Image enhancement-based mouth opening difficulty degree evaluation system for head and neck radiotherapy patient

Abstract

The invention discloses an image enhancement-based mouth opening difficulty degree evaluation system for a head and neck radiotherapy patient. The system comprises an image acquisition module, a mouth opening image enhancement module, a mouth opening image segmentation module and a mouth opening difficulty degree evaluation module. The invention belongs to the field of image processing, and particularly relates to a head and neck radiotherapy patient mouth opening difficulty degree evaluation system based on image enhancement, and the scheme adopts a sensitivity weight-based sampling strategy to alleviate the problem of influence of class imbalance on subsequent evaluation; a local attenuation compensation item is introduced, and correction is carried out based on a multi-scale degradation feature extraction unit and an adaptive modulation parameter generation unit; on the basis of construction of region penalty loss and introduction of boundary numerical value supervision loss of boundary distance weighting, global and local information is considered, and the method adapts to characteristics of edge blur and different local forms generally existing in patient images after radiotherapy; and thus, the mouth opening difficulty degree evaluation effect of the subsequent head and neck radiotherapy patient is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and specifically refers to a system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients based on image enhancement. Background Art

[0002] A system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients generally refers to a comprehensive system that uses medical imaging technology to quantitatively evaluate the degree of mouth opening limitation of patients by collecting oral opening images of patients, finely segmenting the images, and then extracting relevant morphological features. However, the general system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients has problems such as uneven distribution of foreground and background samples, lack of an adaptive correction mechanism for different facial regions, inability to adjust the compensation intensity according to local conditions, and thus poor accuracy in evaluating the subsequent degree of mouth opening difficulty; the general system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients lacks a special processing mechanism for the edge region. In the case of obvious tissue degradation and local morphological changes after radiotherapy, the edge information is easily blurred, resulting in unclear edges of the segmentation result and thus affecting the overall evaluation effect. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides a system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients based on image enhancement. For the general system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients, there are problems such as uneven distribution of foreground and background samples, lack of a mechanism for adaptive correction for different facial regions, and inability to adjust the compensation intensity according to local conditions, which leads to poor accuracy in evaluating the subsequent degree of mouth opening difficulty. This solution adopts a sampling strategy based on sensitivity weights to alleviate the problem that class imbalance affects subsequent evaluations. For the complex local degradation phenomenon of radiotherapy patients, a local attenuation compensation term is introduced to simulate local additional attenuation, and correction is performed based on a multi-scale degradation feature extraction unit and an adaptive modulation parameter generation unit, which can correct both global light reflection and noise, and restore the details of tooth edges and soft tissues; thereby improving the accuracy of subsequent evaluations of the degree of mouth opening difficulty. For the general system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients, there is a lack of a special processing mechanism for edge regions. After radiotherapy, due to obvious tissue degradation and local morphological changes, edge information is easily blurred, resulting in unclear edges in the segmentation result, which in turn affects the overall evaluation effect. This solution adds learnable position encoding through a feature extraction and encoding unit to capture spatial distribution information, so as to more accurately locate local fine regions of oral images; based on multi-scale feature extraction, it can capture details and maintain the global contour while dealing with edge blurring and local degradation, adapting to the complex situation of morphological changes in the oral region after radiotherapy; the skip connection of the decoding and upsampling unit effectively retains low-level details, which is particularly crucial for the problem of local detail loss caused by radiotherapy, so that the final segmentation result is more accurate in terms of details and overall morphology; finally, based on constructing a regional penalty loss and introducing a boundary numerical supervision loss weighted by the boundary distance, taking into account both global and local information, adapting to the characteristics of edge blurring and inconsistent local morphology commonly existing in the images of radiotherapy patients; thereby improving the evaluation effect of the degree of mouth opening difficulty of subsequent head and neck radiotherapy patients.

