Kidney organoid differentiation degree quantitative model construction method and device, medium and program product
By using machine learning models to assess the differentiation degree of kidney organoids and calculating the ratio of renal tubular to non-tubular area, the problem of insufficient assessment methods in existing technologies is solved, and the accurate quantification of the differentiation degree of kidney organoids and batch-to-batch stability comparison are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-24
- Publication Date
- 2026-03-27
AI Technical Summary
Existing technologies lack efficient and non-invasive assessment methods, which hinders the optimization of kidney organoid induction processes, lacks quality control standards, and makes it difficult to quickly and accurately assess the differentiation level of organoids and their stability between individuals and batches.
By extracting features from morphological photographs using a machine learning model, the area ratio of renal tubular-like structural regions to non-renal tubular-like structural regions is calculated, and a quantitative model for the differentiation degree of kidney organoids is established to assess the differentiation degree of kidney organoids.
It provides a unified and objective evaluation standard, eliminates the error of subjective human judgment, ensures the accuracy and repeatability of the evaluation, and lays the foundation for accurate comparison of induced stability between different batches.
Smart Images
Figure CN121746367A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent assisted medical care, and more specifically, to a method, device, medium, and program product for constructing a quantitative model of the degree of differentiation of kidney organoids. Background Technology
[0002] Kidney damage caused by various pathogenic factors and its subsequent progression into chronic kidney disease and end-stage renal disease have become a major global health problem, imposing a heavy economic and medical burden on society. A primary prerequisite for in-depth exploration of the pathogenesis of kidney diseases and the development of effective therapeutic drugs is the establishment of research models that can highly simulate the pathophysiological characteristics of the human kidney. The emergence of human pluripotent stem cells offers a possibility for solving this problem. Compared with traditional two-dimensional cell models, kidney organoids possess a three-dimensional spatial structure that more closely resembles the physiological state. However, to translate this potential into practical industrial applications, the key lies in achieving efficient, stable, and large-scale preparation of kidney organoids. A prerequisite and fundamental requirement for achieving this goal is to establish a technical method that can rapidly and accurately assess the induction efficiency of kidney organoids. Unfortunately, the field currently lacks a simple, high-throughput, and non-invasive assessment method for rapidly evaluating the differentiation level of organoids and their stability between individuals and batches. This severely hinders the optimization of the induction process and the establishment of quality control standards. Summary of the Invention
[0003] This invention aims to solve at least one of the aforementioned technical problems. This invention is a method for constructing a quantitative model of kidney organoid differentiation. First, a machine learning model is used to extract features from morphological photographs to obtain the area of renal tubular-like structures and the area of non-renal tubular-like structures. Then, the ratio of the area of the renal tubular-like structures to the sum of the areas of the renal tubular-like structures and the non-renal tubular-like structures is calculated to obtain a quantitative model of kidney organoid differentiation, which is used to assess the degree of kidney organoid differentiation.
[0004] This invention discloses a method for quantitatively assessing the differentiation degree of kidney organoids, comprising: S1: Obtain a morphological photograph of the kidney organoid differentiation process on day N, where N is 9-16; S2: Input the morphological photograph taken on day N into a machine learning model for segmentation, and obtain renal tubule-like structural regions and non-renal tubule-like structural regions; Furthermore, the specific process of constructing the machine learning model is as follows: Morphological images of kidney organoids were collected on day N of differentiation. The morphological images on day N were labeled by manually dividing them into tubular-like structures and non-tubular-like structures and using them as labels. The morphological photograph taken on day N is input into a machine learning model for training to obtain segmented images; Compare the segmented image with the label and construct a loss function; Optimize the loss function to obtain a well-trained machine learning model.
[0005] Furthermore, S2 also includes performing data augmentation on the morphological photograph of the Nth day to output an enhanced morphological photograph, and inputting the enhanced morphological photograph into a machine learning model for segmentation.
[0006] Preferably, N is 12.
[0007] Optionally, the data augmentation operation includes one or more of the following operations: affine transformation, Gaussian blur, geometric transformation, adding or reducing color channels, GAN, and AutoAugment.
