Follicle region image segmentation method for ovarian tissue analysis

Abstract

The present invention relates to a follicle region image segmentation method for ovarian tissue analysis. A feature extraction algorithm is designed for image characteristics under different imaging modes including ultrasound and magnetic resonance; the spatial offset of imaging angles or tissue deformation between various modes is balanced by using an image alignment algorithm; after feature extraction, a weighted fusion algorithm is used to fuse multiple features of density, texture, and edge according to the weight distribution of each mode in different regions; the fusion mechanism uses the structural characteristics of the follicle region to preliminarily detect the candidate region of the follicle; while retaining the boundary of the follicle region, an anisotropic diffusion filter is used to smooth the noise inside the region; the diffusion direction is controlled by local gradient information; the filter parameters are dynamically adjusted according to the morphology of the candidate region; for follicles of different scales and morphologies, the filter adaptively adjusts the smoothness according to the edge characteristics of the follicle.

Description

Technical Field

[0001] The invention relates to an image segmentation method, in particular to an ovarian follicle region image segmentation method for ovarian tissue analysis. Background Art

[0002] In the prior art, the "Real-time Ultrasound Image Follicle Measurement Method and System Based on Intelligent Image Segmentation" represented by Patent No. 2022105927244 proposes an automated follicle recognition and segmentation method based on an intelligent learner. This technology relies on deep learning models such as convolutional neural networks (CNN), region proposal networks (RPN), region alignment modules, mask prediction networks, and multi-layer perceptrons (MLP), combined with ultrasound image training sets, and automatically segments follicles by learning the follicle features in the image. However, although this method has achieved remarkable results in follicle detection in ultrasound images, there are still some advantages and disadvantages, especially when dealing with the complex image structure of ovarian tissue. There are many challenges.

[0003] First, this method relies on a training set of ultrasound images, and the training effect is heavily dependent on the diversity and quality of the training data. Since the morphology of follicles varies greatly among different individuals and in different physiological cycles, the coverage of the training set directly affects the generalization ability of the model. If there are fewer follicle samples in the training set or they are too concentrated on a certain type of follicle, the model is often difficult to accurately identify other types of follicles, especially those in the early developmental stage or with blurred edges. In addition, since ultrasound images themselves have high noise and low signal-to-noise ratio, the intelligent learner has a weak ability to handle noise, which may lead to problems such as unclear boundary recognition and blurred contours when segmenting follicles with blurred edges.

[0004] Secondly, although this method uses a convolutional neural network to extract features and combines it with a region proposal network to select regions of interest, this method relies more on global feature extraction and lacks a detailed analysis of local features in the follicle region. The developmental stages and internal structural changes of follicles are complex, and it is difficult to accurately depict the detailed features of the follicle edge by simply relying on global feature extraction. Especially when the early follicles are small and the boundaries are unclear, the model is prone to miss the detection of subtle boundaries, resulting in unsatisfactory segmentation results. In addition, the mask prediction network and the region alignment module also have limitations in processing smaller follicle regions, especially when the follicles are irregular in shape or are disturbed by surrounding tissues, the segmentation accuracy of the model decreases. Thirdly, although the technology introduces parallel detectors for multi-task detection and combines statistical information to classify and extract contours of follicles, this multi-task model often has deviations in weight distribution when processing multiple tasks at the same time. For complex ovarian tissue images, multi-task processing can easily lead to an imbalance in the model when processing follicle position recognition and morphology prediction, and the model may sacrifice accuracy in the process of processing different tasks. In addition, since the multi-layer perceptron is used for classification and regression of the region of interest, although it can handle the problem of follicle positioning to a certain extent, in the face of complex ovarian images (such as the presence of multiple overlapping or adjacent follicles), the model's classification and boundary positioning capabilities are weak, and follicles may be missed or misdetected. Fourth, although the convolutional neural network has a good performance in feature extraction, the extracted features mainly depend on the size and number of layers of the convolution kernel, and it is difficult to adapt to the morphological changes of follicles at different stages of development. Follicles are usually small in size, fuzzy at the early stage, and low in density, while mature follicles are large in size, with clear edges and more complex internal structures. In this case, the existing segmentation system based on fixed feature extraction methods performs poorly when dealing with follicles of different morphologies and cannot dynamically adjust the segmentation strategy. Moreover, when aligning the features of different regions, the regional alignment module does not consider the physiological characteristics and texture changes of the follicles, resulting in the model being unable to effectively distinguish the boundaries and internal structures of the follicles when dealing with complex textures or noise interference.

[0005] In general, although the existing follicle measurement method based on intelligent image segmentation has achieved automated segmentation through deep learning networks, it is still insufficient in dealing with the diverse morphology of follicles, edge detail processing, contrast enhancement, and multimodal image adaptability. Future improvements should focus on enhancing the adaptability of the model, refining the local feature processing capabilities, and strengthening the dynamic adjustment of the physiological characteristics of follicles to improve the accuracy and stability of segmentation. Summary of the invention

[0006] The purpose of the present invention is to provide a method for segmenting follicle region images for ovarian tissue analysis, thereby solving some of the drawbacks and deficiencies pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solution, which comprises the following steps:

[0008] S1. Follicle candidate region detection based on multimodal feature fusion:

[0009] S1.1. Design feature extraction algorithms for image characteristics under different imaging modes including ultrasound and magnetic resonance imaging; balance the spatial offset of imaging angles or tissue deformation between different modalities by using image alignment algorithms;

[0010] S1.2. After feature extraction, a weighted fusion algorithm is used to fuse multiple features such as density, texture, and edge according to the weight distribution of each modality in different regions. The fusion mechanism uses the structural characteristics of the follicle region to preliminarily detect the candidate region of the follicle.

[0011] S2. Region boundary refinement based on nonlinear filtering:

[0012] While retaining the boundaries of the follicle region, an anisotropic diffusion filter is used to smooth the noise inside the region; the diffusion direction is controlled by local gradient information; the filter parameters are dynamically adjusted according to the morphology of the candidate region; for follicles of different scales and morphologies, the filter adaptively adjusts the smoothness according to the edge characteristics of the follicle;

[0013] S3. Follicle region morphology fitting based on deformation model:

[0014] S3.1. construct an active contour model to fit the follicle morphology through the forces of internal and external energy; and allow the contour to deform in the image to adaptively adjust the shape to match the actual boundary of the follicle;

[0015] S3.2. During the fitting process, the external energy field of the model is dynamically adjusted according to the edge strength, morphological characteristics, and density distribution of the adjacent tissues in the follicle region;

[0016] S4. Optimization of segmentation model based on specific follicle physiological characteristics:

[0017] According to the changes in the size, shape, and internal texture of follicles at different developmental stages, physiological characteristic parameters are introduced to adjust the segmentation model; for example, early follicles are small in size and have blurred edges, so the edge detection sensitivity is enhanced during segmentation; while mature follicles are larger and have clear structures, so the comparison between the internal and external areas is emphasized during segmentation;

[0018] S5. Optimization of segmentation results based on topological structure preservation:

[0019] S5.1. During the segmentation process, the integrity of the follicle region is controlled to be intact through algorithms based on topological structure preservation, including regional stability and connectivity analysis; the topological structure preservation algorithm identifies the cavity or edge fracture area within the follicle;

[0020] S5.2. Construct a graph structure model of the follicle region, analyze connectivity, and use the minimum spanning tree and connected component algorithm to reconnect or separate the broken parts within the segmented region;

[0021] S6. Follicle region dynamic tracking and adaptive segmentation feedback:

[0022] S6.1. Track the changes in follicle morphology at multiple time points by constructing a time series model of follicle changes over time; compare the segmentation results at different time points, and if significant differences are found, provide feedback and adjust the segmentation algorithm;

[0023] S6.2. Based on dynamic tracking data, segmentation parameters are adjusted in real time, including boundary detection, morphology fitting, and topology preservation, to control the timely update of segmentation results over time.

