Sample constituent detection system

Abstract

The sample formed component detection system provided by each embodiment of the present application uses a second image positioning model constructed by a machine learning algorithm and a first image positioning model constructed by a non-machine learning algorithm to respectively position formed components in the same sample image. Since a plurality of different image positioning models are combined to position the formed components, the formed component omission problem caused by a single positioning method can be avoided to a certain extent. Furthermore, the plurality of positions of the same formed component can be de-duplicated, and the type of the formed component can be identified and the plurality of types of the same formed component can be de-duplicated, so that the multiple detection, position deviation and other problems of the formed component can also be avoided.

Description

[0001] This application claims priority to Chinese Patent Application No. 202211531193.4, filed with the State Intellectual Property Office of China on December 1, 2022, entitled “Sample Formed Components Detection System”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of medical device technology, and more specifically, to a sample formed element detection system. Background Technology

[0003] Body fluids include various types such as urine and vaginal secretions. The detection and identification of formed elements in body fluids is an important part of routine clinical body fluid examinations, and the results of formed element detection can play an important role in the diagnosis of diseases.

[0004] Currently, one method for detecting formed elements is through a formed element image analysis system. This system uses a microscope to capture images of the body fluid sample in a counting chamber, and then employs image segmentation or target detection methods to locate the formed elements in the image, thereby obtaining the positional information of each formed element in the body fluid sample. However, existing formed element detection technologies often suffer from missed detections, resulting in inaccurate results. Summary of the Invention

[0005] In view of this, this application provides a sample formed element detection system to improve the accuracy of the detection results of formed elements in body fluids.

[0006] In a first aspect, embodiments of this application provide a sample formed element detection system, including:

[0007] The counting cell is used to hold the carrier liquid sample;

[0008] An image acquisition device is used to capture images of the body fluid sample in the counting pool to obtain a sample image of the body fluid sample;

[0009] A processing device, communicatively connected to the image acquisition device, is configured to:

[0010] The sample image is input into a first image localization model, which locates the formed elements in the sample image to obtain a first localization result corresponding to the formed elements. The first image localization model is a non-machine learning model.

[0011] The sample image is input into the second image localization model, which locates the formed elements in the sample image to obtain the second localization result corresponding to the formed elements. The second image localization model is a machine learning model.

[0012] Outputting the target localization result of formed elements in the sample image includes: for each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both a first localization result and a second localization result, then output the first localization result or the second localization result corresponding to the formed element according to a preset rule.

[0013] Secondly, embodiments of this application provide a sample formed element detection system, including:

[0014] The counting cell is used to hold the carrier liquid sample;

[0015] An image acquisition device is used to capture images of the body fluid sample in the counting pool to obtain a sample image of the body fluid sample;

[0016] A processing device, communicatively connected to the image acquisition device, is configured to:

[0017] The sample image is input into a first image localization model, which locates the formed elements in the sample image to obtain a first localization result corresponding to the formed elements.

[0018] The sample image is input into a second image localization model, which locates the formed elements in the sample image to obtain a second localization result corresponding to the formed elements; wherein the type of the first formed element is not exactly the same as the type of the second formed element.

[0019] Outputting the target localization result of formed elements in the sample image includes: for each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both a first localization result and a second localization result, then output the first localization result or the second localization result corresponding to the formed element according to a preset rule.

[0020] As can be seen from the above technical solutions, the sample formed element detection system provided in the various embodiments of this application uses a second image localization model constructed with a machine learning algorithm and a first image localization model constructed with a non-machine learning algorithm to locate formed elements in the same sample image. Since the localization results of multiple different image localization models are combined, the problem of missed detection of formed elements caused by a single localization method can be avoided to a certain extent. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1A-1B This is a schematic diagram of the detection results of formed elements by an existing analysis system;

[0023] Figure 2 A schematic diagram of the structure of the sample formed element detection system provided in the embodiments of this application;

[0024] Figure 3 A schematic flowchart of a sample formed element detection method provided in this application embodiment;

[0025] Figure 4 An example diagram of a sample formed element detection method provided in the embodiments of this application;

[0026] Figure 5 A schematic diagram illustrating a method for calculating overlap provided in an embodiment of this application;

[0027] Figure 6 This is another schematic flowchart of the sample formed element detection method provided in the embodiments of this application;

[0028] Figure 7 An example diagram illustrating the types of constituent elements in a sample image provided in this application embodiment;

[0029] Figure 8 This is another schematic flowchart of the sample formed element detection method provided in the embodiments of this application;

[0030] Figure 9 Another flowchart illustrating the sample formed element detection method provided in this application embodiment;

[0031] Figure 10 Another example diagram of the sample formed element detection method provided in the embodiments of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] Body fluid examination refers to the examination of various secretions or excretions from the body in routine clinical testing. It is one of the most commonly used examination items in clinical laboratories, mainly targeting various body fluids such as blood, urine, feces, serous cavity fluid, and cerebrospinal fluid. This article uses urine as an example to illustrate the analysis process of formed elements. It should be noted that the formed element detection systems of the embodiments of this application are not only applicable to the analysis of urine samples, but also applicable to any other type of body fluid sample mentioned above.

[0034] Current bodily fluid formed element analysis systems typically use image segmentation or target detection methods to locate formed elements in the sample image after capturing the image. However, the inventors have found that single detection methods often suffer from problems such as missed detections, location misalignment, and over-detection. For example... Figure 1A The provided image segmentation method detection results are as follows: Figure (a) misses the formed element in the lower left corner; Figure (b) shows the formed element in the lower left corner is located at an inaccurate position; and Figure (c) shows three bounding boxes detected for a formed element in the lower left corner. Research has found that the above-mentioned defects in image segmentation methods mainly apply to formed elements with unclear edges, such as tubules, mucus filaments, and squamous epithelial cells. Figure 1B The example of the target detection method provided cannot locate some extremely small formed elements (such as those indicated by the arrows), such as bacteria and erythrocytes.

[0035] To address at least one of the aforementioned positioning defects, embodiments of this application provide a sample formed element detection system. This system improves the detection accuracy of formed elements by combining multiple image positioning methods. Specifically, the sample formed element detection system can be a urine formed element analyzer, a urine sediment analyzer, or other sample detection equipment.

[0036] like Figure 2 As shown, the sample formed element detection system may include a counting cell 10, an image acquisition device 20, and a processing device 30. It is understood that the sample formed element detection system may also include other conventional components such as a sample introduction device, a sample dispensing device, a driving device, etc., which will not be elaborated upon in the embodiments of this application.

[0037] The counting chamber 10 is used to hold bodily fluid samples, such as urine samples, vaginal secretion samples, fecal samples, or semen samples, etc. In various embodiments of this application, bodily fluid samples may also be referred to as excrement and secretion samples. For example, the counting chamber may be a sealable cavity structure made of a transparent material.

[0038] The image acquisition device 20 is located above the counting cell 10. It can be a microscope imaging device used to acquire and photograph body fluid samples in the counting cell 10 to obtain sample images of the body fluid samples. For example, under the control of the drive device, the image acquisition device 20 is aligned from top to bottom with the imaging field of view of the counting cell 10, and the body fluid sample is magnified, focused, and photographed to obtain a sample image.

[0039] The processing device 30 is coupled to the image acquisition device 20 to achieve a communication connection. The sample image acquired by the image acquisition device 20 can be sent to the processing device 30, which performs detection such as locating and classifying the formed elements in the sample image. It can be understood that formed elements are substances in body fluid samples that have a certain shape and can be identified by microscopic observation. Taking urine samples as an example, the formed elements they contain include red blood cells, white blood cells, white blood cell clusters, crystals, yeast, bacteria, sperm, epithelial cells, casts, mucus filaments, etc.