[0008] Furthermore, between S1 and S2, the method further includes: performing contrast enhancement processing on the morphological photograph taken on day N, specifically the following steps: S11: Divide the morphological photograph of day N into P small blocks, where P is a natural integer. Apply histogram equalization to the P small blocks and then perform interpolation to obtain P contrast-enhanced images. S12: Input the P contrast-enhanced images into a machine learning model for segmentation to obtain renal tubule-like structural regions and non-renal tubule-like structural regions.
[0009] Furthermore, the machine learning model includes one or more of the following: FPN model, U-Net, PSPNet, DeepLab V3+, SegFormer, HRNet.
[0010] S2 specifically involves: inputting the morphological photograph taken on day N into the ResNeXt-101 backbone network for feature extraction to obtain an image feature map; then inputting the image feature map into the FPN model to obtain a multi-scale feature map; and combining the multi-scale feature maps to obtain renal tubular similar structure regions and non-renal tubular similar structure regions. Preferably, the ResNeXt-101 backbone network comprises M layers, where M is a natural integer; The specific steps for inputting the morphological photograph taken on day N into the ResNeXt-101 backbone network for feature extraction to obtain the image feature map are as follows: The morphological photograph taken on day N is input into the first layer of the ResNeXt-101 backbone network for feature extraction to obtain the first image feature map. The first image feature map is input into the second layer of the ResNeXt-101 backbone network for feature extraction to obtain the second image feature map; the process is repeated M times to obtain the Mth image feature map, resulting in a total of M image feature maps. Optionally, the FPN model includes M layers. Specifically, inputting the image feature map into the FPN model to obtain a multi-scale feature map, and obtaining renal tubule-like and non-renal tubule-like structural regions, involves: The Mth image feature map is input into the first layer of the FPN model to perform multi-scale feature extraction to obtain the first FPN feature map. The first FPN feature map and the Mth image feature map are element-wise added to obtain the Mth scale feature map. The (M-1)th image feature map is input into the FPN model for multi-scale feature extraction to obtain the second FPN feature map. The second FPN feature map is then added element-wise to the M-th scale feature map to obtain the (M-1)th scale feature map. This process is repeated M times to obtain the first scale feature map. The renal tubule-like and non-renal tubule-like structural regions are obtained by statistical calculation of the first-scale feature map.
[0011] S3: Calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, input the ratio into the classifier, and determine the degree of differentiation of the kidney organoids based on the output of the classifier.
[0012] Optionally, S3 can be replaced with S3': Calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, and if the ratio is greater than a threshold, the differentiation is good; otherwise, the differentiation is poor.
[0013] Furthermore, the threshold is specifically defined as the value corresponding to the optimal equilibrium point selected from the ROC curve based on the degree of differentiation of the kidney organoids.
[0014] Optionally, the classifier may include one or more of the following: logistic regression, random forest, support vector machine, XGboost, decision tree, and extreme learning machine.
[0015] The present invention also discloses a computer device, the device comprising: a memory and a processor; The memory is used to store program instructions; The processor is used to invoke program instructions, and when the program instructions are executed, the method steps are implemented.
[0016] The present invention also discloses a computer program product, including a computer program that is implemented by a processor to perform the method steps.
[0017] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method steps.
[0018] Advantages of this invention: This invention addresses the core problem in existing technologies—the lack of efficient, non-invasive assessment methods that hinders the optimization of kidney organoid induction processes and results in a lack of quality control standards—by providing a quantitative assessment method for kidney organoid differentiation. The core implementation details involve: extracting and segmenting image features from morphological photographs using machine learning methods, then statistically determining the area of renal tubular-like structures and the area of non-renal tubular-like structures. The ratio of these areas is then calculated to determine the degree of kidney organoid differentiation. By transforming complex morphological features into quantitative, repeatable data indicators, this invention fundamentally eliminates errors caused by subjective human judgment, establishes a unified and objective assessment standard, and lays the foundation for accurate comparison of induction stability between different batches. Attached Figure Description
[0019] 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.