[0024] Furthermore, the region boundary refinement method based on nonlinear filtering includes:

[0025] The local gradient control mechanism is adopted to adjust the diffusion direction by relying on the grayscale changes around the pixel points; when the gradient in the image is at the boundary, the diffusion rate is suppressed; when the gradient is inside the region, a greater degree of diffusion is allowed; the formula is used to describe it:

[0026] Let G(x,y) be the local gradient at position (x,y), which is calculated as follows:

[0027]

[0028] Where: G(x,y) represents the local gradient at the image coordinate (x,y); Δ intensity (z, y) is the change in pixel intensity at that point, i.e., the grayscale difference, which is used to preliminarily determine the change in the area; α is a constant used to control the influence of the second-order derivative on the gradient; a larger α value enhances the effect of the second-order gradient, making the gradient more sensitive to boundary recognition; It is the second-order derivative of the pixel in the horizontal direction, which describes the change intensity of the edge features of the area and is used to determine whether the area is a boundary.

[0029] Furthermore, the region boundary refinement method based on nonlinear filtering includes:

[0030] Introducing a diffusion rate calculation method based on gradient amplitude; dynamically adjusting the diffusion rate according to the magnitude of the local gradient, inhibiting diffusion at the boundary and enhancing diffusion in the internal area; making the segmentation process adapt to the boundary characteristics of the follicle;

[0031] The diffusion rate D(x,y) is defined as follows:

[0032]

[0033] in:

[0034] D(x,y) represents the diffusion rate at position (x,y) and controls the intensity of diffusion;

[0035] β is the adjustment coefficient, which is used to balance the effect of gradient amplitude on diffusion rate. A larger β value makes diffusion more strongly suppressed at the boundary. G(x, y) is the local gradient of the point, and the gradient value is used to determine whether it is a boundary. If the gradient value G(x, y) is at the boundary, the diffusion rate D(x, y) is small. If the gradient value is small, the diffusion rate is relatively large, thereby enhancing the smoothing effect of the internal area.

[0036] Furthermore, the region boundary refinement method based on nonlinear filtering includes:

[0037] An adaptive adjustment strategy of filter parameters based on the morphology of candidate regions is introduced; the filter dynamically adjusts the diffusion intensity and number of iterations according to the morphological characteristics of the follicles, including size and shape, so as to adapt to follicle regions of different morphologies;

[0038] The number of diffusion iterations Dynamically adjusted according to the morphological characteristics of the region, the expression is:

[0039]

[0040] in: Indicates the number of diffusion iterations, that is, the number of steps the filter runs in the diffusion process; by dynamically adjusting the number of iterations, the smoothness of the diffusion process is controlled; D max is the maximum diffusion rate, which is used to set the maximum smoothness allowed in the diffusion process; γ is a control parameter, which is used to adjust the influence of the morphology of the candidate region on the number of diffusion iterations; S f represents the foreground area of ​​the candidate region, that is, the size of the follicle region; S b Represents the area of ​​the background area; by calculating the area difference between the foreground and the background, the filter adaptively adjusts the smoothness of areas of different sizes to ensure that more iterations are allowed for larger follicle areas, while limiting the number of iterations for smaller follicle areas.

[0041] Furthermore, the segmentation model optimization method based on specific follicle physiological characteristics includes:

[0042] The physiological characteristic parameters of the follicles are introduced; the physiological characteristic parameters integrate the volume, density and edge clarity of the follicles. For early follicles, the volume is small and the edges are blurred; while mature follicles have larger volumes and clearer boundaries; by introducing physiological characteristics, the segmentation model dynamically adjusts the sensitivity at different stages; and the physiological characteristic parameters P(x, y, t) are expressed as follows:

[0043]

[0044] Among them: P(x,y,t) is the physiological characteristic parameter at the position (x,y) at time t, and the parameter is dynamically adjusted according to the volume, density and edge clarity of the follicle; V(x,y,t) represents the volume of the follicle, that is, the area size within the follicle region, which increases with the development stage of the follicle; ρ(x,y,t) is the density of the follicle tissue, which is used to reflect the compactness of the internal structure of the follicle; It is the edge clarity gradient of the area, indicating the sharpness or blurriness of the follicle edge and reflecting the boundary visibility; α, β, and γ are the weight coefficients of each parameter, which balance the effects of volume, density, and edge clarity according to actual applications.

[0045] Furthermore, the segmentation model optimization method based on specific follicle physiological characteristics includes:

[0046] By dynamically adjusting the threshold and gradient response weight of the edge detection algorithm, the details of the fuzzy boundary are captured; for mature follicles, the model reduces the edge detection sensitivity; and the sensitivity adjustment function D of the edge detection edge (x,y) is defined as follows:

[0047]

[0048] Where: D edge (x, y) is the edge detection response weight at position (x, y), which controls the sensitivity of the edge detection algorithm to different areas; κ is the adjustment coefficient that controls the sensitivity of edge detection. The larger the value, the more sensitive the model is to edge changes. When κ increases, the edge response becomes more prominent. P(x, y, t) is a physiological characteristic parameter, which is a comprehensive parameter in the above formula. δ is the threshold of edge detection, which is used to determine the starting point of edge detection.

[0049] Furthermore, the segmentation model optimization method based on specific follicle physiological characteristics includes:

[0050] Through the adaptive filtering technology based on morphological changes, the filter is adjusted according to the size and shape of the follicle during development; for early follicles, the filter focuses on removing noise while maintaining the integrity of the edge; in the mature stage, the filter is used for smoothing to reduce noise interference; the adaptive filtering mechanism dynamically adjusts the filtering parameters according to the volume and morphological characteristics of the follicle, and the filter strength adjustment function F smooth (x,y) is expressed as follows:

[0051]

[0052] in:

[0053] F smooth (x,y) is the filter strength at position (x,y), which determines the smoothness of the model in this area; θ is the coefficient that controls the smoothing strength and is used to adjust the strength of the filter; P(x,y,t) is a physiological characteristic parameter that is used to dynamically adjust the filtering operation according to the development stage of the follicle; a large P(x,y,t) value indicates a mature follicle with a greater filter smoothing strength, while a small P(x,y,t) value reduces smoothing and retains edge features.

[0054] Furthermore, the segmentation model optimization method based on specific follicle physiological characteristics includes:

[0055] By analyzing the internal texture and enhancing the regional contrast, the segmentation model distinguishes the internal structure from the external area in the segmentation of mature follicles; the segmentation strategy is dynamically adjusted according to the changes in the internal texture, and the contrast is enhanced to highlight the internal details of the follicle. The contrast enhancement function C between the internal and external areas contrast (x,y) is defined as follows:

[0056]

[0057] Where: C contrast (x,y) represents the contrast between the inner and outer areas at the position (x,y), which is used to distinguish the internal structure of the follicle from the external background; A is the internal area of ​​the follicle, which refers to the part of the model sampled within the follicle area; B is the external area of ​​the follicle, which represents the background around the follicle; I(x,y) is the gray value at the position (x,y), and the average gray value in the area is obtained by integral calculation.

[0058] The follicle region image segmentation method for ovarian tissue analysis of the present invention has the following beneficial effects:

[0059] By introducing the physiological characteristic parameters of follicles (including volume, density and edge clarity), the segmentation model is dynamically optimized, enabling the model to adapt to follicle regions at different developmental stages. Both early follicles with blurred edges and mature follicles with clear edges can be segmented with high precision. Based on the adaptive filtering technology of morphological changes, the filtering strength and number of iterations are dynamically adjusted to ensure that edge details are retained in the early follicle region and to avoid excessive smoothing. In the mature follicle region, the filter can effectively remove noise and ensure the integrity and clarity of the internal structure.

[0060] Through the analysis of internal texture and regional contrast enhancement technology, the present invention can distinguish the internal structure of the follicle from the external area. Especially in mature follicles, the contrast enhancement makes the details inside the follicle stand out, improving the accuracy and stability of segmentation. The sensitivity of edge detection is dynamically adjusted according to the development stage of the follicle. The fuzzy boundaries of early follicles can be captured more sensitively. In the mature follicle area, the model reduces the sensitivity to avoid excessive processing of clear boundaries and reduce false detection and false segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of the follicle region image segmentation method for ovarian tissue analysis of the present invention.

[0062] Figure 2 This is a flow chart of the region boundary refinement method based on nonlinear filtering of the present invention.

[0063] Figure 3 Middle A shows histological images of follicles in ovarian cortical tissues of three groups (fresh, FragDay7, and Frag+cIVADay7), showing the overall morphology of the follicles and the surrounding stromal tissue.

