A method for automated, scalable, high-throughput detection and analysis of objects in biological samples, and a system for implementing the same

Abstract

A method and a system relates to automated, scalable, high-throughput detection of objects in biological samples and comprises acquiring several computer-readable images of biological samples, each of which comprises several objects (S100); identifying the biological sample images one after the other and creating an identifier for each biological sample image (S200); transmitting the biological sample images comprising at least an identifier, creating a detection task associated with each biological sample image and arrange the detection tasks in a list one after the other (S300); processing the biological sample images comprising at least an identifier by detecting the objects on a single biological sample image one at a time in order to create an object image for each detected object comprising at least the identifier of the given biological sample image and position coordinates (S400); transmitting the object images with at least identifiers and position coordinates to a learning management unit, creating a learning task associated with each object image, organizing the learning tasks in a list one after the other (S500); processing, by at least one teaching unit that includes an initial learning model, the object images comprising at least an identifier and position coordinates, by learning the encoding of a single object image one at a time to thereby create a current learning model by improving the initial learning model (S600); transmit ting the current learning model to a learning model database by overwriting the initial learning model and storing the current learning model (S700); transmitting the object images comprising at least an identifier and position coordinates, creating one coding task associated with each object image, organizing the coding tasks one after the other into a list (S800); using the initial or current learning model stored in the learning model database, processing the object images by encoding a single object image at the same time, in order to thereby create an object code that is equipped with at least the identifier of the given object image (S900); transmitting the object code with at least an identifier and store it in an object code database (S1000).

Description

[0001] A method for automated, scalable, high-throughput detection and analysis of objects in biological samples, and a system for implementing the same

[0002] Description

[0003] The invention relates to a method for automated, scalable, high-throughput detection and analysis of objects in biological samples, as well as a system for implementing the method according to the invention.

[0004] Biological samples, such as tissue samples or cell cultures, can comprise a multitude of cells, which can be up to millions of cells per biological sample. For this reason, the fast, automatic and adequate quality processing of computer-readable images of biological samples is a challenge, in particular the analysisgoal of identifying not all, but only acertain number of single cells and / or cell compartments out of millions of cells, and of isolating them in an appropriate manner and of clustering and analyzing according to certain cell characteristics. Until the advent of single-cell image processing technologies, the information about cells could only be given as averaged values for the cell population in the images taken of biological samples with a microscope, and it was not possible to isolate and analyze single cells separately. Sngle-cell imaging technologiesexist today, but they are partially or entirely based on human intervention, primarily for the purpose of correcting cell detection errors during image processing. Fully automatic image processing technologies that also enable the detection of single cells are capable of processing a large number of images quickly, automatically, without manual intervention, however, they are typically more imprecise and, due to quality deterioration, single cells and / or cell compartments cannot be identified, or cannot be properly identified, which may affect the analysisof biological samples based on the cellular characteristics.

[0005] WO 2020168284 A1 describes a system and method for digital pathology purposes, where the method examines an image that presents the entirety of a microscope slide, which is provided with a registration mark that corresponds to a coordinate system. During the method, feature highlighting is performed on the entire slide image, on the basis of which relevant regions are determined, which are provided with coordinates, where the relevant region can be an area of one cell or smaller. After that, according to a given field of view, an image is taken with a microscope at a position corresponding to the coordinates of the relevant region, where a pathological number is assigned to the given field of view image with the help of a classification model, which indicates the probability of the presence of a given disease. The given field of view image can be displayed with a pathology number on a display device. Although the disclosed solution is capable of automatic identification of single cells and cell compartments, the detection is not performed directly on the digital image of the entire slide, but on microscopic images of a given field of view identified based on the coordinates of the slide image. The solution is not suitable for improving the detection of agiven single cell and / or cell compartment identified in the entire microscope slide image or in the microscopic image.

[0006] US 1 1 ,462,032 describes a method for processing digitized tissue samples, primarily human dermatopathology samples, where a computer-readable image of a tissue sample is processed with a convolutional neural network that is trained with image pairs where the first image of the given image pair is limited to the brightness axis of a color space, the second image of the given pair of images is limited to the first or second color axis of a color space, where the output of the processing correspondsto the pathological classof the tissue sample. Although the method is suitable for the automatic processing and classification of digitized tissue samples, even the sample according to the entire digitized slide, the classification is based on the visual similarity of the tissue samples and single cells or cell compartments are not identified, separated and no analysis of the sample is carried out on this basis.

[0007] WO 2022233916 A1 describes a method and system for analyzing histopathological samples, where the images of tissue samples are analyzed using a machine learning model, such that an encoding module uses a single-cell image as input to create the feature vector of the single-cell image, and a similarity module applies a pair of feature vectors as input to a pair of single-cell I mages to generate asout put a score that indicates the similarity between the single-cell images. An advantage of the method is that images of similar single cells and images of different single cells can be used to train the machine learning model without having to label them for use in each training data set, which speeds up and simplifies the process. Furthermore, image analysis according to the method can be effective even without human intervention, but it is not suitable for using the extracted results of the analyzed image data for the development of the machine learning model and thereby improving the quality of image processing.

[0008] None of the above solutions enables the automatic, real-time, scalable detection of objects in computer-readable images depicting biological samples, i.e. single cells or cell compartments, by using the outputs comprising each detected object to continuously improve the quality of image processing, so that the development of the learning model database and the automatic processing of computer-readable images depicting a biological sample could be carried out in parallel. The goal of the solution according to the invention is to implement a method and system that enables the scalable, automatic processing of computer-readable images of biological samples comprising objects, preferably single cells and / or cell compartments, in such a way that , based on the information extracted during the image processing, the image processing ensures - parallel to the image processing - the real-time and continuous improvement of the quality of the image and, where appropriate, can thereby contribute to a more accurate analysis of cell characteristics. Furthermore, the solution according to the invention enables the achievement of the above goal in a retrievable and identifiable manner on the one hand, according to biological samples, and, on the other hand, according to single cells and / or cell compartments.

