Gynecological endocrine index analysis method and system based on image recognition

Abstract

The invention relates to the field of gynecological endocrine, and discloses a gynecological endocrine index analysis method and system based on image recognition, and the method comprises the steps: processing an ovary ultrasonic image, automatically recognizing and cutting an ovary region, and dividing the cut image into a plurality of grid regions; the texture direction of each grid area is analyzed and extracted, texture consistency is evaluated, and whether the follicle form is complete or not is judged; if the texture consistency is lower than a threshold value, removing the region; performing adaptive segmentation in the screened follicle candidate region by adopting a dynamic region growing method to complete follicle labeling; evaluating the roundness characteristics of the follicles by using a multi-scale roundness matching function, and judging the regularity of the follicle forms; performing marginal definition analysis on the regular follicles, and calculating the volume and area of the regular follicles; classifying and counting the regular follicles; based on the space distribution of the follicles, local density analysis is carried out, and ovarian endocrine spectrum feature vectors are generated. The method has the advantage of improving the diagnosis and treatment efficiency.

Description

Technical Field

[0001] The present invention relates to the field of gynecological endocrinology, and in particular to a gynecological endocrinology index analysis method and system based on image recognition. Background Art

[0002] In the clinical diagnosis of gynecological endocrine diseases, especially polycystic ovary syndrome, polycystic ovarian changes are one of the important imaging indicators. At present, the number of follicles is usually counted manually through two-dimensional ultrasound images in clinical practice to determine whether it meets the diagnostic criteria of polycystic ovary morphology. However, the process is highly dependent on the doctor's subjective judgment, especially when the number of follicles is close to the diagnostic threshold, the accuracy of the count has a decisive influence on the final diagnosis. The existing manual counting method is prone to misjudgment, omission or recounting when the image resolution is not high, the follicle boundaries are unclear or the follicles are densely distributed, which seriously affects the accuracy and consistency of the diagnosis. In addition, the clinical image acquisition conditions vary greatly, resulting in significant differences in image features between different patients, further increasing the difficulty of manual identification. Therefore, it is very necessary to design a gynecological endocrine indicator analysis method and system based on image recognition to improve the efficiency of diagnosis and treatment. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a gynecological endocrine index analysis method and system based on image recognition, which has the advantage of improving the efficiency of diagnosis and treatment and solves the problems in the above-mentioned background technology.

[0004] In order to achieve the above-mentioned purpose of improving the efficiency of diagnosis and treatment, the present invention provides the following technical solution: a gynecological endocrine index analysis method based on image recognition, comprising the following steps: By automatically identifying the ovarian region in the ovarian ultrasound image, the boundary clipping of the ovarian ultrasound image is realized to obtain a number of grid regions; Through the consistency analysis of the texture direction in the grid area, the integrity of the follicle morphology is judged, the incomplete follicles are eliminated, and the candidate follicle area is determined; After determining the candidate follicle regions, an adaptive segmentation method based on dynamic region growing is used to mark the follicles. The speed and threshold of region growing are adaptively adjusted by analyzing local image features. The multi-scale circularity matching function was used to evaluate the circularity characteristics of the follicles in the follicle candidate area and determine the regularity of the follicle morphology. The edge clarity of regular follicles was analyzed and combined with the follicle morphology and spatial distribution to classify and count the follicles; According to the spatial distribution of follicles and combined with local density analysis, the feature vector of ovarian endocrine map is generated.

[0005] Preferably, the process of determining whether the follicle morphology is complete is: By calculating the main direction angles of all pixels in the grid area, and counting the mean and mean square error; If the mean square error is less than the threshold, it means that the texture direction of the grid area is consistent, indicating that the follicle morphology is complete; If the mean square error is greater than or equal to the threshold, it means that the texture direction of the grid area is inconsistent, indicating that the follicle morphology is incomplete.

