A target object phenotype determination method based on a SOLOv2 model

Abstract

The application discloses a target object phenotype determination method based on a SOLOv2 model, which comprises the following steps: segmenting muscle fiber cells of a sample image to be determined by a weight-based target segmentation model to obtain a mask image and a first sample image; determining the total number of muscle fiber cells and the cell area of each muscle fiber cell according to muscle fiber cells in the first sample image, which have a cell overlapping area less than a first threshold value and a confidence degree greater than a second threshold value; classifying type I muscle cells and type II muscle cells by a dynamic relative color difference method according to the sample image to be determined to obtain the total number of type I muscle cells, the total number of type II muscle cells and the ratio of the two types of muscle fiber cells; determining the fat proportion of the sample image according to the mask image; and determining the phenotype of a target object according to the total number of muscle fiber cells, the ratio of the two types of muscle fiber cells and the fat proportion, so that the target image can be automatically segmented more quickly, the phenotype information of the target object in the image can be accurately obtained, and the method can be widely applied to the technical field of image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for determining the phenotype of a target object based on the SOLOv2 model. Background Technology

[0002] As the basic structural unit of muscle tissue, muscle fibers directly affect the muscle quality of animals such as pigs, cattle, and sheep. Based on different properties, muscle fibers are classified into Type I and Type II, and the type of muscle fiber is considered a key factor influencing muscle quality. The number, size, and type of muscle fibers are fundamental characteristics. The number and size of muscle fibers determine muscle yield, while the type of muscle fiber is a crucial factor affecting muscle quality. Different muscle fiber types result in variations in post-mortem muscle color, pH value, water-holding capacity, and other quality traits. The fundamental characteristics of muscle fibers largely determine the physiological and biochemical properties of meat and ultimately influence the meat quality of animals.

[0003] Traditional methods for determining muscle fiber phenotypes require manual labeling and counting of each muscle fiber in the sample, as well as subsequent type differentiation. Labeling a single muscle fiber image can take several minutes. This repetitive process not only consumes significant human resources and time, but also results in poor manual segmentation due to blurry boundaries that are difficult for the human eye to distinguish. Failure to obtain accurate boundaries for each muscle fiber will affect subsequent muscle fiber size measurements and may even influence the counting of muscle fibers. Inaccurate segmentation results will also affect the calculation of other subsequent indicators.

[0004] The SOLOv2 model is an instance segmentation model designed based on public datasets such as COCO. While its optimization approach prioritizes general applicability, it has some shortcomings in segmenting myofibroblast cell images. For example, its accuracy in boundary delineation is not high, and it may classify two closely spaced cells as a single cell. These shortcomings can introduce errors into subsequent metric measurements. Therefore, a suitable model is needed for myofibroblast image segmentation that can guarantee segmentation accuracy. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a target object phenotypic determination method based on the SOLOv2 model, which aims to more quickly and automatically segment target images and accurately obtain the phenotypic information of target objects in the images.

[0006] This invention provides a method for determining the phenotype of a target object based on the SOLOv2 model, comprising: segmenting myofibroblasts in a sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image; wherein the target segmentation model is obtained by training a weighted SOLOv2 model; the first sample image includes the cell boundary of each myofibroblast; the mask image includes each myofibroblast and a cell mask; determining the total number of myofibroblasts and the cell area of ​​each myofibroblast based on myofibroblasts in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold; classifying type I and type II myofibroblasts according to the sample image to be measured using a dynamic relative color difference method to obtain the total number of type I, the total number of type II, and the ratio of the two types of myofibroblasts; determining the fat proportion of the sample image based on the mask image; and determining the phenotype of the target object based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the fat proportion.

[0007] Optionally, in the step of segmenting the myofibroblasts of the sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image, the step of training the weight-based SOLOv2 model is as follows: constructing a weight-based SOLOv2 model; obtaining a training image after labeling the fully displayed myofibroblasts in the sample image; inputting the training image into the weight-based SOLOv2 model for model training to obtain a first segmentation model; and evaluating the first segmentation model using a loss function to obtain a target segmentation model.