[0009] The goals set by the invention are achieved by providing a method serving for the automated, scalable, high-throughput detection of objects in biological samples, comprising the steps of

[0010] S100 acquiring several computer-readable images of biological samples, each of which comprises several objects;

[0011] S200 identifying, by a detection coordination unit, the images of biological samples one after the other, creating an identifier for each biological sample image; and further comprising following steps:

[0012] S300 transmitting the biological sample images comprising at least an identifier to a detection management unit , creating respective detection tasks associated with the respective biological sample imagesand arranging the detect ion tasks in a list one aft er the other by the detection management unit ;

[0013] S400 processing, by a detector unit, the biological sample images comprising at least an identifier by detecting the objects on a single biological sample image one at a time in order to create an object image for each detected object , which is associated with at least the identifier and position coordinates of the given biological sample image;

[0014] S500 transmitting the object images comprising at least an identifier and position coordinates to a learning management unit , creating, by the learning management unit, respective learning tasks associated with the respective individual object images, arranging the learning tasks in a list one after the other;

[0015] S600 processing, by at least one teaching unit that includes an initial learning model, the object images comprising at least an identifier and position coordinates, by learning the encoding of a single object image one at a time to thereby create a current learning model by improving the initial learning model;

[0016] S700 transmitting the current learning model to a learning model database comprising an initial learning model, overwriting the initial learning model and storing the current learning model;

[0017] S800 after a step of processing the given biological sample image comprising at least an identifier, transmitting the object images comprising at least an identifier and position coordinatesto acoding management unit, creating a coding task associated with each object image and arranging the coding tasks in a list one after the other;

[0018] S900 processing, by at least one coding unit, the object images, using the initial or current learning model stored in the learning model database, by encoding a single object image at a time to thereby generate an object code with at least the identifier of that object image; and

[0019] S1000 transferring the object code with at least an identifier to an object code database and storing the object code with at least an identifier in the object code database.

[0020] According to a preferred embodiment of the method, metadata are provided to the biological sample images, where during the step of identifying the acquired biological sample images and creating an identifier for each biological sample image and assigning the identifier to the respective biological sample images, the biological sample images are identified one after the other, the metadata is read for each biological sample image and an identifier iscreated for the given biological sample image, and during the subsequent procedural stepsthe given biological sample image, the given object image and the given object code are provided with both metadata and an identifier.

[0021] According to a further preferred embodiment of the method, during the step of identifying the acquired biological sample images and creating an identifier for each biological sample image and assigning the identifier to the respective biological sample images, determining, based on the metadata and / or identifier, if the biological sample image has already been read.

[0022] According to a further preferred embodiment of the method, the step of creating the multiple computer-readable images comprisesthe steps:

[0023] - acquiring several biological samples and storing each biological sample in a sample holder; - obtaining, by an image recording unit, at least one computer-readable image of at least one part of each biological sample and transmitting the computer-readable images to an image storage unit.

[0024] According to a further preferred embodiment of the method, the step of creating the multiple computer-readable images further comprises providing metadata for the sample holder comprising the given biological sample, readingout the metadatafor the given biological sample from the sample holder, transmitting the metadata to a data storage unit, associating the given metadata with the given computer-readable image.

[0025] Computer-readable images and metadata are preferably stored in a single storage unit.

[0026] According to a further preferred embodiment of the method, the step of creating the multiple computer-readable imagesfurther comprisesobtaining acomputer-readable image of the given entire biological sample with the imaging unit.

[0027] According to a further preferred embodiment of the method, additional data associated with the given biological sample istransmitted to the data storage unit.

[0028] According to a further preferred embodiment of the method, the step of creating and lining up respective detection tasks associated with the respective biological sample images further comprises lining up the detection tasks in the list corresponding to order of transmission of the biological sample images to the detection management unit.

[0029] According to a further preferred embodiment of the method, the step of creating and lining up respective detection tasks associated with the respective biological sample images further comprises recognizing, by the detection management unit, at least based on the identifier, if a detection task has already been created for the given biological sample image.

[0030] According to a further preferred embodiment of the method, the step of processing the given biological sample image comprising at least an identifier further comprises:

[0031] - preprocessing the biological sample image comprising at least an identifier;

[0032] - segmenting the preprocessed biological sample image, thereby detecting individual objects;

[0033] - creating an object image for each detected object, which isprovided with at least the identifier and position coordinatesof the given biological sample image; and

[0034] - classifying the object images comprising at least an identifier and position coordinates. According to a further preferred embodiment of the method, the step of creating and lining up learning tasks in the list comprises lining up the learning tasks corresponding to the order of transmission of the object imagesto the learning management unit.

[0035] According to a further preferred embodiment of the method, the step of creating and lining up learning tasks in the list comprises recognizing, by the learning management unit, at least based on the identifier, if a learning task has already been created based on the given object image.

[0036] According to a further preferred embodiment of the method, the step of processing, by at least one teaching unit, the single object image comprising at least an identifier and position coordinates further comprises preprocessing the object image comprising at least an identifier and position coordinates; and learning the coding of the preprocessed object image by creating an actual learning model.

