Image enhancement method for uterine adhesion image

Abstract

The invention relates to an image enhancement method for a uterine adhesion image. The method comprises the following steps: denoising a uterine adhesion image sample; analyzing the gray level distribution characteristics of the denoised image sample by adopting a self-adaptive contrast enhancement technology, and obtaining an enhanced gray level adjustment parameter; constructing a parameter prediction model based on the enhanced gray adjustment parameter; and inputting a to-be-processed uterine adhesion image into the parameter prediction model to obtain an optimal enhancement parameter, and inputting the optimal enhancement parameter into the uterine adhesion image for processing to obtain final enhanced image data. According to the method, the quality of the uterine adhesion image is effectively improved through multi-stage processing, a clearer and more reliable image basis is provided for clinical diagnosis, and the method has important application value.

Description

TECHNICAL FIELD

[0001] The application relates to the field of medical examination method improvement in the field of biological medicine, and in particular to an image enhancement method for uterine adhesion images. BACKGROUND

[0002] Uterine adhesion is an important research topic in the field of gynecology, and its image diagnosis is crucial for early detection and treatment of diseases. High-quality medical images can significantly improve the accuracy of diagnosis, especially in the complex pathological environment of uterine adhesion, image enhancement technology has become an indispensable tool. However, the existing image enhancement methods still face significant limitations in processing uterine adhesion images. Many traditional techniques rely on fixed parameter settings or manual adjustments, which are difficult to adapt to the diversity and complexity of different patient images. In addition, some methods are prone to loss of details or generation of artifacts when processing low-contrast or severely noisy images, affecting the reliability of diagnosis. In this field, the core challenge is how to realize the intelligent and adaptive adjustment of image enhancement parameters. The characteristics of uterine adhesion images are quite different, such as blurred tissue boundaries or insufficient contrast, making it difficult for universal parameters to meet the needs of all scenarios. Inaccurate parameter selection can directly affect the enhancement effect, and may even mask critical pathological information. Furthermore, the complexity of parameter adjustment requires the system to quickly and accurately predict the optimal parameter combination according to the unique characteristics of the input image, but existing methods lack in-depth exploration of the relationship between image characteristics and parameters, making it difficult to achieve efficient automated processing. Therefore, how to analyze a large number of uterine adhesion image samples to establish a model that can automatically predict the optimal enhancement parameters based on image characteristics has become a key problem in this research. SUMMARY

[0003] The purpose of the present application is to provide an image enhancement method for uterine adhesion images to solve the problems existing in the prior art.

[0004] To achieve the above purpose, the present application provides the following scheme:

[0005] An image enhancement method for uterine adhesion images, comprising:

[0006] Denoising uterine adhesion image samples;

[0007] Using adaptive contrast enhancement technology, analyzing the gray scale distribution characteristics of the denoised image samples to obtain enhanced gray scale adjustment parameters;

[0008] Based on the enhanced gray scale adjustment parameters, a parameter prediction model is constructed;

[0009] Input the uterine adhesion image to be processed into the parameter prediction model, obtain the optimal enhancement parameter, and input the optimal enhancement parameter into the uterine adhesion image for processing to obtain the final enhanced image data.

[0010] Optionally, the denoising of the uterine adhesion image sample comprises:

[0011] A feature analysis framework is constructed using a deep learning method to extract the distribution characteristics of the influence of tissue boundary blur and noise interference in the uterine adhesion image sample, and preliminary feature mapping results are obtained;

[0012] According to the preliminary feature mapping results, the image is processed in layers using a convolutional neural network for tissue boundary blur and noise interference, clear image data after denoising is obtained, and the retention degree of key pathological information in the image is determined.

[0013] Optionally, an adaptive contrast enhancement technique is used to analyze the gray scale distribution characteristics of the denoised image sample to obtain enhanced gray scale adjustment parameters, which comprise:

[0014] Through the denoised image data, a Gaussian filtering algorithm is used to obtain smooth image data;

[0015] An adaptive contrast enhancement technique is used to analyze the gray scale distribution characteristics of the smooth image data to obtain a gray scale histogram and determine the gray scale distribution range;

[0016] According to the gray scale distribution range, an adaptive enhancement parameter is calculated to obtain a contrast adjustment parameter set;

[0017] If the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained to obtain the final adjustment parameter, i.e., the enhanced gray scale adjustment parameter.

