Ovarian section automatic identification method and system based on artificial intelligence
Through an artificial intelligence-based method, the deep learning feature encoder and multi-layer perceptron model are used to achieve accurate positioning and contour segmentation of ovaries and follicles, solving the problem of difficulty in ovarian section positioning in ultrasound images, and improving diagnostic efficiency and information richness.
Patent Information
- Application Number
- CN202510680209.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-26
- Publication Date
- 2025-08-15
AI Technical Summary
The prior art is difficult to accurately locate the ovarian section in ultrasound images, especially for physicians who lack clinical experience. The images are susceptible to noise interference and lack of interpretation, which makes it difficult to evaluate ovarian function and follicle development.
Using an artificial intelligence-based method, a deep learning feature encoder and multi-layer perceptron model is combined with Gaussian probability distribution map and iterative segmentation to achieve precise positioning and contour segmentation of ovaries and follicles, providing detailed anatomical structure information.
It improves the accuracy and diagnostic efficiency of ovarian and follicle localization, reduces the dependence on doctors' experience, and provides richer and more objective clinical information.
Smart Images
Figure CN120496065A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image recognition technology, and in particular to an artificial intelligence-based automatic recognition method and system for ovarian sections. Background Art
[0002] During the examination of ovarian function and follicle development, the doctor uses a transvaginal ultrasound probe to scan the patient's abdominal cavity and locate the ovaries among many structures such as the uterus and bladder. By analyzing the ovarian volume and follicle size, the doctor can understand the ovarian function and follicle development.
[0003] Standard ovarian sections are key to accurately assessing ovarian function and follicular development, so obtaining standard ovarian sections is particularly important. However, due to the limitations of ultrasound imaging quality, images are susceptible to noise interference and blurred edge structures. In addition, accurate ovarian section positioning relies heavily on the doctor's clinical experience. Therefore, accurately locating the ovarian section remains a challenge for primary care physicians who lack clinical experience. The current use of deep classification models for ultrasound section recognition cannot provide doctors with detailed section and anatomical structure information, so the model's prediction results lack interpretability and are less persuasive for sections with blurred boundaries. The current use of target detection models to roughly locate structures and determine section types can only provide rough anatomical structure detection frames, and cannot obtain detailed anatomical structure contours.
[0004] In recent years, artificial intelligence technology has been widely used in various industries, including ultrasound. Therefore, using artificial intelligence to analyze and detect transvaginal ultrasound images in real time and capture ovarian ultrasound sections is expected to significantly reduce reliance on operator experience and improve screening efficiency. Summary of the Invention
[0005] Purpose of the invention: The purpose of the present invention is to provide an artificial intelligence-based automatic recognition method and system for ovarian sections to locate the positions of the ovaries and follicles, while providing contour segmentation of the ovaries and follicles, so as to provide doctors with more anatomical structure information while automatically capturing the ovarian sections, thereby facilitating rapid clinical diagnosis.
[0006] Technical solution: An artificial intelligence-based automatic ovarian section recognition method includes the following steps:
[0007] S1, preprocessing of acquired ultrasound images;
[0008] S2, uses a deep learning feature encoder to analyze and encode the preprocessed ultrasound image to predict the coordinates of the ovarian follicle center;
[0009] S3, determining the approximate outlines of the ovaries and follicles based on the coordinates of the center points of the ovarian follicles and the corresponding extracted center point image features, and obtaining the rough outline coordinates of each ovary or follicle;
[0010] S4, fitting the ovarian and follicle contours according to the rough contour coordinates of each ovary or follicle.
[0011] Furthermore, the detailed steps of ultrasound image preprocessing are as follows:
[0012] S11, performing binarization processing on the ultrasound image to convert the ultrasound image into a binary image;
[0013] S12, performing connected domain analysis on the binary image, identifying and extracting the largest connected domain in the binary image, and ensuring that the most important imaging area in the image is extracted;
[0014] S13, for non-square shapes of the ultrasound image, calculate the difference between the shorter side and the longer side of the image, and then evenly add black borders to the corresponding edges of the image until the width and height of the image are equal, achieving an aspect ratio of 1:1; and scale the ultrasound image after adjusting the aspect ratio to adjust the ultrasound image to a uniform pixel size;
[0015] S14, performing normalization processing on the ultrasound image pixels.