[0064] Figure 4 Panels B and C show the survival and growth rates of follicles in different treatment groups, with statistical analysis performed for healthy and atretic follicles, respectively.

[0065] Figure 4 Middle D shows the comparative analysis of the oocyte diameter in the secondary follicle and the overall follicle diameter.

[0066] Figure 5 Middle E shows DNA damage detected by TUNEL and cell proliferation by Ki67 staining, demonstrating the levels of apoptosis and proliferation of follicles under different treatment conditions.

[0067] Figure 6 Middle (F) shows the statistical analysis results of Ki67 and γH2AX positive cells in different treatment groups. DETAILED DESCRIPTION

[0068] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0069] The S1 step of the follicle region image segmentation method for ovarian tissue analysis is the detection of follicle candidate regions based on multimodal feature fusion. Its core purpose is to combine the image characteristics under multiple imaging modes, extract features related to follicles and detect candidate regions. According to the image characteristics of different imaging modes (such as ultrasound and magnetic resonance imaging), a feature extraction algorithm is first designed. In ultrasound images, density and texture features are mainly extracted, while magnetic resonance images focus on extracting structure and edge features. Since the spatial information of each imaging mode is different, the imaging angle or tissue morphology may be offset due to different imaging technologies, so an image alignment algorithm is required. The image alignment algorithm spatially aligns images of two or more modes and corrects the differences caused by imaging angles and tissue deformation to ensure alignment and consistency between images of different modes.

[0070] After feature extraction is completed, a weighted fusion algorithm is used to fuse the features in the multimodal image. The key to the weighted fusion algorithm is to assign weights to the image features of different modalities, and the size of the weight depends on the performance of each modality in different areas. For example, magnetic resonance imaging performs better in capturing tissue structure, so higher weights are given to feature areas related to structure (such as edges); while ultrasound images perform better in extracting density and texture features, so higher weights are given to these areas. Through the reasonable allocation of weights and the complementarity of different modalities, the fusion algorithm can comprehensively process multiple features such as density, texture, and edges in multimodal images.

[0071] This weighted fusion mechanism is designed to highlight the specific structural characteristics of the follicle region, making the follicle morphology more obvious in the image. In the fused image, the characteristics of the follicle region (such as edge clarity and internal density distribution) will be amplified, while other irrelevant background information will be weakened.

[0072] The S2 step is a regional boundary refinement based on nonlinear filtering. Its core goal is to remove the noise inside the region while maintaining the clarity of the follicle region boundary through nonlinear filtering technology. In this step, an anisotropic diffusion filter is used to smooth the noise inside the follicle region. The advantage of anisotropic diffusion filtering is that it can control the direction and intensity of diffusion according to the gradient information of the local area in the image. This filter can effectively suppress diffusion near the boundary to prevent boundary blurring, while allowing a greater degree of diffusion in the internal uniform area, thereby reducing noise and maintaining the structural integrity of the image.

[0073] The key mechanism of anisotropic diffusion is to control the direction of diffusion based on local gradient information. The local gradient refers to the intensity and direction of the change in pixel value in a certain area. Higher gradients usually mean boundaries or significant structural changes. Therefore, diffusion needs to be suppressed in these areas to prevent the smoothing of boundaries from causing edge loss. Conversely, in areas with smaller gradients, diffusion can proceed more freely to effectively remove noise. This feature enables anisotropic diffusion to retain the edges and morphological features of the follicle area while denoising.

[0074] In addition, the filtering parameters are dynamically adjusted according to the morphology of the candidate regions to adapt to the structural characteristics of different follicles. The morphology of follicles (including size, edge clarity, etc.) varies significantly at different stages of development. To this end, the anisotropic diffusion filter dynamically adjusts the diffusion coefficient and number of iterations to adapt to these differences by automatically detecting the morphological features of the candidate regions (such as edge strength, region size, etc.). Specifically, for smaller, early follicles with blurred edges, the filter enhances the protection of weak boundaries and reduces over-smoothing of internal areas; while for larger, mature follicles with clear boundaries, the filter allows a wider range of diffusion to effectively remove internal noise and ensure edge integrity.

[0075] The S3 step is the morphological fitting of the follicle region based on the deformation model. Its core is to use the active contour model (also known as the Snake model) to achieve accurate fitting of the follicle region morphology. The active contour model controls the deformation of the curve through the action of internal energy and external energy, so that the contour can be dynamically adjusted in the image and finally accurately match the actual boundary of the follicle. Specifically, the internal energy mainly controls the smoothness and continuity of the contour to prevent unreasonable distortion or breakage of the contour; while the external energy is mainly responsible for pushing the contour close to the edge of the follicle, ensuring that the contour can gradually approach the actual follicle boundary in the image.

[0076] The active contour model allows its contour to deform freely in the image to adapt to the shape changes of the follicle. This is very critical because the morphology of the follicle may show great differences at different developmental stages and imaging conditions. By deforming and adjusting in the image, the contour can flexibly respond to the actual shape changes of the follicle, ultimately ensuring that the fitted morphology is highly consistent with the actual boundary of the follicle. This morphological fitting process is dynamic. With the action of internal and external forces of the model, the contour gradually approaches the target area, thereby achieving accurate boundary fitting.

[0077] During the morphological fitting process, the active contour model dynamically adjusts the external energy field according to the edge strength, morphological characteristics and density distribution of the adjacent tissues in the follicle area. Edge strength refers to the image gradient at the boundary of the follicle, which is usually used to guide the contour line to shrink toward the target area. Morphological characteristics refer to the size, smoothness or other geometric features of the follicle. This information is used to adjust the behavior of the model to better adapt to the shape of the follicle. The density distribution of the adjacent tissue is used to distinguish the boundary between the follicle and the surrounding tissue. Especially when the edge of the follicle is blurred, the fitting ability of the model is further enhanced by considering the difference in density distribution.

[0078] The S4 step is to optimize the segmentation model based on the physiological characteristics of specific follicles. Its core is to dynamically adjust the segmentation model by introducing the physiological characteristic parameters of follicles at different developmental stages. The size, shape, internal texture and other characteristics of follicles will change significantly at different developmental stages. Therefore, a simple static segmentation model cannot accurately capture these changes. To solve this problem, the segmentation model needs to be adjusted according to the physiological characteristics of the follicles. First, in the early developmental stage, the follicles are small in size and the edges are usually not clear, so the sensitivity of edge detection needs to be enhanced at this stage. By improving the detection ability of fuzzy edges, the model can better capture the boundaries of early follicles and avoid missed detection or incorrect segmentation due to unclear edges.

[0079] As the follicle enters the mature stage, its volume increases, the structure becomes clearer, and the boundary becomes more obvious. In this case, the focus of the segmentation model will shift to the contrast analysis between the internal and external areas of the follicle. In the segmentation of mature follicles, the model needs to enhance the accuracy of extracting internal texture features, because the difference between the internal structure of the follicle and the external background is more obvious at this time, and accurate contrast analysis can help the model more accurately separate the follicle from the surrounding tissue. At the same time, the intensity of the edge of the mature follicle is also relatively high, so there is no need to over-enhance the sensitivity of edge detection, but to pay more attention to the integrity of the overall structure and the processing of details.

[0080] By introducing these physiological characteristic parameters, the model can adaptively adjust its processing strategy at different developmental stages. For example, the early follicle stage focuses on improving the response of edge detection and fine processing of weak edges; while in the mature follicle stage, it focuses on analyzing internal textures and enhancing regional contrast. This dynamic adjustment process enables the segmentation model to flexibly respond to changes in the characteristics of different follicle developmental stages, avoiding differences in details that cannot be captured by traditional static segmentation models.

[0081] Step S5 is the optimization of the segmentation results based on topological structure preservation. Its core is to ensure that the topological structure of the follicle area is not destroyed during the segmentation process. This step optimizes the segmentation results through the algorithm of topological structure preservation, mainly including the stability and connectivity analysis of the region. Due to the complex morphology of the follicle in the image, internal cavities or edge breaks may appear during the segmentation process, and these changes in topological structures may lead to inaccurate segmentation results. Therefore, the task of this algorithm is to identify the possible cavities or edge break areas in the follicles and ensure that these areas can maintain a reasonable structure in the final segmentation results.

