Grape tire slice image analysis method and device based on decision tree model

Abstract

The invention discloses a grape tire slice image analysis method and device based on a decision tree model. The method comprises the following steps: A, constructing a distribution map and a feature extraction module; b, extracting a feature vector of the grape embryo slice image, and obtaining a villus, edema and hyperplasia distribution diagram of the grape embryo slice image; c, acquiring a distribution map of edema villus, hyperplasia villus, normal villus and abnormal villus and a proportion of edema villus, hyperplasia villus and abnormal villus; d, performing feature fusion to obtain a final feature vector; e, constructing and training a decision tree model; and F, analyzing the new pathological image by using the decision tree model to generate a pathological analysis report. The method provided by the invention assists a clinician in more efficient case screening, and solves the problem of low clinical diagnosis and detection efficiency of grape embryos caused by manual observation of pathological characteristics of section tissues in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and particularly to a method and device for analyzing hydatidiform mole section images based on a decision tree model. Background Art

[0002] Hydatidiform mole (HM) is a pregnancy disease characterized by abnormal proliferation of placental cells, forming grape-like vesicles. The diagnosis of hydatidiform mole mainly relies on histopathological examination, that is, microscopic observation and analysis of placental tissue sections. However, traditional pathological examination methods have some limitations. First, this method is time-consuming and highly subjective, requiring professional pathologists for diagnosis, and there may be diagnostic differences among different doctors. Second, due to the limited number of pathologists, this method is difficult to apply to large-scale screening and early screening.

[0003] To solve these problems, in recent years, some researchers have begun to explore the use of computer vision and machine learning technologies to automatically identify and analyze hydatidiform mole section images. These methods mainly include three steps: image preprocessing, feature extraction, and image segmentation. In the image preprocessing step, it is usually necessary to perform operations such as enhancement and denoising on the section images to improve the quality and analyzability of the images. In the feature extraction step, information that can reflect the characteristics of hydatidiform mole, such as color, texture, and shape, needs to be extracted from the preprocessed images. Finally, in the image segmentation step, machine learning algorithms are used to segment the extracted features and determine whether it is a hydatidiform mole.

[0004] Although these methods have achieved certain results, there are still some problems and challenges. For example, due to the complexity and variability of hydatidiform mole section images, some methods are difficult to effectively extract useful information in the preprocessing and feature extraction steps, and the segmentation accuracy and robustness of some machine learning algorithms also need to be improved. In addition, there is currently a lack of analysis methods and devices for the results of hydatidiform mole pathological image processing, and doctors still need to further judge the results of pathological image processing algorithms based on experience. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for analyzing hydatidiform mole section images based on a decision tree model, which can automatically identify, segment, and analyze the lesions in hydatidiform mole pathological images, thereby assisting clinicians to perform case screening more efficiently and solving the problem of low efficiency of clinical diagnosis and detection of hydatidiform mole caused by manually observing the pathological features of section tissues in the prior art.

[0006] The present invention adopts the following technical solutions:

[0007] A method for analyzing hydatidiform mole section images based on a decision tree model includes the following steps:

[0008] A: Construct a distribution map and feature extraction module that includes a villus distribution extraction sub-module, an edema distribution extraction sub-module, and a hyperplasia distribution extraction sub-module; each sub-module includes an image encoding large model and a corresponding task head segmentation network;

[0009] B: Use the distribution map and feature extraction module to extract the feature vectors of the hydatidiform mole section images, and obtain the villus distribution map, edema distribution map, and hyperplasia distribution map of the hydatidiform mole section images; the feature vectors include villus feature vectors, edema feature vectors, and hyperplasia feature vectors;

[0010] C: Input the villus distribution map, edema distribution map, and hyperplasia distribution map into the distribution map fusion analysis module, and output the edema-villus distribution map, hyperplasia-villus distribution map, normal villus distribution map, abnormal villus distribution map, proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi of the hydatidiform mole section;

[0011] D: Perform feature fusion on the villus feature vectors, edema feature vectors, and hyperplasia feature vectors, integrate them into a fused feature vector, and merge the proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi as feature values into the fused feature vector to obtain the final feature vector;

[0012] E: Construct a decision tree model for classifying the pathological conditions based on the final feature vector, and use the final feature vector to train the decision tree model to obtain a trained decision tree model;

[0013] F: Use the trained decision tree model to analyze new pathological images, classify them according to the feature vectors of the lesions in the images, and generate a pathological analysis report, which includes the abnormal villus distribution map, the number of lesions, and features.

[0014] Step A includes:

[0015] A1: Collect and organize a set of hydatidiform mole section images with villus distribution annotation, edema distribution annotation, and hyperplasia distribution annotation, and construct an image training set and a test set;

[0016] A2: Construct an image encoding large model and perform pre-training;

[0017] A3: Corresponding add a task head segmentation network at the output end of each image encoding large model to construct a distribution map and feature extraction module; the three task head segmentation networks are respectively used to distinguish the background and villus regions, background and edema lesions, and background and hyperplastic lesions in the pathological sections, so as to obtain the villus distribution map, edema distribution map, and hyperplasia distribution map;

[0018] A4: Input the training set data of hydatidiform mole slices into the corresponding large image encoding model, extract the feature vectors of the hydatidiform mole slice images and input them into the corresponding task head segmentation network, and generate a predicted distribution map through the corresponding task head segmentation network; the distribution map and the feature extraction module calculate the loss between the predicted distribution map and the true annotation distribution map, and update the parameters of the distribution map and the feature extraction module through the backpropagation algorithm;

[0019] A5: For the trained distribution map and feature extraction module, use the test set of hydatidiform mole slice images for testing, and calculate the segmentation accuracy of the complete distribution map and the feature extraction module.

