Gynecological endoscope image intelligent analysis and precise diagnosis system for cervical lesions

Abstract

The present application relates to the technical field of medical image processing, in particular to a gynecological endoscope image intelligent analysis and cervical lesion accurate diagnosis system, comprising an image enhancement module, a differential geometry feature extraction module, a topological preserving manifold learning module, a curvature-vascular correlation analysis module, a biopsy point accurate positioning module, a real-time intelligent auxiliary diagnosis module, a clinical feedback module and a self-adaptive optimization module, the system models the cervical endoscope image surface as a Riemann manifold, extracts the micro morphological change features through multi-scale curvature analysis, adopts a topological sensitive encoder and a manifold alignment decoder to maintain the key topological structure, establishes a correlation model of the tissue surface curvature change and the vascular morphology, analyzes the differential morphological features, determines the optimal biopsy position based on the comprehensive features, generates the diagnosis results and basis, provides visual display, and significantly improves the detection rate of cervical lesions, especially early micro lesions, reduces the missed diagnosis rate and reduces the multiple point biopsy.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a system for intelligent analysis of gynecological endoscopic images and precise diagnosis of cervical lesions, used for intelligent analysis of cervical endoscopic images and precise diagnosis of cervical lesions. Background Technology

[0002] Cervical cancer is one of the most common malignant tumors in women, and early diagnosis and treatment are crucial for improving patient survival rates. Currently, the diagnosis of cervical lesions mainly relies on endoscopy and biopsy pathological analysis. However, traditional endoscopy has the following problems: on the one hand, endoscopic images are often affected by reflections, shadows, and color distortion, reducing image quality; on the other hand, the identification of lesion areas and the selection of biopsy points are highly dependent on the doctor's experience, resulting in strong subjectivity and poor consistency. In addition, traditional image analysis methods mainly rely on color and texture features, making it difficult to capture subtle morphological changes in the early stages of cervical lesions, leading to a high rate of missed diagnoses of early lesions.

[0003] While some deep learning-based image analysis systems have been applied to the diagnosis of cervical lesions, these systems typically employ general convolutional neural network architectures and lack specific optimizations for the characteristics of cervical tissue. In particular, these systems often lose crucial topological information during feature extraction, leading to insufficient accuracy in identifying lesions with irregular shapes. Furthermore, existing systems lack comprehensive analysis of the correlation between tissue surface differential properties and vascular morphology, failing to fully capture the biological characteristics of cervical lesions and impacting diagnostic accuracy.

[0004] Therefore, there is an urgent need for a gynecological endoscopic image intelligent analysis and cervical lesion accurate diagnosis system that can overcome the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent analysis system for gynecological endoscopic images and a precise diagnostic system for cervical lesions. This system uses innovative technologies such as differential geometric feature extraction, topology-preserving manifold learning, and curvature-vascular correlation analysis to achieve precise analysis of cervical endoscopic images and accurate diagnosis of lesions.

[0006] This invention proposes a gynecological endoscopic image intelligent analysis and cervical lesion precision diagnosis system, comprising:

[0007] The image enhancement module is used to process the acquired cervical endoscopy images, improve image contrast, eliminate reflections and shadows, and perform color correction to obtain enhanced cervical images.

[0008] The differential geometric feature extraction module, connected to the image enhancement module, is used to receive the enhanced cervical image, model the surface of the enhanced cervical image as a Riemannian manifold, calculate multi-scale curvature features, extract features of abnormal regions, and output the curvature feature map and preliminary segmentation results of the cervical image.

[0009] The topology-preserving manifold learning module, connected to the differential geometry feature extraction module, receives the curvature feature map and preliminary segmentation results, processes the feature map through a topology-sensitive encoder and a manifold-aligned decoder, preserves key topological structures, and outputs refined segmentation results.

[0010] The curvature-vascular correlation analysis module, connected to the topology-preserving manifold learning module, is used to receive the fine segmentation results, establish a correlation model between tissue surface curvature changes and vascular morphology, analyze differential morphological features, and construct a comprehensive feature descriptor.

[0011] The biopsy point precise positioning module is connected to the curvature-vascular correlation analysis module and is used to receive the comprehensive feature descriptor, analyze the abnormal area features, and determine the optimal biopsy location.

[0012] The real-time intelligent auxiliary diagnosis module is connected to the biopsy point precise positioning module. It is used to receive the optimal biopsy location and the comprehensive feature descriptor, generate cervical lesion diagnosis results and diagnostic basis, and provide auxiliary diagnostic information.

[0013] Preferably, the differential geometric feature extraction module includes:

[0014] The manifold modeling unit is used to map the enhanced cervical image surface into a parameterized surface, establish a local differential coordinate system, and calculate the local metric tensor.

[0015] The curvature calculation unit, connected to the manifold modeling unit, is used to calculate the Gaussian curvature and average curvature of the parameterized surface at multiple scales and generate a curvature feature map.

[0016] An anomaly region preliminary location unit, connected to the curvature calculation unit, is used to determine the preliminary location of the anomaly region based on the curvature feature map through adaptive threshold segmentation and curvature gradient field analysis.

[0017] Preferably, the curvature calculation unit is implemented in the following manner:

[0018] For the parametric surface, a scale sequence is set. ,in ;

[0019] For each scale The parameterized surface is processed using a Gaussian filter with the corresponding kernel width;

[0020] Calculate the curvature map at each scale to obtain the curvature feature set corresponding to the scale sequence;

[0021] By fusing curvature features at multiple scales using an adaptive weighting function, a fused curvature feature map is obtained.

[0022] Preferably, the topology-preserving manifold learning module includes:

[0023] A topology-sensitive encoder is used to perform multi-layer feature extraction on the curvature feature map and the enhanced cervical image, and to calculate continuous homology features after each layer extraction to supervise the topological structure of the feature space.

