Method, system and equipment for counting organoid in bright field image and medium

Abstract

The invention provides a method, a system and equipment for counting organoid in a bright field image and a medium, and belongs to the field of biomedical image processing, and the method comprises the following steps: obtaining a non-fluorescence labeled organoid bright field image data set; adding double pooling in the multi-scale convolutional neural network model, and adding a multi-scale branch convolution module after pooling processing to obtain an improved multi-scale convolutional neural network model; and inputting the image data of the data set into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model. Inputting a to-be-detected organ-like image into the image segmentation model to obtain a segmentation result; using an ellipse fitting algorithm to fit the organoid contours of the overlapping areas in the segmented image, and counting the number of organoid in the image after fitting. A complex image structure is captured through the model, and the accuracy of counting the overlapping region of the similar organs is improved through the ellipse fitting algorithm, so that the deep tissue structure of a human body can be observed.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical image processing, and in particular relates to a method, system, device and medium for counting organoids in bright field images. Background Art

[0002] Organoid imaging technology plays a key role in helping researchers understand the development of diseases and drug screening by non-invasively observing the deep tissue structures of the human body. Bright field images are a type of image in microscopy technology. Images obtained through bright field microscopy are usually used to observe the basic structure and morphology of samples. Accurate segmentation and counting of organoids in bright field images has important practical value in helping researchers understand tissue structure and function. However, due to problems such as overlapping areas, complex organoid structures and low contrast in organoid bright field images, how to achieve accurate counting remains a daunting task.

[0003] Currently, in many fields of computer vision, including object detection, image classification, and semantic segmentation, models based on convolutional neural networks (CNNs) are mainly used. Compared with traditional image processing methods, CNN models can not only automatically learn features in images, but also show strong performance when processing large-scale data sets. However, the CNN models currently used often perform poorly when processing small data sets in medical imaging, and have limited ability to handle complex organoid structures and overlapping area problems, resulting in low accuracy in counting organoid overlaps, affecting the effectiveness of subsequent disease analysis. Summary of the invention

[0004] In order to solve the problem of counting organoids in bright field images, the present invention provides a method, system, device and medium for counting organoids in bright field images.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for counting organoids in a bright field image comprises the following steps:

[0007] A non-fluorescently labeled organoid bright field image dataset is obtained to construct a training set; maximum pooling and average pooling are introduced into the encoder of the multi-scale convolutional neural network model, and a multi-scale branched convolution module is added after the pooling process to obtain an improved multi-scale convolutional neural network model; the image data of the training set is input into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model;

[0008] Inputting the bright field image of the organoid to be detected into the image segmentation model, wherein the encoder in the image segmentation model extracts features respectively by using maximum pooling and average pooling, concatenating the extracted features and inputting them into a multi-scale branched convolution module to obtain feature information of images at different scales, performing feature enhancement on the image after feature extraction by using an attention gate mechanism, and decoding the feature-enhanced image by a decoder to output a segmented organoid image;

[0009] The organoid contours in the overlapping areas of the segmented organoid images are fitted using an ellipse fitting algorithm to obtain a fitted organoid image, and the number of organoids in the fitted organoid image is counted and an organoid counting result is output.

[0010] Preferably, the step of fitting the organoid contours of the overlapping regions in the segmented organoid images using an ellipse fitting algorithm to obtain the fitted organoid images is performed by the following steps:

[0011] Performing edge detection on the segmented organoid image to extract edges of various organs in the image;

[0012] Locating concave points formed at the contact points of various organoids in the segmented organoid image according to the edge detection result; the concave points are locations where two or more organoids contact each other;

[0013] The ellipse fitting method was used to fit the contours of various organs, and a fitting elliptical curve was constructed to separate the overlapping areas of organoids to obtain the fitted organoid image.

[0014] Preferably, the locating of the concave points formed at the contact points of various organs in the segmented organoid image according to the edge detection result is specifically performed by the following steps:

[0015] Measure the distance from each middle point to the contour to determine its position relative to the contour. If all middle points meet this condition, the pixel is preliminarily classified as a concave point. Repeat the measurement along the entire boundary to identify all potential concave points.