[0020] Step C includes:

[0021] C1: Overlay the villus distribution map, edema distribution map, and hyperplasia distribution map of the same hydatidiform mole slice image to form a comprehensive distribution map;

[0022] C2: Traverse each villus area and obtain the pixel area of each villus through pixel counting, and count the number of edema distribution pixels and hyperplasia distribution pixels in the villus area;

[0023] C3: Calculate the proportion of edema pixels and hyperplasia pixels in each villus area. If the proportion of edema pixels in a certain villus area exceeds the critical threshold of villus edema proportion, it is marked as an edematous villus; if the proportion of hyperplasia pixels exceeds the critical threshold of villus hyperplasia proportion, it is marked as a hyperplastic villus; if a villus is marked as an edematous villus or a hyperplastic villus, the villus is marked as an abnormal villus, otherwise it is marked as a normal villus;

[0024] C4: Generate an edematous villus distribution map using the comprehensive distribution map and the edema mark, generate a hyperplastic villus distribution map using the comprehensive distribution map and the hyperplasia mark, generate an abnormal villus distribution map using the comprehensive distribution map and the abnormal mark, and generate a normal villus distribution map using the comprehensive distribution map and the normal mark;

[0025] C5: According to the edematous villus distribution map, hyperplastic villus distribution map, abnormal villus distribution map, normal villus distribution map, and the set villus area threshold, calculate the proportion of edematous villi, hyperplastic villi, and abnormal villi respectively;

[0026] C6: Output the edematous villus distribution map, hyperplastic villus distribution map, abnormal villus distribution map, and normal villus distribution map, and at the same time output the proportion of edematous villi, hyperplastic villi, and abnormal villi.

[0027] In step C5:

[0028] Count the number of edematous villi with an area exceeding the villus area threshold in the edematous villus distribution map; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of edematous villi, that is, the number of edematous villi divided by the total number of villi;

[0029] Count the number of hyperplastic villi with an area exceeding the villus area threshold in the hyperplastic villus distribution map; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of hyperplastic villi, that is, the number of hyperplastic villi divided by the total number of villi;

[0030] Count the number of abnormal villi with an area exceeding the villus area threshold in the abnormal villus distribution map; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of abnormal villi, that is, the number of abnormal villi divided by the total number of villi.

[0031] Step E includes:

[0032] E1: Construct a decision tree model;

[0033] E2: Use the decision tree algorithm to grow the decision tree;

[0034] E3: Perform decision tree pruning;

[0035] E4: Use the validation set or cross-validation method to evaluate the performance of the decision tree model;

[0036] E5: Obtain the trained decision tree model.

[0037] The task head segmentation network includes: an input layer for receiving the feature map output by the image encoding large model; a convolutional layer for extracting local features to increase the non-linear expression ability of the network, and each convolutional layer is followed by batch normalization and the ReLU activation function; an upsampling layer for performing upsampling to restore the feature map to the original image size, and each upsampling layer is followed by batch normalization and the ReLU activation function; an output layer for mapping the feature map to the required number of segmentation channels, and the final output uses the Softmax activation function for multi-class segmentation or the Sigmoid activation function for binary classification segmentation.

[0038] When training the villus distribution map and feature extraction module, after cutting the corresponding training images into several training image chunks of a set size, the training image chunks are input into the corresponding image encoding large model. The output of the image encoding large model is used as the input of the corresponding task head segmentation network. The actual segmentation output corresponding to each training image chunk is obtained by processing the corresponding villus annotation file, edema annotation file, and hyperplasia annotation file. Freeze the parameters of the image encoding large model, and only update the parameters of the task head segmentation network during training. Finally, the corresponding villus distribution map extraction sub-module, edema distribution map extraction sub-module, and hyperplasia distribution map extraction sub-module are trained.

[0039] The image encoding large model is the DINOv2 model based on the Vision Transformers architecture; Vision Transformers divides the image into tiles of a fixed size and treats each tile as an embedding vector for processing, and models the relationships between these tiles through the self-attention mechanism; DINOv2 uses the Vision Transformers architecture as the backbone network and adopts a self-supervised learning method to train Vision Transformers.

[0040] In step A5, the segmentation accuracy is represented by the intersection over union (IoU), Dice coefficient, or distribution center distance.

[0041] A hydatidiform mole section image analysis device based on a decision tree model, characterized by comprising:

[0042] A section image extraction module for obtaining a hydatidiform mole pathological section image;

[0043] A distribution map and feature extraction module for extracting the villus distribution map and villus feature vector, edema distribution map and edema feature vector, and hyperplasia distribution map and hyperplasia feature vector of the hydatidiform mole section image;

[0044] The distribution map and feature extraction module is constructed based on an image encoding large model and a task head segmentation network, and includes a villus distribution extraction sub-module, an edema distribution extraction sub-module, and a hyperplasia distribution extraction sub-module; each sub-module includes an image encoding large model and a corresponding task head segmentation network; the task head segmentation network is respectively used to distinguish the background and villus regions, background and edema lesions, and background and hyperplasia lesions in the pathological section to obtain the villus distribution map, edema distribution map, and hyperplasia distribution map;

[0045] A distribution map fusion analysis module for superimposing the villus distribution map, edema distribution map, and hyperplasia distribution map to generate a comprehensive distribution map, and obtaining the edematous villus distribution map, hyperplastic villus distribution map, normal villus distribution map, abnormal villus distribution map, proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi after analysis;

[0046] The feature fusion module is used to integrate the villus feature vector, edema feature vector and hyperplasia distribution vector, and merge the proportion of edema villi, the proportion of hyperplasia villi and the proportion of abnormal villi to form a fused feature vector;

[0047] The hydatidiform mole analysis module is used to analyze new pathological images using a decision tree model and output a pathological analysis report based on the lesion feature vector in the pathological image.

[0048] The present invention can assist doctors in improving the accuracy and efficiency of lesion assessment in hydatidiform mole pathological images, reduce the interference of subjective factors, and achieve rapid and objective clinical diagnosis and detection of hydatidiform mole based on the generated pathological analysis report. The present invention uses an image feature encoder to extract features from hydatidiform mole slice images, effectively mining a decision tree trained by integrating features into the actual decision-making process. The intelligent decision-making process of the present invention is interpretable, providing excellent auxiliary opinions and improving the reliability of the final diagnosis results. Furthermore, the method can also be used for early and large-scale screening of hydatidiform mole, helping to improve the detection rate and treatment efficacy of hydatidiform mole. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the process of the hydatidiform mole slice image analysis method of the present invention;

[0050] Figure 2 Schematic diagram of logic for analyzing hydatidiform mole slice images in the present invention;

[0051] Figure 3 Schematic diagram of the structure of the hydatidiform mole slice image analysis device of the present invention. DETAILED DESCRIPTION

[0052] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0053] like Figure 1 and Figure 2 As shown, the method for analyzing hydatidiform mole slice images based on a decision tree model of the present invention comprises the following steps:

[0054] A: Construct a distribution map and feature extraction module, including the villus distribution extraction submodule, the edema distribution extraction submodule, and the hyperplasia distribution extraction submodule. Each submodule includes an image encoding model and a corresponding task head segmentation network.