[0024] A manifold-aligned decoder, connected to the topology-sensitive encoder, is used to construct a feature manifold and ensures geometric consistency during the decoding process through a manifold alignment mechanism.

[0025] An adaptive skip connection unit, connecting the topology-sensitive encoder and the manifold-aligned decoder, is used to dynamically adjust the weights of skip connections based on feature distribution similarity and topological feature similarity.

[0026] Preferably, the topology-sensitive encoder calculates the continuous homology features in the following manner:

[0027] Filter complexes are constructed on each convolutional feature map to capture the topological structure of the feature space;

[0028] Calculate the 0-dimensional and 1-dimensional Betti numbers, representing the number of connected components and cycles;

[0029] Construct a topological feature vector and compare it with the topological feature vector of the original image to calculate the topological similarity index;

[0030] A topology loss term is constructed based on the aforementioned topology similarity index to guide the network in maintaining key topological structures.

[0031] Preferably, the curvature-vascular correlation analysis module includes:

[0032] The blood vessel-curvature co-detection unit is used to extract blood vessel feature maps and curvature feature maps, and the two types of features are mutually enhanced through a collaborative attention mechanism.

[0033] The differential morphology analysis unit, connected to the blood vessel-curvature co-detection unit, is used to calculate the Minkowski functional and fractal dimension through multi-scale morphological processing, and quantify the morphological complexity of abnormal regions.

[0034] The feature tensor construction unit, connected to the differential morphology analysis unit, is used to integrate all extracted features, construct a feature association graph, and analyze the interaction strength between different features.

[0035] Preferably, the blood vessel-curvature coordinated detection unit is implemented in the following manner:

[0036] The enhanced cervical image was processed using a dedicated vascular enhancement filter to extract vascular feature maps.

[0037] The curvature feature map is used as the curvature channel input;

[0038] Design transformation functions to map curvature features to the vascular feature space and vascular features to the curvature feature space, respectively;

[0039] Calculate attention weights to generate vascular features with enhanced curvature and vascular curvature features with enhanced curvature, respectively;

[0040] A collaborative feature map is generated by fusing enhanced features through adaptive weighting factors.

[0041] Preferably, the biopsy point precise positioning module includes:

[0042] The feature synthesis analysis unit is used to analyze the synthesized feature descriptor and assess the severity and saliency of the abnormal region.

[0043] A biomarker association unit, connected to the feature comprehensive analysis unit, is used to perform association analysis between abnormal region features and known cervical lesion biomarkers;

[0044] The biopsy site optimization unit, connected to the biomarker association unit, is used to determine the optimal biopsy location and number of biopsies based on the characteristics of the abnormal region and the biomarker association results, combined with clinical biopsy rules.

[0045] Preferably, the real-time intelligent auxiliary diagnostic module includes:

[0046] The lesion classification unit is used to classify the cervical condition into normal, cervical intraepithelial neoplasia, and cervical cancer based on the comprehensive feature descriptor.

[0047] The diagnostic basis generation unit, connected to the lesion classification unit, is used to extract key features and evidence supporting the diagnostic results and generate diagnostic basis information.

[0048] The visualization unit, connected to the diagnostic basis generation unit, is used to intuitively and visually display the cervical lesion diagnosis results, diagnostic basis, and optimal biopsy location on the enhanced cervical image.

[0049] Preferably, the system further includes:

[0050] The clinical feedback module, connected to the real-time intelligent auxiliary diagnosis module, is used to receive feedback information from doctors on diagnostic results.

[0051] An adaptive optimization module, connected to the clinical feedback module, is used to adjust the parameters of the differential geometric feature extraction module, the topology-preserving manifold learning module, the curvature-vascular correlation analysis module, the biopsy point precise positioning module, and the real-time intelligent auxiliary diagnosis module based on the feedback information, thereby improving the system's diagnostic accuracy.

[0052] The beneficial effects of this invention include:

[0053] 1. It improves the accuracy of cervical lesion detection, especially significantly increasing the detection rate of early microlesions, which helps in the early diagnosis and treatment of cervical cancer.

[0054] 2. Differential geometry modeling can capture minute morphological changes on the surface of cervical tissue, providing richer feature information for lesion identification.

[0055] 3. The topology-preserving manifold learning mechanism ensures that key structural information is not lost during deep learning, thus improving the accuracy of identifying irregularly shaped lesions.

[0056] 4. Through curvature-vascular correlation analysis, the intrinsic link between tissue morphological changes and vascular abnormalities was revealed, enhancing the ability to express pathological characteristics from a biological mechanism perspective.

[0057] 5. It enables precise positioning of biopsy points, reduces unnecessary multi-point biopsies, and improves biopsy efficiency and the representativeness of tissue samples.

[0058] 6. It provides intuitive visualizations and explanations of diagnostic criteria, enhancing the interpretability of the system and assisting doctors in making more accurate diagnostic decisions. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention;

[0060] Figure 2 This is a schematic diagram of the differential geometric feature extraction module in this invention;

[0061] Figure 3 This is a schematic diagram of the topology-preserving manifold learning module in this invention;

[0062] Figure 4 This is a schematic diagram of the curvature-vascular correlation analysis module in this invention;

[0063] Figure 5This is a schematic diagram of the biopsy point precise positioning module in this invention;

[0064] Figure 6 This is a schematic diagram of the real-time intelligent auxiliary diagnosis module in this invention;

[0065] Figure 7 This is a schematic diagram of the processing flow of the system of the present invention. Detailed Implementation

[0066] Please refer to Figure 1 - Figure 7 The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0067] like Figure 1 As shown, the intelligent analysis system for gynecological endoscopic images and precise diagnosis of cervical lesions provided by the present invention includes: an image enhancement module 1, a differential geometric feature extraction module 2, a topology-preserving manifold learning module 3, a curvature-vascular correlation analysis module 4, a biopsy point precise positioning module 5, a real-time intelligent auxiliary diagnosis module 6, and a clinical feedback module 7.