[0008] Optionally, the image processing steps of the weight-based SOLOv2 model include: obtaining the original dimensional feature map of the input; transforming the original dimensional feature map according to the average pooling formula to obtain a first feature map; calculating spatial information weights and channel information weights through a fully connected layer; performing convolutional feature extraction on the first feature map according to the spatial information weights and the channel information weights to obtain image features; and in the merging branch stage, performing mask prediction and instance segmentation on the image features through convolutional kernels.

[0009] Optionally, the step of segmenting the myofibroblasts of the sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image includes: inputting the sample image to be measured into the target segmentation model for automatic segmentation to obtain the cell boundary and cell mask of each myofibroblast; wherein the cell mask constitutes the mask image.

[0010] Optionally, in the step of determining the total number of myofibrils and the cell area of ​​each myofibrils based on the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold, the step of determining the total number of myofibrils includes: identifying the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold as effective myofibrils; performing a number count on the effective myofibrils and determining the result of the number count as the total number of myofibrils; and calculating the area of ​​each effective myofibrils and determining the result of the area calculation as the cell area of ​​each myofibrils.

[0011] Optionally, the step of classifying type I and type II myofibroblasts based on the image of the sample to be measured using a dynamic relative color difference method to obtain the total number of type I, the total number of type II, and the ratio of the two types of myofibroblasts includes: corresponding the mask image to the image of the sample to be measured and determining the connected components of the mask image; determining the first pixel value of each pixel in the myofibroblast image domain corresponding to the connected component, and determining the average pixel value of the myofibroblast image domain; clustering the connected components according to the domain color corresponding to the average pixel value using a dynamic relative color difference method; using the pixel values ​​of the three channels as three-dimensional points on a three-dimensional plane and calculating the Euclidean distance between the three-dimensional points; determining the type I and type II myofibroblasts in the sample to be measured based on the Euclidean distance; and calculating the total number of type I myofibroblasts and the total number of type II myofibroblasts to obtain the ratio of the two types of myofibroblasts.

[0012] Optionally, determining the fat percentage of the sample image based on the mask image includes: connecting all the connected components in the sample image to be measured based on the mask image to obtain a first connected component; determining the largest connected component based on the area of ​​the first connected component; and determining the fat percentage in the largest connected component using the maximum inscribed circle method.

[0013] This invention also provides an electronic device, including a processor and a memory; the memory is used to store a program; the processor executes the program to implement the method described above.

[0014] This invention also provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described above.

[0015] The embodiments of the present invention have the following beneficial effects: By training a SOLOv2 model with weight-based improvements, the present invention obtains a target segmentation model. This model is able to extract important feature distributions, making the neural network focus more on features in the center of the image, thus minimizing the appearance of incomplete cells around the edges. This allows for better segmentation of complete cells in the central part of the image to be measured, which is then used for myofibroblast classification. When classifying two types of myofibroblasts, the colors of the same type of myofibroblast in different images are not necessarily the same, making it impossible to determine the myofibroblast type using fixed or range-of pixel values. The present invention uses a dynamic relative color difference method to dynamically classify the two types of myofibroblasts based on their color differences, obtaining the parameters of each type, which is beneficial for phenotypic determination of the target object. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0017] Figure 1 This is a flowchart of the method steps provided in the embodiments of the present invention;

[0018] Figure 2 This is a structural diagram of the SOLOv2 optimized part provided in the embodiment of the present invention;

[0019] Figure 3 This is a diagram of the computational space weight matrix structure provided in an embodiment of the present invention;

[0020] Figure 4 This is a diagram showing the structure of the calculation channel weight matrix provided in an embodiment of the present invention;

[0021] Figure 5 This is a diagram showing the segmentation results of myofibroblasts provided in this embodiment;

[0022] Figure 6 This is a myofibroblast segmentation mask provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] To address at least one problem in existing technologies, this invention provides a method for determining the phenotype of a target object based on a SOLOv2 model, comprising: segmenting myofibroblasts in a sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image; wherein the target segmentation model is obtained by training a weighted SOLOv2 model; the first sample image includes the cell boundary of each myofibroblast; the mask image includes each myofibroblast and a cell mask; determining the total number of myofibroblasts and the cell area of ​​each myofibroblast based on myofibroblasts in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold; classifying type I and type II myofibroblasts using a dynamic relative color difference method based on the first sample image to obtain the total number of type I, the total number of type II, and the ratio of the two types of myofibroblasts; determining the fat proportion of the sample image based on the mask image; and determining the phenotype of the target object based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the fat proportion.