[0082] To achieve this goal, we first need to construct a graph structure model of the follicle region. In the graph structure model, the follicle region is considered to be composed of multiple connected pixels, each node in the graph represents a pixel, and the edge represents the connectivity relationship between pixels. In this way, the overall morphology of the follicle can be analyzed for connectivity. The purpose of connectivity analysis is to identify the broken parts caused by blurred edges or mis-segmentation during the segmentation process, and ensure that the various parts within the follicle region remain coherent.

[0083] Once the graph structure model is established, the minimum spanning tree (MST) and connected component algorithm are used to optimize the segmentation results. The purpose of the minimum spanning tree algorithm is to reduce the number of edges while ensuring node connectivity, thereby maintaining the simplest structure of the follicle region. Through the minimum spanning tree, the shortest path in the graph structure can be found so that the various parts of the follicle region can be reconnected with the least number of connections. The connected component algorithm is used to identify different connected regions in the graph. If an area is mistakenly separated due to missegmentation, the connected component algorithm can detect these separated parts and determine whether these parts need to be reconnected by analyzing their topological relationships.

[0084] When edge breaks or cavities appear in the segmented area, the topology-preserving algorithm will automatically make adjustments to ensure that these broken parts can be reasonably connected or separated. For example, for the cavity areas in the follicle, the algorithm will ensure that these areas are not mistaken for external areas to avoid confusing the cavity parts with the external areas. In the case of edge breaks, the algorithm will reconnect the broken parts through connectivity analysis to ensure the integrity of the follicle boundary.

[0085] Step S6 is the dynamic tracking and adaptive segmentation feedback of the follicle area. Its core is to track the morphological changes of follicles at multiple time points by constructing a time series model of the changes of follicles over time, and dynamically adjust the segmentation algorithm according to these changes. The time series model can capture the changes of follicles at different times, including features such as morphology, size, and boundary clarity, so as to monitor the development process of follicles throughout. In this process, the model will record the segmentation results at each time point, and compare the segmentation results between different times to identify significant changes or differences. If the segmentation results at a certain moment are significantly different from the segmentation results at the previous and next moments, the system will automatically trigger the feedback mechanism and adjust the segmentation algorithm to ensure the consistency and accuracy of the segmentation results.

[0086] The feedback mechanism plays a vital role in the process of comparing segmentation results. By analyzing time series data, the system can identify potential errors in the segmentation process in real time, such as blurring of boundaries, abnormal changes in morphology, or distortion of topological structure. Once these anomalies are identified, the feedback mechanism will immediately start to automatically adjust the parameters of the segmentation algorithm to optimize key steps such as boundary detection, morphological fitting, and topological structure preservation. For example, if the segmentation results at a certain time point show that the edge of the follicle is blurred, the feedback mechanism will enhance the sensitivity of the boundary detection to ensure that the boundary can be identified more accurately.

[0087] Based on dynamic tracking data, the system will adjust the key parameters of the segmentation in real time to ensure that the segmentation results can be updated in time over time. Dynamic tracking allows the system to gradually adjust the parameters of the segmentation algorithm according to the actual morphological changes of the follicles, so that they match the developmental stage and morphological characteristics of the follicles. For example, early follicles may require more sensitive boundary detection parameters, while mature follicles need to pay more attention to the morphological fitting of the internal structure and the maintenance of the topological structure. Through this adaptive segmentation feedback mechanism, the segmentation algorithm can be adjusted according to the differences in follicle morphology at different time points to ensure the accuracy and consistency of the segmentation results.

[0088] In addition, dynamic tracking combined with a feedback mechanism not only improves segmentation accuracy, but also ensures that the model responds quickly to changes in follicle morphology. Over time, follicle morphology may change slowly or significantly, and the system continuously captures these changes through dynamic tracking, ensuring that the segmentation results are always consistent with the latest morphological information.

[0089] Embodiment 1:

[0090] In actual ovarian tissue analysis, image segmentation methods are needed to detect the boundaries of the follicle region. In this example, the image data comes from ultrasound imaging and magnetic resonance imaging (MRI) of ovarian tissue, and the goal is to accurately detect the boundaries of the follicles and remove image noise.

[0091] At the initial stage of the analysis, a grayscale image of ovarian tissue was imported. The follicle area in the image was partially covered by noise, resulting in unclear boundaries, especially in the early follicular stage, which made edge detection difficult. At this time, it was decided to use a regional boundary refinement method based on nonlinear filtering to enhance the boundary. First, the local gradient G(x,y) of each pixel in the image was calculated to identify the difference between the boundary and the interior of the region.

[0092] Starting from a small patch of the image, this part of the area includes the edge of the follicle and the background area. To calculate the local gradient G(x,y), use the following formula:

[0093]

[0094] In this formula, Δ intensity (x, y) represents the grayscale difference of the pixel at that point, which is used to preliminarily judge the change between pixels. For example, if the pixel intensity I(x1, y1) is 150 (grayscale value) at a certain area (x1, y1) of the image, and the average intensity of the surrounding pixels is 140, then Δ intensity (x1,y1)=10.

[0095] Next, calculate the horizontal second-order derivative of the point Assume that the grayscale change near I(x1,y1) is not large in the horizontal direction, that is, the grayscale difference of nearby pixels is small, and the second-order derivative value in the horizontal direction is In order to enhance the sensitivity of the gradient to the boundary, α is set to 0.8, so the local gradient is calculated:

[0096] G(x1,y1)=10+0.8·5=14

[0097] This local gradient value is used to determine whether the point is a boundary. A larger gradient value means that there may be a boundary in the area, so in the boundary area, the diffusion rate needs to be suppressed to prevent the smoothing operation from destroying the edge features.

[0098] Next, set a pixel point (x2, y2) inside the follicle area, whose grayscale value does not change much with the surrounding pixels, and set Δ intensity (x2,y2)=2, and its second-order derivative value in the horizontal direction is small, for example So we get:

[0099] G(x2,y2)=2+0.8·1=2.8

[0100] The smaller value indicates that this is the inner region of the follicle, where the gradient is lower, thus allowing for a greater degree of diffusion to remove noise inside the region.

[0101] In actual operation, this local gradient control mechanism can adaptively adjust the diffusion rate. When a large gradient value such as G(x1,y1) is detected, the diffusion is suppressed to protect the boundary; while in the internal area such as G(x2,y2) with a small gradient value, the diffusion is enhanced to smooth the noise.

[0102] In order to further verify the effectiveness of the method, this algorithm was applied to the actual image and the parameter range was set: the value range of α was 0.5 to 1.5. When α was larger, the boundary recognition was enhanced, and when α was smaller, the smoothing of the internal area was enhanced. After experiments, when α was set to 0.8, the follicle boundary could be well identified and the internal noise of the image was effectively smoothed. Through such nonlinear filtering processing, the boundary of the follicle area in the image became clearer and the noise interference was removed, proving the feasibility and accuracy of this method in follicle area segmentation.

[0103] Next, we will further introduce a method for calculating the diffusion rate based on the gradient amplitude and explain in detail how to dynamically adjust the diffusion rate D(x,y) so that the segmentation process can adapt to the boundary characteristics of the follicle.

[0104] To better explain this process, we continue to use the ovarian tissue image mentioned above. In this image, a certain area is assumed to contain multiple follicles, which have different boundary strengths and internal textures at different stages of development. The gradient amplitude G(x,y) will be introduced to control the diffusion rate, and the diffusion rate will be dynamically adjusted according to the gradient values ​​at different points. At the boundary, the diffusion rate should be small to protect the edge details; in the internal area, the diffusion rate should be large to smooth the noise.

[0105] According to the calculation formula of diffusion rate:

[0106]

[0107] in:

[0108] D(x,y) represents the diffusion rate at position (x,y) and is used to control the diffusion intensity;

[0109] β is the adjustment coefficient, which is used to balance the effect of gradient amplitude on diffusion rate;

[0110] G(x,y) is the local gradient of the point, and the gradient value is used to determine whether the point is a boundary.

[0111] First, set a pixel point (x1, y1) at the edge of the follicle, and calculate the local gradient G(x1, y1) = 14. At this time, select the adjustment coefficient β = 0.5, because it is hoped that the diffusion at the boundary will be significantly suppressed to avoid boundary blur. Calculate according to the formula:

[0112]

[0113] Therefore, the diffusion rate at the boundary D(x1,y1)≈0.0101 is very small, which indicates that diffusion is strongly suppressed at this boundary point, thereby protecting edge details and preventing key boundary information from being blurred during image segmentation.