[0068] Image enhancement module 1 processes the acquired cervical endoscopic images to improve image contrast, eliminate reflections and shadows, and perform color correction to obtain an enhanced cervical image. In one embodiment of the invention, image enhancement module 1 employs a multi-step processing strategy. First, it improves image contrast through adaptive histogram equalization. Then, it eliminates reflections and shadows using a combination of Gaussian filtering and median filtering. Finally, it performs color correction based on a color correction algorithm to ensure the tissue color matches the actual color. Preferably, the contrast enhancement step uses a local region adaptive processing method, setting different enhancement parameter coefficients for different regions. For example, for regions with low brightness, the enhancement coefficient can be set to 1.2-1.5, while for regions with high brightness, the enhancement coefficient can be set to 0.8-1.0, thereby achieving a more balanced image effect.

[0069] like Figure 2 As shown, the differential geometric feature extraction module 2 is connected to the image enhancement module 1. It is used to receive the enhanced cervical image, model the surface of the enhanced cervical image as a Riemannian manifold, calculate multi-scale curvature features, extract features of abnormal regions, and output the curvature feature map and preliminary segmentation results of the cervical image.

[0070] The differential geometric feature extraction module 2 includes a manifold modeling unit 21, a curvature calculation unit 22, and an anomaly region preliminary localization unit 23.

[0071] The manifold modeling unit 21 is used to map the enhanced cervical image surface to a parametric surface, establish a local differential coordinate system, and calculate the local metric tensor. Specifically, a mapping function from a two-dimensional image to a three-dimensional manifold is first established. For each pixel Calculate its corresponding point on the manifold. In a preferred embodiment of the invention, the mapping function can use image intensity values ​​as the third-dimensional coordinates to construct a parametric surface. Subsequently, in the manifold Establish a local tangent plane at each point. Calculate the local coordinate basis vectors and Construct local metric tensors To eliminate noise and minor artifacts while preserving the true tissue morphology, an adaptive Gaussian filter can be applied to smooth the surface. The filter kernel width... It automatically adjusts based on local curvature. For example, in areas of high curvature, It can be set to a smaller value (such as 1.0-1.5) to preserve details; while in areas of lower curvature, It can be set to a larger value (such as 2.0-3.0) to suppress noise more effectively.

[0072] The curvature calculation unit 22 is connected to the manifold modeling unit 21 and is used to calculate the Gaussian curvature and mean curvature of the parameterized surface at multiple scales, generating a curvature feature map. The curvature calculation unit 22 is implemented in the following manner:

[0073] 1) For parametric surfaces, set a scale sequence. ,in ;

[0074] 2) For each scale The parameterized surface is processed using a Gaussian filter with the corresponding kernel width;

[0075] 3) Calculate the curvature map at each scale to obtain the curvature feature set corresponding to the scale sequence;

[0076] 4) The curvature features of multiple scales are fused by an adaptive weighting function to obtain a fused curvature feature map.

[0077] In one embodiment of the present invention, the scale sequence It can be set to This corresponds to organizational features of different sizes. For each scale... First, use a kernel width of Gaussian filter processing surface Then calculate the principal curvature. and This leads to the Gaussian curvature. and mean curvature Generate a Gaussian curvature map at this scale. and mean curvature diagram .

[0078] To integrate curvature features at different scales, an adaptive weighting function is designed. Weights are assigned based on the discriminative power of curvature features at different scales. In practical applications, the weight function can be optimized using a validation set, initially set to... The formula for calculating the fused curvature feature map is:

[0079] ,

[0080] in, To integrate curvature feature maps, Let be the weight of the i-th scale. This is the Gaussian curvature plot at the i-th scale. The total number of scales, This represents the summation operation from i=1 to i=n, ​​used to accumulate the weighted curvature feature maps at each scale. Finally, the fused features are normalized to obtain the standardized curvature feature map. .

[0081] The preliminary anomaly region localization unit 23 is connected to the curvature calculation unit 22, and is used to determine the preliminary location of the anomaly region based on the curvature feature map through adaptive threshold segmentation and curvature gradient field analysis. In the specific implementation process, firstly based on... Based on the histogram characteristics, design a dynamic threshold function. :

[0082] ,

[0083] in, Let be the dynamic threshold at point (x, y). This represents the mean curvature of a local region (typically a 15×15 pixel window) centered at the point (x,y). The standard deviation of curvature for this local region. This is an adjustment factor that can be set according to the sensitivity requirements for lesion detection, typically ranging from 2.0 to 3.0. For each point (x, y), if If it is, then it is marked as a potential outlier.

[0084] Subsequently, the gradient field of the curvature feature map is calculated. Identify gradient magnitude Greater than the threshold The region is a potential boundary. In a preferred embodiment of the invention, Can be set to The range of 15% to 20% of the maximum value has been validated by a large number of clinical images and can effectively balance the sensitivity and specificity of boundary detection.

[0085] Finally, the level set method is used to refine the preliminary segmentation results. Using the preliminary segmentation results as the initial contour, the velocity field F is defined as follows:

[0086] ,

[0087] in, For the velocity field of the level set evolution, Let be the stopping function based on image gradient, defined as For image, For image gradient, Describes the Euclidean norm. This is the balloon force term, controlling the expansion or contraction of the balloon's shape; it is typically set to 0.5-1.5. The curvature term, used to smooth the contour, is calculated as follows: ,in For level set functions, This represents the divergence operation.

[0088] Iteratively update the level set function ,

[0089] Stop iterating when the value is 0.001 or the maximum number of iterations is reached (usually set to 100).

