An image enhancement processing method for pediatric laryngeal neck development monitoring

Abstract

The application provides an image enhancement processing method for pediatric laryngeal neck development monitoring, relates to the field of image enhancement, improves the traditional attention calculation method, combines pathological masks and adaptive Gaussian variational attention to calculate local pathological attention, highlights the attention of important areas while reducing the calculation complexity, constructs a hierarchical recursive distillation module to gradually extract and enhance the features through recursive distillation, divides the features into two parts through adaptive feature segmentation, uses the important part for attention calculation, directly transmits the remaining part to the last for cross-scale structure fusion to reduce unnecessary calculation, improves the traditional information distillation method, combines feature separation, hierarchical recursive distillation and cross-scale structure guided fusion to construct an adaptive information distillation network, optimizes the image enhancement task for pediatric laryngeal neck development monitoring, improves the model efficiency and retains high-order information through multi-level learning.

Description

Technical Field

[0001] This invention belongs to the field of image enhancement, and specifically relates to an image enhancement processing method for monitoring the development of the larynx and neck in children. Background Technology

[0002] With the development of medical imaging technology, especially in the field of monitoring pediatric laryngeal and neck development, image enhancement has been widely used in clinical diagnosis and has become an important means of assessing the developmental status of children's larynx, vocal cords and trachea. However, existing image enhancement technologies still face many challenges when dealing with the complex anatomical structure and variable developmental status of children's larynx and neck, mainly in the problem of restoring complex image details. The anatomical structure of children's larynx is more delicate and complex than that of adults, especially the incompletely developed larynx, which presents image features that are difficult to identify. Traditional image enhancement methods often cannot accurately enhance different anatomical structures.

[0003] Current image enhancement methods are generally applied to the entire image, lacking fine-grained enhancement of key areas. This makes it difficult for doctors to accurately identify signs of early developmental delays during developmental monitoring. Therefore, how to accurately restore the details of the laryngeal structure and enhance the visualization of key areas while ensuring image quality has become an important research direction in medical image processing. To address the problems of unstable image quality, inaccurate detail restoration, and insufficient local area information enhancement in existing technologies, this paper proposes a method that effectively improves the detail restoration capability of images through local pathological attention and adaptive information distillation. Furthermore, it automatically adjusts the enhancement strategy according to the characteristics of different imaging modalities, thereby providing doctors with more accurate diagnostic information. Summary of the Invention

[0004] This invention provides an image enhancement processing method for monitoring pediatric laryngeal and cervical development. It improves upon traditional attention calculation methods by combining pathological masks and adaptive Gaussian variational attention to calculate local pathological attention, reducing computational complexity while highlighting attention in important regions. A hierarchical recursive distillation module is constructed to progressively extract and enhance features through recursive distillation. Adaptive feature segmentation divides the features into two parts: the important part is used for attention calculation, and the remaining part is directly passed to the final cross-scale structural fusion, reducing unnecessary computation. Furthermore, it improves upon traditional information distillation methods by combining feature separation, hierarchical recursive distillation, and cross-scale structure-guided fusion to construct an adaptive information distillation network. This network is optimized for image enhancement tasks related to pediatric laryngeal and cervical development monitoring, improving model efficiency and preserving high-order information through multi-level learning.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an image enhancement processing method for monitoring the development of the larynx and neck in children, comprising the following steps.

[0006] S1. Obtain images of children's laryngeal and neck development and create a dataset.

[0007] S2. Construct a pediatric laryngeal and neck data enhancement module, design adaptive nonlinear transformation, contrast adaptive enhancement based on structural features, and simulation noise based on the physical properties of medical imaging to simulate the impact of pediatric laryngeal and neck image data acquisition process.

[0008] S3. Introduce Gaussian variational inference to design adaptive Gaussian variational attention, construct a pathological mask to guide attention calculation, calculate local attention based on the pathological mask and improved traditional attention, and combine local attention and adaptive Gaussian variational attention to calculate local pathological attention.

[0009] S4. Based on a recursive optimization strategy, a recursive distillation method is designed to extract and enhance features step by step. An adaptive feature segmentation method is designed based on a self-designed adaptive segmentation coefficient. A hierarchical recursive distillation module is constructed through recursive distillation and adaptive feature segmentation to retain important features for attention calculation in the next layer.

[0010] S5. Improve the traditional information distillation method to construct an adaptive information distillation network, including feature separation, hierarchical recursive distillation and cross-scale structure-guided fusion. Design attention masks based on local gradients for feature separation, and design adaptive fusion weights based on attention gating mechanism to fuse distilled features from different layers.

[0011] S6. Construct an image enhancement model for monitoring pediatric laryngeal and neck development, and complete data simulation and image enhancement for monitoring pediatric laryngeal and neck development through a pediatric laryngeal and neck data enhancement module and an adaptive information distillation network.