[0055] In the present invention, the villus distribution extraction sub-module, the edema distribution extraction sub-module, and the hyperplasia distribution extraction sub-module respectively output a villus distribution map and a villus feature vector, an edema distribution map and an edema feature vector, and a hyperplasia distribution map and a hyperplasia feature vector.

[0056] In the present invention, the task head segmentation network can be trained with a training set of hydatidiform mole slice images with distribution annotations.

[0057] In the present invention, step A includes the following specific steps:

[0058] A1: Collect and organize a set of hydatidiform mole slice images with villus distribution annotations, edema distribution annotations, and hyperplasia distribution annotations, and construct an image training set and a test set; the image training set and the test set include a villus training set and a test set, an edema training set and a test set, and a hyperplasia training set and a test set.

[0059] The set of hydatidiform mole slice images can be obtained by a digital microscope, that is, the stained hydatidiform mole slice is placed on the microscope stage. After the microscope is focused, the camera is used to obtain a scanned image of the hydatidiform mole slice under the microscope field of view, and by moving the hydatidiform mole slice, finally all the scanned images of the hydatidiform mole slice are stitched together to form a complete hydatidiform mole slice image. The set of hydatidiform mole slice images can also be obtained by a pathological slice scanner. After calibrating and cleaning the scanner, the stained hydatidiform mole slice is fixed on the slide holder, the scanner is started and the control software is opened, the high-resolution scanning mode is selected and the image is previewed, and after adjusting the scanning area, the scanning is started, and after the scanning is completed, the hydatidiform mole slice image is saved;

[0060] The set of hydatidiform mole slice images is obtained by manual annotation. Professional pathologists respectively outline the contours of all villi, edema, and hyperplasia on the hydatidiform mole slice images to generate villus distribution annotations, edema distribution annotations, and hyperplasia distribution annotations; the splitting ratio of the hydatidiform mole slice image training set and the test set is 4:1, that is, the images are randomly assigned to the training set and the test set folders according to the ratio of 4:1.

[0061] In the set of hydatidiform mole slice images with distribution annotations, any hydatidiform mole slice image has villus distribution annotations, edema distribution annotations, and / or hyperplasia distribution annotations, which serve as the corresponding villus, edema, and / or hyperplasia training set or test set;

[0062] In step A1, preprocessing is performed on the hydatidiform mole slice image training set, including but not limited to image preprocessing operations such as image normalization, size adjustment, image cutting, and noise removal, to ensure that the image is suitable for input into the image coding large model in the subsequent steps.

[0063] Since the scanned images of hydatidiform mole slices obtained are of large size, and the existing large image encoding models have requirements for the image size during input, it is necessary to perform slicing preprocessing on the images input to the large image encoding model. For the training of different distribution extraction sub-modules in the subsequent steps, the preprocessing methods of the image training set can be different. The preprocessing operation of the villus set is denoted as a-p, the preprocessing operation of the edema set is denoted as b-p, and the preprocessing operation of the hyperplasia set is denoted as c-p.

[0064] The slicing of the hydatidiform mole slice image is to cut the slice into several slices of size size at intervals of s, and there may be partial area overlap between two adjacent slices in the up, down, left, and right directions. Generally, after the slicing preprocessing operation a-p of the villus set, the image size size1 is a pixel size of 3000×3000 to 18000×18000. After the slicing preprocessing operation b-p of the edema set, the image size size2 is a pixel size of 1500×1500 to 9000×9000. After the slicing preprocessing operation c-p of the hyperplasia set, the image size size3 is a pixel size of 1500×1500 to 9000×9000.

[0065] A2: Construct a large image encoding model and perform pre-training.

[0066] In the present invention, the large image encoding model can adopt classic image encoding models such as ResNet, VGG, EfficientNet, and VisionTransformer. The villus distribution extraction sub-module uses the large image encoding model a-m, the edema distribution extraction sub-module uses the large image encoding model b-m, and the hyperplasia distribution extraction sub-module uses the large image encoding model c-m.

[0067] The villus distribution extraction sub-module uses the large image encoding model a-m, and the large image encoding model a-m is the DINOv2 model based on the VisionTransformers architecture. Vision Transformers divides the image into tiles of a fixed size and treats each tile as an embedding vector for processing, and models the relationships between these tiles through the self-attention mechanism. DINOv2 adopts the Vision Transformers architecture as its backbone network and uses self-supervised learning methods to train Vision Transformers. Since DINOv2 is trained on a large-scale and diverse dataset, it can generalize well to various visual tasks.

[0068] The edema distribution extraction sub-module uses the large image encoding model b-m, and the large image encoding model b-m is also the DINOv2 model. The hyperplasia distribution extraction sub-module uses the large image encoding model c-m, and the large image encoding model c-m is also the DINOv2 model.

[0069] A3: Add a task head segmentation network at the output end of the large image coding model to construct a distribution map and a feature extraction module. The task head segmentation network includes multiple convolutional layers, upsampling layers, and activation functions for fine segmentation of images. And design the loss function of the task head segmentation network. Commonly used loss functions include cross-entropy loss, Dice loss, IoU loss, etc. to measure the accuracy of network segmentation.

[0070] In the present invention, the task head segmentation network a-net is used to distinguish the background and villus regions in the pathological section to obtain a villus distribution map; the task head segmentation network b-net is used to distinguish the background and edema lesions in the pathological section to obtain an edema distribution map, and the task head segmentation network c-net is used to distinguish the background and hyperplastic lesions in the pathological section to obtain a hyperplasia distribution map.