[0040] The processing device 30 can be any device with image processing capabilities, such as including at least a processing component with data processing capabilities, such as a processor, and a computer-readable storage medium. The computer-readable storage medium stores a computer program and data associated with the computer program. When the computer program is invoked by the processor, the processing device 30 can execute actions such as... Figure 3 The sample shown is a step S301-S303 in the formed element detection method.

[0041] S301: Input the sample image into the first image localization model. The first image localization model locates the formed elements in the sample image and obtains the first localization result corresponding to the formed elements.

[0042] S302: Input the sample image into the second image localization model. The second image localization model locates the formed elements in the sample image and obtains the second localization result corresponding to the formed elements.

[0043] The first image localization model is a non-machine learning model, while the second image localization model is a machine learning model. In other words, the processing device 30 uses both a machine learning model and a non-machine learning model to locate formed elements in the same sample image, obtaining two sets of localization results. It should be noted that the formed elements contained in the two sets of localization results may be completely identical or not completely identical.

[0044] For ease of distinction, the localization results obtained from the two image localization models are referred to as the first localization result (e.g., first position indication information) and the second localization result (e.g., second position indication information), respectively. The localization result (e.g., position indication information) is used to indicate the position of formed elements in the sample image. In some specific embodiments, the localization result may include various forms such as detection boxes and contour lines.

[0045] It should be noted that a machine learning model is a model trained using machine learning algorithms. A trained machine learning model can perform inference and prediction on unknown data. Specifically, in the embodiments of this application, the second image localization model is obtained by training the initial machine learning model using training sample images. That is, the second image localization model is a trained machine learning model that can predict on sample images not used for training, such as determining the location of constituent elements in the sample image. Machine learning models can be divided into various types, such as supervised learning models, semi-supervised learning models, and unsupervised learning models. The second image localization model can specifically be a supervised learning model built based on deep learning algorithms, including but not limited to the YOLO model, SSD model, R-CNN model, etc.

[0046] The following combination Figure 4 The provided examples illustrate the localization and recognition process of the first image localization model and the second image localization model, respectively.

[0047] like Figure 4 As shown, the sample image is a urine sample image obtained by the image acquisition device 30 using a microscope (20x). The sample image includes multiple formed elements. The same sample image is input into the first image localization model and the second image localization model respectively, and the first localization result (the detection box used to mark the position of formed elements in the figure) and the second localization result (the detection box used to mark the position of formed elements in the figure) are obtained respectively.

[0048] The first image localization model can be a traditional image processing algorithm, which uses features such as color, edge, and texture of the sample image to locate each formed element in the sample image. In a specific embodiment, the first image localization model can be an image segmentation method, that is: the processing device 30 uses an image segmentation method to segment the contours of the formed elements from the sample image to obtain the first localization result.

[0049] Specifically, image segmentation methods segment formed elements in a sample image using color features and / or edge features to obtain the boundary contours of the formed elements. It should be noted that color features can refer to the color space format of the sample image, such as RGB, YCbCr, or HSV. Edge features can be edge information extracted by edge detection algorithms. In one possible implementation, edge detection-based image segmentation methods determine edge information by detecting the degree of pixel value change at the boundaries of different regions in the sample image. First, edge operators are used to determine which pixels in the sample image are edge pixels; these operators include, but are not limited to, the Canny operator, Sobel operator, Prewitt operator, and Laplacian operator. Then, the edge pixels in the sample image are connected together to obtain the edge regions. Finally, pixel filling and binarization are performed on the sample image to obtain a binarized sample image of the foreground and background, from which the boundary contours of the formed elements are obtained. Figure 4 As shown, the first image localization model can generate a binary image of the sample image with respect to the foreground and background of each formed element. The region enclosed by the boundary contour in the binary image represents the position of the formed element in the sample image.

[0050] Image segmentation methods are segmentation-based image processing methods that primarily segment the foreground and background using color features such as HSV and RGB. In one specific implementation, the first image localization model can be an image feature extraction method. This method can extract features of candidate boxes, such as Haar features, HOG features, grayscale features, SIFT features, SURF features, etc., and then match the extracted candidate box features with target regions to locate the target. In this embodiment, specifically, an image feature extraction method can be used to extract features of each candidate box in the sample image, and the extracted candidate box features are matched with each formed element to locate the position of each formed element in the sample image.

[0051] In one specific implementation, the boundary contour can be directly used as the positioning result obtained by the first image positioning model. In another specific implementation, after obtaining the contour of the formed elements, the outer rectangular box of the formed element contour is further determined, and the position of the outer rectangular box is used as the positioning result obtained by the first image positioning model. For example... Figure 4 As shown, the outer rectangular box of the formed element (first positioning result) can be generated based on the boundary contour of the formed element.

[0052] The second image localization model can be a machine learning model pre-trained based on a deep learning algorithm. To distinguish it from other machine learning models, this model can be called a component localization model. Specifically, the sample image is input into the component localization model, which determines the locations of formed elements in the sample image to obtain the second localization result. In one specific implementation, the component localization model is a supervised machine learning model. The training sample images of the component localization model are labeled, that is, location labels are added to the formed elements in the training sample images. After training with labeled training sample images, the component localization model has the ability to locate formed elements. Figure 4 As shown, after inputting the sample image into the component localization model, the component localization model can determine the location of formed elements in the sample image and obtain the detection box corresponding to the formed elements (second localization result).

[0053] It should be noted that image segmentation methods can locate formed elements with well-defined edges in a sample image. Taking urine samples as an example, this includes, but is not limited to, red blood cells, white blood cells, white blood cell clusters, crystals, yeast, bacteria, sperm, and epithelial cells (squamous epithelial cells, clue cells, transitional epithelial cells, etc.). Meanwhile, component localization models built based on deep learning algorithms can locate formed elements with less-than-ideal edges. Taking urine samples as an example, this includes, but is not limited to, casts (hyaluronic acid, pathological casts, etc.), mucus filaments, and epithelial cells (squamous epithelial cells, clue cells, transitional epithelial cells, etc.).

[0054] The processing device 30 uses a component localization model constructed by machine learning algorithms and an image segmentation method to locate formed elements with clear edges and unclear edges in the same sample image, respectively. This ensures that each formed element in the sample image can be located and identified, avoiding the problem of missed detection of formed elements caused by a single detection method.

[0055] In practical applications, different image localization models may locate the same formed element, resulting in at least two localization results for the same formed element. Figure 4 Taking the localization results as an example, both the first and second localization results are detection boxes. Detection boxes 401 and 402 are localization results for formed elements of the same size in the sample image. Since the localization results play a crucial role in the detection of formed elements—for example, identifying the type of formed element, counting the quantity of formed elements, and cropping and displaying the image of formed elements based on the localization results—the uniqueness and accuracy of the localization results directly affect the detection accuracy of formed elements in body fluid samples. Therefore, the processing device 30 performs deduplication of localization results for the same formed element in step S302 to ensure that the same formed element has only one corresponding localization result.

[0056] S303: Output the target localization result of the formed elements in the sample image. The target localization result of the output formed elements has the following characteristics: for each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both a first localization result and a second localization result, then output either a first localization result or a second localization result corresponding to the formed element according to a preset rule.

[0057] The target localization result output by the formed element detection system is unique for each formed element, which avoids confusion caused by multiple localization results and can reduce the occurrence of multiple detections to a certain extent, thereby improving the accuracy of formed element detection.

[0058] A single formed element may correspond to one first location result or one second location result, or it may correspond to both. Therefore, it is necessary to perform deduplication operations such as filtering and / or deleting on the location results corresponding to the formed element to ensure the uniqueness of the location results for the same formed element. In some embodiments, deduplication operations can be performed on the location results for the same formed element in the first and second location results to obtain the target location result. For example, the first and second location results can be combined into a location result set, and then deduplication operations can be performed on the location results for the same formed element in the location result set. Of course, in practical applications, it is also possible not to combine them into a set and directly perform deduplication operations on the first and second location results.