[0018] Optionally, based on the enhanced gray scale adjustment parameter, a parameter prediction model is constructed, which comprises:

[0019] Original data is obtained from the uterine adhesion image sample; a convolutional neural network is used to extract features of the original data to obtain a feature set;

[0020] If the dimension of the feature set is higher than a preset threshold, principal component analysis is used for dimension reduction to obtain a reduced dimension feature set;

[0021] According to the reduced dimension feature set and the gray scale adjustment parameter, a correlation mapping model is constructed to determine the feature parameter mapping relationship;

[0022] Based on the feature parameter mapping relationship, a deep neural network model is constructed;

[0023] The deep neural network model is optimized using a pre-established training data set to obtain a parameter prediction model.

[0024] Optionally, the optimization of the deep neural network model with the pre-established training data set comprises:

[0025] A preliminary classification processing is performed on the image features by obtaining the training data set from a preset repository, to obtain a classified feature set;

[0026] According to the classified feature set, a support vector machine algorithm is used to preliminarily optimize the model to determine the basic prediction ability of the model under different categories of features;

[0027] If the basic prediction ability is lower than a preset threshold, secondary feature extraction is performed on the classified feature set to obtain a more fine-grained feature subset;

[0028] The deep neural network model is adjusted in parameters through the feature subset, and it is determined whether the prediction accuracy of the model after the parameter adjustment is improved;

[0029] If the prediction accuracy improvement amplitude does not reach a preset standard, the correlation between the image features and the adaptive performance is further analyzed from the feature subset to obtain key influencing factors;

[0030] According to the key influencing factors, the weight distribution in the deep neural network model is adjusted to determine the adaptive performance of the model under different image features;

[0031] Through continuous monitoring of the adaptive performance, the accuracy change trend of the model under various image features is recorded to obtain the optimized model configuration parameters.

[0032] An image enhancement system for uterine adhesion images for implementing the method as described above, the system comprising:

[0033] A denoising module: denoising the uterine adhesion image sample;

[0034] An analysis module: using an adaptive contrast enhancement technique to analyze the gray scale distribution characteristics of the denoised image sample to obtain enhanced gray scale adjustment parameters;

[0035] A construction module: constructing a parameter prediction model based on the enhanced gray scale adjustment parameters;

[0036] An output module: inputting the uterine adhesion image to be processed into the parameter prediction model to obtain optimal enhancement parameters, inputting the optimal enhancement parameters into the uterine adhesion image for processing to obtain the final enhanced image data.

[0037] Optionally, the denoising module comprises:

[0038] The extraction unit: a deep learning method is used to construct a feature analysis framework, extract the distribution characteristics of the influence of tissue boundary blur and noise interference in the uterine adhesion image sample, and obtain preliminary feature mapping results;

[0039] The hierarchical processing unit: according to the preliminary feature mapping results, for the influence of tissue boundary blur and noise interference, the image is processed in layers by using a convolutional neural network, clear image data after denoising is obtained, and the retention degree of key pathological information in the image is determined.

[0040] Optionally, the analysis module comprises:

[0041] The smoothing unit: by denoising the image data, a Gaussian filtering algorithm is used to obtain smoothed image data;

[0042] The determination unit: using adaptive contrast enhancement technology, the gray distribution characteristics of the smoothed image data are analyzed, the gray histogram is obtained, and the gray distribution range is determined;

[0043] The calculation unit: according to the gray distribution range, the adaptive enhancement parameters are calculated, and the contrast adjustment parameter set is obtained;

[0044] The judgment unit: if the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained, and the final adjustment parameter, i.e., the enhanced gray adjustment parameter, is obtained.

[0045] The beneficial effects of the present application are:

[0046] The present application discloses a kind of image enhancement method flow chart for uterine adhesion image, for the problems such as tissue boundary blur, noise interference and contrast deficiency existing in uterine adhesion image, feature analysis framework is constructed by deep learning, image features are extracted and preliminary mapping is carried out.Then the image is processed in layers by using convolutional neural network and denoising, clear image data is obtained.On this basis, adaptive contrast enhancement technology is used to analyze gray distribution characteristics, and parameter prediction model is constructed to realize intelligent adjustment.Finally, the input image is processed by optimal enhancement parameter, and clear enhanced image that meets the diagnostic requirement is obtained.The quality of uterine adhesion image is effectively improved by the present application through multi-stage processing, which provides clearer and more reliable image basis for clinical diagnosis, and has important application value. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0048] Figure 1 A flow chart of an image enhancement method for uterine adhesion images according to an embodiment of the present application is shown in FIG. 1.