[0025] Specifically, an embodiment of the present invention provides a method for determining the phenotype of a target object based on the SOLOv2 model, comprising the following steps S100 to S500:

[0026] S100. The target segmentation model is trained to segment the myofibroblasts in the sample image to be measured, resulting in a mask image and a first sample image. The target segmentation model is trained on a weighted SOLOv2 model. The first sample image includes the cell boundary of each myofibroblast. The mask image includes each myofibroblast and a cell mask.

[0027] Specifically, in this embodiment of the invention, the muscle of the animal whose phenotype needs to be determined is first sliced ​​to obtain an experimental sample. Then, the experimental sample is stained and photographed to obtain an image of the sample to be determined. The experimental sample is a well-preserved slice of animal muscle, free from decay or deterioration. The myofibroblasts of the animal described in this embodiment can originate from pigs, cattle, sheep, or dogs. The steps for training the weighted SOLOv2 model include the following steps S110–S140:

[0028] S110. Construct a weight-based SOLOv2 model.

[0029] Specifically, to address some shortcomings of the SOLOv2 model in boundary segmentation, this embodiment of the invention improves and optimizes the SOLOv2 model. The main improvement is to optimize the SOLOv2 backbone network by adding weight calculations to the fully connected layers, improving channel relationships, and enhancing spatial information. The improved weight-based SOLOv2 model performs wired processing on the input image features, enhancing the channel and spatial information of the features, extracting important channel and spatial information, and assigning weights to them. These weights are then used in subsequent network layers for calculations, improving the performance of the original model. This embodiment of the invention constructs a SOLOv2 model with added weight calculations as a segmentation model for subsequent training and image segmentation.

[0030] S120. Obtain the training image after labeling the fully displayed myofibroblasts in the sample image.

[0031] Specifically, since staining is typically required when using muscle fiber samples as the target for phenotypic determination, it is understandable that the training image is also a stained, colored image. The training samples need to be well-preserved slides, free from decay or deterioration. Therefore, data cleaning is necessary to remove images corresponding to deteriorated experimental samples to avoid interference during training. Then, the remaining images containing intact muscle fiber cells are labeled to obtain the training image.

[0032] S130. Input the training image into the weight-based SOLOv2 model for model training to obtain the first segmentation model.

[0033] Specifically, refer to Figure 2 In the weight-based SOLOv2 model, this embodiment of the invention learns spatial and channel information by incorporating global features into fully connected layers of two branches. For the spatial feature set S = {s1, s2, s3…s…} n ,} and channel feature set C={c1,c2,c3…c n Spatial information weights are learned. Channel information weight It is applied to the subsequent computation of the network.

[0034] The weighted SOLOv2 model in this embodiment of the invention includes a calculation process for spatial information weights and a calculation process for channel information weights in the image processing process. The spatial information weights... The calculation process is as follows:

[0035] Reference Figure 3For an original feature map with input dimensions H×W×C, where H is the height, W is the width, and C is the number of channels, the channel information is processed simultaneously through max pooling and average pooling to obtain two H×W×1 features, namely the average channel feature F. avg and maximum channel feature F max Then, the two features are concatenated and convolved to obtain the spatial weights. in The formula for obtaining it is The plus sign here is the concat operation.

[0036] Channel information weight The calculation process is as follows:

[0037] Reference Figure 4 For an input feature map with original dimensions of H×W×C, where H is the height, W is the width, and C is the number of channels, the feature is first transformed into a 1×1×C feature map using average pooling to capture the global perspective. The average pooling formula is as follows: Where u is the feature map to be averaged and pooled, and then connected through a fully connected layer to obtain channel information weights.