[0071] In the present invention, the network architecture of the task head segmentation network a-net is as follows: an input layer that receives the feature map from the large image coding model; a convolutional layer that uses 3 layers of 3x3 convolutional layers to extract local features and increase the non-linear expression ability of the network. After each convolutional layer, batch normalization (Batch Normalization) and ReLU activation functions are connected; an upsampling layer that uses 2 layers of transposed convolution layers (Transposed Convolution) or bilinear interpolation (Bilinear Interpolation) for upsampling to restore the feature map to the original image size. After each upsampling layer, batch normalization and ReLU activation functions are connected; an output layer that maps the feature map to the required number of segmentation channels using the last convolutional layer with 1x1 convolution. The final output uses the Softmax activation function for multi-class segmentation or the Sigmoid activation function for binary classification segmentation. In this way, accurate recognition and segmentation of multiple types of lesions can be achieved.

[0072] The network architectures of the task head segmentation networks b-net and c-net are the same as that of the task head segmentation network a-net.

[0073] In this embodiment, the loss functions of the task head segmentation networks a-net, b-net, and c-net can adopt one of the following two loss functions:

[0074]

[0075] where N is the number of samples, y i and are the true label and the predicted value respectively.

[0076] In this embodiment, the task head segmentation network (a-net / b-net / c-net) adopts a unified architecture design, and realizes the multi-dimensional feature decoupling of medical images through a hierarchical structure of 3 convolutional layers + 2 transposed convolutional layers: the convolutional layer adopts a combination of batch normalization and ReLU activation, effectively suppressing the abnormal gradient fluctuations in small-sample training, enabling the model to converge stably; the transposed convolutional layer upsamples to accurately reconstruct the villous serrated edges, overcoming the problem of detail blurring of traditional interpolation methods; the Softmax / Sigmoid output layer provides a threshold-adjustable segmentation probability map, supporting the dynamic control of the sensitivity of edema / hyperplasia determination. This design realizes the sharing of computing resources while ensuring the independence of the three tasks, taking into account the requirements of high-precision segmentation and clinical real-time performance.

[0077] A4: Input the data of the hydatidiform mole slice training set into the corresponding image encoding large model to extract the corresponding feature vectors of the hydatidiform mole slice images, including the villous feature vector V a , the edema feature vector V b and the hyperplasia feature vector V c ; then input the corresponding feature vectors into the corresponding task head segmentation network, and generate a predicted distribution map through the corresponding task head segmentation network; the distribution map and the feature extraction module calculate the loss between the predicted distribution map and the true annotation distribution map, and update the parameters of the distribution map and the feature extraction module through the backpropagation algorithm, including the parameters of the image encoding large model and the parameters of the task head segmentation network.

[0078] In the present invention, the image encoding large model a-m and the task head segmentation network a-net for obtaining the villous distribution map are trained using the training set of hydatidiform mole slice images with villous distribution annotations, the image encoding large model b-m and the task head segmentation network b-net for obtaining the edema distribution map are trained using the training set of hydatidiform mole slice images with edema distribution annotations, and the image encoding large model c-m and the task head segmentation network c-net for obtaining the hyperplasia distribution map are trained using the training set of hydatidiform mole slice images with hyperplasia distribution annotations.

[0079] In this embodiment, the parameters of the pre-trained image encoding large model can be set to the frozen mode, that is, its parameters are not updated when training the task head segmentation network; it can also be set to the thawed mode, that is, its parameters are updated when training the task head segmentation network, making the image encoding large model more adaptable to the hydatidiform mole slice image processing task.

[0080] This embodiment also provides a method for network training:

[0081] Before training the network, it is necessary to obtain the training pictures for inputting into the network and training.

[0082] Combined with the morphological characteristics of hydatidiform mole required for actual diagnosis in this embodiment, through nearly one year of training on hydatidiform mole section annotation, from hydatidiform mole section annotation to annotation review, the annotation results of 157 scanned images of typical hydatidiform mole sections are obtained. Each section needs to be annotated in three ways. For villus annotation, the villus area of the section is circled. For edema annotation, the edematous part in the villus area of the section is circled. For hyperplasia annotation, the diffusely hyperplastic area of trophoblast cells in the villus area is circled.

[0083] The training method of the villus distribution map and feature extraction module is as follows:

[0084] Cut the training images into several training image blocks with size size1, input the training image blocks into the image encoding large model a-m, and use the output of the image encoding large model a-m as the input of the task head segmentation network a-net. The actual segmentation output corresponding to each training image block is obtained by processing the villus annotation file. Freeze the parameters of the image encoding large model a-m, and only update the parameters of the task head segmentation network a-net during training. Finally, the villus distribution map extraction sub-module is trained;

[0085] The training method of the edema distribution map and feature extraction module is as follows:

[0086] Cut the training images into several training image blocks with size size2, input the training image blocks into the image encoding large model b-m, and use the output of the image encoding large model b-m as the input of the task head segmentation network b-net. The actual segmentation output corresponding to each training image block is obtained by processing the edema annotation file. Freeze the parameters of the image encoding large model b-m, and only update the parameters of the task head segmentation network b-net during training. Finally, the edema distribution map extraction sub-module is trained;

[0087] The training method of the hyperplasia distribution map and feature extraction module is as follows:

[0088] Cut the training images into several training image blocks with size size3, input the training image blocks into the image encoding large model c-m, and use the output of the image encoding large model c-m as the input of the task head segmentation network c-net. The actual segmentation output corresponding to each training image block is obtained by processing the hyperplasia annotation file. Freeze the parameters of the image encoding large model c-m, and only update the parameters of the task head segmentation network c-net during training. Finally, the hyperplasia distribution map extraction sub-module is trained;

[0089] A5: For the trained distribution map and feature extraction module, use the test set of hydatidiform mole section images for detailed evaluation, calculate the segmentation accuracy of the complete distribution map and feature extraction module, and save the trained distribution map and feature extraction module as a deployable model file. The segmentation accuracy is represented by the intersection over union (IoU), Dice coefficient, or distance between distribution centers. The distribution map and feature extraction module includes a villous distribution map extraction sub-module, an edema distribution map extraction sub-module, and a hyperplasia distribution map extraction sub-module.