[0016] For the i-th point on the boundary, the point P current and its two neighboring points P next , P prev The vector pair and The local geometric properties are evaluated and the size of the local depression is quantified by the inverse of the radius of curvature r; the radius of curvature r is specifically defined as:

[0017]

[0018] Where K represents the curvature;

[0019] Set a maximum distance threshold to evaluate the distance d between two concave points. By traversing the contour points within the specified index range and calculating the sum Length of the distances between adjacent points, if the threshold condition Length < d is satisfied, the concave point with a larger curvature is retained;

[0020]

[0021] where dist(P i , P i+1 ) represents the Euclidean distance between point P i and point P i+1 , that is:

[0022] The least squares method is used to achieve the ellipse fitting between adjacent concave points to approximate the complete contour shape of each cell:

[0023] ax 2 + bxy + cy 2 + dx + ey + f = 0;

[0024] where a, b, c determine the shape and orientation of the ellipse, b is related to the rotation of the ellipse; d, e control the translation of the ellipse on the x and y axes; f is a constant term; when (b 2 - 4ac) < 0, this equation represents an ellipse.

[0025] Preferably, the contour of various organs is fitted by the ellipse fitting method, and the overlapping region of the organoids is separated by constructing a fitted ellipse curve, specifically through the following steps:

[0026] Calculate the intersection area A E ∩ A C between the ellipse contour and the contour of a single organoid, evaluate whether the ellipse is completely surrounded by the organoid contour, establish a threshold θ, and consider the ellipse valid, only retain P inside ≥ θ, and filter out the fittings that do not meet the specified conditions:

[0027]

[0028] where E is the ellipse contour; C is the cell contour; A C is the area; A E is the area of the ellipse;

[0029] Quantify the ratio between the major axis L and the minor axis l of the ellipse through an equation, set a ratio threshold, and perform a second - layer screening according to the shape characteristics of the ellipse; the specific equation is:

[0030]

[0031] Perform high-precision segmentation of overlapping areas of organoids and identify the edge contours of various organoids.

[0032] Preferably, in the multi-scale convolution module, convolution kernels of different sizes are used, specifically including 1×1 convolution kernel, 3×3 convolution kernel and 5×5 convolution kernel, to extract features of different receptive fields to obtain feature information of images at different scales.

[0033] Preferably, before extracting part of the organoid images in the organoid image dataset and inputting them into the multi-scale convolutional neural network model for training, the method further includes preprocessing the images of the organoid image dataset, wherein the preprocessing specifically includes randomly rotating the images, randomly cropping them, and performing limited adaptive histogram equalization.

[0034] The present invention also provides a system for counting organoids in bright field images, specifically comprising:

[0035] The model module is used to obtain a non-fluorescently labeled organoid bright field image data set to construct a training set; introduce maximum pooling and average pooling in the encoder of the multi-scale convolutional neural network model, and add a multi-scale branched convolution module after the pooling process to obtain an improved multi-scale convolutional neural network model; input the image data of the training set into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model.

[0036] The image segmentation module is used to input the bright field image of the organoid to be detected into the image segmentation model. In the image segmentation model, the encoder uses maximum pooling and average pooling to extract features respectively, splices the extracted features and inputs them into the multi-scale branched convolution module to obtain feature information of images at different scales, and enhances the image after feature extraction through the attention gate mechanism. The decoder decodes the feature-enhanced image and outputs the segmented organoid image.

[0037] The counting module uses an ellipse fitting algorithm to fit the organoid contours of the overlapping areas in the segmented organoid images to obtain a fitted organoid image, counts the number of organoids in the fitted organoid image and outputs an organoid counting result.

[0038] The present invention also provides a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps described in the method for counting organoids in a bright field image.

[0039] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is loaded by a processor, the steps described in the method for counting organoids in a bright field image can be executed.

[0040] The method for counting organoids in bright field images provided by the present invention has the following beneficial effects:

[0041] The image segmentation model of the present invention includes maximum pooling and average pooling, which are used to extract image features respectively, obtain image features from different angles, improve the sensitivity to changes in the edges of single organoids, and obtain feature information of images at different scales through a multi-scale branch convolution module after pooling. It is possible to fully capture the complex morphology of organoids and improve the segmentation accuracy of different sizes and shapes. The organoid contours of the overlapping areas in the segmented organoid images are fitted using an ellipse fitting algorithm to improve the accuracy of morphological analysis and counting of the overlapping areas of organoids, obtain the fitted organoid images, and output the counting results of the number of organoids in the fitted organoid images. Therefore, combining model processing and ellipse fitting methods can obtain more accurate organoid counting results in bright field images, providing data support for subsequent disease analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 The present invention is a flowchart of a method for counting organoids in a bright field image according to an embodiment of the present invention.