[0016] Furthermore, the steps for predicting the coordinates of the center point of the ovarian follicle include:
[0017] S21, a feature encoder analyzes the input ovarian follicle image and extracts key features related to the ovarian follicle; the key features include the morphology, size, position of the follicle, and the relative relationship between the follicle and surrounding tissue;
[0018] S22, mapping the extracted key features related to the ovarian follicles to a high-dimensional feature vector, and assigning a response value to each pixel or image block according to the high-dimensional feature vector to obtain a Gaussian probability distribution map of the ovary or follicle center; the Gaussian probability distribution map contains all possible position responses of the ovary and follicle: coordinates with a response greater than 0.8 are structural center coordinates with a high probability, coordinates with a response between 0.5 and 0.8 are structural center coordinates with a certain probability, and coordinates with a response less than 0.5 are structural center coordinates with a low probability;
[0019] S23 uses the non-maximum suppression algorithm and the argmax operation, combined with the response of the Gaussian probability distribution map, to calculate the center coordinates of each ovary and its internal follicles.
[0020] Furthermore, the steps for calculating the center coordinates of each ovary and its internal follicles are as follows:
[0021] S231, in the Gaussian probability distribution graph, select the coordinates with response values greater than 0.8;
[0022] S232, applying a non-maximum suppression algorithm to reduce redundant, adjacent local maxima in the Gaussian probability distribution graph;
[0023] S233, for each ovarian region, apply the argmax operation separately to find the point with the highest response value as the center coordinate of the ovary; for follicles, it is necessary to apply a fine argmax operation inside or around the ovarian region to identify and locate multiple follicles.
[0024] Furthermore, the steps to determine the approximate outline of the ovary and follicles are as follows:
[0025] S31, inputting the center point image feature into a multi-layer perceptron model to predict an initial contour coordinate offset of the ovary or follicle, wherein the initial contour coordinate offset refers to a predicted distance or direction from the center point to the actual contour edge of the ovary or follicle;
[0026] S32, performing vector addition of the center coordinates of each ovary or follicle and the corresponding initial contour offset, generating a set of new coordinate points for each ovary or follicle. These new coordinate points roughly outline the edge of the ovary or follicle contour, and all the new coordinate points together constitute the rough contour segmentation result of the ovary or follicle.
[0027] Furthermore, the steps for fitting the ovary and follicle contours are as follows:
[0028] S41, inputting the rough outline coordinates of each ovary or follicle and the corresponding outline point image features into a multi-layer perceptron model to predict the fine outline point coordinate offset of the ovary or follicle; the outline point offset is a vector representing the predicted direction and distance from each outline point to the actual outline edge of the ovary or follicle;
[0029] S42, performing vector addition of each contour point offset vector and the rough contour coordinates of each ovary or follicle to further refine the contour of the ovary or follicle;
[0030] S43, repeating steps S41 to S42. In each iteration, step S41 uses the ovary or follicle outline updated in the previous round as the initial outline for this round of update; after multiple such iterations, the outline of each ovary or follicle is finally outlined.
[0031] An artificial intelligence-based automatic ovarian section recognition system, used to implement any of the above-mentioned automatic ovarian section recognition methods, comprising an ultrasound image preprocessing module, an ovarian follicle positioning module, an ovarian follicle outline rough outline module, and an ovarian follicle outline iterative refinement adjustment module;
[0032] The ultrasonic image preprocessing module is used to preprocess the input ultrasonic image, extract the ultrasonic sector imaging area, and unify the size and pixel range of the ultrasonic image;
[0033] The ovarian follicle positioning module takes the ultrasound image processed by the image preprocessing module as the input image, uses a deep learning feature encoder to perform in-depth analysis and encoding of the input image, and predicts the approximate location of each ovary and follicle by analyzing the ultrasound image features;
[0034] The ovarian follicle rough outline module uses the MLP model to achieve rough outline positioning based on the center point position of the ovarian follicle and image features, and quickly determine the approximate outline of the ovary and follicle;
[0035] The ovarian follicle contour iterative refinement adjustment module is used to further fit the ovarian and follicle contours, and achieve high-precision contour segmentation through multiple iterative fitting.