[0090] In the present invention, IoU is used to measure the overlap degree between the predicted distribution area and the true distribution area, and the calculation formula is as follows:

[0091]

[0092] where |A∩B| is the intersection area of the predicted distribution area A and the true distribution area B, and |A∪B| is the union area of the predicted distribution area A and the true distribution area B;

[0093] The Dice coefficient is used to measure the similarity of two sets, and is especially suitable for scenarios such as medical image segmentation. The calculation formula is as follows:

[0094]

[0095] where |A∩B| is the intersection area of the predicted distribution area A and the true distribution area B, |A| is the area of the predicted distribution area A, and |B| is the union area of the true distribution area B;

[0096] The distance between distribution centers is used to measure the Euclidean distance between the centroid of the predicted segmentation area and the centroid of the true segmentation area. The calculation formula is as follows:

[0097] Let the centroid coordinates of the predicted distribution area be (x p , y p ), and the centroid coordinates of the true segmentation area be (x t , y t ). Then the calculation formula for the distance between distribution centers D is:

[0098]

[0099] where (x p , y p ) are the centroid coordinates of the predicted segmentation area, and the calculation method is where P i represents the value of pixel i in the predicted distribution area A; (x t , y t ) are the centroid coordinates of the predicted segmentation area, and the calculation method is where T jRepresents the value of pixel j within the predicted distribution region B.

[0100] B: Using the distribution map constructed in step A and the feature extraction module to extract the feature vectors of the hydatidiform mole section image, including the villus feature vector V a , the edema feature vector V b and the hyperplasia feature vector V c , and obtain the villus distribution map, edema distribution map, and hyperplasia distribution map of the hydatidiform mole section image;

[0101] In this embodiment, the hydatidiform mole section image can be obtained through a pathological section scanner or by stitching microscope fields.

[0102] In step B, the hydatidiform mole section image can be preprocessed according to the preprocessing method in step A, including but not limited to image normalization, size adjustment, image cutting, noise removal, and other image preprocessing operations to ensure that the image is suitable for input into the distribution map and feature extraction module in the subsequent steps.

[0103] The preprocessing operations should be consistent with those in step A, that is, preprocess the hydatidiform mole section image with operations a-p, input it into the villus distribution extraction sub-module to obtain the villus distribution map and the villus feature vector V a , preprocess the hydatidiform mole section image with operations b-p, input it into the edema distribution extraction sub-module to obtain the edema distribution map and the edema feature vector V b , preprocess the hydatidiform mole section image with operations c-p, input it into the hyperplasia distribution extraction sub-module to obtain the hyperplasia distribution map and the hyperplasia feature vector V c ;

[0104] Among them, the villus distribution extraction sub-module includes the image encoding large model a-m and the task head segmentation network a-net, the edema distribution extraction sub-module includes the image encoding large model b-m and the task head segmentation network b-net, and the hyperplasia distribution extraction sub-module includes the image encoding large model c-m and the task head segmentation network c-net.

[0105] C: Input the villus distribution map, edema distribution map, and hyperplasia distribution map obtained in step B into the distribution map fusion analysis module, and output the edema-villus distribution map, hyperplasia-villus distribution map, normal villus distribution map, abnormal villus distribution map, edema-villus ratio, hyperplasia-villus ratio, and abnormal villus ratio of the hydatidiform mole section;

[0106] In the present invention, step C includes the following specific steps:

[0107] C1: Overlay the villus distribution map, edema distribution map, and hyperplasia distribution map of the same hydatidiform mole section image obtained in step B to form a comprehensive distribution map.

[0108] Preferably, when superimposing the villus distribution map, the edema distribution map, and the hyperplasia distribution map, if a preprocessing operation, especially a size adjustment operation, is performed in step B, a reduction transformation needs to be carried out to ensure that the distribution maps during superimposition are aligned with the original hydatidiform mole section image.

[0109] Since the preprocessing operation in step B includes cutting processing, the preprocessing operation in step C1 is cutting fusion. Specifically, all the cut label maps are spliced according to the corresponding positions of the original cuts in the section, and the overlapping parts of adjacent cuts in the up, down, left, and right directions are fitted according to the pixel mean values of the overlapping parts of multiple cut images.

[0110] C2: The villi appear as an independent island-like area on the comprehensive distribution map. Each villus in the comprehensive distribution map is traversed and analyzed. That is, the connected component analysis algorithm in the image processing algorithm is used to traverse each villus area, and all the pixels in each villus area are counted to calculate the pixel area of each villus. At the same time, the number of edema distribution pixels and hyperplasia distribution pixels in the villus area is counted; for the villus area R i (i = 1, …, N), its villus pixel area is Sa i , the number of edema distribution pixels is Sb i , and the number of hyperplasia distribution pixels is Sc i ;

[0111] C3: Calculate the proportion β of edema pixels in each villus area i = Sb i / Sa i and the proportion γ of hyperplasia pixels i = Sc i / Sa i ;

[0112] Set the critical threshold β d of the villus edema proportion and the critical threshold γ d of the villus hyperplasia proportion. If the proportion β i of edema pixels in a certain villus area exceeds the set threshold β d , it is marked as an edematous villus; if the proportion γ i of hyperplasia pixels exceeds the set threshold γ d , it is marked as a hyperplastic villus; if a villus is marked as either an edematous villus or a hyperplastic villus, the villus is marked as an abnormal villus, otherwise it is marked as a normal villus;

[0113] C4: Generate an edematous villus distribution map using the comprehensive distribution map and the edema mark, generate a hyperplastic villus distribution map using the comprehensive distribution map and the hyperplasia mark, generate an abnormal villus distribution map using the comprehensive distribution map and the abnormal mark, and generate a normal villus distribution map using the comprehensive distribution map and the normal mark;

[0114] C5: Set the villus area threshold S d1 , in the edematous villus distribution map, use the connected component analysis algorithm to count the number l1 of all edematous villi with an area exceeding the villus area threshold S d1 ; in the villus distribution map, use the connected component analysis algorithm to count the number l2 of all villi with an area exceeding the villus area threshold S d1 ; calculate the proportion r1 of edematous villi, that is, r1 = l1 / l2, where the proportion r1 is the number of edematous villi divided by the total number of villi.

[0115] Set the villus area threshold S d2 , in the hyperplastic villus distribution map, use the connected component analysis algorithm to count the number m1 of all hyperplastic villi with an area exceeding the villus area threshold S d2 ; in the villus distribution map, use the connected component analysis algorithm to count the number m2 of all villi with an area exceeding the villus area threshold S d2 ; calculate the proportion r2 of hyperplastic villi, that is, r2 = m1 / m2, where the proportion r2 is the number of hyperplastic villi divided by the total number of villi.