[0049] Figure 2 An echo area that is easily confused with a scar diverticulum in a normal SD rat uterus according to an embodiment of the present application is shown in FIG. 2.

[0050] Figure 3 An SD rat uterine body 1 according to an embodiment of the present application is shown in FIG. 3.

[0051] Figure 4 An SD rat uterine body 2 according to an embodiment of the present application is shown in FIG. 4.

[0052] Figure 5 An SD rat uterine horn according to an embodiment of the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0054] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0055] As shown in FIG. 1, the present embodiment provides an image enhancement method for uterine adhesion images, which comprises the following steps. Figure 1

[0056] Denoising the uterine adhesion image sample;

[0057] Using an adaptive contrast enhancement technique, analyzing the gray scale distribution characteristics of the denoised image sample, and obtaining an enhanced gray scale adjustment parameter;

[0058] Based on the enhanced gray scale adjustment parameter, constructing a parameter prediction model;

[0059] Inputting the uterine adhesion image to be processed into the parameter prediction model, obtaining an optimal enhancement parameter, inputting the optimal enhancement parameter into the uterine adhesion image for processing, and obtaining the final enhanced image data.

[0060] Further, denoising the uterine adhesion image sample comprises the following steps.

[0061] Using a deep learning method to construct a feature analysis framework, extracting the distribution characteristics of the tissue boundary blur and noise interference in the uterine adhesion image sample, and obtaining a preliminary feature mapping result;​

[0062] According to the preliminary feature mapping results, in view of the influence of tissue boundary blur and noise interference, the image is processed in layers by using a convolutional neural network to obtain clear image data after denoising, and the retention degree of key pathological information in the image is determined.

[0063] Specifically, in this embodiment, for feature extraction and analysis of uterine adhesion image samples, first, 1000 medical image samples with a resolution of 1024x1024 pixels are preprocessed, and a Gaussian filter algorithm (standard deviation set to 2.0) is used to remove noise and ensure image smoothness. At the same time, a histogram equalization method is used to enhance contrast, and the gray value range of the original image is adjusted from 0-255 to a more uniform distribution to solve the problem of insufficient contrast. Then, for image feature differences, a convolutional neural network (CNN) model ResNet-50 in deep learning is used for feature extraction, the number of input layer channels is set to 3, the convolution kernel size is set to 3x3, the training iteration number is set to 100, and the learning rate is set to 0.001. Through feature response analysis of the tissue boundary blur area in the sample, it is found that the proportion of the area with boundary blur degree greater than 0.5 is about 15%. Subsequently, for the influence of noise interference, an autoencoder (Autoencoder) model is used to denoise and reconstruct the image, the number of hidden layer neurons is set to 256, the reconstruction error is controlled within 0.02, and the noise distribution characteristics are analyzed. It is found that noise is mainly concentrated in the edge area of the image, accounting for about 10%. Finally, based on the extracted features, principal component analysis (PCA) algorithm is used to reduce the dimension of feature mapping, and the first 95% of the variance information is retained to generate preliminary feature mapping results. The mapping results show that the overlap degree of tissue boundary blur area and noise interference area is about 8%, indicating that there is a certain correlation between the two.

[0064] In view of the influence of the fuzzy tissue boundary and noise interference, the implementation method of constructing a noise suppression module and using a convolutional neural network to perform hierarchical processing on the image can be realized in the following manner. First, in the design of the noise suppression module, a denoising autoencoder model based on deep learning is used, the original medical image data (assuming a resolution of 512x512 pixels and a gray value range of 0-255) is input, the image is compressed into a low-dimensional feature vector (dimension set to 128) through the encoder, and then the image is reconstructed through the decoder. During training, the mean square error loss function is used, and the optimization goal is to control the pixel-level difference between the reconstructed image and the noise-free reference image within 5%. The training data set contains 10,000 labeled images, and the number of iterations is set to 50 rounds to ensure the noise suppression effect. Second, hierarchical processing is performed using a convolutional neural network, a 5-layer CNN architecture is designed, each layer includes a convolution operation with a kernel size of 3x3 and a step size of 1, the activation function uses ReLU, and the pooling layer uses 2x2 max pooling. The deep features of the image are gradually extracted, and the output feature map resolution is reduced to 64x64, which is used to separate noise and tissue boundary information. Analysis shows that after this processing, the signal-to-noise ratio of the image is improved from the original 15.2 to 22.7, indicating that the noise is significantly reduced. Next, when obtaining the clear image data after denoising, the feature map is upsampled back to the original resolution through the deconvolution operation, and the residual connection mechanism is used to retain the detail information. The experimental results show that the structural similarity index (SSIM) of the denoised image reaches 0.92, indicating that the image quality is close to the reference standard.