[0036] Compared with the prior art, the present invention has the following significant effects:
[0037] 1. This invention locates the ovaries and follicles by directly predicting the Gaussian probability distribution map of each ovary and follicle, thus resolving the problem that previous deep classification models used to directly identify ultrasound sections could not provide interpretable identification results. This further improves the confidence of the model predictions, while reducing reliance on the ultrasound physician's experience and improving clinical diagnostic efficiency.
[0038] 2. This invention utilizes an iterative ovarian follicle segmentation scheme to identify ultrasound sections and locate ovarian follicles, providing physicians with precise ovarian and follicular contour information. Compared to methods that use object detection models to identify ultrasound sections and locate anatomical structures, this invention can provide richer, more objective, and more accurate clinical information for clinical assessment of ovarian and follicular development. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 A schematic diagram of the system of the present invention;
[0040] Figure 2 It is the overall flow chart of the present invention;
[0041] Figure 3 This is a processing flow chart of the ultrasound image preprocessing module;
[0042] Figure 4 This is a processing flow chart of the ovarian follicle positioning module;
[0043] Figure 5(a) is the ultrasound image of the ovarian section, (b) is the Gaussian probability distribution map of the ovarian center point, and (c) is the Gaussian probability distribution map of the follicle center point;
[0044] Figure 6 Processing flow chart of the module for roughly outlining the outline of ovarian follicles;
[0045] Figure 7 This is a processing flow chart of the iterative ovarian follicle contour refinement module;
[0046] Figure 8 (a) is the initial outline diagram, and (b) is the updated outline diagram. DETAILED DESCRIPTION
[0047] The present invention will be described in further detail below with reference to the accompanying drawings and specific implementations.
[0048] like Figure 1 As shown, an artificial intelligence-based automatic ovarian section recognition system includes an ultrasound image preprocessing module, an ovarian follicle positioning module, an ovarian follicle outline rough outline module and an ovarian follicle outline iterative refinement adjustment module.
[0049] The ultrasonic image preprocessing module is used to preprocess the input ultrasonic image, extract the ultrasonic sector imaging area, and unify the size and pixel range of the ultrasonic image.
[0050] The ovarian follicle positioning module takes the ultrasound image processed by the ultrasound image preprocessing module as the input image, uses a deep learning feature encoder to perform in-depth analysis and encoding of the input image, and predicts the approximate location of each ovary and follicle by analyzing the ultrasound image features.
[0051] The ovarian follicle rough outline module uses the MLP (Multilayer Perceptron) model to achieve rough outline positioning based on the center point position of the ovarian follicle and image features, and quickly determine the approximate outline of the ovary and follicle.
[0052] The ovarian follicle contour iterative refinement adjustment module is used to further fit the ovarian and follicle contours and achieve high-precision contour segmentation through multiple iterative fitting.
[0053] An artificial intelligence-based automatic recognition method for ovarian sections, the flow chart is as follows Figure 2 As shown, the implementation steps are as follows:
[0054] Step 1, ultrasound image preprocessing;
[0055] like Figure 3 As shown in Figure 2, the detailed steps of ultrasound image preprocessing are as follows:
[0056] In step 11, the input ultrasound image is first converted into a binary image using a traditional image binarization method, such as an adaptive threshold or the Otsu method, to highlight the imaging in the image and remove unnecessary details, thereby laying the foundation for subsequent imaging area extraction.
[0057] Step 12: Analyze the binarized ultrasound image using a graph-theory-based connected domain analysis method to identify and extract the largest connected domain in the binary image. This analysis of inter-pixel connectivity ensures that the most important imaging region in the image, namely the ultrasound sector imaging region, is extracted.
[0058] Step S13, for the non-square shape that may appear in the ultrasound image (that is, the aspect ratio is not 1:1), the short edge padding operation is performed in this embodiment. Specifically, by calculating the difference between the shorter side and the longer side of the image, and then uniformly adding black edges on the corresponding edges of the image (in the direction of the short side), that is, the pixel value is 0, until the width and height of the image are equal, achieving an aspect ratio of 1:1. In order to meet the consistency requirements of the subsequent artificial intelligence algorithm for the input image size, an efficient interpolation algorithm, such as bilinear interpolation or bicubic interpolation, is used to scale the ultrasound image after adjusting the aspect ratio, and adjust the ultrasound image to a uniform 512x512 pixel size.