[0059] First, it should be clarified that "the same formed element" here does not refer to formed elements of the same type, but rather to the same formed element. Of course, this same formed element may be identified as formed elements of the same type or different types. In other words, if the localization results obtained by two image localization models both contain localization results for the same formed element, then this embodiment of the application performs a deduplication operation on the localization results corresponding to that same formed element. It can be understood that deduplication means removing duplicates; after deduplication of the localization results for the same formed element, only one localization result for that same formed element is retained.

[0060] Specifically, the processing device 30 first needs to identify the positioning results in the two sets of positioning results that relate to the same formed element, determine one of the positioning results as a duplicate, and then remove the duplicate positioning results from the two sets of positioning results to obtain the target positioning result. It can be seen that the target positioning result is the positioning result after deduplication of the two sets of positioning results. Let's continue with... Figure 4 For example, the first image localization model obtains 13 first localization results for the sample image, and the second image localization model obtains 1 second localization result for the sample image. After deduplication of the localization results for the formed elements of the largest size, 14 target localization results are obtained.

[0061] The processing device 30 can perform a deduplication operation on the first positioning result and the second positioning result. In one specific embodiment, the processing device 30 can combine the two sets of positioning results into a single positioning result set, delete duplicate positioning results from the set, and thus obtain the target positioning result. In another specific embodiment, the processing device 30 can combine the two sets of positioning results into a single positioning result set, select other positioning results besides duplicate ones from the set, and thus obtain the target positioning result.

[0062] In one embodiment, when executing step S302, the processor 30 is specifically configured to: determine the position indicated by each first positioning result on the sample image and the degree of overlap between the position indicated by each second positioning result on the sample image; if there is a target overlap greater than a preset overlap threshold among the overlaps, then perform a deduplication operation on the positioning result corresponding to the target overlap in the first positioning result and the second positioning result to obtain the target positioning result; wherein, the positioning result corresponding to the target overlap is the first positioning result or the second positioning result.

[0063] Duplicate location results may be either the first location information or the second location result.

[0064] In one embodiment, the deduplication operation can specifically be to treat any location result corresponding to the target overlap as a duplicate location result for deduplication. That is, the processing device 30 determines the location result other than the location result corresponding to the target overlap in the first location result and the second location result as the target location result. For example, the location result corresponding to the target overlap can be deleted from the location result set to obtain the target location result; or, other location results can be selected from the location result set as the target location result. Here, the location result corresponding to the target overlap is either the first location result or the second location result. In a specific implementation, the first location result is obtained by image segmentation, and the second location result is determined by a pre-trained component localization model. Therefore, the location result corresponding to the target overlap is the first location result, i.e., the location result obtained by the image segmentation method is removed. This is because, according to empirical data, the localization accuracy of the component localization model trained by the machine learning algorithm is higher than that of the image segmentation method. Therefore, the deduplication operation removes the location result obtained by the image segmentation method and retains the location result obtained by the component localization model.

[0065] In other embodiments, to improve deduplication accuracy, deduplication can be further performed by combining the type probability of the same formed element, as detailed below.

[0066] Combination Figure 4-5 Here's a detailed explanation. There may be multiple first positioning results and multiple second positioning results. First, determine the overlap between the locations indicated by each first positioning result and each second positioning result. It's understood that the location indicated by the positioning result is an image region, and the overlap is used to represent the degree of overlap between two image regions. If the overlap is greater than a preset overlap threshold, this overlap can be called the target overlap. The positioning result corresponding to the target overlap can be considered as the positioning result of the same formed element; therefore, a deduplication operation is needed for the positioning result of the target overlap.

[0067] by Figure 4 For example, determine the overlap between the positions indicated by 13 first positioning results and the position indicated by 1 second positioning result on the sample image. If the overlap between the positions indicated by the two positioning results (detection box 401 and detection box 402) corresponding to the largest formed element is greater than the preset overlap threshold, then it is necessary to deduplicate the two positioning results, such as deleting the first positioning result (detection box 401).

[0068] There are several ways to calculate overlap. For example, it can be calculated by the overlapping area at the indicated location, such as if the overlapping area is greater than a preset area threshold, or if the proportion of the overlapping area to the area of ​​one of the regions is greater than a preset proportion threshold. Alternatively, it can be calculated using the intersection-over-union (IOU) ratio.

[0069] Specifically, the first localization result is a first rectangular region on the sample image, and the second localization result is a second rectangular region on the sample image; the intersection-union ratio (IUR) of each first rectangular region and each second rectangular region is determined, where the IUR represents the degree of overlap. Figure 5 For example, assuming that for Figure 4 The largest dimension in the array has a first rectangular region with area K1 corresponding to detection box 401 and area K2 corresponding to second rectangular region with area K2. The intersection-union ratio (IOU) of the two rectangular regions is S1 / (K1+K2-S1). Assuming the preset overlap threshold is 0.5 and if the IOU is greater than or equal to 0.5, then it is necessary to... Figure 4 The detection boxes 401 and 402 are deduplicated, such as by deleting detection box 401. It should be noted that the preset overlap threshold can be other values ​​set based on practical experience, and this application does not impose specific limitations on it.

[0070] As can be seen from the above technical solutions, the processing device 30 in the sample formed element detection system uses a second image localization model constructed with a machine learning algorithm and a first image localization model constructed with a non-machine learning algorithm to locate formed elements in the same sample image. Since the specific localization methods of different image localization models are different, the problem of missed detection of formed elements caused by a single localization method can be avoided to a certain extent. For example Figure 1A As shown in Figure (a), traditional image segmentation algorithms may miss formed elements with unclear edges. However, the formed element detection system provided in this application, which uses a second image localization model constructed using a machine learning algorithm, has good localization results for such formed elements. For example... Figure 1B As shown, existing deep learning models may fail to detect extremely small formed elements when the image size difference of formed elements in the sample image is large. However, the sample formed element localization model provided in this application uses a non-machine learning first image localization model, such as an image segmentation method, which can detect extremely small formed elements.

[0071] Furthermore, since the second image localization model in this embodiment can be trained using machine learning algorithms on a large number of training sample images, it has higher accuracy in location localization compared to traditional image segmentation algorithms, and can solve problems such as... Figure 1A The positioning offset problem is shown in Figure (b). Furthermore, the sample formed element detection system of this application performs a deduplication operation on the positioning results obtained from the two image positioning models, thereby solving the problem of... Figure 1A The multi-detection problem is illustrated in Figure (c). That is, if the first image localization model obtains one or more localization results for the same formed element, and the second image localization model obtains only one localization result for the same formed element, then after deduplication of the localization results for the same formed element, only one target localization result for that formed element is retained. Figure 1A Taking Figure (c) as an example, the image segmentation method may delete all three detection boxes obtained from the localization of the same formed element, and only retain one detection box obtained from the element localization model, thereby avoiding the problem of multiple detections of the same formed element.

[0072] Furthermore, to visually examine the various formed elements located in the sample image, the processing device 30 is also configured to control the display device to display the sample image and mark the position indicated by the target positioning result based on the sample image. For example... Figure 4As shown, the location corresponding to the positioning result is marked in the sample image in the form of a rectangular box. The sample formed element detection system may further include a display device, which may be integrated with other components of the sample formed element detection system, or the display device may be physically separated from the sample formed element detection system. For example, the display device may be a monitor that is communicatively connected to the sample formed element detection system, wherein the communication connection may be a wired connection or a remote network connection.