[0004] The technical solution adopted by the present invention is as follows: The system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients based on image enhancement provided by the present invention includes an image acquisition module, a mouth opening image enhancement module, a mouth opening image segmentation module, and a mouth opening difficulty degree evaluation module;

[0005] The image acquisition module acquires oral opening images of historical head and neck radiotherapy patients and screens the images using the imbalance ratio and sensitivity weights;

[0006] The mouth opening image enhancement module uses degradation feature extraction and adaptive modulation parameter generation to correct the acquired oral opening images, thereby realizing mouth opening image enhancement;

[0007] The mouth-opening image segmentation module achieves fine pixel-level segmentation of the oral opening area after radiotherapy through multi-scale feature extraction, position encoding, and Transformer global modeling, combined with progressive upsampling and dual-loss optimization.

[0008] The mouth-opening difficulty assessment module uses morphological processing to extract the boundary and contour of the oral opening and generates a quantitative score using a pre-trained regression model to evaluate the degree of mouth-opening difficulty.

[0009] Furthermore, the image acquisition module annotates the oral opening area in the oral opening images of historical head and neck radiotherapy patients; defines the imbalance ratio and introduces a sampling method based on the imbalance ratio; regards the oral opening area as the foreground and the rest of the facial area as the background; introduces a sensitivity weight , and finally the imbalance ratio IR is expressed as: ; ; ; where x and y are pixel coordinate indices; is the weighted area of the oral opening area; is the area of the oral opening area; is the area of the rest of the facial area; is the adjustment factor; and are the horizontal and vertical gradient values at the pixel (x,y), respectively; arrange the oral opening images of historical head and neck radiotherapy patients in ascending order of the imbalance ratio and select N images as the final image set.

[0010] Furthermore, the mouth-opening image enhancement module specifically includes the following:

[0011] The comprehensive simulation unit for the degradation of the oral opening image; introduce the local attenuation compensation term , and the acquired oral opening image of the patient is simulated as: ; ; where is the oral opening image obtained from the patient; is the ideal oral structure image; is the additional noise; is the visibility inside the oral cavity; L is the light reflection intensity of the oral mirror; is the convolution operation; is the degraded feature map extracted at (x,y);

[0012] The oral opening image rearrangement unit; the rearranged oral opening image is expressed as: ; ; ; and are modulation parameters;

[0013] Degradation feature extraction unit; taking the patient's oral opening image, as well as gradient and texture information as inputs, the degradation feature extraction unit consists of T convolutional layers, and each layer is followed by a ReLU activation function and a normalization layer to form a series of local feature representations; the overall representation is: ; where is feature fusion; is a feature concatenation operation; 、 and are to extract features using 3×3, 5×5, and 7×7 convolutional kernels;

[0014] Adaptive modulation parameter generation unit; using Conv1DNet and the embedding block to generate modulation parameters in the channel and spatial dimensions and ; adopting three parallel 1D convolutional branches on the embedded features, with convolutional kernel sizes of 3, 5, and 7 respectively; used to capture context information at different scales; fusing the outputs at different scales to obtain the final modulation parameters, expressed as: ; ; where 、 and are local modulation parameters extracted at position (x,y) through 1D convolutional branches with different convolutional kernel sizes ; 、 and are local modulation parameters extracted at position (x,y) through 1D convolutional branches with different convolutional kernel sizes ; is to use a fully connected layer to perform a fusion operation on multi-scale features;

[0015] Reconstruction error minimization unit; defining a reconstruction loss function, and optimizing the modulation parameters by minimizing the error between the reconstructed image and the ideal image, expressed as: ; where is to select the modulation parameters that minimize the loss function L; adjusting the parameters of Conv1DNet based on the loss gradient, and training the oral opening image enhancement module based on the original image set.