[0065] Further, the adaptive contrast enhancement technique is used to analyze the gray scale distribution characteristics of the denoised image samples, and the enhanced gray scale adjustment parameters are obtained, including:

[0066] By denoising the image data, a Gaussian filter algorithm is used to obtain smoothed image data.

[0067] The adaptive contrast enhancement technique is used to analyze the gray scale distribution characteristics of the smoothed image data, obtain the gray scale histogram, and determine the gray scale distribution range.

[0068] According to the gray scale distribution range, the adaptive enhancement parameters are calculated to obtain the contrast adjustment parameter set.

[0069] If the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained to obtain the final adjustment parameters, i.e., the enhanced gray scale adjustment parameters.

[0070] Specifically, in the present embodiment, by denoising the image data, a Gaussian filtering algorithm is adopted to obtain smooth image data, and the image data is obtained. An adaptive contrast enhancement technique is used to analyze the gray level distribution characteristics of the image data, obtain a gray level histogram, and determine the gray level distribution range. According to the gray level distribution range, the adaptive enhancement parameters are calculated, the contrast adjustment parameter set is obtained, and the parameter set data is obtained. If the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained, the final adjustment parameter is obtained, and the parameter validity is determined.

[0071] Suppose the input image is a 512x512 pixel medical image with a gray value range of 0 to 255. By calculating the gray level histogram, it is found that the gray value of the image is mainly concentrated between 50 and 150, accounting for as high as 85%, indicating that the contrast is low and the detail information is not obvious. Then, based on this distribution characteristic, an adaptive histogram equalization algorithm (CLAHE) is used, the window size is set to 8x8 pixels, and the threshold for limiting contrast enhancement is set to 4.0 to avoid excessive enhancement of noise. The gray adjustment parameter is calculated, wherein the gray mapping function maps the original gray value 50 to 30 and 150 to 200, thereby stretching the gray range to 20 to 220 and enhancing the contrast. Subsequently, the enhanced image is analyzed again for gray level distribution, and it is found that the gray value distribution is more uniform, and the pixel value difference in the detail area is increased from the original 10 to 30, indicating that the texture information is highlighted. Finally, it is determined whether the parameter meets the pathological information retention requirement, and the standard is set that the average value of the enhanced gray gradient should be greater than 15, and the gray value of the key area (such as the suspected lesion) should not be lost by more than 5%. It is found through calculation that the average value of the enhanced gradient is 18.5, and the gray value retention rate of the lesion area is 98%, meeting the requirement. To ensure logical rigor, if it is found that the gray value of the lesion area is lost too much, the threshold of CLAHE can be further adjusted to 3.5, and the mapping parameter is recalculated to ensure the integrity of the pathological information.

[0072] Further, based on the enhanced gray adjustment parameter, the parameter prediction model is constructed, which includes:

[0073] The original data is obtained from the uterine adhesion image sample; the convolutional neural network is used to extract the features of the original data to obtain the feature set;

[0074] If the dimension of the feature set is higher than the preset threshold, the dimensionality reduction is performed by principal component analysis to obtain the reduced feature set;

[0075] According to the reduced feature set and the gray adjustment parameter, an association mapping model is constructed to determine the feature parameter mapping relationship;

[0076] Based on the feature parameter mapping relationship, a deep neural network model is constructed; the input layer is a feature vector, the hidden layer is two fully connected layers (the number of neurons is 64 and 32 respectively), and the output layer is an adjustment parameter value.

[0077] The deep neural network model is optimized by using a pre-established training data set to obtain a parameter prediction model.