[0059] In step S14, since the pixel value ranges of different ultrasound images are inconsistent, the following formula is used to unify the image pixel value range to [0, 1]:
[0060]
[0061] Wherein, image represents an ultrasound image, min(image) represents the minimum pixel value in the ultrasound image, and max(image) represents the maximum pixel value in the ultrasound image.
[0062] This standardization step facilitates consistency and accuracy in subsequent processing.
[0063] Step 2, predict the location of ovarian follicles;
[0064] like Figure 4 As shown, the steps to predict the location of ovarian follicles include:
[0065] In step 21, the ultrasound image processed by the image preprocessing module is used as the input image, and a deep learning feature encoder is used to perform in-depth analysis and encoding on the input image.
[0066] The "feature encoder" here specifically refers to encoder models widely used in the field of deep learning. These models, trained and optimized with extensive data, are capable of efficiently extracting key features from images. Specifically, advanced deep learning models such as ResNet18, MobileNetV3+, and Vision Transformer can be used as feature encoders. These models, with their powerful feature extraction capabilities and generalization performance, have achieved remarkable results in fields such as image classification, object detection, and semantic segmentation. In this step, the feature encoder meticulously analyzes the input ovarian follicle image to extract key features related to the ovarian follicle. These features include, but are not limited to, the follicle's morphology, size, position, and its relative relationship to surrounding tissue. These extracted key features serve as important input for subsequent steps to accurately locate and segment the ovarian follicles.
[0067] In step 22, an ovarian follicle positioning module is used to analyze the key features related to the ovarian follicles extracted in step 21, and Gaussian probability distribution maps of the positioning of the ovaries and follicles are output respectively.
[0068] Specifically, the ovarian follicle positioning module analyzes the similarity or correlation between each pixel or image block and the ovarian and follicular features through a feature decoder. This process can be implemented with the help of a convolutional neural network (CNN) model, which extracts key feature vectors from the image by learning the morphological features of the ovary or follicle. For example, for an input medical image, the CNN model performs convolution aggregation on the ultrasound image features and learns the similarity between the high-dimensional feature vectors of the contrasting ultrasound image and the ovarian or follicular features, assigning a response value to each pixel or image block, namely, a Gaussian probability distribution map of the center point of the ovary or follicle, such as Figure 5 As shown in (a), (b), and (c) in the figure. A higher response value indicates a closer match between the pixel or image block and the ovary or follicle, meaning the ovary or follicle is more likely to be present at that location. The Gaussian probability distribution map contains all possible positional responses of the ovary and follicle in the ultrasound image. Specifically, coordinates with a Gaussian probability distribution map response greater than 0.8 are highly likely structural center coordinates, coordinates with a response between 0.5 and 0.8 are somewhat likely structural center coordinates, and coordinates with a response below 0.5 are less likely structural center coordinates.
[0069] Step 23 uses a comprehensive method to accurately calculate the center coordinates of each ovary and its internal follicles. This step primarily relies on the non-maximum suppression (NMS) algorithm and the argmax operation, combined with the responses of the Gaussian probability distribution map to perform separation and calculation. Specifically, first, within the Gaussian probability distribution map, coordinates with higher response values (i.e., coordinates with a response greater than 0.8) are selected. Coordinates with higher response values generally correspond to more likely ovarian or follicle locations. Next, the non-maximum suppression algorithm is applied to reduce redundant, adjacent local maxima within the Gaussian probability distribution map. Non-maximum suppression compares the response values of each pixel (or candidate point) with those of other pixels (or candidate points) within its neighborhood, retaining only local maxima. This effectively suppresses those less significant response points, helping to reduce the complexity of subsequent processing and improve the accuracy of coordinate calculation. The argmax operation is then used to further determine the precise coordinates of each ovary and its follicle. The argmax operation is the process of finding the pixel (or candidate point) with the maximum response value within the Gaussian probability distribution map. For each ovarian region, an argmax operation is applied to the point with the highest response value, which is used as the coordinate of the ovary's center. For follicles, a more refined argmax operation is applied within or around the ovarian region to identify and locate multiple follicles. By combining non-maximum suppression and the argmax operation, the precise coordinates of each ovary and its follicle center can be effectively separated and calculated from the Gaussian probability distribution map. This approach not only improves the accuracy of the coordinate calculation, but also enhances the robustness and efficiency of the algorithm.