[0037] According to a further preferred embodiment of the method, if there is more than one learning task in the list, after performing a step of processing, by at least one teaching unit that includes an initial learning model, the object images comprising at least an identifier and position coordinates, by learning the encoding of a single object image one at atime to thereby create a current learning model by improving the initial learning model, and performing a step of transmitting the current learning model to a learning model database comprising an initial learning model, overwriting the initial learning model and storing the current learning model:

[0038] - during the following step of processing, processing, by at least one teaching unit, the next object images comprising at least identifiers and position coordinates in such a way that the coding of a single object image is learned one at a time, in order to create a developed current learning model by improving the current learning model according to the previous processing step, and

[0039] - during the next transmitting and storing step, transmitting the developed current learning model to the learning model database by overwriting the learning model stored according to the previoustransmitting and storing step, and storing the developed current learning model, where both steps are executed as many times as there are learning tasks in the list.

[0040] According to a further preferred embodiment of the method, the step of creating respective coding tasks associated with the respective object images further comprises lining up the coding tasks in the list corresponding to order of transmission of the object images to the coding management unit. According to a further preferred embodiment of the method, the step of creating respective coding tasks associated with the respective object images further comprises recognizing, by the coding management unit, at least based on the identifier, if a coding task has already been created based on the given object image.

[0041] According to a further preferred embodiment of the method, the step of processing, by at least one encoding unit, the given object image comprising at least an identifier and position coordinates further comprises preprocessing the object image comprising at least an identifier and position coordinates; performing feature-highlighting on the preprocessed object image; and using the current learning model in the learning model database, coding the preprocessed and feature-highlighted object image to create an object code that is provided with at least the identifier of the given object image.

[0042] According to a further preferred embodiment of the method, after the step of processing, by at least one detector unit, the given biological sample image comprising at least an identifier, the method further comprises performing at least one of the steps S500 - S700 and at least one of the steps S800 - S1000 simultaneously and in parallel.

[0043] According to a further preferred embodiment of the method, the step of processing the object images by encoding a single object image one at a time further comprises:

[0044] - after processing each encoding task in the given list with at least one encoding unit, querying the initial or current learning model stored in the learning model database for processing the encoding tasks in the next list, or

[0045] - querying the initial or current learning model stored in the learning model database periodically, at predetermined intervals during the processing of the coding tasks in the given list with at least one encoding unit to process the remaining tasks in the given list.

[0046] According to a further preferred embodiment of the method, several detector units, several teaching units and / or several coding units are used simultaneously and in parallel.

[0047] According to afurther preferred embodiment the method further comprises a step of analyzing, by at least one analysis unit, the data stored in the object code database after the step of transferring the object code with at least an identifier to an object code database and storing the object code with at least an identifier in the object code database.

[0048] According to a further preferred embodiment of the method, the objects in the biological samples are single cells and / or cell compartments. The computer-readable images are practically 2-dimensional (2D) or 3-dimensional (3D) digital images.

[0049] The goals set by the invention are also achieved by providing a system for implementing the method according to the invention, where the system comprises:

[0050] - a detection coordination unit for handling the computer-readable image of biological samples and data related to biological samples; where the system also includes:

[0051] - a detection management unit, which isconnected to the detection coordination unit;

[0052] - at least one detector unit which isconnected to the detection management unit;

[0053] - a learning management unit which is connected to the at least one detector unit;

[0054] - at least one teaching unit connected to the learning management unit;

[0055] - a learning model database that is connected to at least one teaching unit;

[0056] - a coding management unit which isconnected to the at least one detector unit;

[0057] - at least one encoding unit, which is connected to the encoding management unit and the learning model database;

[0058] - an object code database that isconnected to at least one coding unit.

[0059] According to a preferred embodiment, the system further comprises:

[0060] - an image recording unit, which comprises at least one sample receiving unit for receiving a sample holder comprising a biological sample, and an image and data recording unit for creating a computer-readable image of at least a part of the biological sample; and

[0061] - an image storage unit connected to the image recording unit for storing computer-readable images.

[0062] According to a further preferred embodiment, the sample holder is provided with metadata, where an image and data recording unit issuitable for reading out the metadata, and the system includes a data storage unit that is connected to the image recording unit for storing data, preferably metadata, and isconnected to the detection coordination unit.

[0063] According to a further preferred embodiment, the image storage unit and the data storage unit are an integrated storage unit. According to a further preferred embodiment, the image storage unit and / or the data storage unit is connected to at least one external device for receiving data and / or computer-readable images.

[0064] According to a further preferred embodiment, the system comprises at least one analysis unit that is connected to the object code database.

[0065] According to a further preferred embodiment , the system includes several detector units, several teaching units and / or several coding units.

[0066] According to a further preferred embodiment, the system includes several analysis units.

[0067] According to a further preferred embodiment, the biological samples comprise objects that are single cells and / or cell compartments.

[0068] According to a further preferred embodiment , computer-readable images are 2-dimensional (2D) or 3-dimensional (3D) digital images.

[0069] According to a further preferred embodiment , the imaging unit isa microscope capable of taking a digital image.

[0070] Hereinafter, the advantageous exemplary embodimentsof the invention are described with the help of the attached drawing, in which

[0071] Figure 1 shows a flowchart of a preferred embodiment of the method according to the invention, and a

[0072] Figure 2 schematically shows a preferred embodiment of a system according to the invention.

[0073] The present invention is defined by the appended claims. The present description describes together with the drawing certain aspects and characteristics of some advantageous embodiments, however, they cannot be considered as limiting the scope of the protection.

