectral image classification and estimation from this type of data. This approach models the imaging process in a manner consistent with the hidden nature of the phenomenon under study (a generative approach). This is a model-based strategy that does not rely on the description of spectral features.

[0029] In this area, the device operates a hierarchical Bayesian generative image model (or directed graphical model) in which several layers of abstraction and information are introduced. Unlike traditional data-driven methods, this approach does not rely solely on strong discriminatory assumptions at the pixel level (or spectral signatures). On the contrary, this Bayesian model naturally produces the

[0030] 6

[0031] SUBSTITUTE SHEET (RULE 26) description of each hyperspectral image, at a higher semantic level of abstraction that we call stochastic hyperspectral signatures. This new model-based representation offers a high level of interpretability of the image formation process. Furthermore, training this type of model does not require expensive annotated data (information annotated at the pixel level), and is robust to occlusion, missing spectral observations, and poor spectral detection conditions.

[0032] A common problem when dealing with hyperspectral imaging is the high dimensionality of the data. To address this issue, solutions can be found that reduce dimensionality. For example, in the case of HSI data, nearby bands generally have a high level of correlation, and the acquisition process includes integration phenomena in both the spatial and spectral domains. In this case, the solution proposes a robust analysis approach that considers the identification of features with less dimensionality than the input data, while preserving the structure of the spectral and spatial content. In this case, managing the process with algorithms that reduce the data in real time and instantly encode the energy of the original spectrum.

[0033] The implementation of this methodology has been validated in the context of the estimation of geometallurgical variables on mineral samples mainly from base metal mining, and has then been tested with samples from the concrete and construction industries (aggregates, cement, clays), the steel industry (coke, iron), and recently with pathogen samples from the food industry. This is a software library that implements a processing pipeline consistent with the stochastic-generative approach described above. The process begins with the training of an estimation model based on a set of hyperspectral images, each associated with an image-level label or annotation related to the variable of interest to be estimated. Once trained, the model allows the variable to be estimated in new samples, based on new hyperspectral images acquired on them.The estimation process is rapid—minutes—and achieves accuracy and precision similar to the test or criterion used to define the values ​​associated with the images used for training.

[0034] The present invention relates to a method and device intended to improve the accuracy and response times required for the detection of pathogenic and non-pathogenic cellular biological material present in a sample compared to the standardized laboratory technique or by means of kits, and in particular with respect to detection techniques using PCR, immunoassay, basic microscopy, infrared, Fourier transform (FTIR), Raman laser spectroscopy (RLS) and hyperspectral imaging (HSI). It is a compact device

[0035] 7

[0036] SUBSTITUTE SHEET (RULE 26) that executes the acquisition of hyperspectral images of pathogenic and non-pathogenic unicellular biological material present on the surface of a sampling sponge. The image is subjected to a numerical segmentation process and simultaneously optimized Stochastic Spectroscopy and active learning processes are executed, which together establish the device's capacity to detect, classify and quantify with high precision and in real time the pathogenic and non-pathogenic unicellular biological material present on the surface of the sponge.

[0037] DESCRIPTION OF THE FIGURES

[0038] 1. Figure 1, Method for detection, classification and quantification of unicellular organisms (2), in the stage of development of a new unicellular organism detection model and its training with active learning.

[0039] 2. Figure 2, Method for detecting, classifying and quantifying unicellular organisms (2), in the stage of applying a unicellular organism detection model.

[0040] 3. Figure 3, general view of the device for detection, classification and quantification of unicellular organisms (1).

[0041] 4. Figure 4, Exploded view of components of the device for detecting, classifying and quantifying unicellular organisms (1).

[0042] 5. Figure 5, Description of object carrier cart (11).

[0043] EXPLANATORY MEMORY

[0044] The present invention relates to a method and device for detecting, classifying and quantifying unicellular organisms (1) that uses hyperspectral technology to detect, classify and quantify in real time the presence of pathogenic and non-pathogenic unicellular organisms present on the surface of a sponge containing a sample obtained from a surface, work material or biological tissue by smearing with a sterile sponge that has previously been moistened with peptone water or other similar liquid suitable for preserving the sample.