[0020] Figure 1 This is a schematic flowchart of the method for quantitatively assessing the differentiation degree of kidney organoids provided in an embodiment of the present invention; Figure 2 This is a diagram of the segmentation model framework based on ResNeXt-101-FPN provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the ResNeXt-101 network hierarchy provided in an embodiment of the present invention; Figure 4 This is a comparative schematic diagram of the segmentation effects of different models provided in the embodiments of the present invention; Figure 5 This is a quantitative statistical diagram of the renal tubular structure in kidney organoids labeled on day 12, provided by an embodiment of the present invention; Figure 6 This is a graph showing the results of quantitative statistics on the proportion of renal tubular structure area in kidney organoids on day 12, provided by an embodiment of the present invention. Figure 7 This is a schematic diagram of a computing device for quantitatively assessing the degree of differentiation of kidney organoids provided in an embodiment of the present invention. Detailed Implementation
[0021] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.
[0022] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be performed in the order they appear herein, or they may be performed in parallel. The operation numbers, such as S1, S2, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] Figure 1 This is a schematic flowchart of a method for quantitatively assessing the differentiation degree of kidney organoids according to an embodiment of the present invention. Specifically, it includes the following steps: S1: Obtain a morphological photograph of the kidney organoid differentiation process on day N, where N is 9-16; S2: Input the morphological photograph taken on day N into a machine learning model for segmentation, and obtain renal tubule-like structural regions and non-renal tubule-like structural regions; Furthermore, the specific process of constructing the machine learning model is as follows: Morphological images of kidney organoids were collected on day N of differentiation. The morphological images on day N were labeled by manually dividing them into tubular-like structures and non-tubular-like structures and using them as labels. The morphological photograph taken on day N is input into a machine learning model for training to obtain segmented images; Compare the segmented image with the label and construct a loss function; Optimize the loss function to obtain a well-trained machine learning model.
[0025] In one specific embodiment, the acquisition of a morphological photograph on day N of kidney organoid differentiation, and the labeling of the morphological photograph on day N, specifically involves: In one embodiment, morphological images were acquired using Celldiscoverer 7 on day 12 of kidney organoid differentiation. A dataset was constructed by annotating 62 organoid images on day 12 using Labelme to assess the proportion of tubular structures in developing organoids. The annotations were performed manually to divide the images into two parts: "renal tubule-like structure regions" and "non-renal tubule-like structure regions". The annotated images were then subjected to data augmentation to output morphological images of 1024×1024 pixels.
[0026] Furthermore, S2 also includes performing data augmentation on the morphological photograph of the Nth day to output an enhanced morphological photograph, and inputting the enhanced morphological photograph into a machine learning model for segmentation.
[0027] In one specific embodiment, N is 12.
[0028] Optionally, the data augmentation operation includes one or more of the following operations: affine transformation, Gaussian blur, geometric transformation, adding or reducing color channels, GAN, and AutoAugment.
[0029] Furthermore, between S1 and S2, the method further includes: performing contrast enhancement processing on the morphological photograph taken on day N, specifically the following steps: S11: Divide the morphological photograph of day N into P small blocks, where P is a natural integer. Apply histogram equalization to the P small blocks and then perform interpolation to obtain P contrast-enhanced images. In one specific embodiment, the morphological photographs were acquired by inducing differentiation of human pluripotent stem cells (hPSCs) using a culture medium.
[0030] S12: Input the P contrast-enhanced images into a machine learning model for segmentation to obtain renal tubule-like structural regions and non-renal tubule-like structural regions.
[0031] Optionally, the machine learning model includes one or more of the following: FPN model, U-Net, PSPNet, DeepLab V3+, SegFormer, HRNet.
[0032] In a specific embodiment, S2 specifically involves: inputting the morphological photograph of day N into the ResNeXt-101 backbone network for feature extraction to obtain an image feature map; then inputting the image feature map into the FPN model to obtain a multi-scale feature map; and combining the multi-scale feature maps to obtain renal tubular similar structure regions and non-renal tubular similar structure regions.