[0074] Figure 1 shows a flowchart of a method according to the invention, where the method is used for the automated, scalable, high-throughput detection of objects in biological samples 1 . A biological sample 1 is a biological sample 1 stored in a specific sample container, such as a plate, which can contain up to several million cells. By the objects in the given biological sample 1 , a single-cell and / or a cell compartment is meant . High-throughput detection (high-content screening (HCS) / high-throughput screening (HT)) means a method that can be used to analyze thousands of biological samples 1 per day, while scalability means that the method according to the invention is suit able for increasing the throughput , the specific implementation of which will be covered during the detailed description of the method.

[0075] The method shown includes the following steps:

[0076] In step S100, several computer-readable imagesof biological samples 1 are taken, each of which comprises several objects. Computer-readable images can be 2-dimensional (2D) or 3- dimensional (3D) digital images, and the objects are preferably single cells and / or cell compartments.

[0077] In step S200, a detection coordinating unit 5 is taken, with which the biological sample images are identified one after the other and an identifier is created for each biological sample image. The identifier is essentially an identification code or symbol that is suitable for ensuring the direct traceability of the given biological sample image during subsequent procedural steps.

[0078] In step S200, the biological sample images can be provided with metadata, where the metadata can provide additional information about the given biological sample, for example, it can comprise the location, method, time of sampling, identification of a patient and other information that can be used in later analysis steps, and the metadata is also can be used to ensure the traceability of the given biological sample image. In the case shown, in step S200, the biological sample imagescan be identified one after the other, the metadatacan be read out for each biological sample image and an identifier can be created for the given biological sample image, and preferably during the subsequent process steps, the given biological sample image, the given object image and the given object code can be provided with both metadata and identifier. In step S200, the detection coordinating unit 5 can recognize, based on the metadata and / or identifier, if the biological sample image has already been read out, thus avoiding repeated reading of the given biological sample image.

[0079] In step S300 a detection management unit 6 is taken, to which the biological sample images comprising at least an identifier, preferably an identifier and metadata are transmitted, where with the detection management unit 6 a detection task associated with each biological sample image is created and the detection tasks are arranged in a list in a row one after the other. In step S300, the order of the detection tasks arranged in the list may correspond to the order of transmission of the biological sample images to the detection management unit 6. Preferably, in step S300, the detection management unit 6 can recognize, at least based on the identifier, preferably based on the identifier and / or the metadata, if a detection task has already been created for the given biological sample image. In step S400, at least one detector unit 7 is taken, which is used to process the biological sample images provided with at least an identifier, preferably an identifier and metadata, by simultaneously detect ing objects on a single biological sample image in order to create an object image for each detected object, which isthe given biological sample isprovided with at least the image's identifier, preferably its identifier and metadata, and position coordinates. If only one detector unit 7 is available, the detection tasks can be processed in the order of the detection tasks arranged in the list in step S300. If several detector units 7 are used, the processing can take place regardless of the order of the detection tasks arranged in the list , preferably in such a way that the processing of the detection tasks is continuous with all detector units 7 operating in parallel. The position coordinate enables the exact position of the given object image, which essentially comprises a single object, i.e. single cell or cell compartment, to be traced on the given biological sample image and thus even for the given biological sample 1 . This essentially means that both the biological sample image and the object on the biological sample image can be traced during the entire method.

[0080] In step S400, the given biological sample image comprising at least an identifier, preferably an identifier and metadata, can be processed with at least one detector unit 7 as follows:

[0081] - 9ep S401 , preprocessing the biological sample image comprising at least an identifier;

[0082] - 9ep S402, segmenting the preprocessed biological sample image, thereby detecting the individual objects;

[0083] - 9 ep S403, creating an object image for each object detected which is provided with at least the identifier, preferably its identifier and metadata, and position coordinates of the given biological sample image; and

[0084] - 9ep S404, classifying the object images comprising at least an identifier and a position coordinate.

[0085] In step S500, a learning management unit 8 is taken, to which the object images comprising at least an identifier and position coordinates are transmitted, where by the learning management unit 8 associated with each object image, a learning task is created and the learning tasks are organized in a list one after the other. In step S500, the order of the learning tasks arranged in the list may correspond to the order of transmission of the object images to the learning management unit 8. Furthermore, preferably in step S500, the learning management unit 8 can recognize, at least based on the identifier, preferably the identifier, metadata and / or position coordinates, if a learning task has already been created, based on the given object image, thus avoiding repetitions.

[0086] In step 9600, at least one learning unit 9 istaken, which includes an initial learning model, where the at least one learning unit 9 is used to process the object imageswith at least identifier and position coordinates by learning the coding of a single object image one at a time, so that through the development of the initial learning model a current learning model will be created. The initial learning model Is a model that can be considered a kind of starting model, which can be developed by the teaching unit 9 based on learning how to encode the given object image, and the efficiency of the model can be enhanced. If only one teaching unit 9 is available, the learning taskscan be processed in the order of the learning tasks arranged in a list in step S500. If several teaching units9 are used, the processing can take place regardlessof the order of the learning tasks arranged in the list, preferably in such a way that the processing of the learning tasks iscontinuouswith all teaching units 9 operating in parallel.

[0087] In step 9600, the given object image provided with at least an identifier, preferably an identifier and metadata, and position coordinates can be processed with at least one teaching unit 9 as follows:

[0088] - 9tep 9601 , preprocessing the object image provided with at least an identifier, preferably an identifier and metadata, and position coordinates; and

[0089] - 9tep 9602, learning the encoding of the preprocessed object image by creating an actual learning model.