[0114] Next, consider a pixel point (x2, y2) in the inner area of ​​the follicle, and set the local gradient G(x2, y2) of this point to 2.8, because the inner area usually has small changes and low gradients. The adjustment coefficient β is still selected to be 0.5. Calculate according to the formula:

[0115]

[0116] Therefore, the diffusion rate D(x2,y2)≈0.203 in the internal area is much higher than the diffusion rate at the boundary, which allows a wider range of diffusion in the internal area to effectively smooth the noise without affecting the clarity of the boundary. To further illustrate the feasibility of this method, different gradient ranges are set throughout the image and the response of the diffusion rate is adjusted. Through experiments, the β value range selected is 0.1 to 1.0. Larger β values ​​(such as 1.0) suppress diffusion more strongly when protecting the boundary, while smaller β values ​​(such as 0.1) are more used to enhance the smoothing effect of the internal area. In this case, β=0.5 is used as an intermediate value, which can well balance boundary protection and internal area smoothing in actual image processing.

[0117] This method can also be extended to dynamically adjust the value of β in more complex follicular structures. For example, for more complex follicular morphology, the β value may need to be increased to 0.7 or higher to better suppress diffusion on complex boundaries, while in some areas with relatively smooth boundaries and more internal noise, β can be reduced to 0.3 to enhance noise removal capabilities. Through this method, the follicle area in the ovarian tissue image can be processed more finely, the boundaries are clearly retained, and the noise in the internal area is effectively removed, making the segmentation results more accurate and reliable in clinical analysis or research applications.

[0118] The embodiment now enters a higher level optimization of the region boundary refinement method based on nonlinear filtering. The goal of this stage is to introduce an adaptive adjustment strategy of filter parameters based on the candidate region morphology, so that the filter can dynamically adjust the intensity and number of iterations of diffusion according to the different morphologies (including size, shape, etc.) of the follicle region. In this way, the model can adaptively process various follicle regions of different morphologies, thereby further improving the segmentation accuracy.

[0119] In the analysis of an ovarian tissue image containing multiple follicles, some of which are large with clear edges, while others are small with blurred boundaries. To solve this problem, an adaptive adjustment strategy of filter parameters based on the morphology of the candidate region is introduced. The filter dynamically adjusts the intensity and number of iterations of the diffusion according to the size and morphology of the follicle region, so that it can adapt to these follicle regions with different morphologies.

[0120] Specifically, the number of diffusion iterations It is dynamically adjusted according to the morphological characteristics of the follicle area. The formula is as follows:

[0121]

[0122] in:

[0123] Indicates the number of diffusion iterations, that is, the number of steps the filter runs during the diffusion process; D max is the maximum diffusion rate, which is used to set the maximum smoothness allowed in the diffusion process; γ is a control parameter, which is used to adjust the influence of the morphology of the candidate region on the number of diffusion iterations. Usually, the value of γ ranges from 0.1 to 2.0; S f represents the foreground area of ​​the candidate region, that is, the size of the follicle region; S b Represents the area of ​​the background area; |S f -S b | is the area difference between the foreground and the background. This difference is used to measure the size of the follicle area relative to the background, and then the number of iterations is dynamically adjusted.

[0124] Now let’s take a specific example to illustrate the feasibility of this method. Suppose a candidate follicle region is detected in a certain area of ​​the image, and its foreground area S f =500 pixels, background area S b = 1000 pixels. At this time, the area difference |S f -S b |=500. Set the maximum diffusion rate D max =5 and control parameter γ = 0.5, and then substitute into the formula to calculate the number of diffusion iterations

[0125]

[0126] The calculation results are:

[0127]

[0128] This means that for larger follicle regions, the number of diffusion iterations is relatively small, because larger regions are easier to smooth through diffusion without requiring too many iterations.

[0129] Now consider a smaller follicle region, whose foreground area S f = 100 pixels, the background area is still S b = 1000 pixels. At this time, the area difference |S f -S b |=900. Similarly, set D max =5 and γ=0.5, substitute into the formula to calculate:

[0130]

[0131] The calculation results are:

[0132]

[0133] It can be seen that for smaller follicle regions, the number of diffusion iterations is more limited, because smaller regions do not require too many iterations to avoid over-smoothing and boundary loss. Through this adaptive adjustment mechanism, the filter can dynamically adjust the number of iterations and diffusion strength according to the follicle regions of different sizes and shapes, thereby ensuring that the smoothing of different regions is fine and accurate. Larger follicle regions allow more iterations to ensure smoothing; while for smaller follicle regions, the number of iterations will be reduced accordingly to avoid over-smoothing of the boundaries. This ensures that regardless of the size and shape of the follicles, the segmentation process can maintain accurate and consistent boundary detection and noise smoothing.

[0134] Embodiment 2:

[0135] In the follicle region image segmentation method for ovarian tissue analysis, the optimization method of the segmentation model based on the physiological characteristics of specific follicles is a key step. To illustrate the practical application and feasibility of this method, an image from ovarian ultrasound imaging is taken as an example. The image contains multiple follicle regions, each of which is at a different stage of development, so the volume, density, and edge clarity are different. The physiological characteristic parameters will be used to optimize the segmentation model to ensure that the segmentation can adapt to the characteristics of these follicles at different stages.

[0136] First, the introduction of physiological characteristic parameters P(x,y,t) helps to adjust the sensitivity of the segmentation model according to the characteristics of follicles at different developmental stages. In this image, two follicle regions at different developmental stages are selected for analysis: an early follicle and a mature follicle. The early follicle is smaller in size, lower in density, and has blurred edges; while the mature follicle is larger in size, higher in density, and has clear boundaries. The processing method of the segmentation model is dynamically adjusted by calculating the physiological characteristic parameters P(x,y,t).

[0137] The expression of physiological characteristic parameters is:

[0138]

[0139] in:

[0140] P(x,y,t) is the physiological characteristic parameter at the position (x,y) at time t, which represents the comprehensive physiological characteristics of the point;

[0141] V(x,y,t) represents the volume of the follicle, which gradually increases with the development of the follicle and is used to reflect the area within the follicle region; ρ(x,y,t) is the density of the follicle tissue, reflecting the compactness of the internal structure of the follicle; It is the edge clarity gradient of the area, indicating the sharpness or blurriness of the follicle edge, and directly reflects the boundary visibility; α, β, and γ are weight coefficients for controlling volume, density, and edge clarity, respectively, with specific value ranges of α=0.4 to 0.7, β=0.3 to 0.6, and γ=0.2 to 0.5.

[0142] In order to better illustrate the calculation of these physiological characteristics and their application in practical applications, two specific examples are given.

[0143] 1. Division of early follicles

[0144] An early follicle region in the image has a small volume. Its volume is set to V(x1, y1, t1) = 100 pixels. Its density is low. ρ(x1, y1, t1) = 0.3 (normalized density between 0 and 1 relative to normal tissue). Its edge is blurred. The edge clarity gradient is measured. The weight coefficients selected are α = 0.5, β = 0.4, γ = 0.3. Substituting into the formula:

[0145] P(x1,y1,t1)=0.5·100+0.4·0.3+0.3·0.2=50+0.12+0.06=50.18

[0146] Through this physiological characteristic parameter P(x1,y1,t1), the model determines that the area belongs to the early follicle, which is small in size and has blurred edges. Therefore, the sensitivity of edge detection is enhanced during the segmentation process, especially for the refined processing of weak edges to prevent missed detection, while maintaining appropriate processing of internal noise.

[0147] 2. Division of mature follicles

[0148] In the same image, the volume of another mature follicle area is set to be larger, and the volume V(x2,y2,t2) is measured to be 400 pixels, with a higher density. ρ(x2,y2,t2) is set to be 0.8, and the edge is very clear. The edge clarity gradient is measured Using the same weight coefficients α = 0.5, β = 0.4, γ = 0.3, substitute into the formula:

[0149] P(x2,y2,t2)=0.5·400+0.4·0.8+0.3·0.9=200+0.32+0.27=200.59

[0150] Through this physiological characteristic parameter P(x2, y2, t2), the model determines that the area belongs to a mature follicle with a large volume and clear edges. Therefore, during the segmentation process, it focuses on analyzing the internal texture and enhancing the regional contrast, while reducing the sensitivity of edge detection, preventing excessive processing of boundaries, and ensuring a natural transition of boundaries.