[0012] Preferably, in step S1, images of pediatric laryngeal and neck development are obtained, capturing images of the vocal cords, trachea, and laryngeal soft tissues. The obtained images are labeled with data, including key area labeling, developmental status labeling, and case category labeling. The vocal cords, laryngeal cartilage, and trachea are labeled as key areas. The developmental status is divided into three types: normal development, developmental delay, and developmental abnormality. The case category labeling is classified according to the pathological condition. The labeled data is divided into training set and validation set.

[0013] Preferably, in step S2, the specific steps of the pediatric laryngeal data augmentation module are as follows:

[0014] S21. Patient posture changes, imaging angle deviations, and equipment jitter are simulated through adaptive nonlinear transformation. The specific formula is as follows:

[0015] In the formula, W geo and B geo The spatial mapping matrix is ​​based on affine transformation, simulating angular offset and scale change respectively, where N is the number of harmonic components, and α... iβ is the amplitude parameter, controlling the deformation intensity. i For the frequency parameter, δ controls the scale of local variation. i is the phase parameter, which controls the initial state of deformation; X is the image input to the pediatric laryngeal data augmentation module; and sin is the sine function.

[0016] S22. Emphasize airway, glottis, and muscle tissue details through contrast adaptation based on structural features, while suppressing background noise. The specific formula is as follows:

[0017]

[0018] In the formula, α con For edge enhancement intensity, tanh is the hyperbolic tangent function. For the Laplace operator, τ and ψ(X) are the mean and standard deviation of X, respectively. con β is the contrast threshold. con The contrast enhancement intensity β is the contrast enhancement intensity. con The mathematical formula is:

[0019]

[0020] In the formula, exp is an exponential function, and γ targ The target brightness value;

[0021] S23. Based on the simulated noise according to the physical characteristics of medical imaging, the specific formula is as follows:

[0022]

[0023] In the formula, α sn Let Γ(0.01, 0.05) represent the equipment noise intensity, and β represent the equipment noise level. sn Let γ be the tissue scattering noise intensity, ∏(-0.3, 0.3) be the tissue scattering noise, which follows a uniform distribution, and γ be the scattering noise intensity. sn For dynamic fuzziness intensity, The rate of change over time; the noise intensity of the equipment α sn The mathematical formula is:

[0024]

[0025] In the formula, Be represents the current load of the device. max Ti represents the maximum load of the device, and Ti represents the device usage time. max This refers to the maximum usage time of the device.

[0026] Preferably, in step S2, the pediatric laryngeal data augmentation module designs adaptive nonlinear transformation, structural feature-based adaptive contrast enhancement, and simulation noise based on medical imaging physical characteristics from three aspects: geometric transformation, contrast enhancement, and simulation experiment noise. Adaptive enhancement is performed through geometric transformation, contrast enhancement, and simulation experiment noise to simulate various influences during the acquisition of pediatric laryngeal medical images. Adaptive nonlinear transformation simulates changes in patient posture, imaging angle deviation, equipment jitter, and angle changes, improving the model's adaptability to different angles and sizes in pediatric laryngeal medical images. Structural feature-based adaptive contrast enhancement highlights details of the airway, glottis, and muscle tissue, suppresses background noise, and optimizes the visualization of low-contrast areas. Simulation noise based on medical imaging physical characteristics simulates noise in pediatric laryngeal medical images, enhancing the model's robustness to noise and artifacts. These enhancement techniques provide more diverse data for training. Using the enhanced pediatric laryngeal data to train the model improves its generalization ability in real medical environments, thereby improving the image quality and diagnostic accuracy of pediatric laryngeal development monitoring.

[0027] Preferably, in step S3, the specific steps for calculating local pathological attention are as follows:

[0028] S31. Input the first image feature X i ∈R H×W×C H, W, and C represent the height, width, and channels of the first image, respectively, to construct the pathological mask M. p Guided attention calculations differentiate between critical laryngeal regions such as the vocal cords, trachea, cartilage, irrelevant background areas, and low-information regions, reducing computational complexity. Attention is calculated for important regions using a pathological mask M. p The local attention A is obtained by weighting the optimized Softmax attention. PL The pathological mask M p The specific calculation formula is as follows:

[0029]

[0030] In the formula, Z is the normalization factor, δ controls the local smoothness, and x i (i, j) represents the grayscale value at pixel position (i, j), Ω represents a 5×5 local region, α represents the local smoothing weight, and β represents the edge enhancement weight. and These are the gradient values ​​at x and y, respectively;

[0031] The local attention A PL The specific calculation formula is as follows:

[0032]

[0033] In the formula, G is the expansion order, ⊙ is the weighting operation, Q, K, and V are the query, key, and value, respectively, derived from the linear transformation matrix W. Q W K W V We get O = W O x i K = W K x i V = W V x i T represents transpose, and d is the dimension of a single attention head;

[0034] S32. Introducing Gaussian variational inference to optimize attention calculation yields adaptive Gaussian variational attention A. GVA The specific formula is as follows:

[0035]

[0036] In the formula, Q i Let Q be a Gaussian distributed variable, where Q represents the query. μ Q Let θ be the mean of Q, ε represent the noise of the standard normal distribution, and θ be the mean of Q. Q Let G be the variance of Q. i p(Q) is the adaptive Gaussian compensation factor. i Let σ be the variational probability density of Q. T σ is a medical modality regulator. T =0.5·Var(x) i )+0.5·Mean(x i ), Var(x i ) is X i The local variance, Mean(x) i ) is X i The local mean;

[0037] The adaptive Gaussian compensation factor G i The specific calculation formula is as follows:

[0038]

[0039] In the formula, W G Let b be a linear transformation matrix. G λ1 and λ2 are the Gaussian bias term, which controls the baseline of the compensation mean. λ1 and λ2 are the learning parameters, and λ1 + λ2 = 1.

[0040] S33, Local attention A PL Adaptive Gaussian Variational Attention A GVA Combined with local pathological attention A LPSA The specific formula is as follows:

[0041] ALPSA =τ·A PL +(1-τ)·A GVA ;

[0042] In the formula, τ is the attention fusion weight, which is adaptively learned by the neural network.

[0043] Preferably, in step S3, local pathological attention plays an important role in enhancing pediatric laryngeal and neck images by introducing pathological masking and adaptive Gaussian variational attention. It can accurately focus on key areas such as the vocal cords, laryngeal cartilage, and trachea in monitoring pediatric laryngeal and neck development, improve the detail recovery of local areas, and optimize noise suppression and interference from irrelevant areas. Through adaptive Gaussian variational inference, the model can adaptively adjust the attention calculation according to different medical imaging modalities, which enhances the robustness and generalization ability of the model.

[0044] Preferably, in step S4, the specific steps of the hierarchical recursive distillation module are as follows:

[0045] S41. The specific calculation formula for the t-th recursive distillation is as follows:

[0046]

[0047] In the formula, α HRD To control the hyperparameters of the recursive weights, Let be the output of the t-th recursive distillation, and Conv be the convolution operation;

[0048] S42, Input the second image features H t W t C and C represent the height, width, and channels of the second image, respectively. Adaptive feature segmentation divides the features into X segments based on the adaptive segmentation coefficients. en and X fu The adaptive segmentation coefficient is calculated using the following formula: (Two parts)

[0049]

[0050] In the formula, X c For feature X t The feature of the c-th channel, c∈(1,C), Sig is the sigmoid function;

[0051] The adaptive feature segmentation formula is as follows:

[0052]

[0053] In the formula, X en X serves as the input for the next layer of attention computation. fu Used for final feature fusion.

[0054] Preferably, in step S4, the hierarchical recursive distillation module extracts and enhances features recursively layer by layer, gradually refining key information and structural details in the pediatric laryngeal image at each layer. The layer-by-layer nature of recursive distillation ensures the gradual enhancement of low-level and high-level features, preventing information loss in deep networks and thus maintaining the integrity of details in the pediatric laryngeal image. After each layer of distillation, the module divides the extracted features into two parts through an adaptive feature segmentation mechanism: one part is passed to the next layer for further refinement, and the other part is used for the final cross-scale structural fusion. This not only enhances local details finely in deep networks but also improves the overall quality of the pediatric laryngeal image through global information fusion. The combination of recursive distillation and adaptive feature segmentation effectively improves the accuracy and computational efficiency of the pediatric laryngeal image super-resolution, ensuring that details and global structure are enhanced simultaneously.

[0055] Preferably, in step S5, the adaptive information distillation network performs the following specific steps:

[0056] S51. Decompose the pediatric laryngeal and neck medical image into key regions and background regions, calculate the attention mask of the image, and based on the attention mask, extract the features I∈R of the pediatric laryngeal and neck medical image. H×W×C Divided into key areas I crux and background area I cont H, W, and C represent the height, width, and channel of a pediatric laryngeal and cervical medical image, respectively, and the attention mask M... AM The specific calculation formula is as follows:

[0057] In the formula, For local gradient values, x1∈[0,H], y1∈[0,W], and MAX is the maximum value.

[0058] The key area I crux The calculation formula is:

[0059] I crux =M AM ⊙I;

[0060] The background area I cont The calculation formula is:

[0061] I cont = (1-M) AM )⊙I;

[0062] S52. Information about key areas is more important than that about background areas; key areas I need to be analyzed. crux A higher distillation weight is assigned, and the region-adaptive distillation feature is obtained based on the distillation weight. The formula for the distillation weight is as follows:

[0063] WRAD =Sig(W 1R I crux +W 2R I cont );

[0064] In the formula, W 1R and W 2R The weight matrix is ​​trainable.