[0090] In step 9700 a learning model database 12 comprising an initial learning model istaken and the current learning model isentered into the learning model database 12 by overwriting the initial learning model and storing the current learning model. The learning model database 12 comprises an initial learning model so that step 9900, to be performed by the coding unit 11 to be presented later, can be performed independently of steps 9600 and 9700, thereby ensuring the parallel execution of steps 9500-9700 and 9800-91000, i.e., increasing the scalability of the system.

[0091] If there are more than one learning task in the list, after performing steps 9600 and 9700:

[0092] - during performing the following step 9600, processing the following object imageswith at least identifiers and position coordinates by the at least one teaching unit 9 in such away as to learn the coding of a single object image one at a time, in order to create an improved, developed current learning model through the development of the learning model of the previous step

[0093] S600, and

[0094] - during the next S700 step, transferring the developed current learning model to the learning model database 12 by overwriting the stored learning model and storing the developed current learning model, where steps 9600 and S700 are executed as many times as there are learning tasks in the list.

[0095] If only one teaching unit 9 is available, the learning tasks can be processed in the order of the learning tasks arranged in a list in step S500. If several teaching units 9 are used, the processing can take place regardless of the order of the learning tasks arranged in the list , preferably in such a way that the processing of the learning tasks is continuous with all teaching units 9 operating in parallel.

[0096] Based on the data processing so far, the developed current learning model can be considered the most effective learning model.

[0097] After step S400, step S800 is performed, during which a coding management unit 10 is taken, to which the object images provided with at least identifiers, preferably identifiers and metadata and position coordinates are transmitted, where respective coding tasks associated with the respective object images are created by the coding management unit 10, and the coding tasks are entered in a list one after the other.

[0098] In step S800, the order of the coding tasks arranged in the list may correspond to the order of transmission of the object images to the coding management unit 10. Furthermore, preferably in step S800, the coding management unit 10 isable to recognize, at least based on the identifier, preferably the identifier, metadata and / or position coordinates, if acoding task has al ready been created based on the given object image.

[0099] In step S900, at least one coding unit 1 1 is taken, with the help of which, using the initial or current learning model stored in the learning model database 12, the object images are processed by encoding a single object image one at a time, thereby creating an object code that is associated with at least the identifier of the given object image, preferably it is provided with its metadata and position coordinates. If only a single coding unit 1 1 is available, the coding tasks can be processed in the order of the coding tasks arranged in a list in step S800. In the case of using several coding units 1 1 , the processing can take place regardlessof the order of the coding tasks arranged in the list, preferably in such a way that the processing of the coding tasks is continuouswith all coding units 1 1 operating in parallel.

[0100] In step S900, the given object image comprising at least an identifier, preferably an identifier and metadata and position coordinates, can be processed with at least one coding unit 1 1 as follows:

[0101] - Step S901 , preprocessing the object image comprising at least an identifier, preferably an identifier and metadata, and position coordinates;

[0102] - Step S902, performing feature highlighting on the preprocessed object image; and

[0103] - Step S903, using the current learning model in the learning model database 12, coding the preprocessed and feature-highlighted object image to create an object code that is provided with at least the identifier of the given object image, preferably provided with an identifier, metadata and position coordinates.

[0104] In step S1000, an object code database 13 is taken, to which the object code with at least an identifier, preferably an identifier, metadata and position coordinates, is transferred and stored therein.

[0105] Preferably, after step S400, at least one of steps S500-S700 and at least one of stepsS800-S1000 are performed simultaneously and in parallel. Particularly preferably, following step S400, steps S500-S700 and steps S800-S1000 are executed in parallel. The parallel, simultaneous execution of individual method steps increases the scalability of the method.

[0106] Furthermore, in step S900, it is possible to perform the following steps:

[0107] - Step S904, after processing all encoding tasks in the given list with at least one encoding unit 1 1 , querying the initial or current learning model stored in the database of the learning model 12 for processing the next encoding tasks in the list, or

[0108] - Step S905, during the processing of the coding tasks in the given list with at least one coding unit 1 1 , periodically querying the initial or current learning model stored in the database of the learning model 12 at predetermined intervals to process the remaining tasks in the given list.

[0109] With the coding unit 1 1 , it is possible to query the initial or current learning model stored in the learning model database 12 at any point in the processing of the coding tasks in the list, periodically at predetermined intervals or based on separate instructions. The advantage of this is that it is not necessary to wait for the processing of all the coding tasks in the list, but by querying the given learning model stored in the learning model database12, the quality of processing can also be improved when processing the coding tasks in the list being processed.

[0110] During the method according to the invention, several detector units 7, several teaching units 9 and / or several coding units 1 1 can be used simultaneously and in parallel. This essentially means that several computer-readable biological sample images can be processed simultaneously during the method, thus increasing the scalability of the system .

[0111] Preferably, after step SI 000, in step S1 100, at least one analysis unit 14 is taken, with which the data stored in the object code database 13 is analyzed. S1 100 is not shown in Figure 1 . In step S1 100, several data analysis units 14 can be used depending on what type of data analysis is performed on the data set , for example a search for similar cells can be performed on the individual biological samples 1 . At least one analysis unit 14 can also be suitable for personalized analysis with detailed analysis limited to the given biological sample 1 , so for example it can provide support in establishing a medical diagnosisor developing atreatment plan. With at least one analysis unit 14, a global analysis of all cell types extracted during the method can be performed, which can be effectively used, for example, in the design of drug molecules or in biomedical applications that use single-cell analysis data.