[0151] From the above examples, we can see that the introduction of physiological characteristic parameters P(x, y, t) can effectively help the segmentation model to adaptively adjust the follicle region at different developmental stages. For early follicles, due to the fuzzy boundaries and small volume, the model automatically enhances the sensitivity of edge detection; while for mature follicles, the model focuses on the processing of internal structure and boundary clarity.

[0152] The embodiments further explain the method based on dynamically adjusting the threshold and gradient response weight of the edge detection algorithm, especially when dealing with follicles at different developmental stages, how the model adaptively adjusts the edge detection sensitivity to capture the details of blurred boundaries or reduce over-detection of mature follicle boundaries.

[0153] The previous ovarian tissue image is set to continue to be analyzed. The boundaries of early follicles are fuzzy, while the boundaries of mature follicles are very clear. In order to ensure that the model can be flexibly adjusted at different stages, the edge detection sensitivity adjustment function D is used. edge (x,y), and the sensitivity of edge detection is controlled by the physiological characteristic parameter P(x,y,t). This function can help better handle early follicles with blurred boundaries while avoiding over-processing of mature follicles.

[0154] The expression of the edge detection sensitivity adjustment function is as follows:

[0155]

[0156] in:

[0157] D edge (x, y) is the edge detection response weight at position (x, y), which determines the sensitivity of the edge detection algorithm at this position; κ is the adjustment coefficient that controls the sensitivity of edge detection. The value range of κ is 0.5 to 2.0. The larger the κ value, the more sensitive the model is to edge changes; P(x, y, t) is a comprehensive indicator based on the physiological characteristic parameters of the follicle, which represents the overall physiological state of the point and has been calculated before; δ is the threshold of edge detection. δ controls the starting point of edge detection, and the typical value is between 0.1 and 0.5.

[0158] First, the early follicles are processed, and the physiological characteristic parameters of a certain follicle area are set to P(x1, y1, t1) = 50.18. Since the boundaries of early follicles are fuzzy, it is necessary to increase the sensitivity of edge detection, set the adjustment coefficient κ = 1.5, and the edge detection threshold δ = 0.3. Substitute into the formula for calculation:

[0159]

[0160] In this region, since P(x1,y1,t1) is small and has fuzzy boundaries, D edge The (x1,y1) value is close to 1, which means that the edge detection sensitivity reaches the highest level and the model detects the fuzzy boundaries more carefully, thereby capturing the weak boundaries of the early follicles.

[0161] Next, we process the mature follicles and set their physiological characteristic parameters to P(x2, y2, t2) = 200.59. In this area, the boundary is very clear and high sensitivity is not required. Set κ = 1.0, keep the threshold δ = 0.3, and substitute the formula to calculate:

[0162]

[0163] In the mature follicle area, D edge The (x2, y2) value is close to 0, which means that the edge detection sensitivity is greatly reduced, the model's excessive response to the boundary is suppressed, and unnecessary over-processing of already clear boundaries is avoided.

[0164] Through these examples, it can be seen that the edge detection sensitivity adjustment function D edge (x,y) plays a key role in processing follicles at different developmental stages. For early follicles, their boundaries are blurred, and the adjustment coefficient κ is set to a larger value (such as 1.5) to enhance the edge detection sensitivity of the model; while for mature follicles, the adjustment coefficient is smaller (such as 1.0) to reduce the edge detection sensitivity of the model and ensure that the already clear boundaries are not over-processed. Through these calculations and adaptive adjustments, the model can effectively optimize the segmentation of follicle regions at different stages, whether it is capturing the blurred boundaries of early follicles or reducing unnecessary boundary enhancement processing in mature follicles, which can improve the segmentation accuracy and the overall performance of the model.

[0165] The adaptive filtering technology based on morphological changes is further introduced to adjust the parameters of the filter to adapt to the changes in the morphology and size of the follicles during development. The core purpose of adaptive filtering is to dynamically adjust the follicle area according to different stages. For early follicles, due to their small size and blurred edges, the filter focuses on removing noise in the image and maintaining the integrity of the follicle edges. For mature follicles, due to their clear edges and complex internal structures, the filter focuses more on smoothing to reduce noise interference and ensure a smooth transition of the internal structure.

[0166] To achieve this goal, the filter strength adjustment function F is used. smooth (x,y), this function dynamically adjusts the filter strength according to the physiological characteristic parameters P(x,y,t) of the follicle. The formula is as follows:

[0167]

[0168] in:

[0169] F smooth (x,y) is the filter strength at position (x,y), which determines the degree of smoothing at that point;

[0170] θ is a coefficient that controls the smoothing strength, usually ranging from 0.1 to 1.0. The larger the value, the stronger the smoothing effect.

[0171] P(x,y,t) is the physiological characteristic parameter calculated previously, which is used to reflect the development stage and current morphological state of the follicle;

[0172] Larger P(x,y,t) values ​​correspond to mature follicles, so the filter smoothing intensity is greater; smaller P(x,y,t) values ​​correspond to early follicles, so the filter intensity is smaller and more edge features are retained.

[0173] Taking follicles at two different developmental stages as examples, the application process and effects of adaptive filtering are explained in detail.

[0174] In an early follicle region in the image, its physiological characteristic parameter is P(x1,y1,t1)=50.18. Due to its small volume and blurred edges, the filter should focus on removing noise while maintaining edge features. Set the control coefficient of smoothing intensity θ=0.3 and substitute it into the formula:

[0175]

[0176] The result is F smooth(x1,y1)≈0.0013. This result shows that the filtering strength in the early follicle region is weaker because it is hoped to retain edge features and avoid edge blurring caused by excessive smoothing. At the same time, the main function of the filter is to remove local noise, thereby improving segmentation accuracy.

[0177] In another area of ​​the image, a mature follicle is detected, which is large in size and has very clear edges. The physiological characteristic parameters of this area are P(x2, y2, t2) = 200.59. For such a mature follicle, the filter should focus on smoothing to reduce internal noise interference and ensure a smooth transition of the internal structure. Similarly, the smoothing intensity control coefficient θ = 0.3 is set and substituted into the formula:

[0178]

[0179] The result is F smooth (x2,y2)≈0.000083. For mature follicles, the filter strength is significantly higher, meaning that this area is more strongly smoothed to remove noise and highlight the internal structure.

[0180] From the above two examples, we can see that the adaptive filtering mechanism dynamically adjusts the filter strength according to the morphological characteristics of the follicles. For early follicles, the filter maintains a lower smoothing strength, focusing on preserving edge features, while in the mature follicle area, the filter applies a greater smoothing strength to ensure that the noise in the internal area can be effectively removed.

[0181] This mechanism ensures that follicles at different developmental stages can receive appropriate filtering. For early follicles with blurred boundaries, the model uses weaker smoothing to ensure that the boundaries are not lost; while when dealing with mature follicles with complex internal structures, the model uses stronger smoothing to reduce noise interference and enhance the continuity and contrast of the internal area.

[0182] Further discussion is made on the optimization method based on internal texture analysis and regional contrast enhancement, especially when dealing with the segmentation of mature follicles, how to effectively distinguish the internal structure of the follicle from the external background. Since there are usually strong texture and contrast differences between the internal structure and external area of ​​mature follicles, by dynamically adjusting the segmentation strategy and using contrast enhancement technology to highlight the internal details, the internal structure of the follicle and the external background can be better separated.

[0183] Suppose that we are processing an image from ovarian tissue, where the inner area of ​​a mature follicle shows obvious texture changes, while its outer area is relatively uniform. At this time, by calculating the contrast enhancement function C contrast (x,y), can significantly improve the distinction between the inside and outside of the follicle, thereby improving the accuracy of segmentation.

[0184] The contrast enhancement function is defined as follows:

[0185]

[0186] in:

[0187] C contrast (x,y) represents the contrast between the internal and external areas at the position (x,y), which is used to distinguish the internal structure of the follicle from the surrounding background; A is the internal area of ​​the follicle, which represents the part sampled by the model inside the follicle; B is the external area of ​​the follicle, that is, the background area around the follicle; I(x,y) is the gray value at the position (x,y), and the average gray value in the area is obtained by integral calculation.