[0065] The regional adaptive distillation feature F RAD The calculation formula is:

[0066]

[0067] In the formula, ||I crux ⊙I cont || 2 Let be the L2 norm of the feature map, and e be the base of the natural logarithm;

[0068] S53. Use local pathological attention to perform attention calculations. The specific formula is as follows:

[0069] F LPSA =A LPSA (F RAD );

[0070] In the formula, A LPSA For local pathological attention;

[0071] S54. Use hierarchical recursive distillation to progressively extract and enhance features, and use an adaptive feature segmentation method to divide the features into two parts, which are used for the next layer of attention calculation and the final feature fusion, respectively.

[0072] S55. The distillation characteristics of different layers are fused, and the fusion formula is as follows:

[0073]

[0074] In the formula, L represents the distillation level, and v i For adaptive fusion weights, The distillation characteristics of the i-th layer;

[0075] The adaptive fusion weight v i The specific calculation formula is as follows:

[0076]

[0077] In the formula, and Learning through attention gating mechanisms.

[0078] Preferably, in step S5, the adaptive information distillation network, by combining feature separation, hierarchical recursive distillation, and cross-scale structure-guided fusion, can accurately process key regions and detailed information in pediatric laryngeal and cervical medical images. In the image enhancement task for monitoring pediatric laryngeal and cervical development, the image is first divided into key regions and background regions through a feature separation mechanism to ensure that key regions receive more computational resources. Then, hierarchical recursive distillation progressively extracts and optimizes the detailed information of the image, refining the features of important regions in each recursive layer to avoid information loss. Finally, cross-scale structure-guided fusion effectively combines the global structure with local details, enabling the image to achieve optimal results in both detail and global structure. Adaptive information distillation, by dynamically adjusting information flow and feature processing, can significantly improve the quality of super-resolution in pediatric laryngeal and cervical medical images.

[0079] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0080] This invention provides an image enhancement processing method for monitoring pediatric laryngeal and cervical development. It improves upon traditional attention calculation methods by combining pathological masks and adaptive Gaussian variational attention to calculate local pathological attention, reducing computational complexity while highlighting attention in important regions. A hierarchical recursive distillation module is constructed to progressively extract and enhance features through recursive distillation. Adaptive feature segmentation divides the features into two parts: the important part is used for attention calculation, and the remaining part is directly passed to the final cross-scale structural fusion, reducing unnecessary computation. Furthermore, it improves upon traditional information distillation methods by combining feature separation, hierarchical recursive distillation, and cross-scale structure-guided fusion to construct an adaptive information distillation network. This network is optimized for image enhancement tasks related to pediatric laryngeal and cervical development monitoring, improving model efficiency and preserving high-order information through multi-level learning. Attached Figure Description

[0081] Figure 1 This is a flowchart of an image enhancement processing method for monitoring the development of the larynx and neck in children, provided by the present invention.

[0082] Figure 2 This is a local pathological attention structure diagram provided by the present invention.

[0083] Figure 3 This is a diagram of the adaptive information distillation network structure provided by the present invention.

[0084] Figure 4 This is an image showing the enhanced effect of a coronal CT scan of the larynx provided by the present invention. Detailed Implementation

[0085] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0086] Please see Figures 1 to 4 This invention provides an image enhancement processing method for monitoring pediatric laryngeal and cervical development. It improves upon traditional attention calculation methods by combining pathological masks and adaptive Gaussian variational attention to calculate local pathological attention, reducing computational complexity while highlighting attention in important regions. A hierarchical recursive distillation module is constructed to progressively extract and enhance features through recursive distillation. Adaptive feature segmentation divides the features into two parts: the important part is used for attention calculation, and the remaining part is directly passed to the final cross-scale structural fusion, reducing unnecessary computation. Furthermore, it improves upon traditional information distillation methods by combining feature separation, hierarchical recursive distillation, and cross-scale structure-guided fusion to construct an adaptive information distillation network, optimized for image enhancement tasks in pediatric laryngeal and cervical development monitoring, improving model efficiency, and preserving high-order information through multi-level learning.

[0087] Please see Figure 1 As shown in the embodiment of this application, an image enhancement processing method for monitoring the development of the larynx and neck in children is presented.

[0088] S1. Obtain images of children's laryngeal and neck development and create a dataset.

[0089] Furthermore, 2000 images of pediatric laryngeal and neck development were obtained, capturing images of the vocal cords, trachea, and laryngeal soft tissues. The obtained images were annotated, including key area annotation, developmental status annotation, and case category annotation. The vocal cords, laryngeal cartilage, and trachea were annotated as key areas. The developmental status was divided into three types: normal development, developmental delay, and developmental abnormality. The case category annotation was classified according to the pathological condition. The annotated data was divided into a training set of 1400 images and a validation set of 600 images.

[0090] S2. Construct a pediatric laryngeal and neck data enhancement module, design adaptive nonlinear transformation, contrast adaptive enhancement based on structural features, and simulation noise based on the physical properties of medical imaging to simulate the impact of pediatric laryngeal and neck image data acquisition process.