[0112] Optionally, multiple computer-readable images of the step S100 may be generated as follows:

[0113] - Step 960, taking several biological samples 1 , where each biological sample 1 is stored in a sample container, where the sample containers are preferably provided with metadata;

[0114] - Step S70, taking an image recording unit 2, with which we take at least one computer-readable image of at least one part of each biological sample 1 , preferably of the entire biological sample 1 , transmitting the computer-readable images to an image storage unit 3, and

[0115] - if the sample holders are equipped with metadata, in step S80, reading by the image recording unit 2 the metadata belonging to each biological sample 1 from the sample holders and transmitting the metadata to a data storage unit 4, where the individual metadata is associated with the image that can be read by the given computer, and

[0116] - optionally, in step S90, transmitting additional data associated with the given biological sample 1 to the data storage unit 4, which provide additional information about the given biological sample 1 .

[0117] Steps S70-S90 can also be executed in parallel, thereby increasing the scalability of the process.

[0118] Steps 960-S90 are not depicted in Figure 1 . Figure 2 schematically shows a preferred embodiment of a system according to the invention. The system according to Figure 2 is also suitable for implementing optional procedural steps of the method according to the invention.

[0119] The system according to the invention is suitable for carrying out the method according to the invention, where the shown embodiment of the system comprises:

[0120] - An image recording unit 2, which comprises at least one sample receiving unit for receiving a sample holder comprising biological sample 1 , where the sample holder is preferably equipped with metadata, such asa bar code, and an image and data recording unit for making acomputer- readable image of at least part of the biological sample 1 , and if the sample holder is equipped with metadata, to read the metadata. The image recording unit 2 is preferably a microscope suitable for creating a digital image.

[0121] - An image storage unit 3 connected to the image recording unit 2 for storing computer readable images.

[0122] - A data storage unit 4, which is connected to the image recording unit 2 for storing data, preferably metadata. The image storage unit 3 and the data storage unit 4 can also be trained as integrated storage units. The image storage unit 3 and / or the data storage unit 4 can be connected to at least one external device 20 for receiving data and / or computer-readable images.

[0123] Computer-readable images can be 2-dimensional (2D) or 3-dimensional (3D) digital images.

[0124] The system also includes:

[0125] - a detect ion coordination unit 5 for handling the computer-readable image of biological samples 1 and data related to biological samples 1 ,

[0126] - a det ection management unit 6, which is connected to the detection coordination unit 5;

[0127] - at least one detector unit 7, which is connected to the detection management unit 6;

[0128] - a learning management unit 8, which is connected to at least one detector unit 7;

[0129] - at least one teaching unit 9, which is connected to the learning management unit 8;

[0130] - a learning model database 12, which is connected to at least one teaching unit 9,

[0131] - a coding management unit 10, which is connected to at least one detector unit 7; - at least one coding unit 1 1 , which is connected to the coding management unit 10 and the learning model database 12;

[0132] - an object code database 13, which is connected to at least one coding unit 1 1 .

[0133] The system may also include at least one analysis unit 14 which is connected to the object code database 13.

[0134] According to the invention, the system may comprise several detector units 7, several teaching units 9, several coding units 1 1 and / or several analysis units 14, which enables a large amount of data to be analyzed quickly and efficiently in such a way that the efficiency of the learning model improves parallel to the increase in the amount of data, and thus the quality of analysis as well.

[0135] The biological samples 1 handled by the system comprise objects that can be single cells and / or cell compartments.

[0136] The system according to Figure 2 is suitable for carrying out the method according to the invention, which was presented in detail in the description of Figure 1 , so we will not repeat it .

[0137] The advantage of the method and system according to the invention isthat it can automatically process a large amount of data quickly and efficiently in such a way that it can develop the learning model database in parallel with the data analysis, so that more and more efficient data analysis can continuously be obtained.

[0138] The method and system according to the invention can be scaled, which is achieved on the one hand by the fact that the at least one detector unit 7 performs a single detection task one at a time, the at least one teaching unit 9 performs a single learning task one at a time and the at least one coding unit 1 1 performs a single coding task one at a time. On the other hand, the scalability of the system is also increased if several detector units 7, teaching units 9 and coding units 1 1 are simultaneously used. Also, the parallel execution of individual procedural steps, i.e. the parallel execution of steps S500-S700 and steps S800-S1000, also increases scalability. Finally, if appropriate, multiple analysis units 14 can increase scalability, asmultiple analyzes can be performed simultaneously on the same biological sample and / or multiple parallel analyzes of the same type can be performed on multiple biological samples 1 .

[0139] A further advantage of the solution according to the invention is that individual objects can be traced back and retrieved on the biological sample images and even on the biological samples 1 by the aid of the position coordinates used. List of reference signs