[0006] Preferably, the process of marking the follicles is: In the selected follicle candidate area, pixels with stable brightness value, edge intensity and texture direction are selected as initial seed points; Set the initial grayscale similarity threshold and define the region growth criteria, including the grayscale difference between adjacent pixels and the current region, texture similarity, and edge continuity; In the process of region growth, the neighborhood pixels around the seed point are traversed. If the neighborhood pixels meet the similarity criteria, they are incorporated into the current region and used as new seed points. Dynamically adjust the similarity threshold according to local image features to adapt to grayscale changes and texture fluctuations in the image; When all neighboring pixels do not meet the growth conditions, the region growth terminates; Finally, a closed outline of the candidate follicle region is formed and the region is marked as the follicle region.

[0007] Preferably, the process of adaptively adjusting the speed and threshold of region growth according to local image features is as follows: Apply edge detection algorithm to the candidate follicle region to extract edge gradient information and obtain edge intensity map; In the edge intensity map, pixels with significant edge response and high texture direction consistency are selected as initial seed points; For each seed point, calculate the grayscale difference, texture direction difference and edge gradient change between its surrounding pixels and the marked pixels in the current area; According to the degree of difference, determine whether the adjacent pixels meet the region expansion conditions. If they do, they are merged into the current segmentation region and added to the seed point queue; During the growth process, the similarity threshold and growth step size are dynamically adjusted according to the average grayscale, texture direction consistency and edge density of the pixels in the current area; The growing process continues to iterate until all the pixels in the neighborhood of the seed point do not meet the expansion condition, thus forming the initial segmentation area of ​​the follicle.

[0008] Preferably, the process of judging whether adjacent pixels meet the region expansion condition is as follows: During the region growing process, if the grayscale difference between the neighboring pixel and the current region is less than the preset grayscale similarity threshold, the pixel is considered to be sufficiently similar to the current region and meets the region expansion conditions, and is added to the current segmented region and continues to participate in region growing as a new seed point; If the grayscale difference is greater than or equal to the similarity threshold, the pixel is considered to be too different from the current area and is excluded from the expansion area and not included in the segmentation area.

[0009] Preferably, the process of determining the regularity of follicle morphology is: Extract boundary information from the follicle segmentation region obtained by the region growing process, and use edge detection algorithm to obtain the contour edge of the follicle region; Fit and sample boundary contours at multiple scales to construct a multi-scale circularity matching function; The roundness values ​​were calculated at different scales, and the scale with the smallest fitting error was selected as the best evaluation scale; The roundness value is compared with the preset threshold. If the roundness is greater than the threshold, the shape is considered regular, otherwise it is irregular.

[0010] A gynecological endocrine index analysis system based on image recognition, comprising: Ovarian region recognition module: responsible for processing the original ovarian ultrasound image, automatically identifying the ovarian region and cropping it; Follicle morphology assessment module: Analyze and extract the texture direction of each mesh area and evaluate the texture consistency of the area; Follicle segmentation module: responsible for adaptive segmentation within the selected follicle candidate area; Follicle morphology determination module: evaluates the roundness characteristics of follicles by applying a multi-scale roundness matching function; Secretion map generation module: local density analysis is performed based on the spatial distribution of follicles to evaluate the distribution pattern of follicles.

[0011] Compared with the prior art, the present invention provides a gynecological endocrine index analysis method and system based on image recognition, which has the following beneficial effects: 1. Through the gynecological endocrine index analysis method based on image recognition, the follicle morphology in ovarian ultrasound images can be efficiently and accurately identified and analyzed, which helps to improve the accuracy of ovarian disease diagnosis, especially in the early diagnosis of endocrine abnormalities such as polycystic ovary syndrome, which has important clinical value.

[0012] 2. The grid area texture direction consistency analysis is combined with the adaptive segmentation method of dynamic region growing to effectively improve the accuracy of follicle morphology determination, and can automatically screen out follicles with complete morphology, reducing the possibility of manual intervention and misjudgment, and improving the level of automation.

[0013] 3. The application of multi-scale circularity matching function can more accurately evaluate the regularity of follicle morphology, avoiding the problem of poor sensitivity of traditional methods to irregular follicle morphology, and further improving the accuracy of follicle counting and classification.