[0016] Furthermore, the oral opening image segmentation module performs pixel-level segmentation on the image output by the oral opening image enhancement module; the oral opening image segmentation module includes a feature extraction and encoding unit B(·), a decoding and upsampling unit D(·), and a dual loss optimization unit; the image after being processed by the oral opening image enhancement module As the input to the segmentation network; the output is a pixel-level probability map P F (x), where the value of each pixel represents the probability that it belongs to the oral opening area, and finally the segmentation result is obtained; the entire open-mouth image segmentation module is expressed as: ; Specifically, it includes the following content:

[0017] Feature extraction and encoding unit; extract low-level features from the input image to obtain a feature map : It is expressed as: ; Add positional encoding P(x,y), take the positional encoding as a learnable parameter, initialize it to a random value, and automatically adjust it through backpropagation during training to obtain the embedded feature , which is expressed as: ; Aiming at the scale difference between global and local information in the oral opening image, three parallel 1D convolutional branches are adopted 、 and , which is expressed as: ; ; ; Each branch focuses on different receptive fields, captures the details of the lip edge and dental floss gap and the features of the overall oral contour; fuse the outputs of the multi-scale branches to generate local modulation features , which is expressed as: ; Add Transformer encoding to capture long-range dependencies to obtain a high-dimensional feature map , which is expressed as: ; And take as the final output of the high-dimensional feature map; among them, is the initial convolution operation; is the low-level feature map; is a one-dimensional convolution operation with a convolution kernel size of k, where k = 3, 5, 7 are the convolution kernel sizes; is the ReLU activation function; 、 and are the fusion weights; is the Transformer encoding, which uses the self-attention mechanism to globally reorganize the fused features, highlighting key structures and edge information;

[0018] Decoder upsampling unit; the decoder converts the high-dimensional feature map into a pixel-level prediction map with the same size as the input image; use progressive upsampling to restore the spatial resolution, and at the same time combine skip connections to retain low-level details, which is expressed as: ; Finally, obtain the pixel-level prediction probability through a layer of convolution and activation function , which is expressed as: ; Among them, is the high-dimensional feature at the l-th layer in the encoding stage; is an upsampling operation; is the low-level feature from the corresponding layer of the encoder; is the decoded upsampling output of the l-th layer; is the Sigmoid function; is the last convolutional operation, which converts the upsampled feature into a predicted value; is the total decoded upsampling output;

[0019] The dual loss optimization unit includes:

[0020] Define the region penalty; design the region penalty loss; define the pixel-level label Y ∈ {0, 1}, where 0 represents the background and 1 represents the oral opening region; the predicted probability is P F (x), and perform power modulation on the predicted probability; the region penalty loss is expressed as: ; where, and are weight factors; is the modulation exponent;

[0021] Define the boundary numerical supervision loss ; introduce the boundary distance weighting The boundary numerical supervision loss is expressed as: ; ; where, is the matching probability between the predicted boundary numerical label and the true boundary label; is the image region; and are adjustment parameters; is the distance from the pixel to the true boundary;

[0022] The final loss function L is expressed as: ; where, is the weight factor of the boundary loss.

[0023] Furthermore, the trismus degree evaluation module real-time collects the oral opening images of head and neck radiotherapy patients. After being processed by the oral opening image enhancement module and the oral opening image segmentation module, it uses morphological processing to extract the oral opening boundary and regional contour; and inputs the morphological processing indexes into a pre-trained regression model to generate a quantitative trismus score for trismus degree evaluation.

[0024] The beneficial effects achieved by the present invention using the above solution are as follows:

[0025] (1) Aiming at the problem that the general head and neck radiotherapy patient's mouth opening difficulty assessment system has uneven foreground and background sample distributions, lacks a mechanism for adaptive correction for different facial regions, and cannot adjust the compensation intensity according to local conditions, resulting in poor accuracy of subsequent mouth opening difficulty assessment. This solution adopts a sampling strategy based on sensitivity weights to alleviate the problem that class imbalance affects subsequent assessment. For the complex local degradation phenomenon of radiotherapy patients, a local attenuation compensation term is introduced to simulate local additional attenuation, and correction is performed based on a multi-scale degradation feature extraction unit and an adaptive modulation parameter generation unit, which can correct both global light reflection and noise, and restore the details of tooth edges and soft tissues; thereby improving the accuracy of subsequent mouth opening difficulty assessment.