[0140] 1 biological sample

[0141] 2 image recording unit

[0142] 3 image storage unit

[0143] 4 data storage unit

[0144] 5 detection coordination unit

[0145] 6 detection management unit

[0146] 7 detector unit

[0147] 8 learning management unit

[0148] 9 teaching unit

[0149] 10 coding management unit

[0150] 11 coding unit

[0151] 12 learning model database

[0152] 13 object code database

[0153] 14 analysis unit

[0154] 20 external device

[0155] 960 step

[0156] S70 step

[0157] S80 step

[0158] S90 step

[0159] S100 step

[0160] S200 step

[0161] S300 step

[0162] 5400 step

[0163] 5401 step

[0164] 5402 step

[0165] 5403 step

[0166] 5404 step

[0167] S500 step

[0168] 9600 step

[0169] 9601 step

[0170] 9602 step

[0171] 9700 step 800 step

[0172] 5900 step

[0173] 5901 step

[0174] 5902 step S903 step

[0175] 5904 step

[0176] 5905 step

[0177] S1000 step

[0178] S1 100 step

Claims

Claims1. A method for the automated, scalable, high-throughput detection of objects in biological samples (1 ), comprising the steps of- acquiring several computer-readable imagesof biological samples (1 ), each of which comprises several objects (SI 00);- identifying, by a detection coordination unit (5), images of the obtained biological samples (1 ) one after the other, creating an identifier for each biological sample image (S200); the method comprises the further steps:S300 transmitting the biological sample images comprising at least an identifier to a detection management unit (6), creating respective detection tasks associated with the respective biological sample images and arranging the detection tasks in a list one after the other by the detection management unit (6);S400 processing, by a detector unit (7), the biological sample images comprising at least an identifier by detecting the objects on a single biological sample image one at a time in order to create an object image for each detected object, which is associated with at least the identifier and position coordinates of the given biological sample image;S500 transmitting the object images comprising at least an identifier and position coordinatesto a learning management unit, creating, by the learning management unit, respective learning tasks associated with the respective individual object images, arranging the learning tasks in a list one after the other;9600 processing, by at least one teaching unit that includes an initial learning model, the object images comprising at least an identifier and position coordinates, by learning the encoding of a single object image one at a time to thereby create a current learning model by improving the initial learning model;S700 transmitting the current learning model to a learning model database comprising an initial learning model, overwriting the initial learning model and storing the current learning model;S800 after a step of processing the given biological sample image comprising at least an identifier, transmitting the object images comprising at least an identifier andposition coordinates to a coding management unit, creating a coding task associated with each object image and arranging the coding tasks in a list one after the other;S900 processing, by at least one coding unit, the object images, using the initial or current learning model stored in the learning model database, by encoding a single object image at a time to thereby generate an object code with at least the identifier of that object image; andS1000 transferring the object code with at least an identifier to an object code database and storing the object code with at least an identifier in the object code database.

2. The method according to claim 1 , characterized in that the step (S200) of identifying, by a detection coordination unit (5), imagesof theobtained biological samples(1 ) one after the other and creating an identifier for each biological sample image, further comprises providing metadata to the biological sample images, during the step of identifying the acquired biological sample images and creating an identifier for each biological sample image and assigning the identifier to the respective biological sample images, identifying the biological sample images one after the other, reading the metadata for each biological sample image, creating an identifier for the given biological sample image, then providing the given biological sample image, the given object image and the given object code with both metadata and an identifier.

3. The method according to claim 1 or 2, characterized in that the step (S200) of identifying, by a detection coordination unit (5), images of the obtained biological samples (1 ) one after the other and creating an identifier for each biological sample image, further comprises determining, by a detection coordination unit (5), based on the metadata and / or identifier, if the biological sample image has already been read.

4. The method according to any one of claims 1 -3, characterized in that the step (SI 00) of acquiring computer-readable imagesof biological samples (1 ) further comprises:- acquiring several biological samples (1 ) and storing each biological sample (1 ) in a sample holder (S60);- obtaining, by an image recording unit (2), at least one computer-readable image of at least one part of each biological sample (1 ) and transmitting the computer-readable images to an image storage unit (3) (S70).

5. The method according to claim 4, characterized in that the step of creating the multiple computer-readable images further comprises: providing metadata for the sample holder comprising the given biological sample, reading out the metadata for the given biological sample (1 ) from the sample holder, transmitting the metadata to a data storage unit (4), associating the given metadata with the given computer- readable image (S80).

6. The method according to claim 4 or 5, characterized in that the computer-readable images and the metadata are stored in a single storage unit.

7. The method according to any one of the claims 4-6, characterized in that the step (S70) of creating the multiple computer-readable imagesfurther comprises: acquiring a computer-readable image of the given entire biological sample (1 ) with the image imaging unit (2).

8. The method according to any one of claims 5-7, characterized in that the step (S90) of acquiring computer-readable images of biological samples (1 ) further comprises: transmitting additional data associated with the given biological sample (1 ) to the data storage unit (4).

9. The method according to any one of the preceding claims, characterized in that the step (S300) of creating and lining up respective detection tasks associated with the respective biological sample images further comprises: lining up the detection tasks in the list corresponding to order of transmission of the biological sample imagesto the detection management unit (6).

10. The method according to any one of the preceding claims, characterized in that the step (S300) of creating and lining up respective detection tasks associated with the respective biological sample images further comprises: recognizing, by the detection management unit (6), at least based on the identifier, if adetection task has already been created for the given biological sample image.

11. The method according to any one of the preceding claims, characterized in that the step (9400) of processing, by at least one detector unit (7), the given biological sample image comprising at least an identifier further comprises:- preprocessing the biological sample image comprising at least an identifier (S401 );- segmenting the preprocessed biological sample image, thereby detecting individual objects (S402);- creating an object image for each detected object, which is provided with at least the identifier and position coordinates of the given biological sample image (S403); and- classifying the object images comprising at least an identifier and position coordinates (9404).

12. The method according to any one of the previous claims, characterized in that the step (S500) of creating and lining up learning tasks in the list comprises lining up the learning tasks corresponding to the order of transmission of the object images to the learning management unit (8).

13. The method according to any one of the previous claims, characterized in that the step (S500) of creating and lining up learning tasks in the list comprises recognizing, by the learning management unit (8), at least based on the identifier, if a learning task has already been created based on the given object image.