[0090] like Figure 3 As shown, the topology-preserving manifold learning module 3 is connected to the differential geometry feature extraction module 2. It is used to receive curvature feature maps and preliminary segmentation results, process the feature maps through a topology-sensitive encoder and a manifold alignment decoder, preserve key topological structures, and output fine segmentation results.

[0091] The topology-preserving manifold learning module 3 includes a topology-sensitive encoder 31, a manifold alignment decoder 32, and an adaptive jump connection unit 33.

[0092] A topology-sensitive encoder 31 is used to perform multi-layer feature extraction on the curvature feature map and the enhanced cervical image, and calculates persistent homology features after each layer extraction to supervise the topological structure of the feature space. In one embodiment of the present invention, a three-channel input tensor is first constructed. ,in To enhance the grayscale version of the image, This is the preliminary segmentation result. This is to standardize the curvature feature map. Multi-channel information is fused using a 1×1 convolution to generate the initial feature map. .

[0093] The encoder employs a 5-layer structure, with each layer containing two 3×3 convolutional layers and one 2×2 max-pooling layer. After each convolutional layer, persistent homology features are computed by constructing filter complexes on the feature map to capture the topological structure of the feature space, followed by calculating 0-dimensional and 1-dimensional Betti numbers. and ), representing the number of connected components and the number of cycles, respectively, and storing the topological feature vector of each layer. .

[0094] The topology-sensitive encoder 31 computes persistent cohomology features in the following manner:

[0095] Filter complexes are constructed on each convolutional feature map to capture the topological structure of the feature space;

[0096] Calculate the 0-dimensional and 1-dimensional Betti numbers, representing the number of connected components and cycles;

[0097] Construct a topological feature vector and compare it with the topological feature vector of the original image to calculate the topological similarity index;

[0098] A topology loss term is constructed based on the topology similarity index to guide the network in maintaining key topological structures.

[0099] In practical implementation, the filter complex is constructed using a threshold-based method, applying a series of thresholds to the feature map F. This yields a nested sequence of binary graphs. For each binary graph, the number of its connected components is calculated. and ring number This forms a continuously homogeneous barcode. Topological similarity index The calculation is based on the Wasserstein distance of the barcode:

[0100] ,

[0101] in This is a topological similarity index, with values ​​ranging from [0,1]. A larger value indicates a more similar topological structure. It is a natural exponential function. This is a scaling factor that controls the sensitivity of similarity to changes in distance; it is typically set to 0.1-0.5. Barcode of the original image With encoder feature barcode The Wasserstein distance between the two barcodes measures the degree of difference between them. Topological loss term. Defined as:

[0102] ,

[0103] in, This is the topology loss term, with values ​​ranging from [0,1]. A smaller value indicates better preservation of the topology. This loss term helps the network maintain the crucial topology during feature extraction.

[0104] The manifold-aligned decoder 32 is connected to the topology-sensitive encoder 31 to construct a feature manifold, ensuring geometric consistency during the decoding process through a manifold alignment mechanism. Specifically, in the implementation, the bottleneck layer feature map of the encoder is first... Consider it as a high-dimensional characteristic manifold To preserve the geometry of the manifold in the feature space, a manifold modeling method based on geodesic distance preservation is employed. For each eigenvector in the feature graph... Local neighborhoods are constructed using the K-nearest neighbor algorithm. Neighborhood size Typically, the value is set to 8 to 15. Euclidean distances between feature points are calculated within the local neighborhood, and geodesic distances on the manifold are estimated using the Isomap algorithm to construct the geodesic distance matrix. Based on the geodesic distance matrix, a local connectivity graph of the characteristic manifold is constructed. In the diagram, nodes represent feature vectors, and edge weights reflect geodesic distances, thus accurately depicting the manifold geometric relationships between features.

[0105] To achieve a differentiable mapping from the encoder feature space to the decoder feature space, a lightweight manifold mapping network is designed. The network consists of three fully connected layers, each followed by a LeakyReLU activation function. The mapping network receives encoder features. As input, the corresponding decoder feature space representation is output. The training objective of the mapping network is to minimize the change in geodesic distance between feature points before and after mapping, ensuring that the intrinsic geometry of the manifold is preserved during the mapping process.

[0106] The decoder also uses a 5-layer structure, with each layer containing one The transposed convolutional layer is used for upsampling, skip connections to the corresponding layer features of the encoder, and two... Convolutional layer. The last layer uses... Convolution and sigmoid activation are applied to output a segmentation probability map. To ensure the geometric consistency of the feature space, a manifold alignment loss is designed. The loss function consists of two parts: a geodesic distance preservation term and a local reconstruction term. The geodesic distance preservation term ensures that the geodesic distance relationship between the encoder and decoder features remains consistent on their respective manifolds, and is calculated as follows:

[0107] ,

[0108] in, To preserve the geodesic distance loss, the first summation symbol This indicates that the summation is performed on 5 levels. The weight factor of the i-th layer, and the second summation symbol. Represents the neighborhood set of the i-th layer Accumulate all feature point pairs in the summation. and They are feature point pairs within the neighborhood. Represents the geodesic distance in the encoder feature space. Represents the geodesic distance in the decoder's feature space. Represents a manifold mapping network. This represents the square Euclidean norm.

[0109] The local reconstruction term is based on the principle of local linear embedding, ensuring that the local neighborhood structure of feature points is preserved after mapping. The calculation formula is as follows:

[0110] ,

[0111] in, For local reconstruction loss, summation sign This indicates that the summation is performed on 5 levels. Let i be the weight factor of the i-th layer. Let i be the feature map of the i-th layer of the decoder. Describes the neighborhood set of the features of the i-th layer. The local reconstruction weights are obtained by minimizing the reconstruction error of features within the neighborhood and satisfying the constraints. , For manifold mapping networks, The neighborhood features of the corresponding layer of the encoder Let represent the squared Euclidean norm. The total loss for manifold alignment is:

[0112]

[0113] in, For the total loss of manifold alignment, and The balancing coefficient controls the relative weights of the two loss terms; in a preferred embodiment, it can be set to... , By optimizing the loss function through end-to-end training, the manifold mapping network can learn feature transformations that maintain geometric consistency. At the same time, the entire manifold alignment mechanism is fully differentiable and supports gradient backpropagation, thus achieving end-to-end training optimization.