[0014] 4. By combining edge clarity analysis with follicle spatial distribution for classification and counting, we can understand the status and spatial distribution characteristics of follicles in detail, thereby providing clinicians with comprehensive ovarian function assessment information and helping to formulate more accurate treatment plans.

[0015] 5. The generated ovarian endocrine map feature vector can provide quantitative analysis of ovarian endocrine levels, help track patients' endocrine changes, and provide data support for clinical treatment effect evaluation and decision-making.

[0016] 6. It realizes efficient and automated follicle analysis, reduces interference from human factors, and can be widely used in different patient groups. It has good universality and stability, and provides strong support for large-scale clinical screening and telemedicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the method of the present invention; Figure 2 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1

[0020] See also Figure 1 As shown, the gynecological endocrine index analysis method based on image recognition according to an embodiment of the present invention comprises the following steps: S1: By automatically identifying the ovarian region in the ovarian ultrasound image, the boundary clipping of the ovarian ultrasound image is achieved to obtain a number of grid regions.

[0021] Ovarian ultrasound images are acquired through conventional two-dimensional grayscale ultrasound imaging equipment, and the images are saved in DICOM format or standard image format and imported into this system for analysis through the image acquisition module. In order to achieve localized and refined processing of image analysis, the ovarian area in the image must first be automatically identified.

[0022] Automatic recognition of the ovarian region uses a medical image segmentation algorithm based on deep learning, such as U-Net, DeepLab or its improved network structure. The network takes the ultrasound image as input, extracts multi-scale image features through the encoder, and then gradually restores the spatial resolution through the decoder, and finally outputs the pixel-level segmentation mask of the ovarian region. The traditional method based on the combination of image grayscale threshold, edge detection and morphological operation is used to extract the candidate ovarian region. The recognition results are used to guide the subsequent cropping and feature extraction operations.

[0023] After the ovarian region is identified, its boundary is cropped and the image region within the minimum bounding rectangle is extracted as the target region for subsequent analysis. In order to enhance the ability to describe local features of the image, the cropped image region is divided into several grid regions. The grid division adopts an equal spacing division strategy, that is, according to the set number of rows and columns or grid size, the ovarian region is divided into sub-regions of the same size in equal proportion. Each grid region is used as an independent analysis unit for the calculation and evaluation of indicators such as texture direction, consistency, and edge features, thereby providing a basis for the screening of candidate follicle regions.

[0024] S2: Extract the texture direction of each grid area, evaluate the consistency of the texture direction of the grid area, and determine whether the follicle morphology is complete; if the texture direction consistency of the grid area is lower than the preset threshold, the follicle morphology is considered incomplete and the grid area is removed from the candidate area.

[0025] The process of determining whether the follicle morphology is complete in S2 is as follows: The process of judging whether the follicle morphology is complete is: By calculating the main direction angles of all pixels in the grid area, and counting the mean and mean square error; The main direction angle of all pixels in a grid area Perform statistics and calculate the mean and mean square error , set a texture consistency threshold T to distinguish whether the texture direction is consistent, and judge according to the size of the mean square error: like <T, it means that the texture direction of the grid area is consistent, indicating that the follicle morphology is complete; like ≥T, it means that the texture direction of the grid area is inconsistent, indicating that the follicle morphology is incomplete; It should be noted that the role of judging whether the follicle morphology is complete is: Function 1: It is used to eliminate abnormal areas that are unclear due to factors such as poor image quality, tissue overlap, artifact interference or edge occlusion, so as to ensure that the subsequent follicle segmentation and counting analysis process is only based on the effective follicle area with complete morphological characteristics and clear boundaries, thereby improving the segmentation accuracy and the reliability of endocrine index extraction; Function 2: Completely shaped follicles are easier to fit geometrically, making it easier to quantify their developmental status and degree of maturity, which is critical for evaluating ovarian function and assisting in the diagnosis of gynecological diseases such as polycystic ovary syndrome.