[0188] Continuing to use the above image, a mature follicle area is detected, and the grayscale difference between the inner and outer areas is large. Assuming that the inner area A of the follicle contains 100 pixels, the average grayscale value of the area is measured as ∫ A I(x,y)dx dy=1500, that is, the average gray value of each pixel is 15. The outer area B of the follicle contains 200 pixels, and the average gray value of this area is ∫ B I(x,y)dxdy=800, that is, the average grayscale value of each pixel is 4.

[0189] Substituting this data into the contrast enhancement function:

[0190]

[0191] The contrast enhancement value C calculated by contrast (x,y)=1.875, which indicates that there is a clear contrast difference between the internal structure of the follicle and the external background. The model can use this contrast value to more accurately segment the follicle region.

[0192] Based on the calculated contrast enhancement value, the segmentation model dynamically adjusts the processing strategy. contrast (x,y) values ​​indicate that the internal area is significantly different from the external background. At this time, the model can use a lower threshold for segmentation because the contrast itself is obvious enough and no additional contrast enhancement is required. In areas with lower contrast (such as early follicles), the model needs to increase the contrast to ensure that weak boundaries can be correctly detected.

[0193] For example, when processing another part of the image, the grayscale value of the inner area of ​​another follicle is set to ∫ A I(x,y)dxdy=1200, the gray value of the outer area is ∫ B I(x,y)dx dy=1100, and we can get the following equation:

[0194]

[0195] This indicates that the contrast between the inside and outside of this area is low. At this time, the model needs to increase contrast enhancement processing to better segment the follicle from the surrounding background.

[0196] Through the above examples, we can see that the contrast enhancement function C contrast (x,y) plays a crucial role in distinguishing the internal structure of the follicle from the external area. A larger contrast value can help the model quickly identify the difference between the internal and external areas. For areas with low contrast, the model automatically adjusts the segmentation strategy to increase the contrast so that the boundary can be presented more clearly.

[0197] Embodiment 3:

[0198] The present invention "follicle region image segmentation method for ovarian tissue analysis" aims to provide an automated, high-precision segmentation and analysis solution for ovarian tissue image analysis, and is particularly suitable for follicle region detection and quantification in complex tissue structures. In this embodiment, combined with the given image content, it is further demonstrated how this method is applied in ovarian tissue experiments, especially under different culture conditions, through the analysis of biological indicators such as follicle growth, survival, proliferation and DNA damage, how to use image segmentation technology to provide accurate data. The analysis of each image will be elaborated in detail, and combined with specific calculation formulas, the specific role of image segmentation in data extraction and quantification will be explained.

[0199] In this experiment, ovarian cortical tissue fragments from 33 women were cultured for 7 days and divided into three groups: fresh tissue (Fresh), Frag group that was only fragmented, and Frag+cIVA group that was treated with chemical in vitro activation (cIVA) after fragmentation. The researchers used a variety of techniques, including histological sections, immunofluorescence staining (Ki67, γH2AX, TUNEL detection) to evaluate follicle survival, growth, DNA damage and cell proliferation. The main goal of the experiment was to explore the survival and growth of follicles and evaluate changes at the cellular level by detecting the effects of different treatments on the follicular area.

[0200] The follicle region image segmentation method of the present invention greatly improves the accuracy and efficiency of follicle region identification and data quantification through automated image segmentation and analysis in this process, and provides indispensable technical support, especially when facing follicles with complex tissue structures and diverse morphologies.

[0201] Image 3A: Segmentation and quantification of follicle histological sections

[0202] Figure 3A in the middle shows histological images of follicles in three groups (fresh, FragDay7, Frag+cIVADay7) of ovarian cortical tissue, showing the overall morphology of the follicles and the surrounding matrix tissue. The follicle region image segmentation technology of the present invention can automatically segment the follicle region based on the morphological characteristics of the follicle, and further quantify the growth of the follicle by calculating the geometric parameters such as the diameter and circumference of each follicle.

[0203] Suppose we use this segmentation algorithm to automatically detect the follicle area first. In the image, the boundary of each follicle is identified by anisotropic diffusion filtering and local gradient detection. The segmented follicle area A follicle The area can be obtained by integration:

[0204] A follicle =∫ Ω dx

[0205] Among them, Ω represents the boundary of the follicle. Then, the diameter and circumference of each follicle were calculated by the boundary curve detection algorithm to help quantitatively evaluate the follicle growth of different treatment groups.

[0206] For example, in the Frag+cIVADay7 group, the diameter of the larger follicle can be automatically calculated by the segmentation algorithm of the present invention as D follicle ≈120mum. For the follicles in the Frag group, due to the blurred edges, the segmentation sensitivity was automatically adjusted by the segmentation algorithm, and a diameter of approximately D follicle =80mum, which is highly consistent with the result of manual detection.

[0207] Figure 4 Middle B and 4C: Statistics of follicle survival and growth rate:

[0208] Figure 4 Figures B and 4C show the survival rate and growth rate of follicles in different treatment groups, and statistical analysis is performed on healthy and atretic follicles, respectively. Through the automated image segmentation method of the present invention, the boundaries and morphology of each follicle can be accurately identified, and healthy follicles and atretic follicles can be classified. The segmentation algorithm is dynamically adjusted according to the volume, density and edge clarity of the follicles during the analysis process, and healthy and atretic follicles are automatically distinguished.

[0209] Specifically, the physiological characteristic parameters P(x,y,t) of the follicle are defined as:

[0210]

[0211] Among them, V(x,y,t) represents the volume of the follicle, ρ(x,y,t) is the density of the follicle area, is the follicle edge clarity gradient. This physiological characteristic parameter is used to determine the health status of the follicle. By calculating P(x,y,t) of each follicle, the system can automatically classify healthy and atretic follicles.

[0212] For example, in the Frag+cIVADay7 group, the healthy follicles detected by the segmentation algorithm accounted for more than 80% of the total number of follicles, while in the Frag group, the proportion of healthy follicles dropped to about 50%, indicating that cIVA treatment can significantly improve the survival rate of follicles. Through the automatic segmentation technology of the present invention, researchers can quickly obtain statistical results without manually marking each follicle.

[0213] Figure 4 Middle D: Measurement of follicle and oocyte diameter:

[0214] Figure 4 D in the middle shows the comparative analysis of the diameter of the oocyte in the secondary follicle and the diameter of the whole follicle. Using the automated segmentation method of the present invention, the diameter of the follicle can be accurately measured, and the diameter of the oocyte can be further calculated.

[0215] Specifically, the diameter of the follicle D follicle and the diameter of the oocyte D oocyte Calculated by the following formula:

[0216] Specifically, the diameter of the follicle D follicle and the diameter of the oocyte D oocyte Calculated by the following formula:

[0217]

[0218] Among them, A follicle and A oocyte are the areas of the segmented follicle and oocyte regions, respectively.

[0219] In the Frag+cIVA group, the diameter of the secondary follicle measured by segmentation was D follicle =120 mm, the diameter of the oocyte is D oocyte =80mum, these data support the conclusion that cIVA treatment can promote follicle growth.

[0220] Figure 5 Middle E: Analysis of immunofluorescence staining images:

[0221] Figure 5 E in the figure shows the DNA damage detected by TUNEL and the cell proliferation stained by Ki67, showing the cell apoptosis and proliferation level of follicles under different treatment conditions. The image segmentation method of the present invention identifies positively stained cells by automated segmentation technology, and counts the proportion of Ki67 and γH2AX positive cells in different regions.

[0222] In the Frag+cIVADay7 group, there were relatively few DNA damage positive cells stained by TUNEL, while the number of Ki67 positive cells increased significantly. Through the automatic segmentation technology of the present invention, the percentage of Ki67 positive cells in the follicle area can be accurately calculated and compared with the TUNEL staining area. This automated analysis method greatly improves the data processing efficiency and reduces the error of manual labeling.

[0223] Figure 6 Middle F: Statistical analysis of Ki67 and γH2AX positive cells:

[0224] Figure 6 F shows the statistical analysis results of Ki67 and γH2AX positive cells in different treatment groups. The segmentation technology of the present invention can automatically identify the number of immunofluorescence staining positive cells and perform ratio calculation.