[0116] Set the villus area threshold S d3 , in the abnormal villus distribution map, use the connected component analysis algorithm to count the number n1 of all abnormal villi with an area exceeding the villus area threshold S d3 ; in the villus distribution map, use the connected component analysis algorithm to count the number n2 of all villi with an area exceeding the villus area threshold S d3 ; calculate the proportion r of abnormal villi, that is, r = n1 / n2, where the proportion r is the number of abnormal villi divided by the total number of villi.

[0117] C6: Output the edematous villus distribution map, hyperplastic villus distribution map, abnormal villus distribution map and normal villus distribution map, and at the same time output the proportion ω1 of edematous villi, the proportion ω2 of hyperplastic villi and the proportion ω of abnormal villi, providing a quantitative basis for pathologists.

[0118] In this embodiment, through the pixel-level fusion and object-level statistical analysis of the multi-modal distribution map, the lesion localization accuracy and clinical interpretability are significantly improved: the mean fusion splicing technology of the cut label map is adopted to eliminate the image deformation error caused by preprocessing, ensuring the strict alignment of the villus, edema, and hyperplasia distribution maps in the original coordinate system; the connected component analysis is used to independently traverse and pixel count each villus area, accurately simulating the diagnostic mode of pathologists observing each villus under the microscope; the finally generated abnormal / normal villus distribution map supports the spatial superposition visualization with the original pathological section, and doctors can quickly locate the targeted area marked by the algorithm, realizing the full-transparency verification of the algorithm decision-making process.

[0119] D: For the villus feature vector V obtained in step B a , the edema feature vector V b and the hyperplasia feature vector V cPerform feature fusion and integrate it into the fused feature vector V d , and incorporate the proportion of edematous villi ω1, the proportion of hyperplastic villi ω2, and the proportion of abnormal villi ω obtained in step C into the fused feature vector V as three eigenvalues d . Obtain the final feature vector V for subsequent decision tree analysis

[0120] The said step D includes the following specific steps

[0121] D1: Perform feature fusion on the obtained villus feature vector V a , the edema feature vector V b and the hyperplasia feature vector V c to obtain the fused feature vector V d ;

[0122] In step B, use the trained distribution map and feature extraction module to extract features from the hydatidiform mole section, input the hydatidiform mole section into the image encoding large model, and extract the feature vector of the image. Specifically, the villus distribution extraction sub-module uses the image encoding large model a-m to obtain the villus feature vector V a , the edema distribution extraction sub-module uses the image encoding large model b-m to obtain the edema feature vector V b , and the hyperplasia distribution extraction sub-module uses the image encoding large model c-m to obtain the hyperplasia feature vector V c .

[0123] Specific feature vector fusion methods include but are not limited to methods such as direct splicing, weighted summation, principal component analysis, and deep learning

[0124] D2: Incorporate the proportion of edematous villi ω1, the proportion of hyperplastic villi ω2, and the proportion of abnormal villi ω obtained in step C into the front end of the fused feature vector V as three eigenvalues using the method of direct splicing d to obtain the final feature vector V for subsequent decision tree analysis

[0125] This embodiment provides accurate quantitative indicators for the pathological diagnosis of hydatidiform mole through a hierarchical area screening and multi-dimensional lesion statistics mechanism: set a villus area threshold to filter out small interference regions, use connected component analysis to count the number of independent regions that meet the area standard in edematous, hyperplastic, and abnormal villi, calculate the proportion of edematous / hyperplastic / abnormal villi, and finally output a quantitative report accompanied by a distribution map. This design improves the statistical signal-to-noise ratio through area filtering and combines multi-lesion dimension data for synchronous output, enabling pathologists to quickly cross-verify the abnormal distribution trend

[0126] E: Construct a decision tree model for classifying pathological conditions based on the final feature vector V, and use the final feature vector V in step D to train the decision tree model. During the decision tree training process, the decision tree will perform node splitting according to the importance of features until a preset stopping condition is reached, and output a pathological analysis report including an abnormal villus distribution map, the number of lesions, and features.

[0127] Step E includes the following specific steps:

[0128] E1: Select a decision tree algorithm to construct a decision tree model;

[0129] In this embodiment, a common decision tree algorithm can be selected to construct a decision tree model, such as ID3 (Iterative Dichotomiser 3), C4.5, and CART (Classification And Regression Tree);

[0130] E2: Use the selected decision tree algorithm to grow the decision tree.

[0131] The above process includes selecting the best feature as a node, partitioning the dataset according to the feature values, and recursively constructing subtrees on each subset until a stopping condition is reached. The stopping condition may include the maximum depth of the tree, the minimum number of data samples in a node, or node purity, etc. The above techniques belong to conventional techniques in this field and will not be elaborated here.

[0132] E3: Perform decision tree pruning:

[0133] To prevent overfitting, it is usually necessary to prune the grown decision tree. Pruning can be achieved by setting a certain threshold to remove unnecessary nodes, thereby simplifying the model. Common pruning techniques include pre-pruning and post-pruning. Pre-pruning stops the decision tree growth process in advance, while post-pruning removes unnecessary nodes after the tree is fully grown. The above techniques belong to conventional techniques in this field and will not be elaborated here.

[0134] E4: Use a validation set or cross-validation method to evaluate the performance of the decision tree model.

[0135] In this embodiment, the evaluation metrics include but are not limited to accuracy, recall rate, F1 score, etc. According to the evaluation results, adjust the decision tree, such as adjusting the pruning parameters or feature selection strategy.

[0136] E5: Save the trained decision tree model as a deployable model file.

[0137] F: Analyze the new pathological image using the decision tree model trained in step E. The decision tree model will classify according to the feature vectors of the lesions in the image and generate a pathological analysis report, which includes the abnormal villus distribution map, the number of lesions, and the features. The pathological analysis report can be output in formats such as PDF, Excel, or HTML for easy viewing by doctors and patients.

[0138] As Figure 3 shown, the hydatidiform mole section image analysis device based on the decision tree model of the present invention includes:

[0139] A section image extraction module for obtaining the pathological section image of hydatidiform mole;

[0140] In this embodiment, the section image extraction module can adopt an existing digital microscope or a pathological section scanner.