[0026] The technical solution of this embodiment is: to judge whether the follicle morphology is complete by extracting the texture direction of the grid area and evaluating its consistency. The specific process is: first, calculate the main direction angle of all pixels in the grid area, and count their mean and mean square error; then, set the texture consistency threshold, and judge the integrity of the follicle morphology according to the size of the mean square error. When the mean square error is lower than the threshold, it means that the texture direction is consistent and the follicle morphology is complete; when the mean square error is higher than the threshold, it means that the texture direction is inconsistent, the follicle morphology may be incomplete, and the relevant area will be eliminated. The method can effectively eliminate abnormal areas caused by factors such as poor image quality, tissue overlap or artifacts, thereby ensuring the accuracy of subsequent follicle segmentation and counting analysis, and improving the reliability of endocrine index extraction. At the same time, the complete follicle morphology is easier to perform geometric fitting, which helps to quantify the developmental state and maturity of the follicle, and is of great significance for ovarian function assessment and auxiliary diagnosis of gynecological diseases such as polycystic ovary syndrome.

[0027] Example 2

[0028] like Figure 1 As shown, the gynecological endocrine index analysis method based on image recognition also includes the following steps: S3: After determining the candidate follicle region, an adaptive segmentation method based on dynamic region growing is used to mark the follicle. The initial segmentation is completed by analyzing the edge features and texture features of the follicle, and the speed and threshold of region growing are adaptively adjusted according to the local image features.

[0029] In S3, an adaptive segmentation method based on dynamic region growing is used, and the process of labeling follicles is as follows: In the selected follicle candidate area, pixels with stable brightness value, edge strength and texture direction are selected as initial seed points; pixels with relatively high brightness value and clearness are selected as candidate seed points, because these points usually represent the edge or core area of ​​follicles and have better segmentation effect; the edge of the follicle area is detected by edge detection algorithm, and pixels with obvious performance in edge strength map are selected. Pixels with strong edge strength often have higher resolution and accuracy; Set the initial grayscale similarity threshold and define the region growth criteria, including the grayscale difference between adjacent pixels and the current region, texture similarity, and edge continuity; set a threshold to control the grayscale difference between the seed point and the surrounding neighborhood pixels. This threshold indicates whether there is similarity between the seed point and the adjacent pixels, ensuring that only those pixels similar to the current region can be included in the expansion region; define the criteria for region growth, including: Grayscale difference: The grayscale difference between the adjacent pixel and the marked area of ​​the current area. If the grayscale difference is less than the set threshold, it means that the pixel belongs to the current area.

[0030] Texture similarity: By calculating the consistency of texture direction between the neighboring pixels and the pixels in the current area, if the texture directions are similar, the pixels meet the expansion conditions.

[0031] Edge continuity: Check whether the edges between the neighboring pixels and the marked area are continuous. Pixels with strong edge continuity are usually more suitable for joining the current area to form a more complete follicle outline.

[0032] In the process of region growing, the neighborhood pixels around the seed point are traversed. If the neighborhood pixels meet the similarity criteria, they are incorporated into the current region and used as new seed points. In the process of region growing, the neighborhood pixels around each seed point are traversed to check whether they meet the set similarity criteria. The similarity calculation of neighborhood pixels includes: Grayscale difference: Calculate the grayscale difference between the seed point and the neighboring pixels. If the grayscale difference is less than the preset threshold, the pixel is considered to belong to the current area.

[0033] Texture direction difference: Compare the texture directions of the seed point and the neighboring pixels. If the difference between the texture directions of the two is small, it means that the pixel meets the expansion conditions.

[0034] Edge gradient change: Calculate the edge gradient change of the neighborhood pixels. If the gradient change of the neighborhood pixels is small, it means that the edge continuity with the current area is good and meets the expansion conditions.

[0035] Add pixels that meet the conditions to the current region: If the neighborhood pixels meet the similarity criteria, they are incorporated into the current segmented region, and the pixels are used as new seed points to continue participating in region growth.