[0078] Further, the optimization of the deep neural network model by using the pre-established training data set comprises:

[0079] The training data set is obtained from a pre-set repository, and the image features are preliminarily classified to obtain a classified feature set;

[0080] According to the classified feature set, the model is preliminarily optimized by using a support vector machine algorithm to determine the basic prediction ability of the model under different categories of features;

[0081] If the basic prediction ability is lower than a pre-set threshold, secondary feature extraction is performed on the classified feature set to obtain a finer-grained feature subset;

[0082] The deep neural network model is adjusted by using the feature subset to determine whether the prediction accuracy of the model after parameter adjustment has improved;

[0083] If the improvement in prediction accuracy does not reach a pre-set standard, the correlation between the image features and the adaptive performance is further analyzed from the feature subset to obtain key influencing factors;

[0084] According to the key influencing factors, the weight distribution in the deep neural network model is adjusted to determine the adaptive performance of the model under different image features;

[0085] Through continuous monitoring of the adaptive performance, the accuracy change trend of the model under various image features is recorded to obtain the optimized model configuration parameters.

[0086] Specifically, in the process of constructing the image adaptive parameter prediction model in this embodiment, first, the image samples are processed by the enhanced gray scale adjustment parameters. Assuming that the initial gray scale adjustment parameters are brightness gain 1.5 and contrast gain 1.2, 1000 image samples with a resolution of 1920x1080 are batch-adjusted using an image processing algorithm, the change matrix of pixel gray scale values before and after adjustment is recorded, the average gray scale improvement value is about 20.5, and the standard deviation is 5.3. The gray scale distribution characteristics of the image under different lighting conditions are analyzed, and it is concluded that the influence weight of brightness gain on low-light images is 0.7, and the influence weight of contrast gain on high-light images is 0.6. Then, combined with the analysis results of the image samples, image features such as texture entropy (average 3.8) and edge strength (average 12.4) are extracted, principal component analysis algorithm is used to reduce the dimension of feature data, the first three principal components are retained, and the contribution rate reaches 85%, and a feature vector matrix is constructed. Subsequently, in view of the limitation of general parameters, a correlation mapping model is designed, and a support vector regression algorithm is used to perform nonlinear mapping between the feature vector and the gray scale adjustment parameters, the training set error is controlled within 0.05, the verification set error is 0.08, the correlation coefficient between the texture entropy and the brightness gain is 0.62, and the correlation coefficient between the edge strength and the contrast gain is 0.55. Finally, the adaptive parameter prediction initial framework is determined, based on the above mapping results, a deep neural network model is constructed, the input layer is the feature vector, the hidden layer is two fully connected layers (the number of neurons is 64 and 32 respectively), the output layer is the adjustment parameter value, the mean square error loss function is used for optimization, the learning rate is 0.001, and after 100 rounds of training, the deviation between the predicted parameters and the actual parameters is less than 0.1, ensuring the adaptability of the model under different image scenes, and logically forming a closed loop from feature extraction to parameter prediction.

[0087] The embodiment optimizes the deep neural network model using the pre-established training data set, including the following technical steps:

[0088] Obtain the training data set from the preset storage library, which contains a large number of labeled and unlabeled uterine adhesion image samples, which cover different pathological features and image qualities;

[0089] Preliminary classification processing is performed on the image features, clustering analysis and feature engineering methods are used, the samples are divided into multiple categories according to the similarity of pathological features, and a classified feature set is obtained. The samples in each category have similar pathological features and image features, while there are significant differences between different categories;

[0090] According to the classified feature set, the deep neural network model is preliminarily optimized using a support vector machine algorithm. The support vector machine algorithm determines the basic prediction ability of the model under different category features by analyzing the feature distribution in each category, including classification accuracy, recall rate and F1 score.

[0091] If the base prediction capability is below a preset threshold, for example, the classification accuracy is below 85%, secondary feature extraction is performed on the classified feature set. A more refined feature extraction algorithm, such as Local Binary Pattern (LBP) or Scale-Invariant Feature Transform (SIFT), is used in combination with a feature pyramid network (FPN) in deep learning to obtain a more fine-grained feature subset that can more accurately describe local and global features in the image.

[0092] The parameter adjustment of the deep neural network model is performed through the feature subset, and the hyperparameters (such as learning rate, regularization coefficient, convolution kernel size, etc.) of the model are adjusted using Bayesian optimization algorithm or genetic algorithm to determine whether the prediction accuracy of the model after parameter adjustment has improved, such as calculating the average accuracy of the model on the training set and validation set through cross-validation.