[0045] The method for the detection, classification and quantification of unicellular organisms (2), specifically used in the development stage of the new unicellular organism detection model and its training with active learning (Figure 1), first includes obtaining a data set for an initial characterization (3) of the new unicellular organism objective, it begins with the definition of the pathogenic or non-pathogenic unicellular organism (3A) that is sought to be analyzed and the specific strain to be detected. Preferably, a stable and pure amount of the isolate (3B) of said organism should be obtained by means of a laboratory microbiological technique, which will be used to prepare the solutions of

[0046] 8

[0047] SUBSTITUTE SHEET (RULE 26) inoculation (3C) to be used for the preparation of the data set (3). To improve the accuracy of the detection, classification and quantification model, a large number of sample images are required. Preferably, the sample set will be obtained using standardized analytical methodologies for each unicellular organism, using a growth medium for each isolate, such as Half Fraser Broth or another specific medium where the inoculation solutions will be maintained. With this material, different target isolates are prepared, including previously defined concentration or dilution levels with their respective replicates.Once the inoculation solutions and replicas (3C) have been prepared, they are preferably required to be inoculated on the study surface where they are expected to fix or stabilize, from which a sample is then obtained by sponge smear (3D) of each inoculum solution using a sampling sponge (14). By sampling sponge (14) we mean all the materials or elements for sampling surfaces such as dry sponges, cellulose sponge, sterile free of biocides that are pre-hydrated with some type of solution or medium for sample collection and allows maintaining the viability of microorganisms, such as peptone water, physiological serum or other. With the sponge, by smearing on the surface inoculated with the unicellular organism of known concentration, at least one set of samples is obtained for image acquisition and training (3E).More preferably, an equivalent number of blank samples free of organisms should be provided. To corroborate the content of each sample, they should be analyzed using standardized laboratory techniques, the characteristics of which (3F) will be applied in the sample labeling process in the next step.

[0048] Method (2) in the development stage of the new unicellular organism detection model and its training with active learning, also considers a variant in the data collection methodology, focusing on obtaining a data set for model training (3'). This time, once the unicellular organism (3A) has been defined, it will be necessary to obtain the isolate (3B') of said organism, in order to prepare the inoculation solutions (3C') that will be used for the preparation of the training data set (3'). In this case, to train the detection, classification and quantification model, a large number of samples are required. Preferably, the set of samples will be obtained through standardized analytical methodologies that include several types of growth medium such as Half Fraser Broth or another specific one to prepare the inoculation solutions for each unicellular organism.The target isolate is prepared in different concentrations, and each of these is inoculated onto various non-biological surfaces, such as stainless steel, plastic bands, or ceramics; or biological surfaces, such as fish, beef, chicken, sausages, fruit, etc. Non-biological surfaces can also be sterile or contaminated with other pathogenic or non-pathogenic organisms, sanitizers, water containing food residue, grease, blood, etc.

[0049] 9

[0050] SUBSTITUTE SHEET (RULE 26) In this way, a sample set (3E') will be created that is as varied as possible, emulating the real conditions where the invention operates. More preferably, the sample set should include as many types of pathogenic or non-pathogenic unicellular organisms as possible, and to further increase the precision of the method, other strains of the target organism whose biochemical profile is as similar as possible can also be included; for example, if the target variable is Listeria monocytogenes, the set of isolates could include Listeria innocua, Listeria welshimeri or another. Even more preferably, an equivalent number of blank samples without the presence of organisms should be provided. Once the inoculation solutions and replicas (3C') have been prepared, it is preferably required to wait for them to fix or stabilize on the inoculated surface, then a sample of each solution is obtained (3D') using a sampling sponge (14).Finally, at least one set of sponge samples (14) is obtained for image acquisition and training (3E'). To corroborate the content of each sample, once the hyperspectral image has been recorded, they can be analyzed using standardized laboratory techniques whose characteristics (3F) will be applied in the sample labeling of the next stage.