[0070] Step 3: Determine the general outline of the ovaries and follicles;
[0071] like Figure 6 As shown, the steps to determine the approximate outline of the ovary and follicles are as follows:
[0072] In step 31, based on the coordinates of the center point of the ovary or follicle obtained in step 24, the corresponding center point image features are extracted from the ultrasound image features obtained in step 22. Subsequently, these center point image features are input into the MLP model. MLP is a feedforward artificial neural network that simulates complex nonlinear relationships through connections between multiple layers of nodes (or neurons). In this scenario, the MLP model is trained to predict the initial contour coordinate offset of the ovary or follicle based on the input structural center point image features. The "initial contour coordinate offset" here refers to the predicted distance or direction from the center point to the actual contour edge of the ovary or follicle. Through the prediction of the MLP model, one or more contour point offsets can be obtained, and these offsets will be used in subsequent contour construction or adjustment steps.
[0073] Step 32, using the initial contour offset of each ovary or follicle predicted in step 31, to achieve ovarian follicle contour segmentation. The core of this step is to perform vector addition of the center coordinates of each ovary or follicle (usually a two-dimensional coordinate point, indicating its position in the ultrasound image) and the corresponding initial contour offset. Through vector addition, a set of new coordinate points are generated for each ovary or follicle. These new coordinate points roughly outline the edge of its contour, and all the new coordinate points together constitute the rough contour segmentation result of the ovary or follicle. It is worth noting that since the offset is predicted based on the center point image features, this rough contour may not be completely accurate, but it provides important initial information for subsequent contour optimization and precise segmentation.
[0074] Step 4: Fit the outline of the ovary and follicles;
[0075] like Figure 7 As shown, the steps for fitting the ovary and follicle contours are as follows:
[0076] In step 41, based on the rough outline coordinates of each ovary or follicle and the corresponding contour point image features extracted from the ultrasound image features obtained in step 22, an MLP model is trained to predict the fine contour point coordinate offsets of the ovary or follicle based on the contour point image features. Here, "contour point offset" is a vector that describes the predicted direction and distance from each contour point to the actual contour edge of the ovary or follicle.
[0077] Step 42 , performing vector addition of each contour point offset vector obtained in step 41 and the rough contour coordinates of each ovary or follicle, to further refine the contour of the ovary or follicle.
[0078] Step 43, by repeatedly executing steps 41 and 42, the outline of the ovary or follicle is gradually and meticulously optimized. Specifically, in each iteration, step 41 uses the ovary or follicle outline updated in the previous round as the initial outline of the current round of update, such as Figure 8 Then, the contour offset vector predicted in step 42 is added to the initial contour to further adjust the contour so as to be closer to the actual shape of the ovary or follicle, as shown in FIG. Figure 8 After multiple iterations, the outline of each ovary or follicle can be accurately outlined, providing a reliable basis for subsequent image analysis, disease diagnosis or treatment planning.
Claims
1. An artificial intelligence-based automatic recognition method for ovarian sections, characterized in that: The steps are as follows: S1, preprocessing of acquired ultrasound images; S2, uses a deep learning feature encoder to analyze and encode the preprocessed ultrasound image to predict the coordinates of the ovarian follicle center; S3, determining the approximate outlines of the ovaries and follicles based on the coordinates of the center points of the ovarian follicles and the corresponding extracted center point image features, and obtaining the rough outline coordinates of each ovary or follicle; S4, fitting the ovarian and follicle contours according to the rough contour coordinates of each ovary or follicle.
2. The method for automatic identification of ovarian sections based on artificial intelligence according to claim 1, characterized in that: The detailed steps of ultrasound image preprocessing are as follows: S11, performing binarization processing on the ultrasound image to convert the ultrasound image into a binary image; S12, performing connected domain analysis on the binary image, identifying and extracting the largest connected domain in the binary image, and ensuring that the most important imaging area in the image is extracted; S13, for non-square shapes of the ultrasound image, calculate the difference between the shorter side and the longer side of the image, and then evenly add black borders to the corresponding edges of the image until the width and height of the image are equal, achieving an aspect ratio of 1:1; and scale the ultrasound image after adjusting the aspect ratio to adjust the ultrasound image to a uniform pixel size; S14, performing normalization processing on the ultrasound image pixels.