[0038] Combining spatial and channel weights, the feature map is input into a convolutional network for further feature extraction. During convolution, the actual proportion of spatial and feature weights is... Based on the spatial characteristics of the muscle fiber image, higher weights are given to features in the middle of the image and lower weights are given to features at the edges, which helps to fade out incomplete muscle fiber cells at the edges.

[0039] The backbone network outputs multi-layered features, each with different scale and semantic information, yet sharing weights. Features from each layer are input into subsequent branch convolutional networks for feature extraction. Each branch undergoes convolutional kernel learning and feature learning, with shared weights in each branch for feature extraction. During branch merging, a final convolutional kernel is used for mask prediction, ultimately leading to instance segmentation.

[0040] In this embodiment of the invention, training samples are input into the aforementioned weight-based SOLOv2 model, and the training samples are processed and trained using the image processing method of the SOLOv2 model to obtain the first segmentation model.

[0041] S140. The first segmentation model is evaluated using a loss function to obtain the target segmentation model.

[0042] Specifically, the first segmentation model is evaluated using a loss function. When the segmentation loss value is less than a predetermined threshold, the target segmentation model is obtained. This loss function can be either the FocalLoss loss function or the DiceLoss loss function.

[0043] Step S100 further includes the following step S150:

[0044] S150. The myofibrils of the sample image to be measured are segmented using a target segmentation model to obtain a mask image and a first sample image. The cell mask constitutes the mask image.

[0045] Specifically, refer to Figure 5 and Figure 6 In this embodiment of the invention, the myofibroblasts of the sample image to be measured are segmented using a target segmentation model to obtain a first sample image, as shown below. Figure 5 The resulting mask image is as follows Figure 6 .

[0046] S200. Determine the total number of myofibrils and the cell area of ​​each myofibril based on the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold.

[0047] Specifically, step S200 includes the following steps S210 to S230:

[0048] S210. Determine the effective myofibroblasts as those whose cell overlap area in the first sample image is less than the first threshold and whose confidence level is greater than the second threshold.

[0049] Specifically, the cell overlap area and confidence level can be obtained by the SOLOv2 model during image processing, which will not be elaborated here. This embodiment of the invention configures a first threshold and a second threshold to screen cells in the first sample image, enabling effective myofibroblasts to receive attention and further accurately obtain phenotypic measurement data, which is beneficial for the phenotypic determination of the target object.

[0050] S220. Count the number of effective myofibroblasts and determine the result of the count as the total number of myofibroblasts.

[0051] S230. Calculate the area of ​​each effective myofibroblast cell and determine the cell area of ​​each myofibroblast cell based on the area calculation results.

[0052] S300. Based on the image of the sample to be measured, classify type I and type II myocytes using the dynamic relative color difference method to obtain the total number of type I cells, the total number of type II cells, and the ratio of the two types of myofibrils.

[0053] Specifically, step S300 includes the following steps S310 to S360:

[0054] S310. Match the mask image with the sample image to be measured, and determine the connected components of the mask image.

[0055] Specifically, the mask image is mapped to the sample image to be measured, and the connected components of the mask image are determined. The boundaries of these connected components, as well as the location of the connected components themselves, correspond to the boundaries and regions of myofibroblasts in the sample image to be measured. For ease of description, the region of the myofibroblast in the image to be measured is named the myofibroblast image domain. It should be noted that a sample image to be measured contains one or more myofibroblasts, therefore, there are also one or more myofibroblast image domains obtained.

[0056] S320. Determine the first pixel value of each pixel in the muscle fiber image domain corresponding to the connected component, and determine the average pixel value of the muscle fiber image domain.

[0057] Specifically, the first pixel value S of each pixel in the muscle fiber image domain is obtained. i The average pixel count of the muscle fiber image domain is calculated based on the first set of pixel values ​​in the muscle fiber image domain. The calculation formula is:

[0058]

[0059] in, S is the average pixel count, where n represents the number of pixels. i This represents the pixel value of the i-th pixel.