[0033] Figure 2 This is a framework diagram of a segmentation model based on ResNeXt-101-FPN provided in an embodiment of the present invention. Specifically, the construction process of the segmentation model is as follows: Obtain morphological images of kidney organoids; The adjusted image is obtained by scaling the renal tracheal morphology photograph and then performing an affine transformation. The adjusted image is then enhanced using the CLAHE algorithm and standardized to obtain the enhanced image feature map. The enhanced image feature map is input into the ResNeXt101 network for downsampling to obtain M feature maps of different scales, where M is a natural integer. The M feature maps at different scales are input into the FPN network for upsampling to obtain M upsampled feature maps; the Mth feature map is iteratively weighted and summed with the (M-1)th feature map to obtain the segmented image.
[0034] In one specific embodiment, the principle of the CLAHE algorithm is to divide the image into P non-overlapping small blocks; calculate the histogram of each small block; calculate histogram equalization for each small block and introduce a limiting parameter to control the degree of contrast enhancement; if the histogram of a certain small block exceeds this limit, then the contrast enhancement will be limited to avoid over-enhancement.
[0035] In one specific embodiment, the ResNeXt-101 backbone network includes M layers, where M is a natural integer; The specific steps for inputting the morphological photograph taken on day N into the ResNeXt-101 backbone network for feature extraction to obtain the image feature map are as follows: The morphological photograph taken on day N is input into the first layer of the ResNeXt-101 backbone network for feature extraction to obtain the first image feature map. The first image feature map is input into the second layer of the ResNeXt-101 backbone network for feature extraction to obtain the second image feature map; the process is repeated M times to obtain the Mth image feature map, resulting in a total of M image feature maps.
[0036] Figure 3This is a schematic diagram of the layer distribution of the ResNeXt-101 network provided in an embodiment of the present invention. Specifically, the ResNeXt-101 network refers to a ResNet network with 101 layers and a cardinality of 32. Each step consists of multiple ResNeXt modules, and the total number of layers is equal to one one-dimensional convolutional layer. Two-dimensional convolution, Layer 3D convolution, Four-dimensional convolution, The sum of five-dimensional convolutional layers and one pooling layer.
[0037] In a specific embodiment, the FPN model includes M layers. The step of inputting the image feature map into the FPN model to obtain a multi-scale feature map, and obtaining renal tubule-like structural regions and non-renal tubule-like structural regions, specifically involves: The Mth image feature map is input into the first layer of the FPN model to perform multi-scale feature extraction to obtain the first FPN feature map. The first FPN feature map and the Mth image feature map are element-wise added to obtain the Mth scale feature map. The (M-1)th image feature map is input into the FPN model for multi-scale feature extraction to obtain the second FPN feature map. The second FPN feature map is then added element-wise to the M-th scale feature map to obtain the (M-1)th scale feature map. This process is repeated M times to obtain the first scale feature map. The renal tubule-like and non-renal tubule-like structural regions are obtained by statistical calculation of the first-scale feature map.
[0038] Figure 4 This is a comparative diagram of the segmentation performance of different models provided in this embodiment of the invention. Comparative experiments evaluated various semantic segmentation models, including U-Net, FPN, PSPNet, and DeepLab V3+. The loss function was obtained by calculating the weighted sum of IoU loss, Dice loss, and BCE loss, and the loss function was iteratively optimized to output the optimal segmentation result. Figure 4 As shown, the FPN model (in bold) performs best in the evaluation metrics of mean intersection-over-union ratio (mIoU) and Dice, achieving an mIoU of 66.4% and a Dice value of 79.5%. This is because FPN allows the model to not only understand the "overall picture" (macrostructure) of the image, but also pay attention to the "local details" (microscopic boundaries), which is beneficial for accurately segmenting the boundaries of small renal tubules. Therefore, it was selected as the final segmentation model.
[0039] S3: Calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, input the ratio into the classifier, and determine the degree of differentiation of the kidney organoids based on the output of the classifier.
[0040] In one embodiment, the classifier includes one or more of the following: logistic regression, random forest, support vector machine, XGboost, decision tree, and extreme learning machine.
[0041] Preferably, the classifier is a decision tree model.
[0042] In one embodiment, the specific process by which the classifier obtains the degree of differentiation is as follows: The ratio is input into the decision tree model, and the decision path is traversed from the root node until the leaf node is reached. Based on the classification results corresponding to the leaf nodes, the differentiation level of the kidney organoid is output.