[0073] In practical applications, the analysis and detection of formed elements in a sample image by a formed element detection system includes not only locating the formed elements but also determining their type. See [link / reference]. Figure 6 The processing device 30 can be configured to perform the sample formed element detection method shown in steps S601-604.

[0074] S601: Input the sample image into the first image localization model. The first image localization model locates the formed elements in the sample image and obtains the first localization result corresponding to the formed elements.

[0075] S602: Input the sample image into the second image localization model. The second image localization model locates the formed elements in the sample image and obtains the second localization result corresponding to the formed elements.

[0076] S603: Output the target localization result of the formed elements in the sample image. In some embodiments, the process of outputting the target localization result may include performing a deduplication operation on the localization results of the same formed element to obtain the target localization result. Specifically, a deduplication operation is performed on the localization results of the first localization result and the second localization result regarding the same formed element, such as deleting the first localization result obtained by the image segmentation method, thereby obtaining the target localization result.

[0077] For explanations of steps S601-S603, please refer to the corresponding steps above; they will not be repeated here.

[0078] S604: Input the sample image and target localization result into the pre-trained component classification model to obtain the type of the target formed component. The training sample images of the component classification model can be labeled with formed components, and the labeled formed components have both location and type labels.

[0079] Specifically, the formed elements corresponding to the target localization result can be called target formed elements. The component classification model classifies the target formed elements in the sample image to obtain their type. To facilitate differentiation from other component classification models, this model can be called a third-component classification model. The third-component classification model is trained using training sample images, and the formed elements in the training sample images have location and type labels. Taking urine samples as an example, the type labels include at least one of red blood cells, white blood cells, white blood cell clusters, crystals, yeast, bacteria, sperm, epithelial cells, casts, and mucus filaments. The trained third-component classification model has the ability to identify the type of formed elements. After inputting the sample image and target localization result into the third-component classification model, it can determine the type of target formed elements in the sample image.

[0080] In one specific embodiment, the third component classification model can be a neural network model, such as a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), or a Generative Adversarial Network (GAN). The following explanation uses a CNN as an example to illustrate the third component classification model. Specifically, the third component classification model is a component classification model constructed using a CNN, which includes, but is not limited to, the ResNet series, GoogLeNet series, AlexNet series, DenseNet series, and Inception series. CNNs have requirements regarding the size of the input image; therefore, a portion of the sample image to be identified can be cropped from the sample image and input into the CNN for classification. It should be noted that the cropping size can be set to other values ​​according to actual needs, and this embodiment does not impose specific limitations.

[0081] Specifically, the processing device 30 first obtains partial sample images corresponding to the target localization result from the sample images. For example, the sample image is cropped to a size of m*m*3, centered on the coordinate center of the location indicated by the target localization result, to obtain various partial sample images. Alternatively, after obtaining the boundary contours of formed elements using an image segmentation method, the sample image is cropped to a size of m*m*3, centered on the region center of the boundary contour, to obtain various partial sample images. Typically, m is 224 pixels; in this embodiment, m can be 60 pixels, and 3 represents three color channels (e.g., if the image format is RGB, then 3 represents the R, G, and B color channels). Of course, if the size of a formed element is larger than m*m, it can be directly cropped using the smallest bounding rectangle of the formed element. Then, the partial sample images are input into a pre-trained third-component classification model, which classifies the target formed elements in the partial sample images. The classification inference process of the third component classification model is as follows: the convolutional layer extracts features from a portion of the sample images to obtain an N-dimensional feature matrix of the sample images; the classifier classifies the N-dimensional feature matrix to obtain the classification type corresponding to the sample images.

[0082] like Figure 7 As shown, according to Figure 4 The obtained target localization results are used to crop the sample images, for example, by cropping them in an m*m*3 manner (where m is 60 pixels), to obtain partial sample images. These partial sample images are then input into a component classification model constructed using a convolutional neural network. The component classification model classifies each partial sample image to determine the types of formed elements contained in each partial sample image, such as white blood cells, impurities, cocci, red blood cells, and pathological casts.

[0083] As can be seen from the above technical solutions, the sample formed element detection system provided in this application embodiment, after the processing device 30 determines the position information of each formed element in the sample image, can further identify the type of each formed element, thereby obtaining the position information and type information of each formed element.

[0084] It should be noted that the above embodiments first deduplicate the location information of formed elements, and then classify the formed elements after location deduplication. Of course, the formed elements can also be classified first, and then the location information can be deduplicated. The following describes the classification process of the formed elements in the sample through various embodiments.

[0085] Another embodiment of this application provides a sample formed element detection system, which includes a counting cell 10, an image acquisition device 20, and a processing device 30. The descriptions of the counting cell 10 and the image acquisition device 20 are provided above and will not be repeated here. The processing device 30 will be described in detail below. The processing device 30 is coupled to the image acquisition device 20 to achieve a communication connection, for realizing the following... Figure 8 The sample shown has a formed element detection method, which includes steps S801-S804.

[0086] S801: Input the sample image into the first image localization model. The first image localization model locates the formed elements in the sample image and obtains the first localization result corresponding to the formed elements. Input the sample image into the second image localization model. The second image localization model locates the formed elements in the sample image and obtains the second localization result corresponding to the formed elements.

[0087] S802: Perform deduplication on the positioning results of the same formed element to obtain the target positioning result.

[0088] For details regarding S801-S802, please refer to the corresponding steps above; they will not be repeated here.

[0089] S803: Determine the type of the first formed element corresponding to each first positioning result and the type of the second formed element corresponding to each second positioning result.

[0090] For ease of distinction, the formed element corresponding to the first positioning result is referred to as the first formed element, and the formed element corresponding to the second positioning result is referred to as the second formed element.

[0091] The processing device 30 determines the type of each first formed element and the type of each second formed element.

[0092] In one specific implementation, the processing device 30 can determine the type of the first formed element and the type of the second formed element in the same way, such as by using the same pre-trained machine learning model to determine the type of the first formed element and the type of the second formed element respectively. The machine learning model can be a neural network model such as a convolutional neural network, or a non-neural network machine learning model.

[0093] In another specific implementation, the processing device 30 uses different methods to determine the type of the first formed element and the type of the second formed element, respectively. For example, different machine learning models trained on the machine learning device can be used to determine the type of the first formed element and the type of the second formed element. The machine learning model can be a neural network model or a non-neural network model. Specific classification methods can be described as follows.

[0094] Specifically, the processing device 30 inputs the sample image and the first positioning result into the pre-trained first component classification model, and the first component classification model classifies the first formed element corresponding to the first positioning result in the sample image to obtain the type of the first formed element.

[0095] To facilitate differentiation from other component classification models, this component classification model can be referred to as the first component classification model. The first component classification model is trained using training sample images. The formed elements in these training sample images are labeled, and each labeled formed element has both a location label and a type label. The trained first component classification model has the ability to recognize any sample image. Therefore, after the processing device 30 inputs the sample image acquired by the image acquisition device 20 into the first component classification model, the first component classification model can identify the type of formed elements based on the location results of the formed elements in the sample image.

[0096] The first component classification model can be a convolutional neural network (CNN). Accordingly, the classification by the processing device 30 specifically includes: obtaining a partial sample image corresponding to the first localization result from the sample image; inputting the partial sample image into the pre-trained first component classification model, and having the first component classification model classify the first formed elements in the partial sample image. It should be noted that the process of cropping a partial sample image from the sample image can be found in the above description and will not be repeated here.

[0097] The classification process of the second formed elements can be as follows: the processing device 30 inputs the sample image into the second component classification model that has been pre-trained based on a deep learning algorithm, and the second component classification model identifies the formed elements in the sample image to obtain the type and type probability of the second formed elements.

[0098] To distinguish it from other component classification models, this component classification model can be referred to as the second component classification model. For example, the first component classification model is a component classification model built using a convolutional neural network, while the second component classification model is a component classification model based on a deep learning algorithm.