[0114] An adaptive skip connection unit 33 connects the topology-sensitive encoder 31 and the manifold-aligned decoder 32, and is used to dynamically adjust the weights of the skip connections based on feature distribution similarity and topological feature similarity. In the traditional U-Net network, skip connections are static and cannot adapt to different lesion types. This invention designs an adaptive weight function that, for encoder feature maps... and decoder feature map Calculate the similarity of feature distributions Similarity to topological features The overall similarity is obtained by combining the results:

[0115] ,

[0116] in, Let be the total similarity of the i-th layer, with a value ranging from [0,1]. This is a balancing factor that controls the relative importance of feature distribution similarity and topological feature similarity; it is typically set to 0.5-0.7. For feature distribution similarity, it is calculated based on the cosine similarity of the feature vectors. For topological feature similarity, similarity is calculated based on the Betti number. The adaptive weighting function is defined as:

[0117] ,

[0118] in, represents the weight of the skip connection in the i-th layer, with a value range of (0,1). The sigmoid activation function is defined as follows: , This is a sensitivity factor that controls how sensitive the weights are to changes in similarity; it is typically set to 2.0-5.0. The feature fusion formula for skip connections is:

[0119] ,

[0120] in, This is the feature map after fusion at the i-th layer. Let the weights be the skip connections in the i-th layer. Let i be the feature map of the encoder layer i. Let be the feature map of the i-th layer of the decoder. The features are integrated and fused through a 1×1 convolution to generate an updated decoder feature map.

[0121] like Figure 4 As shown, the curvature-vascular correlation analysis module 4 is connected to the topology-preserving manifold learning module 3. It is used to receive fine segmentation results, establish a correlation model between tissue surface curvature changes and vascular morphology, analyze differential morphological features, and construct a comprehensive feature descriptor.

[0122] The curvature-vascular correlation analysis module 4 includes a vascular-curvature co-detection unit 41, a differential morphology analysis unit 42, and a feature tensor construction unit 43.

[0123] The blood vessel-curvature co-detection unit 41 is used to extract blood vessel feature maps and curvature feature maps, and the two types of features are mutually enhanced through a collaborative attention mechanism. The blood vessel-curvature co-detection unit 41 is implemented in the following way:

[0124] The enhanced cervical image was processed using a dedicated vascular enhancement filter to extract vascular feature maps.

[0125] Use the curvature feature map as the curvature channel input;

[0126] Design transformation functions to map curvature features to the vascular feature space and vascular features to the curvature feature space, respectively;

[0127] Calculate attention weights to generate vascular features with enhanced curvature and vascular curvature features with enhanced curvature, respectively;

[0128] A collaborative feature map is generated by fusing enhanced features through adaptive weighting factors.

[0129] In a preferred embodiment of the present invention, the dedicated vascular enhancement filter is an improvement upon the Frangi filter and is optimized for the characteristics of cervical blood vessels:

[0130] ,

[0131] in, For scale The vascular response value is set below, ranging from [0,1]. A larger value indicates a higher probability that the point belongs to a blood vessel. and The eigenvalues ​​of the Hessian matrix , The eigenvalue ratio reflects the shape characteristics of the structure. The magnitude of the second derivative structure reflects the contrast of the structure. It is a natural exponential function. and These are parameters that control sensitivity to blood vessel shape and contrast. For cervical blood vessels, the preferred settings are... , The filter operates at multiple scales, capturing blood vessels of different diameters, and ultimately generating a vascular feature map. .

[0132] For the collaborative attention mechanism, the following transformation function is designed:

[0133] ,

[0134] ,

[0135] in, This represents the result of mapping curvature features to the vascular feature space. This represents the result of mapping vascular features to curvature feature space. and It is a learnable transformation matrix with sizes of 1 and 2. and ,in and These are the dimensions of curvature features and vascular features, respectively. and These are the curvature features and blood vessel features after 3×3 convolution processing, respectively. The formula for calculating the attention weights is:

[0136] ,

[0137] ,

[0138] in, This is an attention weight map for vascular features. Attention weight map for curvature features. The sigmoid activation function maps values ​​to the (0,1) interval. The enhanced features are calculated as follows:

[0139] ,

[0140] ,

[0141] in, This is a vascular feature characterized by increased curvature. The curvature characteristics of enhanced blood vessels This represents element-wise multiplication (Hadamard product). The final collaborative features are fused using adaptive weighting factors:

[0142] ,

[0143] in, For the synergistic features after fusion, and It is an adaptive weighting factor that is dynamically adjusted based on image characteristics to satisfy... In practical applications, weight factors can be designed based on the information entropy of the feature map:

[0144] ,

[0145] ,

[0146] in, express Information entropy express Information entropy, the formula for calculating information entropy is: ,in This is a normalized histogram of eigenvalues. Features are integrated and fused through 1x1 convolutions to generate the final collaborative feature map. .

[0147] The differential morphological analysis unit 42 is connected to the blood vessel-curvature co-detection unit 41, and is used to calculate the Minkowski functional and fractal dimension through multi-scale morphological processing to quantify the morphological complexity of abnormal regions. In the specific implementation, a set of morphological operators is first defined. Erosion, dilation, opening operation, closing operation and set of structural elements It contains structural elements of different sizes and shapes. (Regarding the segmentation results...) By applying various morphological operators and combinations of structuring elements, morphological feature sequences are generated. .