[0044] Figure 2 Schematic diagram of the overall structure of the overlapping area in the bright field image of the organoid in an embodiment of the present invention.

[0045] Figure 3 4 is a structural diagram of a multi-scale convolutional neural network model in an embodiment of the present invention.

[0046] Figure 4 Schematic diagram of the structure of the ellipse fitting module in the embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0048] Example

[0049] The present invention provides a method for counting organoids in bright field images, such as Figure 1 As shown, the specific steps include:

[0050] Step 1: Obtain a dataset of bright field images of non-fluorescently labeled organoids and perform preprocessing. The preprocessing of the dataset includes random rotation, random cropping, and limited adaptive histogram equalization of the images.

[0051] Step 2: Select some organoid image data from the processed organoid bright field image dataset as the training set, input the organoid images of the training set into the improved multi-scale convolutional neural network model for training, and obtain the image segmentation model and optimal parameters. Introduce maximum pooling and average pooling in the encoder of the multi-scale convolutional neural network model, and add a multi-scale branched convolution module after pooling to obtain an improved multi-scale convolutional neural network model, which can effectively encode image features and decode to obtain preliminary segmentation results.

[0052] S21. In the encoding stage, the input layer first performs a preliminary 3×3 convolution operation on the input image with a step size of 1. Then, the output of the convolution layer is normalized through the batch normalization (BN) layer to improve the stability of training. Finally, the ReLU activation function is applied. This step of convolution processes the low-level features of the image and provides a good foundation for subsequent more complex feature extraction. It is processed by the following equation:

[0053] I out =Relu(Conv 3×3 (I in ));

[0054] Among them, I in represents the input image, and I out Represents the output of a convolution operation. Within the same image, the density of organoids varies significantly, with some areas being dense and others being relatively sparse, and the uneven distribution of levels during acquisition results in organoids being at different depths.

[0055] Next, the extracted features are subjected to multi-scale convolution and pooling modules respectively. The pooling module combines the maximum pooling and average pooling operations to extract features from different perspectives. Maximum pooling captures the most salient features, especially emphasizing the edges of individual organoids in dense areas, while average pooling smoothes and integrates overall features, especially in sparser areas. By concatenating the outputs of maximum pooling and average pooling, the model retains key information and improves adaptability at different levels.

[0056] F i =Concat(Maxpool(M i ),Avgpool(M i ));

[0057] Among them, M iIt is the output of the previous convolution operation. This method of maximum pooling Maxpool and average pooling Avgpool enhances the segmentation performance in high-density and overlapping areas and increases the sensitivity to changes in the edges of individual organoids.

[0058] The multi-scale branched convolution module uses convolution kernels of different sizes to extract features of different receptive fields. The 1×1 convolution kernel captures detailed features, such as the edges and internal structures of individual organoids, and is suitable for heterogeneous morphology. The 3×3 convolution kernel focuses on medium-scale features, such as the spatial relationship between individual organoids, which helps to deal with areas with significant density changes. The 5×5 convolution kernel captures larger-scale overall features, solving the problem of inconsistent levels and dense feature extraction.

[0059] M(X) = Concat(Relu(Conv 1×1 (X)),Relu(Conv 3×3 (X)),Relu(Conv 5×5 (X)));

[0060] H(X)=Relu(BN(Conv 1×1 (M(X))));

[0061] The feature extraction of scales can fully capture the complex morphology of organoids, thereby improving the segmentation accuracy of different sizes and shapes. Subsequently, the feature map generated by the convolution operation is concatenated with the feature map generated by the multi-scale convolution. After obtaining the new features G(X), multi-scale convolution is applied in the next layer to concatenate the feature map of the unified convolution operation with the convolution map of the subsequent layer.