[0099] Similarly, the second component classification model is trained from training sample images. The formed elements in these training sample images are labeled, and each labeled formed element has both a location label and a type label. The training sample images for the two classes can be the same or different. The difference can lie in the fact that the training sample images are different, or that the training sample images are the same but the types of labeled formed elements are different. The trained second component classification model also has the ability to identify the type of formed elements in any sample image. After the processing device 30 inputs the sample images acquired by the image acquisition device 20 into the second component classification model, the second component classification model can identify the type of formed elements based on the location results of the formed elements in the sample images.

[0100] In one specific implementation, the second component classification model and the component localization model are integrated into a single deep learning model, which can be called a component detection model. The trained component detection model can both identify the type of formed elements and determine their location. For example... Figure 10 In the provided example, after the sample image is input into the component detection model, the component detection model determines the localization result (detection box) and type (pathological cast).

[0101] S804: Perform a deduplication operation on the types of the same formed element in each of the first formed elements and the types of each of the second formed elements to obtain the type of the target formed element.

[0102] In this context, the first formed element and the second formed element are located from the same sample image. Since the first and second formed elements may overlap, the identification of the first and second formed element types may also be redundant. That is, the same formed element may be identified more than once, resulting in multiple type identification results for the same formed element. Figure 10 For example, the component detection model identifies the type of the second formed element corresponding to the detection box 402 as a pathological cast, and the component classification model can identify the type of the first formed element corresponding to the detection box 401 as non-squamous epithelium.

[0103] Therefore, it is necessary to perform a deduplication operation on the types of each first formed element and each second formed element obtained in step S803. The formed element after the deduplication operation is called the target formed element, that is, the type of the target formed element obtained after deduplication. It should be noted that there is a correspondence between the type of the target formed element and the target positioning result. The correspondence is reflected in the fact that the location indicated by the target positioning result is the target formed element, so the type of the target formed element can be considered as the type corresponding to the target positioning result, or in other words, the target positioning result and the type of the target formed element are bound one-to-one. Figure 10For example, there is a one-to-one correspondence between the target localization result (detection box) and the type of the target formed element. It is also understandable that there is a one-to-one correspondence between the type of the first formed element and the first localization result, and between the type of the second formed element and the second localization result.

[0104] In one specific implementation, since there is a one-to-one correspondence between the location result and the type, the type deduplication operation can be based on the location deduplication result of step S802 above. The type corresponding to the removed location result is then removed, thus obtaining the type of the target formed element. In some embodiments, the types of the first formed element and the second formed element can be considered as a type set. Duplicate location results removed from the first and second location results are identified, and the types corresponding to these duplicate location results are removed, such as deleting the type from the type set, or extracting other types from the type set besides the duplicate location results, thereby obtaining the type of the target formed element. Figure 10 As shown, after merging the first positioning result and type with the second positioning result and type, duplicate positioning results are deduplicated, and duplicate types are deduplicated to obtain the target positioning result and the type corresponding to the target positioning result.

[0105] As can be seen from the above technical solutions, the sample formed element detection system provided in this application uses two methods to identify the position of formed elements. On the one hand, it locates the position of formed elements and removes duplicates from multiple positions of the same formed element. On the other hand, it identifies the type of formed elements and removes duplicates from multiple types of the same formed element. This ensures more comprehensive detection of formed elements, avoids missed detection of formed elements, and avoids problems such as multiple detections and positional shifts of formed elements.

[0106] Another embodiment of this application provides a sample formed element detection system, which includes a counting cell 10, an image acquisition device 20, and a processing device 30. The descriptions of the counting cell 10 and the image acquisition device 20 are provided above and will not be repeated here. The processing device 30 will be described in detail below. The processing device 30 is used to achieve the following... Figure 9 The sample shown has a formed element detection method, which includes steps S901-S907.

[0107] S901: Input the sample image into the first image localization model. The first image localization model locates the formed elements in the sample image and obtains the first localization result corresponding to the formed elements.

[0108] S902: Determine the type and type probability of the first formed element corresponding to each first positioning result.

[0109] S903: Input the sample image into the second image localization model. The second image localization model locates the formed elements in the sample image and obtains the second localization result corresponding to the formed elements.

[0110] S904: Determine the type and type probability of the second formed element corresponding to each second positioning result.

[0111] For explanations of S901-S904, please refer to the corresponding steps above, which will not be repeated here.

[0112] It should be noted that the type probabilities of the first formed element and the second formed element can be found in the explanation of step S803 above. In step S803, the type of the first formed element is obtained through the first component classification model, and the type of the second formed element is obtained through the second component classification model. Each type can have a corresponding type probability, which represents the likelihood that the formed element belongs to that type. Figure 10 For example, the probability of the type corresponding to the cast in the case is 95%, and the probability of the type corresponding to the non-squamous epithelium is 87%, and so on.

[0113] S905: Determine the overlap between the positions indicated by each first positioning result on the sample image and the positions indicated by each second positioning result on the sample image.

[0114] S906: If there is a target overlap greater than the preset overlap threshold among the various overlap degrees, then determine the first type probability corresponding to the target overlap degree from the type probability of the first formed element, and determine the second type probability corresponding to the target overlap degree from the type probability of the second formed element.

[0115] Specifically, first, the first localization result corresponding to the target overlap is determined. Then, from the type probabilities corresponding to each first formed element, the type probability corresponding to this first localization result is determined. This type probability is the first type probability corresponding to the target overlap. Similarly, the second type probability corresponding to the target overlap is determined. For example... Figure 10 As shown, if the overlap between detection boxes 401 and 402 is greater than a preset overlap threshold, the type probability of the type corresponding to detection box 401 (referred to as the first type probability for easy distinction) and the type probability of the type corresponding to detection box 402 (referred to as the second type probability for easy distinction) are determined.

[0116] S907: Based on the first type probability and the second type probability, perform a deduplication operation on the positioning results corresponding to the target overlap in the first positioning result and the second positioning result to obtain the target positioning result.

[0117] This step can be found in the above description. Figure 3Explanation of step S302. It should be noted that this embodiment combines two dimensions, overlap degree and type probability, to perform a deduplication operation on the positioning results corresponding to the target overlap degree in the first positioning result and the second positioning result.

[0118] In one specific implementation, the target location result is determined from the first and second location results, excluding the location result corresponding to the type with the lower probability. The lower type probability is the type with the smaller probability between the first and second type probabilities. This implementation directly removes the location results with the lower type probability during deduplication. For example... Figure 10 As shown, if the probability of the type corresponding to the cast in the case is 95% and the probability of the type corresponding to the non-squamous epithelium is 87%, then the localization result corresponding to the non-squamous epithelium is directly deleted.

[0119] Of course, the type of formed elements can also be determined, and the method for deduplication of the positioning results can be determined based on whether the types are consistent.

[0120] In another specific implementation, we can first determine whether the type corresponding to the first type probability is the same as the type corresponding to the second type probability.

[0121] If the type corresponding to the first type probability is the same as the type corresponding to the second type probability, then the other localization results in the first and second localization results, excluding the localization result corresponding to the target overlap, are determined as the target localization result. It should be noted that the localization result corresponding to the target overlap is the localization result of either of the two image localization models with respect to the target overlap. Therefore, if the types are the same, the localization result obtained by either image localization model can be removed. For example, the localization result obtained by the first image localization model, such as an image segmentation method, can be removed.

[0122] If the type corresponding to the first type probability is different from the type corresponding to the second type probability, then the other positioning results between the first and second positioning results, except for the positioning result corresponding to the smaller type probability, are determined as the target positioning result. The smaller type probability is the type with the lower probability between the first and second type probabilities. In this implementation, during deduplication, if the type corresponding to the first type probability is different from the type corresponding to the second type probability, the positioning result with the lower type probability corresponding to the target overlap is removed.