[0091] Furthermore, the pediatric laryngeal data enhancement module designs adaptive nonlinear transformation, contrast-adaptive enhancement based on structural features, and simulation noise based on the physical characteristics of medical imaging from three aspects: geometric transformation, contrast enhancement, and simulation experiment noise. These simulate the impact of pediatric laryngeal image data acquisition process. The enhanced pediatric laryngeal data is then used to train the model. The specific steps of the pediatric laryngeal data enhancement module are as follows.

[0092] S21. Patient posture changes, imaging angle deviations, and equipment jitter are simulated through adaptive nonlinear transformation. The specific formula is as follows:

[0093]

[0094] In the formula, W geo and B geo The spatial mapping matrix is ​​based on affine transformation, simulating angular offset and scale change respectively. N is the number of harmonic components, set to 10, and the amplitude parameter α... i The initial value is set to 0.1, and the range is [0.05, 0.2]. The frequency parameter β... i The initial value is set to 1, and the value range is [0.5, 2]. The phase parameter δ i The initial value is set to The range of values ​​is X is the image input to the pediatric laryngeal data augmentation module, and sin is the sine function.

[0095] S22. Emphasize airway, glottis, and muscle tissue details through contrast adaptation based on structural features, while suppressing background noise. The specific formula is as follows:

[0096]

[0097] In the formula, the edge reinforcement intensity α con The initial value is set to 2, and the range of values ​​is [0.5, 3]. tanh is the hyperbolic tangent function. For the Laplace operator, ψ(X) and ψ(X) are the mean and standard deviation of X, respectively, and the contrast threshold τ is the standard deviation of X. con Set to 0.3, contrast enhancement intensity β con The initial value is set to 1, and the value range is [0, 2]. The specific enhancement intensity β con The mathematical formula is:

[0098]

[0099] In the formula, exp is an exponential function, and γ targ The target brightness value is initially set to 10, with a range of [0, 50].

[0100] S23. Based on the simulated noise according to the physical characteristics of medical imaging, the specific formula is as follows:

[0101]

[0102] In the formula, the equipment noise intensity α sn The initial value is set to 0.1, with a range of [0.1, 0.4]. Γ(0.01, 0.05) represents the equipment noise, and the tissue scattering noise intensity β... sn The initial value is set to 0.1, with a range of [0, 0.3]. Π(-0.3, 0.3) represents tissue scattering noise, which follows a uniform distribution. The dynamic fuzziness intensity γ sn The initial value is set to 0.1, and the range is [0, 0.2]. The rate of change over time;

[0103] The noise intensity α of the equipment sn The mathematical formula is:

[0104]

[0105] In the formula, Be represents the current load of the equipment, initially set to 10kW, with a value range of [0, 50]kW. max Let Ti be the maximum load of the equipment, set to 50kW, and let Ti be the equipment usage duration, initially set to 0.5 hours, with a range of [0, 12] hours. max The maximum usage time for the device is set to 12 hours.

[0106] S3. Introduce Gaussian variational inference to design adaptive Gaussian variational attention, construct a pathological mask to guide attention calculation, calculate local attention based on the pathological mask and improved traditional attention, and combine local attention and adaptive Gaussian variational attention to calculate local pathological attention.

[0107] Furthermore, such as Figure 2 As shown, the attention of key regions is calculated by combining local attention and adaptive Gaussian variational attention. The specific steps for calculating local pathological attention are as follows.

[0108] S31. Input the first image feature X i ∈R H×W×C H, W, and C are the height, width, and channels of the first image, respectively, set to 640×640×3, to construct the pathological mask M. p Guided attention calculations differentiate between critical laryngeal regions such as the vocal cords, trachea, cartilage, irrelevant background areas, and low-information regions, reducing computational complexity. Attention is calculated for important regions using a pathological mask M. p The local attention A is obtained by weighting the optimized Softmax attention. PL The pathological mask Mp The specific calculation formula is as follows:

[0109]

[0110] In the formula, the normalization factor Z is set to 15.6, the initial value of the local smoothness δ is set to 2, and the value range is [1,3]. X i (i, j) represents the grayscale value at pixel position (i, j), Ω represents a 5×5 local region, the initial value of the local smoothing weight α is set to 1, and the value range is [0.5, 2], and the initial value of the edge enhancement weight β is set to 1, and the value range is [0.3, 1.5]. and These are the gradient values ​​at x and y, respectively;

[0111] The local attention A PL The specific calculation formula is as follows:

[0112]

[0113] In the formula, the expansion order G is set to 4, ⊙ represents the weighting operation, and Q, K, and V are the query, key, and value, respectively, determined by the linear transformation matrix W. Q W K W V We get Q = W Q x i K = W K x i V = W V x i T represents transpose, and the dimension d of a single attention head is set to 64.