[0062] G(X)=Concat(I out ,H(X));

[0063] In the skip connections of encoding and decoding, an attention gate mechanism is introduced to dynamically filter and enhance the features of key areas. The output feature maps of each layer of the encoding part are integrated with the upsampled feature maps of the decoding part through attention gates. These gates dynamically adjust the weights, based on the importance of different input features, so that the model can focus on the key features of organoids, such as boundaries and internal structures. This dynamic weighting reduces the focus on background noise, thereby enhancing the focus on the area containing organoids.

[0064] psi=Sigmoid(Upsample(Conv 1×1 (g i ))+(Conv 1×1 (x i ));

[0065] Attention = x⊙psi;

[0066] Among them, x i represents the feature output of each layer of the encoding part, g i Represents the output after upsampling of the decoding part. This mechanism enables the model to emphasize salient organoid features during decoding and improve segmentation accuracy, especially in the case of blurred boundaries.

[0067] Then, the fused feature map is combined with the upsampled feature g i The convolutional layer is used to further extract features and enhance semantics, generating g i+1 :

[0068] g i+1 =Conv 3×3 (Concat(Upsample(g i );

[0069] To ensure that organoid features are fully recovered, segmentation feature maps are generated layer by layer during the decoding process. Each decoding layer produces feature maps with different levels of detail and semantic information, which are finally concatenated to form a comprehensive segmentation result.

[0070] Output = Conv 1×1 (Conv 3×3 (Concat(out1,out2,out3,out4)));

[0071] By gradually connecting the feature maps generated by each decoding layer, multi-scale and multi-level information fusion is achieved, significantly enhancing the model's global perception of organoid images. This information fusion strategy enables the model to more comprehensively understand and capture the internal structure and morphological characteristics of organoids, improving the consistency and stability of segmentation results.

[0072] Step 3: Use the remaining data in the organoid bright field image dataset as the test set, input the test set into the image segmentation model to obtain preliminary segmentation results, evaluate the segmentation performance of the model, and compare the segmentation results output by the model with the manually labeled results to verify accuracy and reliability.

[0073] Step 4: Based on the organoid segmentation image output by the model, an ellipse fitting algorithm is applied to process the overlapping area of ​​the organoids to obtain the counting results of the organoids in the bright field image.

[0074] S41. Based on the organoid segmentation image output by the model, edge detection is performed on the organoid to extract the edges of each organoid in the image. Edge detection is crucial to obtain the exact shape of each organoid and capture the outline of the organoid, thus providing the necessary data for subsequent concave point detection.

[0075] S42. Detect and locate the concave points formed at the contacts of various types of organoids in the segmented organoid images, and use the ellipse fitting method to fit the contour of each organoid; the concave points are usually the positions where two or more organoids contact each other.

[0076] Measure the distance from each intermediate point to the contour to determine its position relative to the contour. If all intermediate points meet this condition, then preliminarily classify this pixel as a concave point; repeat the measurement along the entire boundary to identify all potential concave points.

[0077] For any i-th point on the boundary, from this point P current and its two adjacent points P next , P prev constitute the vector pairs and to evaluate the local geometric properties. The size of the local depression is quantified by the reciprocal of the radius of curvature r; the radius of curvature r is specifically defined as:

[0078]

[0079] where K represents the curvature.

[0080] Set a maximum distance threshold to evaluate whether the distance d between two concave points is too large. By traversing the contour points within the specified index range and calculating the sum Length of the distances between adjacent points, if the threshold condition Length < d is satisfied, then retain the concave point with a larger curvature.

[0081]

[0082] where, dist(P i , P i+1 ) represents the Euclidean distance between point P i and point P i+1 , that is:

[0083] The least squares method is used to achieve the ellipse fitting between adjacent concave points to approximate the complete contour shape of each cell:

[0084] ax 2 + bxy + cy 2 + dx + ey + f = 0;

[0085] where, a, b, c determine the shape and orientation of the ellipse, b is related to the rotation of the ellipse; d, e control the translation of the ellipse on the x and y axes; f is a constant term that affects the overall position of the ellipse. Satisfy (b 2-4ac)<0. In the method based on concave point detection, two concave points are first selected as the key points of the ellipse, and the equation is constructed by combining the three points on the edge line between the two concave points.