[0051] For both the option of obtaining a data set for an initial characterization (3) or a data set for model training (3'), the next step will be to obtain a hyperspectral image (4): for this purpose, the location of each sponge (4A) that makes up the set for an initial characterization (3) or data set for training (3') in the sample tray (15) is required. Preferably, the hyperspectral image acquisition component (9) is calibrated (4B) considering that it can focus on the entire surface of the sample tray (15), and the image capture parameters are set, considering adjusting the shutter opening, integration time (frames per second), tray speed and lighting. Preferably, the acquisition equipment (9) may have the capacity to record spectral layers of 400-1000 / 400-770 nm, at a scanning speed of 0.1 mm / s - 99 mm / s.In parallel, the samples are labeled (4C) by assigning a metadata to each sponge with a specific code. This metadata may contain information about the origin of the sample, location in the tray, date the sample was obtained, dilution, type of inoculation surface, results of laboratory analysis, among others; This labeling is done by means of a device software (1) installed on a computer; All the characteristics of the samples identified with the labeling are stored in a database (4E). Finally, the hyperspectral image of the samples is acquired by scanning (4D) of the entire set of samples (3E), thereby obtaining an irradiance image of the entire set of samples.

[0052] Image preprocessing (5): During preprocessing the irradiance image obtained is processed using the "Flat Field" correction algorithm (5A).

[0053] 10

[0054] SUBSTITUTE SHEET (RULE 26) Correction" that subtracts the spectrum of the incident light and the effect of thermal noise from the image sensor to generate what is called a reflectance image (5B), which only contains the spectral response of each sample with a certain degree of normalization sufficient for the method. The reflectance images (5 B) are cropped to generate a file for each sample (5C), this set of images is stored in a database of preprocessed samples (5D) where the smear characterization data (3F) and sample labeling (4C) are incorporated, this database of processed samples (5D) will be used for model training (6), each processed image is stored in the processor (13) with a format, hdf5 or similar.

[0055] Model training (6): This uses training algorithms that model the imaging process in a way that is consistent with the hidden nature of the phenomenon under study (a generative approach). This is a model-based strategy, which does not rely on spectral feature descriptions. The device operates a hierarchical Bayesian generative image modeling (or directed graphical model) in which several layers of abstraction and information are introduced. Unlike traditional data-driven methods, this approach does not rely solely on strong discriminatory assumptions at the pixel level (or spectral signatures). Instead, this Bayesian model naturally produces the description of each hyperspectral image, at a higher semantic level of abstraction that we call stochastic hyperspectral signatures.This new model-based representation offers a high level of interpretability of the imaging process. Furthermore, training this type of model does not require expensive annotated data (information annotated at the pixel level) and is robust to occlusion, missing spectral observations, and poor spectral detection conditions. It can preferably incorporate detection-classification models (6B), where the model is trained on samples, assigning a value of 1 to those with some concentration of the target organism and a 0 to those that were, by construction, sterile-impregnated or with a different content. The images are used to train a hierarchical classification scheme. This scheme is structured in a stage that detects pixel populations for a new image to be classified and applies a classification model to each population.Thus, at the end of the process, each image has an experimental distribution of the binomial type of values ​​for the response variable and the maximum likelihood estimator is applied, which is counting in this case. Preferably, in addition, the training model can include a quantification model (6C), where it is trained with the initial inoculum characteristics (3F). In each case, the model is trained with the samples labeled with the concentration values ​​obtained in the laboratory for each sample (3F). Finally, model validation tests are run (6C) that could incorporate statistical analysis; thus obtaining a validated detection, classification, quantification estimator model (7).

[0056] 11

[0057] SUBSTITUTE SHEET (RULE 26) Where said model (7) develops stochastic image spectroscopy processes whose model is able to deliver the result of detection, classification and quantification of unicellular organisms. Once trained, the model allows estimating the variable in new samples, from new hyperspectral images acquired on them. The estimation process is fast with a response in minutes and achieves an accuracy and precision similar to those of the test or criterion with which the values ​​associated with the images used for training were defined.