[0036] The similarity threshold is dynamically adjusted according to the local image features to adapt to the grayscale changes and texture fluctuations in the image; during the region growing process, the grayscale and texture in the image may change, so the similarity threshold needs to be dynamically adjusted. The method is: Local grayscale changes: When the grayscale changes in the image area are large, the grayscale similarity threshold should be appropriately increased to avoid misjudgment.

[0037] Texture fluctuation: According to the texture fluctuation of the local area, the texture similarity threshold is adjusted to ensure that the region growing process does not mistakenly expand to irrelevant areas.

[0038] Edge density change: In the edge area, the edge density may change greatly. The threshold should be adjusted appropriately to adapt to this change and ensure the continuity of the edge.

[0039] When all neighboring pixels do not meet the growth conditions, the region growth terminates; when all neighboring pixels do not meet the similarity criteria during the continuous expansion of the region growth, the growth stops. This means that the current region has reached its maximum expansion range and cannot continue to merge more pixels that meet the conditions.

[0040] Finally, the outline of the closed candidate follicle region is formed, and the region is marked as the follicle region; after the expansion process of regional growth, a closed follicle region is finally formed. The boundary of this region is fully annotated in the image and is consistent with the actual morphology of the follicle. The region is marked as a follicle and a clear follicle outline is provided for subsequent analysis.

[0041] In S3, the initial segmentation is completed by analyzing the edge features and texture features of the follicle, and the speed and threshold of the region growth are adaptively adjusted according to the local image features. The process is: Apply edge detection algorithm to the candidate follicle region to extract edge gradient information and obtain edge intensity map; In the edge intensity map, pixels with significant edge response and high texture direction consistency are selected as initial seed points; For each seed point, calculate the grayscale difference, texture direction difference and edge gradient change between its surrounding pixels and the marked pixels in the current area; According to the degree of difference, determine whether the adjacent pixels meet the region expansion conditions. If they do, they are merged into the current segmentation region and added to the seed point queue; During the growth process, the similarity threshold and growth step size are dynamically adjusted according to the average grayscale, texture direction consistency and edge density of the pixels in the current area; The growing process continues to iterate until all the pixels in the neighborhood of the seed point do not meet the expansion condition, thus forming the initial segmentation area of ​​the follicle.

[0042] The edge detection algorithm is applied to the candidate follicle region to extract the edge gradient information and obtain the edge intensity map. At the same time, the Gabor filter is used to extract the main texture direction, texture energy and consistency index of the region to form a complete texture feature map. Select pixels with significant edge strength and stable texture direction as seed points. Define the growth criteria as follows: the average edge strength difference, gray value difference, and texture direction difference between the pixel to be expanded and the current growth area should be lower than the corresponding threshold to be included in the current area; For each seed point, calculate the grayscale difference between its surrounding pixels and the pixels in the marked area in the current area. The grayscale difference calculation formula is: ; In the formula, is the gray value of the pixel (x, y) to be judged, is the gray value of the seed point, is the grayscale difference between the pixel and the seed point; In some preferred embodiments, the grayscale values ​​of pixels in the neighborhood of the seed point are compared with the grayscale values ​​of pixels in the current area: During the region growing process, if the grayscale difference between the neighboring pixel and the current region is less than the preset grayscale similarity threshold, the pixel is considered to be sufficiently similar to the current region and meets the region expansion conditions, and is added to the current segmented region and continues to participate in region growing as a new seed point; If the grayscale difference is greater than or equal to the similarity threshold, the pixel is considered to be too different from the current area and is excluded from the expansion area and not included in the segmentation area.

[0043] S4: Apply the multi-scale circularity matching function to the annotated follicle candidate region to evaluate the matching degree between the circularity characteristics of the candidate region and the standard circle and determine the regularity of the follicle morphology.