[0086] S43. Separate the overlapping regions of organoids by constructing a fitted elliptical curve and count the overlapping organoids.

[0087] Calculate the intersection area A between the ellipse outline and the outline of a single organoid E ∩A C , evaluate whether the ellipse is completely surrounded by the organoid outline, establish a threshold θ, consider the ellipse valid, and only retain P inside ≥θ, filter out the fits that do not meet the specified conditions:

[0088]

[0089] Among them, E is the ellipse outline; C is the cell outline; A C is the area; A E is the area of ​​the ellipse.

[0090] The ratio between the major axis L and the minor axis l of the ellipse is quantified by the equation, the ratio threshold is set, and the second-level screening is performed according to the shape characteristics of the ellipse; the specific equation is:

[0091]

[0092] Perform high-precision segmentation of organoid regions, identify the edges of various organoids, and output organoid counting results.

[0093] The present invention also provides a system for counting organoids in bright field images, specifically comprising:

[0094] The model module is used to obtain a non-fluorescently labeled organoid bright field image dataset to construct a training set; introduce maximum pooling and average pooling in the encoder of the multi-scale convolutional neural network model, and add a multi-scale branched convolution module after the pooling process to obtain an improved multi-scale convolutional neural network model; input the image data of the training set into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model.

[0095] The image segmentation module is used to input the bright field image of the organoid to be detected into the image segmentation model. In the image segmentation model, the encoder uses maximum pooling and average pooling to extract features respectively, concatenates the extracted features and inputs them into the multi-scale branched convolution module to obtain the feature information of images at different scales, and enhances the image after feature extraction through the attention gate mechanism. The decoder decodes the feature-enhanced image and outputs the segmented organoid image.

[0096] The counting module uses an ellipse fitting algorithm to fit the organoid contours in the overlapping areas of the segmented organoid images to obtain the fitted organoid images, counts the number of organoids in the fitted organoid images and outputs the organoid counting results.

[0097] Each module in the above-mentioned bright field image organoid counting system can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0098] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps in an embodiment of a method for counting organoids in a bright field image. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0099] Furthermore, the present invention also provides a non-temporary computer-readable storage medium containing instructions, and a computer program is stored on the storage medium. For example, a memory containing instructions, the above instructions can be executed by a processor of a computer device to complete the above method. For example, a non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a tape, a floppy disk, and an optical data storage device. When the computer program is executed by the processor, it can implement the steps in an embodiment of a method for counting organoids in a bright field image. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0100] It will be appreciated by those skilled in the art that embodiments of the present invention may provide methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0104] It should be pointed out that the specific implementation methods described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although the invention has been described in detail in this specification and embodiments, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the protection scope of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved. Any simple change or equivalent replacement of the technical solution that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belongs to the protection scope of the present invention.

Claims

1. A method for counting organoids in bright field images, characterized in that: Including the following steps: Obtain a dataset of bright-field images of non-fluorescently labeled organoids to construct a training set; introduce max pooling and average pooling in the encoder of the multi-scale convolutional neural network model, and add a multi-scale branch convolutional module after the pooling process to obtain an improved multi-scale convolutional neural network model; input the image data of the training set into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model; Input the bright-field image of the organoid to be detected into the image segmentation model. In the encoder of the image segmentation model, max pooling and average pooling are used to extract features respectively. The extracted features are concatenated and input into the multi-scale branch convolutional module to obtain the feature information of the image at different scales. The image after feature extraction is feature-enhanced through the attention gate mechanism, and the decoder decodes the feature-enhanced image to output the segmented organoid image; Use the ellipse fitting algorithm to fit the organoid contours in the overlapping regions of the segmented organoid image to obtain the fitted organoid image, and count the number of organoids in the fitted organoid image and output the organoid counting result.

2. The method for counting organoids in a bright field image according to claim 1, characterized in that: The step of using the ellipse fitting algorithm to fit the organoid contours in the overlapping regions of the segmented organoid image to obtain the fitted organoid image is specifically as follows: Perform edge detection on the segmented organoid image to extract the edges of each organoid in the image; Locate the concave points formed at the contact points of each organoid in the segmented organoid image according to the edge detection result; the concave points are the positions where two or more organoids are in contact; Use the ellipse fitting method to fit the contours of each organoid, and construct a fitted ellipse curve to separate the overlapping regions of the organoids to obtain the fitted organoid image.