[0148] For each morphologically processed image Calculate the Minkowski functional, including area. ,perimeter And Euler number The area represents the total area of ​​the lesion region, the perimeter represents the boundary length of the lesion region, and the Euler number is equal to the number of connected components minus the number of holes, reflecting the topological complexity of the region. Construct the Minkowski feature vector. .

[0149] In addition, box counting was used to estimate the fractal dimension of the lesion region. :

[0150] ,

[0151] in, For fractal dimension, Indicates when The limit when approaching 0, Let be the side length of the box. The required side length to cover the lesion area is The number of boxes, This represents the natural logarithm. In practical calculations, estimations are performed at multiple scales, and a more stable fractal dimension value is obtained through linear regression. The fractal dimension reflects the irregularity and complexity of diseased tissue; it is typically between 1.2 and 1.4 for normal tissue, while it can reach 1.7 to 1.9 for cancerous tissue.

[0152] The feature tensor construction unit 43 is connected to the differential morphology analysis unit 42 to integrate all extracted features, construct a feature correlation graph, and analyze the interaction strength between different features. In the specific implementation, all extracted features are first collected: segmentation results... Synergistic features Minkowski characteristics and fractal dimension Construct feature vectors To reduce dimensionality, principal component analysis was applied, retaining 95% of the variance information.

[0153] Then, construct the feature association graph. Nodes represent features, and edges represent the correlations between features. Mutual information between features... Used to quantify nonlinear correlation:

[0154] ,

[0155] in, Features and Mutual information between them Indicates the feature and Double summation of all possible values, It is a feature and The joint probability distribution, and It is a marginal probability distribution. This represents the natural logarithm. By setting a mutual information threshold (typically 0.3-0.5), highly correlated feature groups are identified, forming feature clusters.

[0156] Finally, a unique identifier is assigned to each anomalous region, and the center coordinates, direction, and size of the region are recorded. A spatial relationship tree is constructed to describe the relative positions between anomalous regions. The output is a comprehensive feature descriptor D, which includes a segmentation mask, feature vectors, spatial information, and a relationship graph.

[0157] like Figure 5 As shown, the biopsy point precise positioning module 5 is connected to the curvature-vascular correlation analysis module 4, which is used to receive comprehensive feature descriptors, analyze abnormal area features, and determine the optimal biopsy location.

[0158] The biopsy site precise positioning module 5 includes a feature comprehensive analysis unit 51, a biomarker association unit 52, and a biopsy site optimization unit 53.

[0159] The feature synthesis analysis unit 51 is used to analyze the comprehensive feature descriptor to assess the severity and saliency of the abnormal region. In one embodiment of the present invention, a risk scoring function for the abnormal region is constructed by comprehensively considering curvature features, vascular features, and morphological features.

[0160] ,

[0161] in, The risk score for region R ranges from [0,1], with higher values ​​indicating a greater risk of disease. , and Let represent the curvature abnormality, vascular abnormality, and morphological abnormality of region R, respectively, all normalized to the [0,1] interval. , and These are weighting coefficients that control the contribution of each feature to the overall risk score, satisfying... For cervical lesions, the preferred weighting is set to... =0.3, =0.4, =0.3, this configuration has shown good risk assessment capability in clinical validation.

[0162] Biomarker association unit 52 is connected to feature comprehensive analysis unit 51 to perform association analysis between abnormal region features and known cervical lesion biomarkers. In cervical lesions, the main biomarkers include abnormal vascular patterns, atypical glands, and abnormal epithelial changes. This unit establishes feature-marker mapping relationships; for example, high curvature regions are associated with abnormal epithelial changes, vascular network complexity is associated with abnormal vascular patterns, and specific morphological features are associated with atypical glands. Through this mapping, the extracted features are transformed into clinically meaningful biomarker indicators.

[0163] The biopsy site optimization unit 53 is connected to the biomarker association unit 52. It is used to determine the optimal biopsy location and number of biopsies based on the characteristics of the abnormal region and the biomarker association results, combined with clinical biopsy rules. This unit employs a multi-objective optimization strategy, with the objective function including:

[0164] Maximize the representativeness of biopsy sites: Select locations that best reflect the characteristics of the lesion;

[0165] Minimize the number of biopsies: Reduce unnecessary biopsies while ensuring diagnostic accuracy;

[0166] Follow clinical biopsy guidelines, such as the Reid Colposcopic Index, prioritizing areas with obvious vascular abnormalities and significant color changes.

[0167] The mathematical expression of the optimization problem is:

[0168] ,

[0169] ,

[0170] in, This means finding the maximum value among all possible subsets P of region R. The representativeness measure of the biopsy site set P is defined as follows: , It refers to the number of biopsy sites. and It is a balancing factor that controls the relative importance of representativeness and the number of biopsies. This represents a clinical rule constraint, a Boolean value that is true when the biopsy point set P satisfies the clinical rule, and false otherwise. In practical applications, It can be set to 0.7-0.8. The value should be 0.2-0.3 to achieve a good balance between representativeness and the number of biopsies.

[0171] like Figure 6 As shown, the real-time intelligent auxiliary diagnosis module 6 is connected to the biopsy point precise positioning module 5, which is used to receive the optimal biopsy location and comprehensive feature descriptor, generate cervical lesion diagnosis results and diagnostic basis, and provide auxiliary diagnostic information.

[0172] The real-time intelligent auxiliary diagnosis module 6 includes a lesion classification unit 61, a diagnostic basis generation unit 62, and a visualization display unit 63.