[0044] The process of evaluating the matching degree between the roundness feature of the candidate region and the standard circle and determining the regularity of the follicle morphology in S4 is as follows: Extract boundary information from the follicle segmentation region obtained by the region growing process, use edge detection algorithm to extract the contour edge of the follicle region, and perform smoothing on it; The boundaries of the follicles are analyzed at multiple scales, and a multi-scale circularity matching function is defined. The function is based on the geometric characteristics of the standard circle for comparison. The formula is: ; In the formula, Indicated in scale The roundness of the is the boundary length of the candidate region, The radius is used to calculate the roundness; the function measures the similarity between the follicle area and the standard circle. The closer the roundness is to 1, the closer the shape is to the ideal circle. At different scales, calculate the roundness match and select the most suitable scale for roundness evaluation. By adjusting the scale, ensure that the slight changes in the follicle morphology and the circular features at different sizes can be captured to avoid missing the judgment of morphological changes due to a single scale. The roundness match at different scales can be obtained by the following steps: at each scale, adjust the radius of the candidate area and calculate the corresponding boundary length; calculate the roundness at the scale to obtain the roundness match results at multiple scales; Based on the calculated roundness value, a preset threshold is set to judge the regularity of the follicle morphology; In some preferred embodiments, the circularity matching value is compared with a preset threshold: If the circularity matching value is greater than or equal to the pre-screening threshold, it indicates that the follicle morphology is regular; If the circularity matching value is less than the pre-screening threshold, it indicates that the follicle morphology is irregular.

[0045] S5: For follicles that are assessed as regular by the circularity matching function, edge clarity analysis is performed, and the volume and area are calculated based on their morphological characteristics. The regular follicle areas are classified, and the follicles are divided into different categories based on their size, morphology, and distribution, and the follicle count is completed.

[0046] The process of classifying regular follicles and completing the follicle counting in S5 is as follows: The edge clarity analysis was performed on the follicle area that was evaluated as regular by the roundness matching function. The Canny edge detection algorithm was applied to the follicle area to extract the follicle boundary. The gradient intensity, continuity and curvature of the edge were further calculated to evaluate whether the follicle edge was smooth and clear. The edge clarity formula was: ; Where E represents edge clarity, N is the number of edge pixels, is the gradient value of the i-th edge pixel; After confirming that the follicle morphology is regular and the edges are clear, the volume and area of ​​the follicle are calculated, and the area of ​​the follicle is estimated by the number of pixels in the binary image; Follicles are divided into different categories according to their size, shape and distribution characteristics. Specific classification criteria include: Size classification: Based on the area and volume of the follicles, the follicles are divided into small follicles, medium follicles and large follicles, and are classified according to their size range.

[0047] Morphological classification: Based on the roundness, edge smoothness and other characteristics of the follicles, the follicles are further subdivided into regular follicles and irregular follicles.

[0048] Distribution classification: Based on the spatial distribution of follicles in the ovaries, it is determined whether the follicles are evenly distributed, whether they are concentrated in a specific area of ​​the ovaries, and the distribution density. For example, follicles may be divided into evenly distributed follicles, clustered follicles, or individual follicles.

[0049] The follicles are counted by marking and classifying each type of follicle area. The counting method is: traverse all eligible follicle areas and record the classification results of each follicle; count each type of follicle separately to obtain the number of follicles of each type; count the total number of all follicles and the number of follicles in each category to provide a basis for subsequent endocrine analysis.

[0050] The above technical solution can accurately calculate the volume and area of ​​follicles, and classify them based on their morphology, size, distribution and other characteristics, providing a detailed basis for subsequent follicle development analysis, endocrine index evaluation and clinical decision-making. In addition, accurate follicle counting and classification can also help doctors better understand the ovarian function status and the diagnosis process of diseases such as polycystic ovary syndrome.

[0051] S6: Based on the spatial distribution of regular follicles, local density analysis is performed to evaluate the distribution pattern of follicles and generate the ovarian endocrine map feature vector.