[0141] A distribution map and feature extraction module for extracting the villus distribution map and villus feature vectors, the edema distribution map and edema feature vectors, and the hyperplasia distribution map and hyperplasia feature vectors of the hydatidiform mole section image;

[0142] The distribution map and feature extraction module can be constructed based on an image coding large model and a task head segmentation network. The distribution map and feature extraction module includes a villus distribution extraction sub-module, an edema distribution extraction sub-module, and a hyperplasia distribution extraction sub-module; each sub-module includes an image coding large model and a corresponding task head segmentation network; the three task head segmentation networks are respectively used to distinguish the background and villus areas, the background and edema lesions, and the background and hyperplastic lesions in the pathological section to obtain the villus distribution map, the edema distribution map, and the hyperplasia distribution map;

[0143] The specific structure of the distribution map and feature extraction module has been described in detail above and will not be elaborated here;

[0144] A distribution map fusion analysis module for superimposing the villus distribution map, the edema distribution map, and the hyperplasia distribution map to generate a comprehensive distribution map, and obtaining the edematous villus distribution map, the hyperplastic villus distribution map, the normal villus distribution map, the abnormal villus distribution map, the proportion of edematous villi, the proportion of hyperplastic villi, and the proportion of abnormal villi after analysis;

[0145] A feature fusion module for integrating the villus feature vectors, the edema feature vectors, and the hyperplasia distribution vectors, and combining the proportion of edematous villi, the proportion of hyperplastic villi, and the proportion of abnormal villi to form a fusion feature vector;

[0146] A hydatidiform mole analysis module for analyzing the new pathological image using the decision tree model and outputting a pathological analysis report according to the lesion feature vectors in the pathological image.

[0147] The above-mentioned each module has been described in detail above and will not be elaborated here.

[0148] Those skilled in the art can also understand that the various illustrative logical blocks and steps listed in the embodiments of the present application can be implemented by electronic hardware, computer software, or a combination of both. Whether such a function is implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art can use various methods to implement the described function for each specific application, but such implementation should not be construed as exceeding the scope protected by the embodiments of the present application.

[0149] In addition, the term "exemplary" is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as "exemplary" is not necessarily to be understood as being advantageous compared to other aspects or designs. Instead, the use of the term exemplary is intended to present concepts in a concrete manner. As used herein, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless otherwise specified or clear from the context, "X applies A or B" is intended to mean any arrangement in a natural inclusive listing. That is, if X applies A; X applies B; or X applies both A and B, then "X applies A or B" is satisfied in any of the foregoing instances. Additionally, unless otherwise specified or clear from the context indicating a singular form, the articles "a" and "an" as used in this application and the appended claims are generally understood to mean "one or more".

[0150] Similarly, although the present disclosure has been shown and described with respect to one or more implementations, those skilled in the art will envision equivalent variations and modifications after reading and understanding this specification and the drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the claims. Specifically with respect to the various functions performed by the components (e.g., elements, resources, etc.) described above, unless otherwise indicated, the terms used to describe such components are intended to correspond to any component (functionally equivalent) that performs the specific function of the described component, even if not structurally equivalent to the disclosed structure. Additionally, although a particular feature of the present disclosure may have been disclosed with respect to only one of several implementations, such a feature may, as may be desired and advantageous for any given or particular application, be combined with one or more other features of other implementations. Further, with respect to the use of "comprises", "comprising", "has", "having", or variants thereof in the detailed description or claims, such terms are intended to be inclusive in a manner similar to the term "includes".

[0151] Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common general knowledge or conventional technical means in the technical field not disclosed herein. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.

[0152] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A method for analyzing hydatidiform mole slice images based on a decision tree model, characterized in that, It includes the following steps: A: Construct a distribution map and feature extraction module including a villus distribution extraction sub-module, an edema distribution extraction sub-module, and a hyperplasia distribution extraction sub-module; each sub-module includes an image encoding large model and a corresponding task head segmentation network; B: Use the distribution map and feature extraction module to extract the feature vectors of the hydatidiform mole section images, and obtain the villus distribution map, edema distribution map, and hyperplasia distribution map of the hydatidiform mole section images; the feature vectors include villus feature vectors, edema feature vectors, and hyperplasia feature vectors; C: Input the villus distribution map, edema distribution map, and hyperplasia distribution map into the distribution map fusion analysis module, and output the edema villus distribution map, hyperplasia villus distribution map, normal villus distribution map, abnormal villus distribution map, proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi of the hydatidiform mole section; D: Perform feature fusion on the villus feature vectors, edema feature vectors, and hyperplasia feature vectors, integrate them into a fused feature vector, and merge the proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi as feature values into the fused feature vector to obtain the final feature vector; E: Construct a decision tree model for classifying the pathological conditions based on the final feature vector, and use the final feature vector to train the decision tree model to obtain a trained decision tree model; F: Use the trained decision tree model to analyze new pathological images, classify them according to the feature vectors of the lesions in the images, and generate a pathological analysis report, which includes an abnormal villus distribution map, the number of lesions, and features.

2. The method for analyzing hydatidiform mole section images according to claim 1, wherein Step A includes: A1: Collect and organize a set of hydatidiform mole section images with villus distribution annotation, edema distribution annotation, and hyperplasia distribution annotation, and construct an image training set and a test set; A2: Construct an image encoding large model and perform pre-training; A3: Corresponding add a task head segmentation network at the output end of each image encoding large model to construct a distribution map and feature extraction module; the three task head segmentation networks are respectively used to distinguish the background and villus regions, background and edema lesions, and background and hyperplastic lesions in the pathological sections, so as to obtain the villus distribution map, edema distribution map, and hyperplasia distribution map; A4: Input the hydatidiform mole section training set data into the corresponding image encoding large model, extract the feature vectors of the hydatidiform mole section images and input them into the corresponding task head segmentation network, and generate a predicted distribution map through the corresponding task head segmentation network; the distribution map and feature extraction module calculates the loss between the predicted distribution map and the true annotation distribution map, and updates the parameters of the distribution map and feature extraction module through the backpropagation algorithm; A5: For the trained distribution map and feature extraction module, use the hydatidiform mole section image test set for testing, and calculate the segmentation accuracy of the complete distribution map and feature extraction module.