[0123] As can be seen from the above technical solutions, the sample constituent element detection system provided in this application embodiment combines the two dimensions of overlap degree and type probability to perform deduplication operation on the positioning results corresponding to the target overlap degree in the first positioning result and the second positioning result, thereby further ensuring the accuracy of the deduplication operation.

[0124] Furthermore, in another embodiment of this application, the sample formed element detection system, based on the above steps S901-S907, can also perform a type deduplication step in step S908, that is, perform a deduplication operation on the types of the same formed element among the types of each first formed element and each second formed element to obtain the type of the target formed element. In some embodiments, a type set of the types of the first formed element and the types of the second formed element can be constructed, and the deduplication operation can be performed on the type set; of course, it is also possible not to construct a type set. For the specific implementation of this step, please refer to the relevant descriptions of the above embodiments, which will not be repeated here. It should be noted that the type of the target formed element has a corresponding relationship with the target positioning result. In this way, the sample formed element detection system can not only accurately locate the position of the formed element, but also accurately obtain the type of the formed element, thereby avoiding problems such as multiple detections, missed detections, and positional offsets of formed elements, and the component detection results are more comprehensive.

[0125] In practical applications, in order to facilitate a direct and intuitive view of the detection results of formed elements in the sample image, the detection results (location results + type) can be displayed after the processing device 30 obtains the detection results of formed elements.

[0126] Display Method 1: Overall Image Display. Specifically, the processing device 30 is further configured to display a sample image, and, based on the sample image, indicate the location indicated by the target positioning result and display the type of formed elements of the target corresponding to the target positioning result. For example... Figure 7 and Figure 10 As shown, the sample image is displayed, and the location of the formed elements of the target is marked using a detection box. The type of formed elements of the target is also shown.

[0127] Display Method Two: Cropping and Categorization Display. Specifically, the processing device 30 is further configured to acquire partial sample images corresponding to the target localization results from the sample images; and to categorize and display the partial sample images corresponding to each type based on the type of the target formed elements corresponding to the partial sample images. Figure 7 and Figure 10 For example, each target formed element can be cropped from the sample image to obtain a partial sample image, and the partial sample images of each type of target formed element can be displayed according to five types: red blood cells, cocci, white blood cells, impurities, and pathological casts.

[0128] In practical applications, the detection results of formed elements can be used to further count the formed elements of the target. That is, the processing device 30 is further configured to: obtain counting information for the same type of formed element based on its type and the corresponding target location result. For example, the number of target location results corresponding to each type of formed element is counted, and this number is used as the counting result of the formed elements. Still using... Figure 7 and Figure 10 For example, after counting the white blood cells in the detection box, the count result is 2; after counting the red blood cells in the detection box, the count result is 2; after counting the cocci in the detection box, the count result is 3; after counting the impurities in the detection box, the count result is 5; and after counting the pathological casts in the detection box, the count result is 1.

[0129] In addition, this application embodiment also provides a sample formed element detection system, which includes: a counting pool, an image acquisition device, and a processing device.

[0130] A counting chamber is used to hold body fluid samples. In one specific implementation, the body fluid sample includes a urine sample, a vaginal secretion sample, a fecal sample, or a semen sample.

[0131] An image acquisition device is used to photograph body fluid samples in a counting pool to obtain sample images of the body fluid samples.

[0132] The processing unit, which is communicatively connected to the image acquisition unit, is configured to:

[0133] The sample image is input into a first image localization model, which locates the formed elements in the sample image to obtain a first localization result corresponding to the formed elements. The sample image is then input into a second image localization model, which locates the formed elements in the sample image to obtain a second localization result corresponding to the formed elements. The types of the first and second formed elements are not entirely the same. For example, taking a urine sample as an example, the types of the first formed elements include at least one of the following: red blood cells, white blood cells, white blood cell clusters, crystals, yeast, bacteria, sperm, and epithelial cells. The types of the second formed elements include at least one of the following: casts, mucus filaments, and epithelial cells.

[0134] The target localization results of formed elements in the output sample image include: For each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both one first localization result and one second localization result, then output one first localization result or one second localization result corresponding to the formed element according to a preset rule.

[0135] In some embodiments, the positioning results of the same formed element in the first positioning result and the second positioning result can be deduplicated to obtain the target positioning result, wherein the same formed element is either the first formed element or the second formed element.

[0136] It should be noted that the formed element detection system provided in this application uses two different image localization models to locate the formed elements. The types of formed elements obtained by localization are not completely the same. In other words, formed elements that are missed by one image localization model can be located by another image localization model, thereby solving the problem of missed detection of formed elements in the prior art.

[0137] In one specific implementation, the sample image includes formed elements with clear edges and formed elements with unclear edges. A first image localization model can locate formed elements with clear edges. The sample image also includes formed elements of extremely small size and formed elements of non-extremely small size. A second image localization model can locate formed elements of non-extremely small size. The inclusion of extremely small and non-extremely small formed elements in the sample image indicates a significant difference in size among the formed elements. It should be noted that extremely small and non-extremely small sizes can be defined by those skilled in the art based on the conventional definitions of formed element size in laboratory medicine for bodily fluid samples. Of course, if the size difference among the formed elements in the sample image is within a certain range, that is, if the size of each formed element is relatively uniform, then even if each formed element is extremely small, the second image localization model can still locate each formed element.

[0138] In one specific implementation, the processing device is further configured to: determine the type of a first formed element corresponding to each first positioning result, and determine the type of a second formed element corresponding to each second positioning result; output the type of the target formed element in the sample image that corresponds to the target positioning result, including, for each output type of the target formed element: if the target formed element corresponds to only one type of the first formed element, then output the type of the first formed element that the target formed element corresponds to; if the target formed element corresponds to only one type of the second formed element, then output the type of the second formed element that the target formed element corresponds to; if the target formed element corresponds to both one type of the first formed element and one type of the second formed element, then output the type of the first formed element or the type of the second formed element that the target formed element corresponds to according to a preset rule. In some embodiments, a deduplication operation is performed on the types of the first formed elements and the types of the second formed elements regarding the same formed element to obtain the type of the target formed element; wherein, the type of the target formed element corresponds to the target positioning result.

[0139] In one specific implementation, the processing device is further configured to: input the sample image and the target localization result into a pre-trained third component classification model, and classify the target formed elements corresponding to the target localization result in the sample image by the third component classification model to obtain the type of the target formed elements; wherein, the formed elements labeled in the training sample image of the third component classification model have location labels and type labels.

[0140] In one specific implementation, the counting information of target formed elements of the same type is obtained based on the type of the target formed elements and the target positioning result corresponding to the target formed elements.

[0141] In one specific implementation, the positioning results obtained by the first image positioning model for the same formed element include one or more, and the positioning results obtained by the second image positioning model for the same formed element include one. Then, after performing a deduplication operation on the positioning results for the same formed element, the target positioning result for the same formed element is one.

[0142] It should be noted that the descriptions of each step in the embodiments of this application can be found in the relevant steps described above, and will not be repeated here.

[0143] The features described above in the disclosed embodiments can be substituted or combined with each other to enable those skilled in the art to implement or use this application. This document has been described with reference to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications can be made to the exemplary embodiments without departing from the scope of this document. For example, various operational steps and components for performing operational steps can be implemented in different ways depending on the specific application or considering any number of cost functions associated with the operation of the system (e.g., one or more steps can be deleted, modified, or combined with other steps).

[0144] The terms "first," "second," etc., used in this specification, claims, and the accompanying drawings are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, or apparatuses.