3. The method for counting organoids in a bright field image according to claim 2, characterized in that: The step of locating the concave points formed at the contact points of each organoid in the segmented organoid image according to the edge detection result is specifically as follows: Measure the distance from each intermediate point to the contour to determine its position relative to the contour. If all intermediate points meet this condition, then preliminarily classify this pixel as a concave point; repeat the measurement along the entire boundary to identify all potential concave points; For the i-th point on the boundary, the point P current and its two neighboring points P next , P prev The vector pair and The local geometric properties are evaluated and the size of the local depression is quantified by the inverse of the radius of curvature r; the radius of curvature r is specifically defined as: Where K represents curvature; Set a maximum distance threshold to evaluate the distance d between two concave points. By traversing the contour points in the specified index range and calculating the sum Length of the distances between adjacent points, if the threshold condition Length < d is met, then retain the concave point with a larger curvature; Among them, dist(P i ,P i+1 ) indicates P i Point and P i+1 The Euclidean distance between points is: The least squares method is used to achieve the ellipse fitting between adjacent concave points to approximate the complete contour shape of each cell: ax 2 +bxy+cy 2 +dx+ey+f=0; Among them, a, b, c determine the shape and direction of the ellipse, b involves the rotation of the ellipse; d, e control the translation of the ellipse on the x and y axes; f is a constant term; when (b 2 -4ac)<0, the equation represents an ellipse.

4. The method for counting organoids in a bright field image according to claim 2, characterized in that: The step of using the ellipse fitting method to fit the contours of each organoid and construct a fitted ellipse curve to separate the overlapping regions of the organoids is specifically as follows: Calculate the intersection area A between the ellipse outline and the outline of a single organoid E ∩A C , evaluate whether the ellipse is completely surrounded by the organoid outline, establish a threshold θ, consider the ellipse valid, and only retain P inside ≥θ, filter out the fits that do not meet the conditions: Among them, E is the ellipse outline; C is the cell outline; A C is the area; A E is the area of ​​the ellipse; Quantify the ratio between the major axis L and the minor axis l of the ellipse through an equation, set a ratio threshold, and perform a second-level screening according to the ellipse shape characteristics; the equation is specifically: Perform high-precision segmentation on the overlapping regions of the organoids to identify the edge contours of each organoid.

5. The method for counting organoids in a bright field image according to claim 1, characterized in that: In the multi-scale convolution module, convolution kernels of different sizes are used, including 1×1 convolution kernel, 3×3 convolution kernel and 5×5 convolution kernel, to extract features of different receptive fields, so as to obtain feature information of images at different scales.

6. The method for counting organoids in bright field images according to claim 1, characterized in that: Before extracting some organoid images from the organoid image dataset and inputting them into the multi-scale convolutional neural network model for training, the method also includes preprocessing the images of the organoid image dataset, and the preprocessing specifically includes randomly rotating the images, randomly cropping them, and performing limited adaptive histogram equalization.

7. A system for counting organoids in bright field images, characterized in that: include: A model module is used to obtain a non-fluorescently labeled organoid bright field image data set to construct a training set; introduce maximum pooling and average pooling into the encoder of the multi-scale convolutional neural network model, and add a multi-scale branched convolution module after the pooling process to obtain an improved multi-scale convolutional neural network model; input the image data of the training set into the improved multi-scale convolutional neural network model for training to obtain an image segmentation model; An image segmentation module is used to input the bright field image of the organoid to be detected into the image segmentation model, in which the encoder uses maximum pooling and average pooling to extract features respectively, splices the extracted features and inputs them into the multi-scale branched convolution module to obtain feature information of images at different scales, enhances the image after feature extraction through the attention gate mechanism, and decodes the feature-enhanced image by the decoder to output the segmented organoid image; The counting module uses an ellipse fitting algorithm to fit the organoid contours of the overlapping areas in the segmented organoid images to obtain a fitted organoid image, counts the number of organoids in the fitted organoid image and outputs an organoid counting result.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is loaded into a processor, it can execute the steps of the method according to any one of claims 1 to 6.

Images