[0052] The local density analysis process based on the spatial distribution of regular follicles in S6 is as follows: extracting the spatial position of the follicles from the classified regular follicle region, each regular follicle is identified in the ovarian image by its center coordinates, and the spatial coordinates of the follicles can be converted from the pixel coordinate system of the image to the actual spatial coordinate system to obtain the precise position of the follicle in the ovary; The local density analysis of the spatial distribution of regular follicles aims to evaluate the distribution pattern of follicles in the ovarian region. The local density value of each follicle is obtained by calculating the number of other follicles within a certain range near the center point of a follicle in a local window.

[0053] The process of evaluating the distribution pattern of follicles and generating the ovarian endocrine map feature vector in S6 is as follows: The spatial distribution pattern of follicles was analyzed by statistically analyzing the local density of all follicles. According to the local density value, the follicles were divided into the following distribution patterns: Uniform distribution: The follicles are evenly distributed in the ovarian area, with little local density variation.

[0054] Clustered distribution: The follicles are densely distributed in certain areas of the ovaries, with higher local density.

[0055] Scattered distribution: The follicles are scattered in the ovarian area, and the local density fluctuates greatly.

[0056] Based on the results of local density analysis, the ovarian endocrine map feature vector is generated. The feature vector is used to describe the distribution characteristics and density pattern of ovarian follicles. The steps are: Feature extraction: A set of feature values ​​are extracted through local density values, spatial distribution density of follicles, distribution type and other information. These feature values ​​include: Mean follicle density: reflects the overall density level of follicles distributed in the ovaries.

[0057] Standard deviation of follicle density: describes the degree of dispersion of follicle distribution.

[0058] Distribution pattern category: Based on the analysis results of local density, the distribution pattern of the marked follicles is determined, such as uniform distribution, clustered distribution, or scattered distribution.

[0059] The extracted eigenvalues ​​are combined into a eigenvector, which may include multiple dimensions, such as the local density value of the follicle, the distribution pattern marker, the spatial position and other information, to form a comprehensive ovarian endocrine map eigenvector.

[0060] Through the technical solution of this embodiment, the spatial distribution characteristics of follicles can be accurately analyzed, and the local density and distribution pattern of follicles can be evaluated, thereby providing a more accurate evaluation of the endocrine state of the ovaries. The generated ovarian endocrine map feature vector provides strong data support for subsequent ovarian health analysis, disease prediction and diagnosis, and can serve as an important basis for clinicians to assist in decision-making.

[0061] Example 3 See also Figure 2 As shown, a gynecological endocrine index analysis system based on image recognition according to an embodiment of the present invention includes: Ovarian region recognition module: responsible for processing the original ovarian ultrasound image, automatically identifying the ovarian region and cropping it; Follicle morphology assessment module: Analyze and extract the texture direction of each mesh area and evaluate the texture consistency of the area; Follicle segmentation module: responsible for adaptive segmentation within the selected follicle candidate area; Follicle morphology determination module: evaluates the roundness characteristics of follicles by applying a multi-scale roundness matching function; Secretion map generation module: local density analysis is performed based on the spatial distribution of follicles to evaluate the distribution pattern of follicles.

[0062] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0063] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A gynecological endocrine index analysis method based on image recognition, characterized in that: The following steps are involved: By automatically identifying the ovarian region in the ovarian ultrasound image, the boundary clipping of the ovarian ultrasound image is realized to obtain a number of grid regions; Through the consistency analysis of the texture direction in the grid area, the integrity of the follicle morphology is judged, the incomplete follicles are eliminated, and the candidate follicle area is determined; After determining the candidate follicle region, an adaptive segmentation method based on dynamic region growing is used to mark the follicle, and the speed and threshold of region growing are adaptively adjusted by analyzing local image features. The multi-scale circularity matching function was used to evaluate the circularity characteristics of the follicles in the follicle candidate area and determine the regularity of the follicle morphology. The edge clarity of regular follicles was analyzed and combined with the follicle morphology and spatial distribution to classify and count the follicles; According to the spatial distribution of follicles and combined with local density analysis, the feature vector of ovarian endocrine map is generated.