[0043] In one embodiment, S3 is replaced by S3': calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, and if the ratio is greater than a threshold, the differentiation is good; otherwise, the differentiation is poor.
[0044] In a specific embodiment, the threshold is specifically: the value corresponding to the optimal balance point selected from the ROC curve based on the degree of differentiation of the kidney organoids is used as the threshold.
[0045] In one embodiment, the data augmentation operation includes one or more of the following operations: affine transformation, Gaussian blur, geometric transformation, adding or reducing color channels, GAN, and AutoAugment.
[0046] S2: Calculate the area of the renal tubule-like structure region and the area of the non-renal tubule-like structure region in the segmentation results; calculate the ratio of the area of the renal tubule-like structure region to the sum of the areas of the renal tubule-like structure region and the non-renal tubule-like structure region to obtain a quantitative model of the degree of differentiation of kidney organoids.
[0047] Figure 5This is a quantitative statistical diagram of renal tubular structure in kidney organoids labeled on day 12, provided in an embodiment of the present invention. Ctrl-Orgs represents kidney organoid products without optimization treatment; ANX-Orgs represents kidney organoid products treated with optimized conditions during differentiation. Dark red represents renal tubular structures, and green represents non-tubular structures. The scale bar is 30 µm. The optimized conditions refer to the additional treatment of kidney organoids with 200 ng / mL annexin A2 during day 6 to day 16 of the Ctrl-Orgs induction process.
[0048] Figure 6 The figure shows the results of quantitative statistics on the proportion of renal tubular structure area in kidney organoids on day 12, as provided in this embodiment of the invention. It can be seen that the proportion of renal tubular structure area in the unoptimized kidney organoid products is lower than that in the optimized kidney organoid products, indicating that the degree of differentiation of kidney organoids is significantly improved after optimization.
[0049] The present invention also discloses a computer device, the device comprising: a memory and a processor, such as... Figure 7 As shown: The memory is used to store program instructions; The processor is used to invoke program instructions, and when the program instructions are executed, the method steps are implemented.
[0050] The present invention provides a computer program product, including a computer program, which is implemented by a processor to implement the method steps.
[0051] The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method steps.
[0052] The present invention also discloses a computer system, the system comprising: Acquisition Unit: Used to acquire morphological images of kidney organoids on day N during differentiation, where N is 9-16; Segmentation unit: used to input the morphological photograph of day N into the machine learning model for segmentation, and to obtain renal tubule-like structural regions and non-renal tubule-like structural regions; Output unit: used to count the area of the renal tubule-like structure region and the area of the non-renal tubule-like structure region; calculate the ratio of the area of the renal tubule-like structure region to the sum of the areas of the renal tubule-like structure region and the non-renal tubule-like structure region, input the ratio into the classifier, and determine the degree of differentiation of kidney organoids based on the output of the classifier.
[0053] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0054] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.
[0055] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0056] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The aforementioned integrated modules can be implemented in hardware or as software functional modules.
[0057] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0058] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0059] The computer device provided by the present invention has been described in detail above. For those skilled in the art, there will be changes in the specific implementation and application scope based on the ideas of the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for quantitatively assessing the differentiation degree of kidney organoids, characterized in that, include: S1: Obtain a morphological photograph of the kidney organoid differentiation process on day N, where N is 9-16; S2: Input the morphological photograph taken on day N into a machine learning model for segmentation, and obtain renal tubule-like structural regions and non-renal tubule-like structural regions; S3: Calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, input the ratio into the classifier, and determine the degree of differentiation of the kidney organoids based on the output of the classifier.
2. The method for quantitatively assessing the differentiation degree of kidney organoids according to claim 1, characterized in that, S3 can be replaced by S3': Calculate the first area of the renal tubule-like structure region and the second area of the non-renal tubule-like structure region; calculate the sum of the first area and the second area to obtain the total area; calculate the ratio of the first area to the total area, and if the ratio is greater than a threshold, the differentiation is good; otherwise, the differentiation is poor. The threshold is specifically defined as the value corresponding to the optimal equilibrium point selected from the ROC curve based on the degree of differentiation of the kidney organoids.