[0145] Furthermore, as those skilled in the art will understand, the principles herein can be reflected in a computer program product on a computer-readable storage medium pre-loaded with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-ROMs, DVDs, Blu-ray discs, etc.), flash memory, and / or the like. These computer program instructions may be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine, such that instructions, which execute on the computer or other programmable data processing apparatus, can generate means for performing a specified function. These computer program instructions may also be stored in a computer-readable storage medium that can instruct the computer or other programmable data processing apparatus to operate in a particular manner, such that instructions stored in the computer-readable storage medium can form an article of manufacture, including means for implementing the specified function. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to perform a series of operational steps on the computer or other programmable apparatus to produce a computer-implemented process, such that instructions, which execute on the computer or other programmable apparatus, can provide steps for implementing the specified function.

[0146] The foregoing specific descriptions have been described with reference to various embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of this disclosure. Therefore, considerations for this disclosure are to be illustrative rather than restrictive, and all such modifications are to be included within its scope. Similarly, advantages, other advantages, and solutions to problems with respect to various embodiments have been described above. However, benefits, advantages, solutions to problems, and any elements that produce these, or make them more explicit, should not be construed as critical, essential, or necessary. The term “comprising” and any other variations thereof as used herein are non-exclusive inclusion, meaning that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed or not part of the process, method, system, article, or apparatus. Furthermore, the term “coupled” and any other variations thereof as used herein refer to physical connections, electrical connections, magnetic connections, optical connections, communication connections, functional connections, and / or any other connections.

[0147] The above embodiments are merely illustrative of several implementation methods, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this invention, and these all fall within the protection scope of this invention. Therefore, the protection scope of this invention should be determined by the appended claims.

Claims

1. A sample formed element detection system, characterized in that, include: The counting cell is used to hold the carrier liquid sample; An image acquisition device is used to capture images of the body fluid sample in the counting pool to obtain a sample image of the body fluid sample; A processing device, communicatively connected to the image acquisition device, is configured to: The sample image is input into a first image localization model, which locates the formed elements in the sample image to obtain a first localization result corresponding to the formed elements. The first image localization model is a non-machine learning model. The sample image is input into the second image localization model, which locates the formed elements in the sample image to obtain the second localization result corresponding to the formed elements. The second image localization model is a machine learning model. Outputting the target localization result of formed elements in the sample image includes: for each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both a first localization result and a second localization result, then at least based on the overlap between the position indicated by the first localization result on the sample image and the position indicated by the second localization result on the sample image, perform a deduplication operation on the first localization result and the second localization result, and output the first localization result or the second localization result corresponding to the formed element.

2. The sample formed element detection system as described in claim 1, characterized in that, When executing the first image localization model to locate the formed elements in the sample image and obtain the first localization result corresponding to the formed elements, the processing device is configured to specifically be used for: The contours of formed elements are segmented from the sample image using an image segmentation method to obtain the first localization result corresponding to the formed elements.

3. The sample formed element detection system as described in claim 1, characterized in that, When executing the second image localization model to locate the formed elements in the sample image and obtain the second localization result corresponding to the formed elements, the processing device is configured to specifically be used for: The second image localization model determines the location of formed elements in the sample image and obtains the second localization result corresponding to the formed elements; wherein the second image localization model is pre-trained based on a deep learning algorithm, and the formed elements marked in the training sample images of the second image localization model have location labels.

4. The sample formed element detection system as described in claim 1, characterized in that, When performing the step of deduplicating the positioning results of the first positioning result and the second positioning result regarding the same formed element, the processing device is configured to specifically be used for: Determine the overlap between the positions indicated by each of the first positioning results on the sample image and the positions indicated by each of the second positioning results on the sample image; If there is a target overlap greater than a preset overlap threshold among the various overlap degrees, then the positioning results corresponding to the target overlap degree are deduplicated in the first positioning result and the second positioning result to obtain the target positioning result; wherein the positioning result corresponding to the target overlap degree is the first positioning result or the second positioning result.

5. The sample formed element detection system as described in claim 4, characterized in that, The first positioning result is a first rectangular region on the sample image, and the second positioning result is a second rectangular region on the sample image; When performing the step of determining the overlap between the positions indicated by each of the first positioning results on the sample image and the positions indicated by each of the second positioning results on the sample image, the processing device is configured to specifically be used for: Determine the intersection-union ratio (IUGR) between each of the first rectangular regions and each of the second rectangular regions, wherein the IUGR is used to represent the degree of overlap.

6. The sample formed element detection system as described in claim 4, characterized in that, When performing the step of deduplicating the positioning results corresponding to the target overlap in the first positioning result and the second positioning result to obtain the target positioning result, the processing device is configured to specifically be used for: The positioning results other than the positioning results corresponding to the target overlap in the first positioning result and the second positioning result are determined as the target positioning results; wherein, the positioning result corresponding to the target overlap is the first positioning result or the second positioning result.

7. The sample formed element detection system as described in claim 4, characterized in that, The first localization result is obtained by image segmentation, and the second localization result is determined by a pre-trained second image localization model. The localization result corresponding to the target overlap is the first localization result.

8. The sample formed element detection system as described in claim 4, characterized in that, When performing the step of deduplicating the positioning results corresponding to the target overlap in the first positioning result and the second positioning result to obtain the target positioning result, the processing device is configured to specifically be used for: Determine the type and type probability of the first formed element corresponding to each of the first positioning results, and determine the type and type probability of the second formed element corresponding to each of the second positioning results; determine the first type probability corresponding to the target overlap from the type probabilities of the first formed elements, and determine the second type probability corresponding to the target overlap from the type probabilities of the second formed elements; Based on the first type probability and the second type probability, the target overlap corresponding to the first positioning result and the second positioning result are deduplicated to obtain the target positioning result.

9. The sample formed element detection system as described in claim 8, characterized in that, When performing the step of deduplicating the positioning results corresponding to the target overlap in the first positioning result and the second positioning result according to the first type probability and the second type probability to obtain the target positioning result, the processing device is configured to specifically be used for: The positioning results other than the positioning results corresponding to the smaller type probability in the first positioning result and the second positioning result are determined as the target positioning result, wherein the smaller type probability is the smaller probability type between the first type probability and the second type probability.

10. The sample formed element detection system as described in claim 8, characterized in that, When performing the step of deduplicating the positioning results corresponding to the target overlap in the first positioning result and the second positioning result according to the first type probability and the second type probability to obtain the target positioning result, the processing device is configured to specifically be used for: If the type corresponding to the first type probability is the same as the type corresponding to the second type probability, then the other positioning results in the first positioning result and the second positioning result, other than the positioning result corresponding to the target overlap, are determined as the target positioning result. If the type corresponding to the first type probability is different from the type corresponding to the second type probability, then the other positioning results between the first positioning results and the second positioning results, except for the positioning results corresponding to the smaller type probability, are determined as the target positioning results, wherein the smaller type probability is the smaller probability type between the first type probability and the second type probability.

11. The sample formed element detection system as described in claim 8, characterized in that, When performing the step of determining the type and type probability of the first formed element corresponding to each of the first positioning results, the processing device is configured to specifically be used for: The sample image and the first localization result are input into a pre-trained first component classification model. The first component classification model classifies the first formed element corresponding to the first localization result in the sample image to obtain the type and type probability of the first formed element. The formed elements labeled in the training sample image of the first component classification model have location labels and type labels.

12. The sample formed element detection system as described in claim 11, characterized in that, The first component classification model is a convolutional neural network; When executing the step of inputting the sample image and the first localization result into a pre-trained first component classification model, and having the first component classification model classify the first formed element corresponding to the first localization result in the sample image, the processing device is specifically configured as follows: From the sample images, obtain a portion of the sample image corresponding to the first positioning result; The partial sample images are input into a pre-trained first component classification model, which then classifies the first formed elements in the partial sample images.