[0060] S330. Based on the domain color corresponding to the average pixel value, the connected domain is clustered and classified using the dynamic relative color difference method.

[0061] Specifically, after obtaining the average pixels in all muscle fiber image domains, a histogram is constructed to represent the color of the muscle fiber image domain using the average pixels. The two peaks in the histogram are the color values ​​of the two types of muscle fiber cells in the sample image to be tested. The muscle fiber image domain is then clustered and classified using a clustering method.

[0062] S340. Treat the pixel values ​​of the three channels as three-dimensional points on a three-dimensional plane, and calculate the Euclidean distance between the three-dimensional points.

[0063] Specifically, the pixel values ​​of the three channels are treated as three-dimensional points on a three-dimensional plane, and the Euclidean distance between the three-dimensional points is calculated using the following formula:

[0064]

[0065] Where dist is the Euclidean distance, S i S is the pixel value of the i-th pixel. jS is the pixel value of the j-th pixel. ix S iy S iz S represents the x, y, and z values ​​of the i-th pixel in the three-dimensional plane. jx S jy S jz These represent the x, y, and z values ​​of the j-th pixel on the three-dimensional plane, respectively.

[0066] S350. Determine the type I and type II myofibroblasts in the sample to be tested based on the Euclidean distance.

[0067] Specifically, based on the Euclidean distance calculated above, the 3D points are divided into two clusters. First, two 3D points are randomly selected as the cluster centers of the two clusters. Then, each 3D point is added to the nearest cluster center, and the cluster centers are updated according to the Euclidean distance. This process of adding cluster centers is iterated until the objective function W is minimized. The objective function W is:

[0068]

[0069] Where W represents the objective function, k represents the number of cluster centers, C represents the cluster center, and S represents each pixel.

[0070] When the cluster centers do not change with iteration, the two clusters obtained are the final classification results.

[0071] S360. Calculate the total number of type I myofibroblasts and the total number of type II myofibroblasts to obtain the ratio of the two types of myofibroblasts.

[0072] Specifically, the total number of type I myofibroblasts obtained from the statistical classification is recorded as the type I total; the total number of type II myofibroblasts is recorded as the type II total. The type I total and the type II total are summed to obtain the ratio of the two types of myofibroblasts.

[0073] S400. Determine the fat percentage of the sample image based on the mask image.

[0074] Specifically, step S400 includes the following steps S410 to S430:

[0075] S410. Based on the mask image, connect all connected components in the sample image to be measured to obtain the first connected component.

[0076] Specifically, based on the mask image obtained from the sample to be measured, each muscle fiber mask domain is connected to obtain all connected components in the graph. Since noise points may exist, there may be one or more first connected components.

[0077] S420. Determine the largest connected region based on the area of ​​the first connected region.

[0078] Specifically, for noise points with too small a filtering area, the largest connected component in the remaining first connected component is determined based on its area and becomes the largest connected component.

[0079] S430. Determine the fat percentage in the largest connected region using the maximum inscribed circle method.

[0080] Specifically, the largest inscribed circle is generated in the largest connected region. The fat and cells in the largest inscribed circle are identified by the color value of the pixels. The black area is fat and the white area is a cell mask. The fat ratio of the sample image under test is calculated based on the ratio of fat to cells.

[0081] S500. Determine the phenotype of the target subject based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the proportion of fat.

[0082] Specifically, the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the proportion of fat are all phenotypic indicators of myofibroblasts. The phenotype of the target object, the myofibroblast, can be determined based on these phenotypic indicators.

[0083] This invention also provides a target object phenotyping system based on the SOLOv2 model, comprising: a first module for segmenting myofibroblasts in a sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image; wherein the target segmentation model is obtained by training a weighted SOLOv2 model; the first sample image includes the cell boundary of each myofibroblast; the mask image includes each myofibroblast and a cell mask; a second module for determining the total number of myofibroblasts and the cell area of ​​each myofibroblast based on myofibroblasts in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold; a third module for classifying type I and type II myofibroblasts in the sample image to be measured using a dynamic relative color difference method to obtain the total number of type I, the total number of type II, and the ratio of the two types of myofibroblasts; a fourth module for determining the fat proportion of the sample image based on the mask image; and a fifth module for determining the phenotype of the target object based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the fat proportion.