14. The method according to any one of the preceding claims, characterized in that the step(9600) of processing, by at least one teaching unit (9), the single object image comprising at least an identifier and position coordinates further comprises:- preprocessing the object image comprising at least an identifier and position coordinates(9601 ); and- learning the coding of the preprocessed object image by creating an actual learning model(9602).

15. The method according to any one of the preceding claims, characterized in that if there are more than one learning tasks in the list, after performing a step 9600 and a step 9700,- during the next step 9600, processing, by at least one teaching unit (9), the next object images comprising at least identifiers and position coordinates in such a way that the coding of a single object image is learned one at a time, in order to create a developed current learning model by improving the current learning model according to the previous step 9600, and- during the next step S700, transmitting the developed current learning model to the learning model database (12) by overwriting the learning model stored according to the previous step S700, and storing the developed current learning model, where steps 9600 and S700 are executed as many times as there are learning tasks in the list.

16. The method according to any one of the preceding claims, characterized in that the step (S800) of creating respective coding tasks associated with the respective object images further comprises: lining up the coding tasks in the list corresponding to order of transmission of the object images to the coding management unit (10).

17. The method according to any one of the preceding claims, characterized in that the step (S800) of creating respective coding tasks associated with the respective object images further comprises: recognizing, by the coding management unit (10), at least based on the identifier, if a coding task has already been created based on the given object image.

18. The method according to any one of the preceding claims, characterized in that the step (S900) of processing, by at least one encoding unit (11 ), the given object image comprising at least an identifier and position coordinates further comprises:- preprocessing the object image comprising at least an identifier and position coordinates (9901 );- performing feature-highlighting on the preprocessed object image (9902); and- using the current learning model in the learning model database (12), coding the preprocessed and feature-highlighted object image to create an object code that is provided with at least the identifier of the given object image (9903).

19. The method according to any one of the preceding claims, characterized in that after step (9400) of processing, by at least one detector unit (7), the given biological sample image comprising at least an identifier, the method further comprises: performing at least one of the steps 9500 - 9700 and at least one of the steps 9800 - 91000 simultaneously and in parallel.

20. The method according to any one of claims 1 -19, characterized in that the step (9900) of processing the object images by encoding a single object image one at a time further comprises:- after processing each encoding task in the given list with at least one encoding unit (11 ), querying the initial or current learning model stored in the learning model database (12) for processing the encoding tasks in the next list (S904), or- querying the initial or current learning model stored in the learning model database (12) periodically, at predetermined intervals during the processing of the coding tasks in the given list with at least one encoding unit (11 ) to process the remaining tasks in the given list (S905).

21. The method according to any one of the preceding claims, characterized in that several detector units (7), several teaching units (9) and / or several coding units (11 ) are used simultaneously and in parallel.

22. The method according to any one of the previous claims, characterized in that the method further comprises a step (S1100) of analyzing, by at least one analysis unit (14), the data stored in the object code database (13) after the step (S1000) of transferring the object code with at least an identifier to an object code database and storing the object code with at least an identifier in the object code database.

23. The method according to any one of the preceding claims, characterized in that the objects in the biological samples (1) are single cells and / or cell compartments.

24. The method according to any one of the preceding claims, characterized in that the computer-readable images are 2-dimensional or 3-dimensional digital images.

25. System for implementing the method according to any of the preceding claims, the system comprises:- a detection coordination unit (5) for handling the computer-readable image of biological samples (1 ) and data related to biological samples (1 ); characterized in that the system further comprises:- a detection management unit (6), which isconnected to the detection coordination unit (5);- at least one detector unit (7), which isconnected to the detection management unit (6);- a learning management unit (8) which isconnected to the at least one detector unit (7);- at least one teaching unit (9), which isconnected to the learning management unit (8);- a learning model database (12), which isconnected to at least one teaching unit (9);- a coding management unit (10) which isconnected to the at least one detector unit (7);- at least one encoding unit (11 ), which is connected to the encoding management unit (10) and the learning model database (12);- an object code database (13) which is connected to at least one coding unit (11 ).

26. The system according to claim 25, characterized in that the system further comprises:- an image recording unit (2), which comprises at least one sample receiving unit for receiving a sample holder comprising a biological sample (1 ), and an image and data recording unit for creating a computer-readable image of at least a part of the biological sample (1 ); and- an image storage unit (3) which is connected to the image recording unit (2) for storing computer-readable images.

27. The system according to claim 26, characterized in that the sample holder is provided with metadata, where an image and data recording unit issuitable for reading out the metadata, and the system includes a data storage unit (4) which is connected to the image recording unit (2) for storing data, preferably metadata, and isconnected to the detection coordination unit (5).

28. The system according to claim 27, characterized in that the image storage unit (3) and the data storage unit (4) are an integrated storage unit.

29. The system according to any one of the claims 26-28, characterized in that the image storage unit (3) and / or the data storage unit (4) is connected to at least one external device (20) for receiving data and / or computer-readable images.

30. The system according to any one of the claims 25-29, characterized in that the system comprises at least one analysis unit (14) which is connected to the object code database (13).

31. The system according to any one of claims25-30, characterized in that the system comprises several detector units (7), several teaching units (9) and / or several coding units (11 ).

32. The system according to any one of claims 30-31 , characterized in that it comprises several analysis units (14).

33. The system according to any one of claims 25-32, characterized in that the biological samples (1 ) comprise objects that are single cells and / or cell compartments.

34. The system according to any one of claims 25-33, characterized in that the computer- readable images are 2-dimensional or 3-dimensional digital images.

35. The system according to any one of the claims 26-34, characterized in that the image recording unit (2) is a microscope capable of creating a digital image.

Images