[0058] Figure 2 details the method of detection, classification and quantification of unicellular organisms (2), specifically when the detection of pathogenic or non-pathogenic unicellular organism is sought, whose model has already been established, said method includes, when the need is identified, the use of an active learning algorithm to detect, classify and quantify the target organism, requires the application of a more limited methodology considering obtaining a data set for sampling (3") which begins with obtaining a sponge smear (3D") directly from the surface to be studied and does not require going through the previous process of inoculation preparation. Then the set of smear samples is prepared for image acquisition and real-time detection (3E") to obtain hyperspectral images (4) and image preprocessing (5) where it is required to execute the same process defined in Figure 1.In this case, the methodology applies after preprocessing (5) the execution of stochastic spectroscopy processes of images of detection, classification, quantification estimator (7), from which it may or may not be possible to identify data or atypical values ​​(outlier), or simply deliver the results of detection, classification and quantification of unicellular organisms. In the case of identifying outlier, the algorithm must report the requirement to enter a new variable for active learning in model training (6), and at the same time it will be necessary to analyze the atypical sample with standardized laboratory techniques, to provide the characteristics or parameters of the atypical variable.

[0059] Figure 3 corresponds to a general view of the device for detection, classification and quantification of unicellular organisms (1) that uses the methodology for detection, classification and quantification of unicellular organisms (2) for both models with new organisms and models with established organisms. The device (1) of compact size, is made up of at least one component support frame (8), it is a fixed structure in whose upper part at least one hyperspectral image acquisition component (9) is located. The frame (8) also allows the support of at least one standardized lighting system (10), preferably halogen; and in the lower part of the frame (8) on a horizontal axis, at least one object carrying carriage with controlled horizontal movement (11) is structured, which allows the location of a sampling set (12) that projects on a perpendicular axis with respect to the lens of the image acquisition component.

[0060] 12

[0061] SUBSTITUTE SHEET (RULE 26) hyperspectral image (9). The device (1) also has a built-in processor (13) that stores the data set generated with the hyperspectral image capture system, as described in point (4) of the method for detecting, classifying and quantifying unicellular organisms (2), and said processor (13) in terms of image processing and stochastic spectroscopy operates: a) software for acquiring and sorting the hyperspectral image data; b) implementation of training, detection, classification and quantification algorithms; c) data set of the spectral signature of each family or species of target unicellular organism; and d) a digital platform in suite for viewing the results.

[0062] The object carrying carriage (11), Figure 5, is formed by at least one base for moving the sample carrying tray (16); and at least one sample carrying tray (15). Where said at least one tray (15) allows the location of a plurality of sampling sponges (14) that make up the sampling set (12). The number of sampling sponges (14) is defined based on the size of the sample carrying tray (15), where said tray (15) can contain a series of reliefs or grooves that allow each sponge (14) to be adjusted.

[0063] 13

[0064] SUBSTITUTE SHEET (RULE 26)