[0084] This invention also provides an electronic device, including a processor and a memory; the memory is used to store a program; the processor executes the program to implement the method described above.

[0085] This invention also provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described above.

[0086] The embodiments of the present invention have the following beneficial effects:

[0087] 1. This invention incorporates spatial and channel information weights into the backbone network of the SOLOv2 model, and optimizes it. The original backbone network used ResNet, which directly inputs H×W×C image features into the network for convolution. In this embodiment, however, the input image features are processed preferentially, and the channel and spatial information of the features are enhanced. Weights of important information, i.e., feature distribution, are extracted, and the weights of less influential channels are reduced, resulting in a better fit between channels. By processing the target sample using the weight-based SOLOv2 model of this invention, the neural network's emphasis on central image information is utilized, reducing the appearance of incomplete cells around the image and better segmenting the complete cells in the middle of the muscle fiber image, which is beneficial for subsequent processing. This embodiment also utilizes spatial information to achieve a more precise segmentation of cell edges in the image, improving the overall measurement accuracy.

[0088] 2. When classifying two types of muscle fibers, the colors of the same type of muscle fibers in different images are not the same, making it impossible to determine the muscle fiber type using fixed or range-based pixel values. This invention dynamically classifies the two types of muscle fiber cells based on their color difference, obtaining the number of each type of muscle fiber cell. By using the proportion of fat in the largest inscribed circle, samples of different muscle fibers are scaled down for comparison, aiding in the assessment of sample quality.

[0089] The following is an application example of an embodiment of the present invention:

[0090] The target segmentation model is obtained by training the weighted SOLOv2 model using training samples. The image of the sample to be measured is acquired, and the myofibrils in the image are segmented using this target segmentation model to obtain a mask image and a first sample image. The total number of myofibrils and the cell area of ​​each myofibril are determined based on the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold. Based on the image of the sample to be measured, type I and type II myofibrils are classified using the dynamic relative color difference method to obtain the total number of type I and type II myofibrils and the ratio of the two types of myofibrils. The fat proportion of the sample image is determined based on the mask image. The phenotype of the target object is determined based on the total number of myofibrils, the ratio of the two types of myofibrils, and the fat proportion.

[0091] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0092] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0093] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0094] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0095] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0096] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for measuring a phenotype of a target object based on a SOLOv2 model, characterized by, include: The target segmentation model is trained to segment the myofibrils in the sample image to be measured, resulting in a mask image and a first sample image. The target segmentation model is trained on a weighted SOLOv2 model. The first sample image includes the cell boundary of each myofibril. The mask image includes each myofibril and a cell mask. The total number of myofibrils and the cell area of ​​each myofibril are determined based on the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold. Based on the image of the sample to be measured, type I myocytes and type II myocytes are classified by dynamic relative color difference method to obtain the total number of type I cells, the total number of type II cells, and the ratio of the two types of myofibrils. The fat percentage of the sample image is determined based on the mask image; The phenotype of the target object is determined based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the fat percentage. The step of classifying type I and type II myocytes based on the image of the sample to be measured using a dynamic relative color difference method to obtain the total number of type I and type II myocytes and the ratio of the two types of myofibrils includes: The mask image is matched with the sample image to be measured to determine the connected components of the mask image; Determine the first pixel value of each pixel in the muscle fiber image domain corresponding to the connected component, and determine the average pixel value of the muscle fiber image domain; Based on the domain color corresponding to the average pixel value, the connected domain is clustered and classified using the dynamic relative color difference method. The pixel values ​​of the three channels are used as three-dimensional points on a three-dimensional plane, and the Euclidean distance between the three-dimensional points is calculated. The type I and type II myofibroblasts in the sample to be tested are determined based on the Euclidean distance. Calculate the total number of type I myofibroblasts and the total number of type II myofibroblasts to obtain the ratio of the two types of myofibroblasts.