[0114] S32. Introducing Gaussian variational inference to optimize attention calculation yields adaptive Gaussian variational attention A. GVA The specific formula is as follows:

[0115]

[0116] In the formula, Q i Let Q be a Gaussian distributed variable, where Q represents the query. μ Q Let θ be the mean of Q, ε represent the noise of the standard normal distribution, and θ be the mean of Q. Q Let G be the variance of Q. i p(Q) is the adaptive Gaussian compensation factor. i Let σ be the variational probability density of Q. T σ is a medical modality regulator. T =0.5·Var(x) i )+0.5·Mean(x i ), Var(x i ) is xi The local variance, Mean(x) i ) is X i The local mean;

[0117] The adaptive Gaussian compensation factor G i The specific calculation formula is as follows:

[0118]

[0119] In the formula, W G Let b be the linear transformation matrix and the Gaussian bias term be b. G The initial value is set to 0.1, with a range of [0.05, 0.2]. The learning parameters λ1 and λ2 are both initially set to 0.5, with a range of [0.3, 0.7]. λ1 + λ2 = 1.

[0120] S33, Local attention A PL Adaptive Gaussian Variational Attention A GVA Combined with local pathological attention A LPSA The specific formula is as follows:

[0121] A LPSA =τ·A PL +(1-τ)·A GVA ;

[0122] In the formula, the initial value of the attention fusion weight τ is set to 0.4, and the value range is [0.2, 0.8].

[0123] S4. Based on a recursive optimization strategy, a recursive distillation method is designed to extract and enhance features step by step. An adaptive feature segmentation method is designed based on a self-designed adaptive segmentation coefficient. A hierarchical recursive distillation module is constructed through recursive distillation and adaptive feature segmentation to retain important features for attention calculation in the next layer.

[0124] Furthermore, the specific steps of the hierarchical recursive distillation module are as follows.

[0125] S41. The specific calculation formula for the t-th recursive distillation is as follows:

[0126]

[0127] In the formula, the hyperparameter α controls the recursive weights. HRD The initial value is set to 0.1, and the range of values ​​is (0, 1). is the output of the t-th recursive distillation, and Conv is a 3×3 convolution operation.

[0128] S42, Input the second image features H t W tC and C represent the height, width, and channels of the second image, respectively, set to 320×320×3. Adaptive feature segmentation divides the features into X segments based on the adaptive segmentation coefficients. en and X fu The adaptive segmentation coefficient is calculated using the following formula: (Two parts)

[0129]

[0130] In the formula, X c For feature X t The feature of the c-th channel, c∈(1,C), Sig is the sigmoid function;

[0131] The adaptive feature segmentation formula is as follows:

[0132]

[0133] In the formula, X en X serves as the input for the next layer of attention computation. fu Used for final feature fusion.

[0134] S5. Improve the traditional information distillation method to construct an adaptive information distillation network, including feature separation, hierarchical recursive distillation and cross-scale structure-guided fusion. Design attention masks based on local gradients for feature separation, and design adaptive fusion weights based on attention gating mechanism to fuse distilled features from different layers.

[0135] Furthermore, such as Figure 3 As shown, the specific steps of the adaptive information distillation network are as follows.

[0136] S51. Decompose the pediatric laryngeal and neck medical image into key regions and background regions, calculate the attention mask of the image, and based on the attention mask, extract the features I∈R of the pediatric laryngeal and neck medical image. H×W×C Divided into key areas I crux and background area I cont H, W, and C represent the height, width, and channels of a pediatric laryngeal and cervical medical image, respectively, and are set to 640×640×3. The attention mask M... AM The specific calculation formula is as follows:

[0137]

[0138] In the formula, For local gradient values, x1∈[0,H], y1∈[0,W], and MAX is the maximum value.

[0139] The key area I crux The calculation formula is:

[0140] I crux =M AM⊙I;

[0141] The background area I cont The calculation formula is:

[0142] I cont = (1-M) AM )⊙I.

[0143] S52. Information about key areas is more important than that about background areas; key areas I need to be analyzed. crux A higher distillation weight is assigned, and the region-adaptive distillation feature is obtained based on the distillation weight. The formula for the distillation weight is as follows:

[0144] W RAD =Sig(W 1R I crux +W 2R I cont );

[0145] In the formula, W 1R and W 2R The weight matrix is ​​trainable.

[0146] The regional adaptive distillation feature F RAD The calculation formula is:

[0147]

[0148] In the formula, ||I crux ⊙I cont ||2 is the L2 norm of the feature map, and e is the base of the natural logarithm.

[0149] S53. Use local pathological attention to perform attention calculations. The specific formula is as follows:

[0150] F LPSA =A LPSA (F RAD );

[0151] In the formula, A LPSA For local pathological attention.

[0152] S54. Use hierarchical recursive distillation to progressively extract and enhance features, and use an adaptive feature segmentation method to divide the features into two parts, which are used for the next layer of attention calculation and the final feature fusion, respectively.