3. The method for quantitatively assessing the differentiation degree of kidney organoids according to claim 1, characterized in that, The classifiers include one or more of the following: logistic regression, random forest, support vector machine, XGboost, decision tree, and extreme learning machine.
4. The method for quantitatively assessing the differentiation degree of kidney organoids according to claim 1, characterized in that, The specific process of constructing the machine learning model is as follows: Morphological images of kidney organoids were collected on day N of differentiation. The morphological images on day N were labeled by manually dividing them into tubular-like structures and non-tubular-like structures and using them as labels. The morphological photograph taken on day N is input into a machine learning model for training to obtain segmented images; Compare the segmented image with the label and construct a loss function; Optimize the loss function to obtain a well-trained machine learning model.
5. The method for quantitatively assessing the differentiation degree of kidney organoids according to claim 1, characterized in that, S2 further includes performing data augmentation on the morphological photograph of the Nth day to output an enhanced morphological photograph, and inputting the enhanced morphological photograph into a machine learning model for segmentation; Preferably, N is 12; Optionally, the data augmentation operation includes one or more of the following operations: affine transformation, Gaussian blur, geometric transformation, adding or reducing color channels, GAN, and AutoAugment.
6. The method for quantitatively assessing the differentiation degree of kidney organoids according to claim 1, characterized in that, Between S1 and S2, the method further includes: performing contrast enhancement processing on the morphological photograph of day N, specifically the following steps: S11: Divide the morphological photograph of day N into P small blocks, where P is a natural integer. Apply histogram equalization to the P small blocks and then perform interpolation to obtain P contrast-enhanced images. S12: Input the P contrast-enhanced images into a machine learning model for segmentation to obtain renal tubule-like structural regions and non-renal tubule-like structural regions; Optionally, the machine learning model includes one or more of the following: FPN model, U-Net, PSPNet, DeepLab V3+, SegFormer, HRNet.
7. The method for quantitatively assessing the degree of differentiation of kidney organoids according to claim 1, characterized in that, S2 specifically involves: inputting the morphological photograph taken on day N into the ResNeXt-101 backbone network for feature extraction to obtain an image feature map; then inputting the image feature map into the FPN model to obtain a multi-scale feature map; and combining the multi-scale feature maps to obtain renal tubular similar structure regions and non-renal tubular similar structure regions. Preferably, the ResNeXt-101 backbone network comprises M layers, where M is a natural integer; The specific steps for inputting the morphological photograph taken on day N into the ResNeXt-101 backbone network for feature extraction to obtain the image feature map are as follows: The morphological photograph taken on day N is input into the first layer of the ResNeXt-101 backbone network for feature extraction to obtain the first image feature map. The first image feature map is input into the second layer of the ResNeXt-101 backbone network for feature extraction to obtain the second image feature map; the process is repeated M times to obtain the Mth image feature map, resulting in a total of M image feature maps. Optionally, the FPN model includes M layers. Specifically, inputting the image feature map into the FPN model to obtain a multi-scale feature map, and obtaining renal tubule-like and non-renal tubule-like structural regions, involves: The Mth image feature map is input into the first layer of the FPN model to perform multi-scale feature extraction to obtain the first FPN feature map. The first FPN feature map and the Mth image feature map are element-wise added to obtain the Mth scale feature map. The (M-1)th image feature map is input into the FPN model for multi-scale feature extraction to obtain the second FPN feature map. The second FPN feature map is then added element-wise to the M-th scale feature map to obtain the (M-1)th scale feature map. This process is repeated M times to obtain the first scale feature map. The renal tubule-like and non-renal tubule-like structural regions are obtained by statistical calculation of the first-scale feature map.
8. A computer device, comprising: Memory and processor; The memory is used to store program instructions; The processor is used to invoke program instructions, which, when executed, implement the method steps of any one of claims 1-7.
9. A computer program product, comprising a computer program, characterized in that, The computer program is implemented by a processor using the method steps described in any one of claims 1-7.
10. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method described in any one of claims 1-7.