[0173] The lesion classification unit 61 is used to classify cervical status into normal, cervical intraepithelial neoplasia, and cervical cancer based on a comprehensive feature descriptor. This unit employs an ensemble learning method, combining the results of multiple classifiers to improve classification robustness. Specifically, it uses classifiers such as random forest, support vector machine, and gradient boosting tree, with each classifier trained on a different feature subset. The final classification result is determined based on weighted voting.

[0174] ,

[0175] in, The classification results for region R, This indicates finding the category c that maximizes the following expression. This represents the summation operation from i=1 to i=k, where k is the number of classifiers. The weights of the i-th classifier satisfy the following condition: , This represents the probability prediction of the i-th classifier that region R belongs to class c. The weights can be determined based on the performance of each classifier on the validation set. For example, the weights of classifiers that perform well can be set to 0.4-0.5, while the weights of classifiers that perform poorly can be set to 0.2-0.3.

[0176] The diagnostic criteria generation unit 62 is connected to the lesion classification unit 61 and is used to extract key features and evidence supporting the diagnostic results, generating diagnostic criteria information. This unit identifies the features that contribute most to the classification results through feature importance analysis and associates these features with clinical diagnostic criteria. For example, for cervical intraepithelial neoplasia, key features might include specific vascular patterns, abnormal glandular openings, and irregular edges of the transformation zone. Diagnostic criteria are ranked by importance to form a structured diagnostic report.

[0177] The visualization unit 63 is connected to the diagnostic evidence generation unit 62, and is used to intuitively visualize the cervical lesion diagnosis results, diagnostic evidence, and optimal biopsy location on the enhanced cervical image. This unit adopts a multi-layer visualization strategy, including:

[0178] Base layer: Displays an enhanced original image of the cervix;

[0179] Segmentation layer: Uses semi-transparent colors to cover and display abnormal areas, with different colors representing different risk levels;

[0180] Feature layer: Marks the location of key features, such as abnormal blood vessels, atypical glands, etc.;

[0181] Biopsy layer: Marks recommended biopsy site locations and indicates priority.

[0182] The visual interface also provides interactive features, allowing doctors to adjust the displayed content, view detailed feature analysis results, and assist in making more accurate diagnostic decisions.

[0183] like Figure 7 As shown, the system of the present invention also includes a clinical feedback module 7 and an adaptive optimization module 8.

[0184] The clinical feedback module 7 connects to the real-time intelligent auxiliary diagnosis module 6 to receive feedback from doctors on diagnostic results. This module features a structured feedback interface that allows doctors to evaluate the system's diagnostic accuracy, the appropriateness of biopsy point recommendations, and to provide specific improvement suggestions.

[0185] The adaptive optimization module 8, connected to the clinical feedback module 7, adjusts the parameters of the differential geometry feature extraction module 2, the topology-preserving manifold learning module 3, the curvature-vessel correlation analysis module 4, the biopsy point precise localization module 5, and the real-time intelligent auxiliary diagnosis module 6 based on feedback information, thereby improving the system's diagnostic accuracy. This module employs an incremental learning strategy to continuously optimize system parameters, enabling the system to adapt to different clinical environments and equipment conditions. The optimization process considers the consistency and reliability of feedback, giving higher weight to highly consistent feedback.

[0186] Figure 7 The overall data flow and processing workflow of the system of this invention are demonstrated. First, the image enhancement module 1 processes the acquired cervical endoscopic image and outputs an enhanced cervical image. Then, the differential geometric feature extraction module 2 receives the enhanced image, extracts curvature features, and outputs a curvature feature map and preliminary segmentation results. Next, the topology-preserving manifold learning module 3 performs deep learning based on these features to preserve key topological structures and outputs refined segmentation results. Subsequently, the curvature-vascular correlation analysis module 4 establishes the correlation between tissue morphology and vascular features, constructing a comprehensive feature descriptor. The biopsy point precise positioning module 5 determines the optimal biopsy location based on feature analysis. Finally, the real-time intelligent auxiliary diagnosis module 6 generates diagnostic results and evidence, providing a visual display. The entire process forms a closed-loop system, and performance is continuously optimized through clinical feedback.

[0187] In a preferred embodiment of the present invention, the acquired gynecological endoscopic image data typically has a resolution of 512×512 pixels, RGB color format, and includes different types such as normal cervical tissue, inflamed areas, low-grade lesions, and high-grade lesions. The system's image processing time is controlled within 2 seconds, ensuring support for real-time analysis. The curvature feature extraction accuracy reaches the pixel level, enabling the identification of minute morphological changes. The topology-preserving network is trained using 5000 labeled images, and the verification accuracy reaches over 92%. The overall diagnostic accuracy of the system is 23.5% higher than traditional methods, especially in the detection of early lesions, where the sensitivity is improved by 31.2%.

[0188] In summary, the intelligent analysis system for gynecological endoscopic images and the accurate diagnosis system for cervical lesions provided by this invention, through innovative technologies such as differential geometric feature extraction, topology-preserving manifold learning, and curvature-vascular correlation analysis, achieves accurate analysis of cervical endoscopic images and accurate diagnosis of lesions, significantly improving the detection rate of cervical lesions, especially early small lesions, and providing clinicians with a powerful intelligent auxiliary diagnostic tool.