13. The sample formed element detection system as described in claim 8, characterized in that, When performing the step of determining the type and type probability of the second formed element corresponding to each of the second positioning results, the processing device is configured to specifically be used for: The sample image is input into a second component classification model that has been pre-trained based on a deep learning algorithm. The second component classification model identifies formed elements in the sample image and obtains the type and type probability of the second formed elements. The formed elements labeled in the training sample image of the second component classification model have location labels and type labels.

14. The sample formed element detection system as described in claim 13, characterized in that, The second component classification model and the second image localization model are integrated into the same deep learning model.

15. The sample formed element detection system as described in claim 8, characterized in that, The processing device is further configured to display the sample image, and to mark the location indicated by the target positioning result based on the sample image and to display the type of the target formed element corresponding to the target positioning result.

16. The sample formed element detection system as described in claim 8, characterized in that, The processing device is further configured to obtain a partial sample image corresponding to the target localization result from the sample image; and to classify and display the partial sample images corresponding to each type according to the type of the target formed elements corresponding to the partial sample images.

17. The sample formed element detection system as described in claim 8, characterized in that, The body fluid sample is a urine sample. The first formed element includes at least one of red blood cells, white blood cells, white blood cell clusters, crystals, yeast, bacteria, sperm, and epithelial cells. The second formed element includes at least one of casts, mucus filaments, and epithelial cells.

18. The sample formed element detection system as described in claim 1, characterized in that, The processing device is further configured to input the sample image and the target localization result into a pre-trained third component classification model, and the third component classification model classifies the target formed elements corresponding to the target localization result in the sample image to obtain the type of the target formed elements; wherein, the formed elements marked in the training sample image of the third component classification model have location labels and type labels.

19. The sample formed element detection system as described in claim 18, characterized in that, The third component classification model is a component classification model constructed using a convolutional neural network; When executing the step of inputting the sample image and the target localization result into a pre-trained third-component classification model, and having the third-component classification model classify the formed components of the target corresponding to the target localization result in the sample image, the processing device is specifically configured as follows: From the sample images, obtain a portion of the sample images corresponding to the target localization result; The sample images are input into a pre-trained third-component classification model, which then classifies the formed components of the target in the sample images.

20. The sample formed element detection system as described in claim 1, characterized in that, The processing device is further configured to: Determine the type of the first formed element corresponding to each of the first positioning results, and determine the type of the second formed element corresponding to each of the second positioning results; Output the type of the formed element of the target in the sample image that corresponds to the target localization result, including, for each output type of the formed element: if the formed element of the target corresponds to only one type of the first formed element, then output the type of the first formed element to which the formed element of the target corresponds; if the formed element of the target corresponds to only one type of the second formed element, then output the type of the second formed element to which the formed element of the target corresponds; if the formed element of the target corresponds to both one type of the first formed element and one type of the second formed element, then output the type of the first formed element or the type of the second formed element to which the formed element of the target corresponds according to a preset rule.

21. The sample formed element detection system as described in claim 20, characterized in that, When performing the step of outputting the type of target formed element in the sample image that corresponds to the target localization result, the processing device is configured to: The types of the first formed element and the second formed element are deduplicated to obtain the type of the target formed element.

22. The sample formed element detection system as described in claim 18 or 20, characterized in that, The processing device is further configured to display the sample image, and to mark the location indicated by the target positioning result based on the sample image and to display the type of the target formed element corresponding to the target positioning result.

23. The sample formed element detection system as described in claim 18 or 20, characterized in that, The processing device is further configured to obtain a partial sample image corresponding to the target localization result from the sample image; and to classify and display the partial sample images corresponding to each type according to the type of the target formed elements corresponding to the partial sample images.

24. The sample formed element detection system as described in claim 18 or 20, characterized in that, The processing device is further configured to: Based on the type of the target formed elements and the target positioning result corresponding to the target formed elements, the counting information of the target formed elements of the same type is obtained.

25. The sample formed element detection system as described in claim 1, characterized in that, The processing device is further configured to display the sample image and, based on the sample image, mark the location indicated by the target localization result.

26. The sample formed element detection system as described in claim 1, characterized in that, If the first image localization model obtains one or more localization results for the same formed element, and the second image localization model obtains one localization result for the same formed element, then after performing a deduplication operation on the localization results for the same formed element, the target localization result for the same formed element is one.

27. The sample formed element detection system as described in claim 1, characterized in that, The bodily fluid samples include urine samples, vaginal secretion samples, fecal samples, serous cavity fluid, cerebrospinal fluid, or semen samples.

28. A sample formed element detection system, characterized in that, include: The counting cell is used to hold the carrier liquid sample; An image acquisition device is used to capture images of the body fluid sample in the counting pool to obtain a sample image of the body fluid sample; A processing device, communicatively connected to the image acquisition device, is configured to: The sample image is input into a first image localization model, which locates the formed elements in the sample image to obtain a first localization result corresponding to the formed elements. The sample image is input into a second image localization model, which locates the formed elements in the sample image to obtain a second localization result corresponding to the formed elements; wherein the type of formed elements located by the first image localization model is not exactly the same as the type of formed elements located by the second image localization model. Outputting the target localization result of formed elements in the sample image includes: for each formed element of the output target localization result: if the formed element corresponds to only one first localization result, then output the only first localization result corresponding to the formed element; if the formed element corresponds to only one second localization result, then output the only second localization result corresponding to the formed element; if the formed element corresponds to both a first localization result and a second localization result, then at least based on the overlap between the position indicated by the first localization result on the sample image and the position indicated by the second localization result on the sample image, perform a deduplication operation on the first localization result and the second localization result, and output the first localization result or the second localization result corresponding to the formed element.

29. The sample formed element detection system as described in claim 28, characterized in that, The sample image includes formed elements with clear edges and formed elements with unclear edges, and the first image localization model can locate the formed elements with clear edges. The sample image includes formed elements of extremely small size and formed elements of non-extremely small size, and the second image localization model is able to locate the formed elements of non-extremely small size.

30. The sample formed element detection system as described in claim 28, characterized in that, The processing device is further configured to: Determine the type of the first formed element corresponding to each of the first positioning results, and determine the type of the second formed element corresponding to each of the second positioning results; Output the type of the formed element of the target in the sample image that corresponds to the target localization result, including, for each output type of the formed element: if the formed element of the target corresponds to only one type of the first formed element, then output the type of the first formed element to which the formed element of the target corresponds; if the formed element of the target corresponds to only one type of the second formed element, then output the type of the second formed element to which the formed element of the target corresponds; if the formed element of the target corresponds to both one type of the first formed element and one type of the second formed element, then output the type of the first formed element or the type of the second formed element to which the formed element of the target corresponds according to a preset rule.

31. The sample formed element detection system as described in claim 28, characterized in that, The processing device is further configured to: The sample image and the target localization result are input into a pre-trained third component classification model. The third component classification model classifies the target formed elements corresponding to the target localization result in the sample image to obtain the type of the target formed elements. The formed elements labeled in the training sample image of the third component classification model have location labels and type labels.

32. The sample formed element detection system as described in claim 30 or 31, characterized in that, Based on the type of the target formed elements and the target positioning result corresponding to the target formed elements, the counting information of the target formed elements of the same type is obtained.

33. The sample formed element detection system as described in claim 28, characterized in that, If the first image localization model obtains one or more localization results for the same formed element, and the second image localization model obtains one localization result for the same formed element, then after performing a deduplication operation on the localization results for the same formed element, the target localization result for the same formed element is one.

34. The sample formed element detection system as described in claim 28, characterized in that, The bodily fluid samples include urine samples, vaginal secretion samples, fecal samples, serous cavity fluid, cerebrospinal fluid, or semen samples.

Images