Claims

CLAIMS 1 - Device for detecting, classifying and quantifying unicellular organisms (1) that uses hyperspectral technology to detect, classify and quantify in real time the presence of pathogenic and non-pathogenic unicellular organisms present on the surface of a sponge that has been rubbed on a surface, material or biological tissue under study, CHARACTERIZED in that it has at least one component holding frame (8); at least one hyperspectral image acquisition component (9); at least one standardized lighting system (10); and at least one object carrying carriage with controlled horizontal movement (11). 2.- Method for the detection, classification and quantification of unicellular organisms (2) intended to detect, classify and quantify accurately and in real time the presence of pathogenic and non-pathogenic unicellular organisms CHARACTERIZED in that it comprises a model for the detection of a new unicellular organism and its training with active learning; or in a model for the detection of a trained unicellular organism. 3.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 2 CHARACTERIZED because the method of detection, classification and quantification of unicellular organisms (2) comprises the achievement of the following steps to detect, classify and quantify unicellular organisms: 1) Obtain data set for initial characterization (3), obtain data set for training (3') or obtain data set for sampling (3"); 2) Obtain hyperspectral image (4); 3) Preprocess the images (5); 4) Model training (6); 5) Apply estimator model for detection, classification, quantification (7). 4.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 3 CHARACTERIZED because obtaining a data set for initial characterization (3), considers a model with a new unicellular organism and includes: 1) Define unicellular organism (3A); 2) Obtain target isolate (3B); 4) Inoculate with known concentration and replicates (3C); 5) Obtain sponge smears (3D); 6) Obtain sample set for image acquisition and initial training (3E); 7) Characterize smears by standardized laboratory analysis (3F). 5.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 4 CHARACTERIZED because obtaining a data set for initial characterization (3) includes inoculating with known concentration and replicas (3C) including concentration, dilution and replica levels. 6.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 5 3 CHARACTERIZED because obtaining a data set for initial characterization (3'), trains the detection model and includes: 1) Obtain target isolate (3B'); 2) Inoculate and generate replicas (3C'); 3) Obtain sponge smears (3D'); 4) Obtain sample set for image acquisition and training (3E'). 7.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 96 CHARACTERIZED because obtaining a data set for training (3') based on the unicellular organism or pathogen includes different types of growth medium to inoculate and generate replicas (3C'), also includes different concentrations and inoculation on different surfaces, work material or biological tissue; each with at least one replica. 8.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 3 CHARACTERIZED because obtaining a data set for sampling preferably considers real operating conditions, includes 1) Obtain sponge smears (3D); 2) Create a set of smear samples for image acquisition and initial training (3E"). 9.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 93 CHARACTERIZED because obtaining hyperspectral image (4) includes: 1) Place each sponge (4A) in the sample tray (8); 2) Calibrate (4B) the hyperspectral image acquisition component (3); 3) Label samples (4C); 4) Acquire the hyperspectral image of samples by scanning (4D); 5) Get Database (4E). 10.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 3 CHARACTERIZED because preprocessing the image (5) includes: 1) Process images using correction algorithm (5A); 2) Generate reflectance image (5B); 3) Crop image to generate a file for each sample (5C); 4) Obtain Database of preprocessed samples (5D). 11.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 3 CHARACTERIZED because training the model (6) includes: 1) Run a detection-classification model (6A); 2) Run a quantification model (6B); 3) Validate the model (6C). 12.- Method for detecting, classifying and quantifying unicellular organisms (2), according to claim N 9 2 CHARACTERIZED because when there are atypical values in the image, the method (2) can consider training the model (6) and at the same time register a new characteristic of the organism in the training model. 13.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 1 CHARACTERIZED because in the upper section of the al at least one component holding frame (8) is located the at least one hyperspectral image acquisition component (9). 14.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 1 CHARACTERIZED in that the at least one component holding frame (8) allows the holding of at least one standardized lighting system (10). 15.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 1 CHARACTERIZED in that at least one horizontally moving object-carrying carriage (11) is structured in the lower section of the at least one component holding frame (8). 16.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 915 CHARACTERIZED in that the at least one object-carrying carriage (11) is formed by at least one sample-carrying tray movement base (16); and at least one sample-carrying tray (15). 17.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 16 CHARACTERIZED in that the at least one sample-holding tray (15) allows the location of a plurality of sampling sponges (14) that make up the sampling set (12). 18.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 17 CHARACTERIZED because the number of sampling sponges (14) is defined based on the size of the sample tray (15), where said tray (15) can contain a series of reliefs or grooves that allow each sponge (14) to be adjusted. 19.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 1 CHARACTERIZED the device (1) has at least one processor component (13). 20.- Device for detecting, classifying and quantifying unicellular organisms (1), according to claim N 9 17 CHARACTERIZED because the sampling sponges (14) have a sterile characteristic, free of biocides that prior to sampling are prehydrated with some type of solution or medium that allows maintaining the viability of the microorganisms

Images