[0189] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A gynecological endoscopic image intelligent analysis and cervical lesion precision diagnosis system, characterized in that, include: The image enhancement module is used to process the acquired cervical endoscopy images, improve image contrast, eliminate reflections and shadows, and perform color correction to obtain enhanced cervical images. The differential geometric feature extraction module, connected to the image enhancement module, is used to receive the enhanced cervical image, model the surface of the enhanced cervical image as a Riemannian manifold, calculate multi-scale curvature features, extract features of abnormal regions, and output the curvature feature map and preliminary segmentation results of the cervical image. The topology-preserving manifold learning module, connected to the differential geometry feature extraction module, receives the curvature feature map and preliminary segmentation results, processes the feature map through a topology-sensitive encoder and a manifold-aligned decoder, preserves key topological structures, and outputs refined segmentation results. The curvature-vascular correlation analysis module, connected to the topology-preserving manifold learning module, is used to receive the fine segmentation results, establish a correlation model between tissue surface curvature changes and vascular morphology, analyze differential morphological features, and construct a comprehensive feature descriptor. The biopsy point precise positioning module is connected to the curvature-vascular correlation analysis module and is used to receive the comprehensive feature descriptor, analyze the abnormal area features, and determine the optimal biopsy location. The real-time intelligent auxiliary diagnosis module is connected to the biopsy point precise positioning module. It is used to receive the optimal biopsy location and the comprehensive feature descriptor, generate cervical lesion diagnosis results and diagnostic basis, and provide auxiliary diagnostic information.

2. The system according to claim 1, characterized in that, The differential geometric feature extraction module includes: The manifold modeling unit is used to map the enhanced cervical image surface into a parameterized surface, establish a local differential coordinate system, and calculate the local metric tensor. The curvature calculation unit, connected to the manifold modeling unit, is used to calculate the Gaussian curvature and average curvature of the parameterized surface at multiple scales and generate a curvature feature map. An anomaly region preliminary location unit, connected to the curvature calculation unit, is used to determine the preliminary location of the anomaly region based on the curvature feature map through adaptive threshold segmentation and curvature gradient field analysis.

3. The system according to claim 2, characterized in that, The curvature calculation unit is implemented in the following way: For the parametric surface, a scale sequence is set. ,in ; For each scale The parameterized surface is processed using a Gaussian filter with the corresponding kernel width; Calculate the curvature map at each scale to obtain the curvature feature set corresponding to the scale sequence; By fusing curvature features at multiple scales using an adaptive weighting function, a fused curvature feature map is obtained.

4. The system according to claim 1, characterized in that, The topology-preserving manifold learning module includes: A topology-sensitive encoder is used to perform multi-layer feature extraction on the curvature feature map and the enhanced cervical image, and to calculate continuous homology features after each layer extraction to supervise the topological structure of the feature space. A manifold-aligned decoder, connected to the topology-sensitive encoder, is used to construct a feature manifold and ensures geometric consistency during the decoding process through a manifold alignment mechanism. An adaptive skip connection unit, connecting the topology-sensitive encoder and the manifold-aligned decoder, is used to dynamically adjust the weights of skip connections based on feature distribution similarity and topological feature similarity.

5. The system according to claim 4, characterized in that, The topology-sensitive encoder calculates the continuous homology feature in the following manner: Filter complexes are constructed on each convolutional feature map to capture the topological structure of the feature space; Calculate the 0-dimensional and 1-dimensional Betti numbers, representing the number of connected components and cycles; Construct a topological feature vector and compare it with the topological feature vector of the original image to calculate the topological similarity index; A topology loss term is constructed based on the aforementioned topology similarity index to guide the network in maintaining key topological structures.

6. The system according to claim 1, characterized in that, The curvature-vascular correlation analysis module includes: The blood vessel-curvature co-detection unit is used to extract blood vessel feature maps and curvature feature maps, and the two types of features are mutually enhanced through a collaborative attention mechanism. The differential morphology analysis unit, connected to the blood vessel-curvature co-detection unit, is used to calculate the Minkowski functional and fractal dimension through multi-scale morphological processing, and quantify the morphological complexity of abnormal regions. The feature tensor construction unit, connected to the differential morphology analysis unit, is used to integrate all extracted features, construct a feature association graph, and analyze the interaction strength between different features.

7. The system according to claim 6, characterized in that, The blood vessel-curvature coordinated detection unit is implemented in the following way: The enhanced cervical image was processed using a dedicated vascular enhancement filter to extract vascular feature maps. The curvature feature map is used as the curvature channel input; Design transformation functions to map curvature features to the vascular feature space and vascular features to the curvature feature space, respectively; Calculate attention weights to generate vascular features with enhanced curvature and vascular curvature features with enhanced curvature, respectively; A collaborative feature map is generated by fusing enhanced features through adaptive weighting factors.

8. The system according to claim 1, characterized in that, The biopsy point precise positioning module includes: The feature synthesis analysis unit is used to analyze the synthesized feature descriptor and evaluate the severity and saliency of the abnormal region; A biomarker association unit, connected to the feature comprehensive analysis unit, is used to perform association analysis between abnormal region features and known cervical lesion biomarkers; The biopsy site optimization unit, connected to the biomarker association unit, is used to determine the optimal biopsy location and number of biopsies based on the characteristics of the abnormal region and the biomarker association results, combined with clinical biopsy rules.

9. The system according to claim 1, characterized in that, The real-time intelligent auxiliary diagnostic module includes: The lesion classification unit is used to classify the cervical condition into normal, cervical intraepithelial neoplasia, and cervical cancer based on the comprehensive feature descriptor. The diagnostic basis generation unit, connected to the lesion classification unit, is used to extract key features and evidence supporting the diagnostic results and generate diagnostic basis information. The visualization unit, connected to the diagnostic basis generation unit, is used to intuitively and visually display the cervical lesion diagnosis results, diagnostic basis, and optimal biopsy location on the enhanced cervical image.

10. The system according to claim 1, characterized in that, The system also includes: The clinical feedback module, connected to the real-time intelligent auxiliary diagnosis module, is used to receive feedback information from doctors on diagnostic results. An adaptive optimization module, connected to the clinical feedback module, is used to adjust the parameters of the differential geometric feature extraction module, the topology-preserving manifold learning module, the curvature-vascular correlation analysis module, the biopsy point precise positioning module, and the real-time intelligent auxiliary diagnosis module based on the feedback information, thereby improving the system's diagnostic accuracy.

Images