[0093] If the prediction accuracy improvement is less than a preset standard, for example, the improvement is less than 5%, the correlation between image features and adaptive performance is further analyzed from the feature subset, and the Pearson correlation coefficient or mutual information method is used to find the key influencing factors that have the greatest impact on the prediction accuracy of the model, such as certain specific gray distribution patterns or texture features.

[0094] According to the key influencing factors, the weight distribution in the deep neural network model is adjusted, and the attention mechanism or adaptive weight adjustment algorithm is used to enhance the adaptive performance of the model under different image features, focusing on improving the recognition ability of key pathological features and the robustness to noise.

[0095] Through continuous monitoring of the adaptive performance, the accuracy trend of the model under various image features is recorded, and learning curves and confusion matrices are used to obtain the optimized model configuration parameters, including the final network structure, weight matrix and bias term, etc., to ensure the high precision and stability of the model in the enhanced parameter prediction task of uterine adhesion images.

[0096] Taking the rat uterus as an example, Figure 2 the echo area of the normal SD rat uterus is easy to be confused with the scar diverticulum, Figure 3 the SD rat uterine body 1, Figure 4 the SD rat uterine body 2, Figure 5 and the SD rat uterine horn.

[0097] The embodiment also proposes an image enhancement system for uterine adhesion images, which includes:

[0098] A denoising module is used to denoise the uterine adhesion image sample.

[0099] The analysis module: using adaptive contrast enhancement technology, analyzing the gray distribution characteristics of the denoised image sample, and obtaining the enhanced gray adjustment parameter;

[0100] The construction module: based on the enhanced gray adjustment parameter, constructing a parameter prediction model;

[0101] The output module: inputting the uterine adhesion image to be processed into the parameter prediction model, obtaining the optimal enhancement parameter, inputting the optimal enhancement parameter into the uterine adhesion image for processing, and obtaining the final enhanced image data.

[0102] Further, the denoising module comprises:

[0103] The extraction unit: using a deep learning method to construct a feature analysis framework, extracting the distribution characteristics of the tissue boundary blur and noise interference in the uterine adhesion image sample, and obtaining the preliminary feature mapping result;

[0104] The hierarchical processing unit: according to the preliminary feature mapping result, for the tissue boundary blur and noise interference, using a convolutional neural network to perform hierarchical processing on the image, obtaining the clear image data after denoising, and determining the retention degree of the key pathological information in the image.

[0105] Further, the analysis module comprises:

[0106] The smoothing unit: through the denoised image data, using a Gaussian filter algorithm to obtain the smoothed image data;

[0107] The determination unit: using adaptive contrast enhancement technology, analyzing the gray distribution characteristics of the smoothed image data, obtaining the gray histogram, and determining the gray distribution range;

[0108] The calculation unit: according to the gray distribution range, calculating the adaptive enhancement parameter, and obtaining the contrast adjustment parameter set;

[0109] The judgment unit: if the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained, and the final adjustment parameter, i.e. the enhanced gray adjustment parameter, is obtained.

[0110] The above-described embodiments are only descriptions of the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the present application as defined by the claims.

Claims

1. An image enhancement method for images of uterine adhesions, characterized in that, include: Denoise removal was performed on the uterine adhesion image samples; Adaptive contrast enhancement technology is used to analyze the gray-level distribution characteristics of the denoised image samples and obtain the enhanced gray-level adjustment parameters. Based on the enhanced grayscale adjustment parameters, a parameter prediction model is constructed; The uterine adhesion image to be processed is input into the parameter prediction model to obtain the optimal enhancement parameters. The optimal enhancement parameters are then input into the uterine adhesion image for processing to obtain the final enhanced image data.

2. The image enhancement method for uterine adhesion images according to claim 1, characterized in that, Denoising of uterine adhesion image samples includes: A feature analysis framework was constructed using deep learning methods to extract the distribution characteristics of tissue boundary blurring and noise interference in uterine adhesion image samples, and to obtain preliminary feature mapping results. Based on the preliminary feature mapping results, and considering the effects of blurred tissue boundaries and noise interference, a convolutional neural network was used to perform layered processing on the images to obtain clear image data after denoising, and to determine the degree of preservation of key pathological information in the images.