[0153] S55. The distillation characteristics of different layers are fused, and the fusion formula is as follows:

[0154]

[0155] In the formula, the distillation level L is set to 16, and v i For adaptive fusion weights, The distillation characteristics of the i-th layer;

[0156] The adaptive fusion weight v i The specific calculation formula is as follows:

[0157]

[0158] In the formula, and Learning through attention gating mechanisms.

[0159] S6. Construct an image enhancement model for monitoring pediatric laryngeal and neck development, and complete data simulation and image enhancement for monitoring pediatric laryngeal and neck development through a pediatric laryngeal and neck data enhancement module and an adaptive information distillation network.

[0160] Furthermore, in step S6, the images from the pediatric laryngeal development monitoring are input into the image enhancement model for pediatric laryngeal development monitoring. First, data enhancement is performed through the pediatric laryngeal data enhancement module. Then, an adaptive information distillation network is used to extract and optimize detail information, significantly improving the quality of the super-resolution of the images from the pediatric laryngeal development monitoring. The image enhancement model for pediatric laryngeal development monitoring is based on the PyTorch framework and implemented using the PyCharm application. The model is trained using a training set of 1200 images from the pediatric laryngeal development monitoring image dataset. The trained model is then tested using a test set. In the image enhancement task for pediatric laryngeal development monitoring, not only is the image enhancement effect improved, but the model efficiency is also optimized.

[0161] Furthermore, such as Figure 4 As shown, Figure 4 The middle left half is the initial coronal CT image of the larynx. Figure 4 The right half of the image shows a coronal CT scan of the larynx after processing with an image enhancement model for monitoring pediatric laryngeal and neck development, which significantly enhances the clarity and recognizability of the images.

[0162] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. An image enhancement processing method for pediatric laryngeal neck development monitoring, characterized by, It comprises the following steps: S1, obtaining pediatric laryngeal neck development monitoring images and making data sets; S2, constructing a pediatric laryngeal neck data enhancement module, designing adaptive nonlinear transformation, contrast adaptive enhancement based on structural features, and simulating noise based on medical imaging physical properties to simulate the influence in the pediatric laryngeal neck image data acquisition process; S3, introducing Gaussian variational inference to design adaptive Gaussian variational attention, constructing pathological mask guided attention calculation, calculating local attention based on pathological mask and improved traditional attention, and combining local attention and adaptive Gaussian variational attention to calculate local pathological attention; S4, designing recursive distillation to gradually extract and enhance features based on recursive optimization strategy, designing adaptive feature segmentation method based on self-designed adaptive segmentation coefficient, constructing hierarchical recursive distillation module through recursive distillation and adaptive feature segmentation, and retaining important features to the next layer for attention calculation; S5, improving the traditional information distillation method to construct an adaptive information distillation network, including feature separation, hierarchical recursive distillation and cross-scale structure guided fusion, designing attention mask based on local gradient for feature separation, and designing adaptive fusion weight based on attention gate mechanism to fuse distillation features of different layers; The adaptive information distillation network first decomposes the pediatric laryngeal neck medical image into key regions and background regions, calculates the attention mask of the image, the key region information is more important than the background region, and higher distillation weight needs to be given to the key region, the region adaptive distillation feature is obtained according to the distillation weight, the local pathological attention is used for attention calculation, the features are gradually extracted and enhanced through hierarchical recursive distillation, and the features are divided into two parts through the adaptive feature segmentation method, which are used for attention calculation of the next layer and feature fusion at the end, and finally the distillation features of different layers are fused; S6, constructing an image enhancement model for pediatric laryngeal neck development monitoring, completing data simulation and image enhancement for pediatric laryngeal neck development monitoring through the pediatric laryngeal neck data enhancement module and the adaptive information distillation network.

2. The image enhancement processing method for pediatric laryngeal neck development monitoring according to claim 1, wherein, In the S1 step, the images for pediatric laryngeal neck development monitoring are obtained, the images of the vocal cords, trachea and laryngeal soft tissue are captured, the obtained images are data labeled, including key region labeling, development state labeling and case category labeling, the vocal cords, laryngeal cartilage and trachea are labeled as key regions, the development state is divided into three types of normal development, development lag and development abnormality, and the case category is labeled according to the pathological condition, and the labeled data is divided into training set and verification set.