2. The gynecological endocrine index analysis method based on image recognition according to claim 1, characterized in that: The process of judging whether the follicle morphology is complete is: By calculating the main direction angles of all pixels in the grid area, and counting the mean and mean square error; If the mean square error is less than the threshold, it means that the texture direction of the grid area is consistent, indicating that the follicle morphology is complete; If the mean square error is greater than or equal to the threshold, it means that the texture direction of the grid area is inconsistent, indicating that the follicle morphology is incomplete.

3. The gynecological endocrine index analysis method based on image recognition according to claim 1, characterized in that: The process of labeling follicles is: In the selected follicle candidate area, pixels with stable brightness value, edge intensity and texture direction are selected as initial seed points; Set the initial grayscale similarity threshold and define the region growth criteria, including the grayscale difference between adjacent pixels and the current region, texture similarity, and edge continuity; In the process of region growth, the neighborhood pixels around the seed point are traversed. If the neighborhood pixels meet the similarity criteria, they are incorporated into the current region and used as new seed points. Dynamically adjust the similarity threshold according to local image features to adapt to grayscale changes and texture fluctuations in the image; When all neighboring pixels do not meet the growth conditions, the region growth terminates; Finally, a closed outline of the candidate follicle region is formed and the region is marked as the follicle region.

4. The gynecological endocrine index analysis method based on image recognition according to claim 3 is characterized in that: The process of adaptively adjusting the speed and threshold of region growth according to local image features is as follows: Apply edge detection algorithm to the candidate follicle region to extract edge gradient information and obtain edge intensity map; In the edge intensity map, pixels with significant edge response and high texture direction consistency are selected as initial seed points; For each seed point, calculate the grayscale difference, texture direction difference and edge gradient change between its surrounding pixels and the marked pixels in the current area; According to the degree of difference, determine whether the adjacent pixels meet the region expansion conditions. If they do, they are merged into the current segmentation region and added to the seed point queue; During the growth process, the similarity threshold and growth step size are dynamically adjusted according to the average grayscale, texture direction consistency and edge density of the pixels in the current area; The growing process continues to iterate until all the pixels in the neighborhood of the seed point do not meet the expansion condition, thus forming the initial segmentation area of ​​the follicle.

5. The gynecological endocrine index analysis method based on image recognition according to claim 4, characterized in that: The process of judging whether adjacent pixels meet the region expansion conditions is: During the region growing process, if the grayscale difference between the neighboring pixel and the current region is less than the preset grayscale similarity threshold, the pixel is considered to be sufficiently similar to the current region and meets the region expansion conditions, and is added to the current segmented region and continues to participate in region growing as a new seed point; If the grayscale difference is greater than or equal to the similarity threshold, the pixel is considered to be too different from the current area and is excluded from the expansion area and not included in the segmentation area.

6. The gynecological endocrine index analysis method based on image recognition according to claim 5, characterized in that: The process of determining the regularity of follicle morphology is: Extract boundary information from the follicle segmentation region obtained by the region growing process, and use edge detection algorithm to obtain the contour edge of the follicle region; Fit and sample boundary contours at multiple scales to construct a multi-scale circularity matching function; The roundness values ​​were calculated at different scales, and the scale with the smallest fitting error was selected as the best evaluation scale; The roundness value is compared with the preset threshold. If the roundness is greater than the threshold, the shape is considered regular, otherwise it is irregular.

7. A gynecological endocrine index analysis system based on image recognition, applied to the method according to any one of claims 1 to 6, characterized in that: include: Ovarian region recognition module: responsible for processing the original ovarian ultrasound image, automatically identifying the ovarian region and cropping it; Follicle morphology assessment module: Analyze and extract the texture direction of each mesh area and evaluate the texture consistency of the area; Follicle segmentation module: responsible for adaptive segmentation within the selected follicle candidate area; Follicle morphology determination module: evaluates the roundness characteristics of follicles by applying a multi-scale roundness matching function; Secretion map generation module: local density analysis is performed based on the spatial distribution of follicles to evaluate the distribution pattern of follicles.

Images