3. The method for automatic identification of ovarian sections based on artificial intelligence according to claim 1, characterized in that: The steps for predicting the coordinates of the center point of the ovarian follicle include: S21, a feature encoder analyzes the input ovarian follicle image and extracts key features related to the ovarian follicle; the key features include the morphology, size, position of the follicle, and the relative relationship between the follicle and surrounding tissue; S22, mapping the extracted key features related to the ovarian follicles to a high-dimensional feature vector, and assigning a response value to each pixel or image block according to the high-dimensional feature vector to obtain a Gaussian probability distribution map of the ovary or follicle center; the Gaussian probability distribution map contains all possible position responses of the ovary and follicle: coordinates with a response greater than 0.8 are structural center coordinates with a high probability, coordinates with a response between 0.5 and 0.8 are structural center coordinates with a certain probability, and coordinates with a response less than 0.5 are structural center coordinates with a low probability; S23 uses the non-maximum suppression algorithm and the argmax operation, combined with the response of the Gaussian probability distribution map, to calculate the center coordinates of each ovary and its internal follicles.
4. The method for automatic identification of ovarian sections based on artificial intelligence according to claim 3, characterized in that: The steps to calculate the center coordinates of each ovary and its internal follicles are as follows: S231, in the Gaussian probability distribution graph, select the coordinates with response values greater than 0.8; S232, applying a non-maximum suppression algorithm to reduce redundant, adjacent local maxima in the Gaussian probability distribution graph; S233, for each ovarian region, apply the argmax operation separately to find the point with the highest response value as the center coordinate of the ovary; for follicles, it is necessary to apply a fine argmax operation inside or around the ovarian region to identify and locate multiple follicles.
5. The method for automatic identification of ovarian sections based on artificial intelligence according to claim 1, characterized in that: The steps to determine the approximate outline of the ovaries and follicles are as follows: S31, inputting the center point image feature into the MLP model to predict the initial contour coordinate offset of the ovary or follicle, wherein the initial contour coordinate offset refers to the predicted distance or direction from the center point to the actual contour edge of the ovary or follicle; S32, performing vector addition of the center coordinates of each ovary or follicle and the corresponding initial contour offset, generating a set of new coordinate points for each ovary or follicle. These new coordinate points roughly outline the edge of the ovary or follicle contour, and all the new coordinate points together constitute the rough contour segmentation result of the ovary or follicle.
6. The method for automatic identification of ovarian sections based on artificial intelligence according to claim 1, characterized in that: The steps to implement the ovarian and follicle contour fitting are as follows: S41, inputting the rough outline coordinates of each ovary or follicle and the corresponding outline point image features into the MLP model to predict the fine outline point coordinate offset of the ovary or follicle; the outline point offset is a vector representing the predicted direction and distance from each outline point to the actual outline edge of the ovary or follicle; S42, performing vector addition of each contour point offset vector and the rough contour coordinates of each ovary or follicle to further refine the contour of the ovary or follicle; S43, repeating steps S41 and S42. In each iteration, step S41 uses the ovary or follicle outline updated in the previous round as the initial outline for this round of update; after multiple such iterations, the outline of each ovary or follicle is finally outlined.
7. An artificial intelligence-based automatic recognition system for ovarian sections, characterized by: Used to execute the automatic ovarian section recognition method described in any one of 1-6, including an ultrasound image preprocessing module, an ovarian follicle positioning module, an ovarian follicle outline rough outline module and an ovarian follicle outline iterative refinement adjustment module; The ultrasonic image preprocessing module is used to preprocess the input ultrasonic image, extract the ultrasonic sector imaging area, and unify the size and pixel range of the ultrasonic image; The ovarian follicle positioning module takes the ultrasound image processed by the image preprocessing module as the input image, uses a deep learning feature encoder to perform in-depth analysis and encoding of the input image, and predicts the approximate location of each ovary and follicle by analyzing the ultrasound image features; The ovarian follicle rough outline module uses the MLP model to achieve rough outline positioning based on the center point position of the ovarian follicle and image features, and quickly determine the approximate outline of the ovary and follicle; The ovarian follicle contour iterative refinement adjustment module is used to further fit the ovarian and follicle contours, and achieve high-precision contour segmentation through multiple iterative fitting.