3. The image enhancement method for uterine adhesion images according to claim 1, characterized in that, Adaptive contrast enhancement technology was employed to analyze the gray-level distribution characteristics of the denoised image samples, and the enhanced gray-level adjustment parameters were obtained, including: Smooth image data is obtained by using Gaussian filtering algorithm on denoised image data; Adaptive contrast enhancement technology is used to analyze the gray-level distribution characteristics of smooth image data, obtain gray-level histograms, and determine the gray-level distribution range. Based on the grayscale distribution range, calculate the adaptive enhancement parameters and obtain the contrast adjustment parameter set; If the parameter set data meets the preset pathological information retention threshold, then the parameter set data is retained, and the final adjustment parameters are obtained, namely the enhanced grayscale adjustment parameters.

4. The image enhancement method for uterine adhesion images according to claim 1, characterized in that, Based on the enhanced grayscale adjustment parameters, the parameter prediction model is constructed as follows: Raw data was obtained from uterine adhesion image samples; features of the raw data were extracted using a convolutional neural network to obtain a feature set; If the dimension of the feature set is higher than a preset threshold, then dimensionality reduction is achieved through principal component analysis to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set and grayscale adjustment parameters, an association mapping model is constructed to determine the mapping relationship of feature parameters; Based on the aforementioned feature parameter mapping relationship, a deep neural network model is constructed; The deep neural network model is optimized using a pre-established training dataset to obtain a parameter prediction model.

5. The image enhancement method for uterine adhesion images according to claim 4, characterized in that, Optimizing deep neural network models using pre-established training datasets includes: By obtaining the training dataset from a pre-defined repository, preliminary classification processing is performed on the image features to obtain a set of classified features. Based on the classified feature set, the support vector machine algorithm is used to perform preliminary optimization of the model to determine the basic predictive ability of the model under different category features. If the basic prediction ability is lower than the preset threshold, a second feature extraction is performed on the classified feature set to obtain a more fine-grained feature subset. By adjusting the parameters of a deep neural network model using feature subsets, we can determine whether the prediction accuracy of the model has been improved after parameter adjustment. If the improvement in prediction accuracy does not reach the preset standard, the correlation between image features and adaptation performance will be further analyzed from the feature subset to obtain the key influencing factors. Based on key influencing factors, the weight allocation in the deep neural network model is adjusted to determine the adaptive performance of the model under different image features; By continuously monitoring the adaptive performance, the accuracy change trend of the model under various image features is recorded, and the optimized model configuration parameters are obtained.

6. An image enhancement system for images of uterine adhesions, characterized in that, The system for implementing the method as described in any one of claims 1-5 includes: Denoising module: Denoises the uterine adhesion image samples; Analysis module: Employs adaptive contrast enhancement technology to analyze the grayscale distribution characteristics of the denoised image samples and obtain the enhanced grayscale adjustment parameters; Construction module: Based on the enhanced grayscale adjustment parameters, construct a parameter prediction model; Output module: Input the uterine adhesion image to be processed into the parameter prediction model to obtain the optimal enhancement parameters, input the optimal enhancement parameters into the uterine adhesion image for processing, and obtain the final enhanced image data.

7. The image enhancement system for images of uterine adhesions according to claim 6, characterized in that, The noise reduction module includes: Extraction Unit: A feature analysis framework is constructed using deep learning methods to extract the distribution characteristics of tissue boundary blurring and noise interference in uterine adhesion image samples, and to obtain preliminary feature mapping results; Layered processing unit: Based on the preliminary feature mapping results, and considering the effects of blurred tissue boundaries and noise interference, the convolutional neural network is used to perform layered processing on the image to obtain clear image data after denoising, and to determine the degree of preservation of key pathological information in the image.

8. The image enhancement system for images of uterine adhesions according to claim 6, characterized in that, The analysis module includes: Smoothing unit: Obtains smoothed image data by using a Gaussian filtering algorithm on the denoised image data; Unit Determination: Adaptive contrast enhancement technology is used to analyze the gray-level distribution characteristics of smooth image data, obtain gray-level histograms, and determine the gray-level distribution range; Calculation unit: Calculates adaptive enhancement parameters based on the grayscale distribution range and obtains a set of contrast adjustment parameters; Judgment unit: If the parameter set data meets the preset pathological information retention threshold, the parameter set data is retained, and the final adjustment parameters are obtained, namely the enhanced grayscale adjustment parameters.

Images