2. The method of claim 1, wherein the SOLOv2 model is based on a model of a target object phenotype. In the step of segmenting the myofibroblasts of the sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image, the step of training the weight-based SOLOv2 model is as follows: Construct a weight-based SOLOv2 model; Obtain a training image after labeling the fully displayed myofibrils in the sample image; The training images are input into the weight-based SOLOv2 model for model training to obtain the first segmentation model; The first segmentation model is evaluated using a loss function to obtain the target segmentation model.

3. The method for determining the phenotype of a target object based on the SOLOv2 model according to claim 2, characterized in that, The image processing steps of the weight-based SOLOv2 model include: Obtain the original dimensional feature map of the input; The original dimensional feature map is transformed according to the average pooling formula to obtain the first feature map; Spatial information weights and channel information weights are calculated through a fully connected layer; Based on the spatial information weights and the channel information weights, convolutional feature extraction is performed on the first feature map to obtain image features; During the merging branch stage, the image features are masked and instances are segmented using convolutional kernels.

4. The method of claim 1, wherein the SOLOv2 model is based on a model of a target object phenotype. The step of segmenting the myofibroblasts of the sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image includes: The sample image to be measured is input into the target segmentation model for automatic segmentation to obtain the cell boundary and cell mask of each myofibroblast; wherein, the cell mask constitutes a mask image.

5. The method of claim 1, wherein the SOLOv2 model is based on a model of a target object phenotype. In the step of determining the total number of myofibrils and the cell area of ​​each myofibril based on the myofibrils in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold, the step of determining the total number of myofibrils includes: The myofibroblasts in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold are identified as effective myofibroblasts. The effective myofibroblasts are counted, and the result of the count is determined as the total number of myofibroblasts. The area of ​​each effective myofibroblast is calculated, and the result of the area calculation is determined as the cell area of ​​each myofibroblast.

6. The method of claim 1, wherein the SOLOv2 model is based on a model of a target object phenotype. Determining the fat percentage of the sample image based on the mask image includes: Based on the mask image, all connected components in the sample image to be measured are connected to obtain the first connected component; The largest connected region is determined based on the area of ​​the first connected region; The fat percentage in the largest connected region is determined by the method of the largest inscribed circle.

7. A target object phenotyping system based on the SOLOv2 model, characterized in that, include: The first module is used to segment the myofibroblasts in the sample image to be measured using a trained target segmentation model to obtain a mask image and a first sample image; wherein, the target segmentation model is obtained by training a weight-based SOLOv2 model; the first sample image includes the cell boundary of each myofibroblast; the mask image includes each myofibroblast and a cell mask; The second module is used to determine the total number of myofibroblasts and the cell area of ​​each myofibroblast based on the myofibroblasts in the first sample image whose cell overlap area is less than a first threshold and whose confidence level is greater than a second threshold. The third module is used to classify type I and type II myocytes based on the image of the sample to be measured using the dynamic relative color difference method, and obtain the total number of type I cells, the total number of type II cells, and the ratio of the two types of myocytes. The fourth module is used to determine the fat ratio of the sample image based on the mask image; The fifth module is used to determine the phenotype of the target object based on the total number of myofibroblasts, the ratio of the two types of myofibroblasts, and the proportion of fat. The third module is specifically used for: The mask image is matched with the sample image to be measured to determine the connected components of the mask image; Determine the first pixel value of each pixel in the muscle fiber image domain corresponding to the connected component, and determine the average pixel value of the muscle fiber image domain; Based on the domain color corresponding to the average pixel value, the connected domain is clustered and classified using the dynamic relative color difference method. The pixel values ​​of the three channels are used as three-dimensional points on a three-dimensional plane, and the Euclidean distance between the three-dimensional points is calculated. The type I and type II myofibroblasts in the sample to be tested are determined based on the Euclidean distance. Calculate the total number of type I myofibroblasts and the total number of type II myofibroblasts to obtain the ratio of the two types of myofibroblasts.

8. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 6.

Images