[0225] The percentage of Ki67 and γH2AX positive cells was calculated by the following formula:

[0226]

[0227] For example, in the Frag+cIVADay7 group, the Ki67-positive cells detected by the system accounted for approximately 3.5% of the total cell population, while in the Frag group this proportion was only 1.5%. These results, obtained through automated image segmentation, support the differences in follicular cell proliferation and DNA damage under different treatment conditions.

[0228] Through the "follicle region image segmentation method for ovarian tissue analysis" of the present invention, researchers can quickly and accurately identify and quantify the growth, survival, proliferation and DNA damage of follicles in complex ovarian tissue images. This technology has a high degree of automation, especially when facing complex follicle morphological changes, it can achieve accurate segmentation by dynamically adjusting the segmentation parameters. By combining image segmentation and quantitative analysis, this method significantly improves experimental efficiency, reduces manual operation errors, and provides strong technical support for the analysis of follicle regions in ovarian tissue.

[0229] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for segmenting follicle regions for ovarian tissue analysis, characterized in that The steps include: S1. Follicle candidate region detection based on multimodal feature fusion: S1.

1. Design feature extraction algorithms for image characteristics under different imaging modes including ultrasound and magnetic resonance imaging; balance the spatial offset of imaging angles or tissue deformation between different modalities by using image alignment algorithms; S1.

2. After feature extraction, a weighted fusion algorithm is used to fuse density, texture, and edge features according to the weight distribution of each modality in different regions; The fusion mechanism uses structural features that highlight the follicle region to preliminarily detect candidate regions of the follicle; S2. Region boundary refinement based on nonlinear filtering: While retaining the boundaries of the follicle region, an anisotropic diffusion filter is used to smooth the noise inside the region; the diffusion direction is controlled by local gradient information; the filter parameters are dynamically adjusted according to the morphology of the candidate region; for follicles of different scales and morphologies, the filter adaptively adjusts the smoothness according to the edge characteristics of the follicle; S3. Follicle region morphology fitting based on deformation model: S3.

1. construct an active contour model to fit the follicle morphology through the forces of internal and external energy; and allow the contour to deform in the image to adaptively adjust the shape to match the actual boundary of the follicle; S3.

2. During the fitting process, the external energy field of the model is dynamically adjusted according to the edge strength, morphological characteristics, and density distribution of the adjacent tissues in the follicle region; S4. Optimization of segmentation model based on specific follicle physiological characteristics: According to the changes in the size, shape, and internal texture of follicles at different developmental stages, physiological characteristic parameters are introduced to adjust the segmentation model; for example, early follicles are small in size and have blurred edges, so the edge detection sensitivity is enhanced during segmentation; while mature follicles are larger and have clear structures, so the comparison between the internal and external areas is emphasized during segmentation; S5. Optimization of segmentation results based on topological structure preservation: S5.

1. During the segmentation process, the integrity of the follicle region is controlled to be intact through algorithms based on topology preservation, including regional stability and connectivity analysis; the topology preservation algorithm identifies the cavity or edge break region within the follicle; S5.

2. Construct a graph structure model of the follicle region, analyze connectivity, and use the minimum spanning tree and connected component algorithm to reconnect or separate the broken parts within the segmented region; S6. Follicle region dynamic tracking and adaptive segmentation feedback: S6.

1. Track the changes in follicle morphology at multiple time points by constructing a time series model of follicle changes over time; compare the segmentation results at different time points, and if significant differences are found, provide feedback and adjust the segmentation algorithm; S6.

2. Based on the dynamic tracking data, the segmentation parameters are adjusted in real time, including boundary detection, morphology fitting, and topology preservation, to control the update of the segmentation results over time.

2. The follicle region image segmentation method for ovarian tissue analysis according to claim 1, characterized in that The region boundary refinement method based on nonlinear filtering comprises: The local gradient control mechanism is adopted to adjust the diffusion direction by relying on the grayscale changes around the pixel points; when the gradient in the image is at the boundary, the diffusion rate is suppressed; when the gradient is inside the region, a greater degree of diffusion is allowed; the formula is used to describe it: set up For location The local gradient at is calculated as follows: , in: Represented in image coordinates The local gradient at ; is the pixel intensity change at that point, i.e., the grayscale difference, which is used to preliminarily determine the change in the area; is a constant used to control the influence of the second-order derivative on the gradient; a larger The value enhances the effect of the second-order gradient, making the gradient more sensitive to boundary recognition; It is the second-order derivative of the pixel in the horizontal direction, which describes the change intensity of the edge features of the area and is used to determine whether the area is a boundary.

3. The follicle region image segmentation method for ovarian tissue analysis according to claim 2, characterized in that The regional boundary refinement method based on nonlinear filtering includes: introducing a diffusion rate calculation method based on gradient amplitude; dynamically adjusting the diffusion rate according to the size of the local gradient, suppressing diffusion at the boundary, and enhancing diffusion in the internal area.

4. The follicle region image segmentation method for ovarian tissue analysis according to claim 3, characterized in that The method for refining the regional boundary based on nonlinear filtering includes: introducing a filter parameter adaptive adjustment strategy based on the candidate region morphology; the filter dynamically adjusts the diffusion intensity and the number of iterations according to the morphological characteristics of the follicles including the size and shape, so as to adapt to the follicle regions with different morphologies; The number of diffusion iterations Dynamically adjusted according to the morphological characteristics of the region, the expression is: , in: Indicates the number of diffusion iterations, that is, the number of steps the filter runs in the diffusion process; by dynamically adjusting the number of iterations, the smoothness of the diffusion process is controlled; is the maximum diffusion rate, which is used to set the maximum smoothness allowed in the diffusion process; is a control parameter used to adjust the influence of the morphology of the candidate region on the number of diffusion iterations; Represents the foreground area of ​​the candidate region, that is, the size of the follicle area; Represents the area of ​​the background area; by calculating the area difference between the foreground and the background, the filter adaptively adjusts the smoothness of areas of different sizes to ensure that more iterations are allowed for larger follicle areas, while limiting the number of iterations for smaller follicle areas.

5. The follicle region image segmentation method for ovarian tissue analysis according to claim 1, characterized in that The segmentation model optimization method based on specific follicle physiological characteristics includes: The physiological characteristic parameters of the follicles are introduced; the physiological characteristic parameters integrate the volume, density and edge clarity of the follicles. For early follicles, the volume is small and the edges are blurred; while mature follicles have larger volumes and clearer boundaries; by introducing physiological characteristics, the segmentation model dynamically adjusts its sensitivity at different stages.

6. The follicle region image segmentation method for ovarian tissue analysis according to claim 5, characterized in that The segmentation model optimization method based on specific follicle physiological characteristics includes: By dynamically adjusting the threshold and gradient response weight of the edge detection algorithm, the details of the fuzzy boundary are captured; for mature follicles, the model reduces the edge detection sensitivity; and the sensitivity adjustment function of edge detection The definition is as follows: , in: is in position The edge detection response weight at controls the sensitivity of the edge detection algorithm to different areas; is the adjustment coefficient that controls the sensitivity of edge detection. The larger the value, the more sensitive the model is to edge changes. When it increases, the edge response becomes more prominent; is the physiological characteristic parameter; It is the threshold of edge detection, which is used to determine the starting point of edge detection.

7. The follicle region image segmentation method for ovarian tissue analysis according to claim 6, characterized in that The segmentation model optimization method based on specific follicle physiological characteristics includes: Through adaptive filtering technology based on morphological changes, adjustments are made according to the size and morphology of the follicles during development; for early follicles, the filter focuses on removing noise while maintaining the integrity of the edges; in the mature stage, the filter is used for smoothing to reduce noise interference; the adaptive filtering mechanism dynamically adjusts the filtering parameters according to the volume and morphological characteristics of the follicles.

8. The follicle region image segmentation method for ovarian tissue analysis according to claim 7, characterized in that The segmentation model optimization method based on specific follicle physiological characteristics includes: through internal texture analysis and regional contrast enhancement, the segmentation model distinguishes the internal structure and the external area in the segmentation of mature follicles; dynamically adjusts the segmentation strategy according to the change of internal texture, and enhances the contrast to highlight the internal details of the follicle.

Images