3. The image enhancement processing method for pediatric laryngeal neck development monitoring according to claim 2, wherein, In the S2 step, the specific steps of the pediatric laryngeal neck data enhancement module are as follows: S21, simulate patient posture changes, imaging angle deviations and device jitter through adaptive nonlinear transformation, and the specific formula is: ； wherein, and is a spatial mapping matrix based on affine transformation, simulating angle offset and scale change respectively, N is the number of harmonic components, is an amplitude parameter, controlling the deformation strength, is a frequency parameter, controlling the local change scale, is a phase parameter, controlling the deformation initial state, is an image input to the pediatric laryngeal neck data enhancement module, is a sine function; S22, highlight airway, glottis and muscle tissue details through contrast adaptive based on structural features, and suppress background noise, and the specific formula is: ； wherein, is an edge enhancement strength, is a hyperbolic tangent function, is a Laplacian operator, and are the mean and standard deviation of respectively, is a contrast threshold, is a contrast enhancement strength, the contrast enhancement strength is mathematically formulated as: ； In the formula, is an exponential function, is a target luminance value; S23, through the simulation noise based on the physical properties of medical imaging, the specific formula is: ； wherein is the device noise intensity, is the device noise, is the tissue scatter noise intensity, is the tissue scatter noise, subject to a uniform distribution, is the dynamic blur intensity, is the rate of change in time; The device noise intensity The mathematical formula is: ； wherein is the current load of the device, is the maximum load of the device, is the usage time of the device, is the maximum usage time of the device.

4. The image enhancement processing method for pediatric laryngeal neck development monitoring according to claim 3, wherein, In the S3 step, the specific steps of local pathological attention calculation are as follows: S31, input the first image feature H, W and C are the height, width and channel of the first image respectively, construct a pathology mask Attention guidance calculation, distinguish between key throat area vocal cords, trachea, cartilage, irrelevant area background and low information area, reduce calculation complexity, calculate attention of important areas, pathology mask With the optimized Local attention is obtained by attention weighting The pathology mask The specific calculation formula is: ； In the formula, Z is the normalization factor. Controlling local smoothness, For pixel position grayscale value at that location It is a 5×5 local region. For local smoothing weights, Strengthen the weights at the edges. and These are the gradient values ​​at x and y, respectively; The local attention The specific calculation formula is: ； In the formula, G is the expansion order, is a weighted operation, Q, K, V are query, key, value respectively, and is a linear transformation matrix , , , , , , T represents transposition, and d is the dimension of a single attention head. S32, introducing Gaussian variational inference to optimize attention computation to obtain adaptive Gaussian variational attention The specific formula is: ； wherein Q is a query, , is a mean of Q, is a noise of a standard normal distribution, is a variance of Q, is an adaptive Gaussian compensation factor, is a variational probability density of Q, is a medical modality adjustment factor, , is a local variance of is a local mean of .​ The adaptive Gaussian compensation factor The specific calculation formula is: ； wherein is a linear transformation matrix, is a Gaussian bias term, controlling the reference for the compensation of the mean, is a learning parameter, ; S33, the local attention is obtained and adaptive Gaussian variational attention combining to obtain local pathological attention The specific formula is: ； In the formula, is an attention fusion weight, which is adaptively learned by a neural network.

5. The image enhancement processing method for pediatric laryngeal neck development monitoring according to claim 4, wherein, In the S4 step, the specific steps of the hierarchical recursive distillation module are as follows: S41, the specific calculation formula for the tth recursive distillation is: ； wherein is a hyperparameter controlling the recurrent weights, is the output of the t-th recursive distillation, is a convolution operation; S42, inputting a second image feature , , and are respectively height, width and channel of the second image, adaptive feature segmentation divides the feature into and two parts according to an adaptive segmentation coefficient, and a specific calculation formula of the adaptive segmentation coefficient is: ； wherein is a feature of the cth channel, , is a sigmoid function; The adaptive feature segmentation formula is: ； ； In the formula, input for next layer attention computation, for final feature fusion.

6. The image enhancement processing method for pediatric laryngeal neck development monitoring according to claim 5, wherein, In the S5 step, the adaptive information distillation network specific steps are: S51, decompose the pediatric laryngeal neck medical image into a key region and a background region, calculate an attention mask of the image, and based on the attention mask, extract a feature of the pediatric laryngeal neck medical image into the key region and the background region , H, W and C are height, width and channel of the pediatric laryngeal neck medical image, and the attention mask The specific calculation formula is: ； wherein is the local gradient value, , , is the maximum value taken; The critical region The calculation formula is: ； The background region The calculation formula is: ； S52, the key region information is more important than the background region, and the key region a higher distillation weight is given, and a region adaptive distillation feature is obtained according to the distillation weight, and the distillation weight formula is: ； wherein and are trainable weight matrices; The region adaptive distillation feature The calculation formula is: ； wherein is the L2 norm of the feature map, e is the base of the natural logarithm; S53, attention calculation is performed using local pathological attention, and the specific formula is: ； In the formula, Local pathology attention; S54, hierarchical recursive distillation is used to extract and enhance features step by step, and an adaptive feature segmentation method is used to divide the features into two parts, which are used for next layer attention calculation and final feature fusion respectively; S55, the distillation features of different layers are fused, and the fusion formula is: ； In the formula, L is a distillation level, is an adaptive fusion weight, is a distillation feature of the i-th layer. The adaptive fusion weight The specific calculation formula is: ； In the formula, and Through attention gate mechanism learning.

Images