3. The hydatidiform mole slice image analysis method according to claim 1, wherein Step C includes: C1: Overlay the villus distribution map, edema distribution map, and hyperplasia distribution map of the same hydatidiform mole section image to form a comprehensive distribution map; C2: Traverse each villus area and obtain the pixel area of each villus through pixel counting, and count the number of pixels with edema distribution and hyperplasia distribution within the villus area; C3: Calculate the proportion of edematous pixels and the proportion of hyperplastic pixels in each villus area. If the proportion of edematous pixels in a certain villus area exceeds the critical threshold of villus edema proportion, it is marked as an edematous villus; if the proportion of hyperplastic pixels exceeds the critical threshold of villus hyperplasia proportion, it is marked as a hyperplastic villus; if a villus is marked as an edematous villus or a hyperplastic villus, then the villus is marked as an abnormal villus, otherwise it is marked as a normal villus; C4: Generate a distribution map of edematous villi using the comprehensive distribution map and the edema mark, generate a distribution map of hyperplastic villi using the comprehensive distribution map and the hyperplasia mark, generate a distribution map of abnormal villi using the comprehensive distribution map and the abnormal mark, and generate a distribution map of normal villi using the comprehensive distribution map and the normal mark; C5: According to the distribution map of edematous villi, the distribution map of hyperplastic villi, the distribution map of abnormal villi, the distribution map of normal villi and the set villus area threshold, calculate the proportion of edematous villi, the proportion of hyperplastic villi and the proportion of abnormal villi respectively; C6: Output the distribution map of edematous villi, the distribution map of hyperplastic villi, the distribution map of abnormal villi and the distribution map of normal villi, and at the same time output the proportion of edematous villi, the proportion of hyperplastic villi and the proportion of abnormal villi.

4. The hydatidiform mole slice image analysis method according to claim 3, characterized in that In step C5: Count the number of edematous villi with an area exceeding the villus area threshold in the distribution map of edematous villi; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of edematous villi, that is, the number of edematous villi divided by the total number of villi; Count the number of hyperplastic villi with an area exceeding the villus area threshold in the distribution map of hyperplastic villi; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of hyperplastic villi, that is, the number of hyperplastic villi divided by the total number of villi; Count the number of abnormal villi with an area exceeding the villus area threshold in the distribution map of abnormal villi; count the number of villi with an area exceeding the villus area threshold in the villus distribution map; then calculate the proportion of abnormal villi, that is, the number of abnormal villi divided by the total number of villi.

5. The method for analyzing hydatidiform mole slice images according to claim 1, wherein Step E includes: E1: Construct a decision tree model; E2: Use the decision tree algorithm to grow the decision tree; E3: Perform decision tree pruning; E4: Use the validation set or cross-validation method to evaluate the performance of the decision tree model; E5: Obtain the trained decision tree model.

6. The method for analyzing hydatidiform mole slice images based on a decision tree model according to claim 1, wherein The task head segmentation network includes: an input layer for receiving the feature map output by the image encoding large model; a convolutional layer for extracting local features to increase the non-linear expression ability of the network, and each convolutional layer is followed by batch normalization and a ReLU activation function; an upsampling layer for performing upsampling to restore the feature map to the original image size, and each upsampling layer is followed by batch normalization and a ReLU activation function; an output layer for mapping the feature map to the required number of segmentation channels, and the final output uses the Softmax activation function for multi-class segmentation or the Sigmoid activation function for binary classification segmentation.

7. The method for analyzing hydatidiform mole slice images based on a decision tree model according to claim 1, wherein: When training the villus distribution map and feature extraction module, after cutting the corresponding training images into several training image patches with a set size, the training image patches are input into the corresponding image encoding large model, and the output of the image encoding large model is used as the input of the corresponding task head segmentation network. The actual segmentation output corresponding to each training image patch is obtained by processing the corresponding villus annotation file, edema annotation file, and hyperplasia annotation file. The parameters of the image encoding large model are frozen, and only the parameters of the task head segmentation network are updated during training. Finally, the corresponding villus distribution map extraction sub-module, edema distribution map extraction sub-module, and hyperplasia distribution map extraction sub-module are trained.

8. The method for analyzing hydatidiform mole section images based on a decision tree model according to claim 1, wherein: The image encoding large model is the DINOv2 model based on the Vision Transformers architecture; Vision Transformers divides the image into fixed-size patches and treats each patch as an embedding vector for processing, and models the relationships between these patches through the self-attention mechanism; DINOv2 uses the Vision Transformers architecture as the backbone network and adopts a self-supervised learning method to train Vision Transformers.

9. The method for analyzing hydatidiform mole slice images based on a decision tree model according to claim 1, characterized in that: In step A5, the segmentation accuracy is represented by the intersection over union (IoU), Dice coefficient, or distance to the distribution center.

10. A hydatidiform mole section image analysis device applying the method according to any one of claims 1 to 9, characterized in that, It includes: A sliced image extraction module for obtaining the pathological section image of hydatidiform mole; A distribution map and feature extraction module for extracting the villus distribution map and villus feature vector, edema distribution map and edema feature vector, and hyperplasia distribution map and hyperplasia feature vector of the hydatidiform mole section image; The distribution map and feature extraction module is constructed based on the image encoding large model and the task head segmentation network, and includes a villus distribution extraction sub-module, an edema distribution extraction sub-module, and a hyperplasia distribution extraction sub-module; each sub-module includes an image encoding large model and the corresponding task head segmentation network; the task head segmentation network is respectively used to distinguish the background and villus regions, background and edema lesions, and background and hyperplasia lesions in the pathological section to obtain the villus distribution map, edema distribution map, and hyperplasia distribution map; A distribution map fusion analysis module for overlaying the villus distribution map, edema distribution map, and hyperplasia distribution map to generate a comprehensive distribution map, and obtaining the edematous villus distribution map, hyperplastic villus distribution map, normal villus distribution map, abnormal villus distribution map, proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi after analysis; A feature fusion module for integrating the villus feature vector, edema feature vector, and hyperplasia distribution vector, and combining the proportion of edematous villi, proportion of hyperplastic villi, and proportion of abnormal villi to form a fused feature vector; A hydatidiform mole analysis module for analyzing new pathological images using a decision tree model and outputting a pathological analysis report based on the lesion feature vector in the pathological image.

Images