[0026] (2) Aiming at the problem that the general head and neck radiotherapy patient's mouth opening difficulty assessment system lacks a special processing mechanism for edge regions. After radiotherapy, due to obvious tissue degradation and local morphological changes, edge information is easily blurred, resulting in unclear edges in the segmentation result, and further affecting the overall assessment effect. This solution adds learnable position encoding to the feature extraction and encoding unit to capture spatial distribution information, so as to more accurately locate local fine regions of oral images; based on multi-scale feature extraction, it can capture details and maintain the global contour while dealing with edge blur and local degradation, adapting to the complex situation of morphological changes in the oral region after radiotherapy; the skip connection of the decoding and upsampling unit effectively retains low-level details, which is particularly crucial for the problem of local detail loss caused by radiotherapy, so that the final segmentation result is more accurate in terms of details and overall morphology; finally, based on constructing a regional penalty loss and introducing a boundary numerical supervision loss weighted by boundary distance, taking into account both global and local information, adapting to the characteristics of edge blur and inconsistent local morphology commonly existing in the images of radiotherapy patients; thereby improving the assessment effect of mouth opening difficulty in subsequent head and neck radiotherapy patients. Description of the Drawings

[0027] Figure 1 It is a schematic flowchart of the mouth opening difficulty assessment system for head and neck radiotherapy patients based on image enhancement provided by the present invention;

[0028] Figure 2 It is a schematic flowchart of the mouth opening image segmentation module.

[0029] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. Detailed Embodiments

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0032] Embodiment 1. Refer to Figure 1 , the system for evaluating the degree of mouth opening difficulty of head and neck radiotherapy patients based on image enhancement provided by the present invention includes an image acquisition module, a mouth opening image enhancement module, a mouth opening image segmentation module, and a mouth opening difficulty degree evaluation module;

[0033] The image acquisition module acquires the oral opening images of historical head and neck radiotherapy patients, and screens the images using the imbalance ratio and sensitivity weight; and sends the data to the mouth opening image enhancement module;

[0034] The mouth opening image enhancement module uses the extraction of degradation features and the generation of adaptive modulation parameters to correct the acquired oral opening images, thereby realizing the enhancement of the mouth opening images; and sends the data to the mouth opening image segmentation module;

[0035] The mouth opening image segmentation module realizes the fine pixel-level segmentation of the oral opening area after radiotherapy through multi-scale feature extraction, position encoding, and Transformer global modeling, combined with progressive upsampling and dual loss optimization; and sends the data to the mouth opening difficulty degree evaluation module;

[0036] The mouth opening difficulty degree evaluation module uses morphological processing to extract the boundary and contour of the oral opening, and uses a pre-trained regression model to generate a quantitative score to realize the evaluation of the degree of mouth opening difficulty.

[0037] Embodiment 2. Refer to Figure 1, this embodiment is based on the above embodiment. The image acquisition module annotates the oral opening area in the oral opening images of historical head and neck radiotherapy patients; defines the imbalance ratio. Since the oral opening area usually accounts for a relatively small proportion in the overall facial image, direct training may face serious class imbalance problems; introduces a sampling method based on the imbalance ratio; regards the oral opening area as the foreground and the remaining facial area as the background; introduces a sensitivity weight , magnifies the proportion of the key structure. Even if the overall area does not change much, the changes in the key parts can be fully reflected, so as to more accurately measure the degree of mouth opening limitation of the patient. Finally, the imbalance ratio IR is expressed as: ; ; ; where x and y are pixel coordinate indices; is the weighted area of the oral opening area; is the area of the oral opening area; is the area of the remaining facial area; is the adjustment factor; and are the gradient values in the horizontal and vertical directions at the pixel (x, y) respectively; when the patient has difficulty opening the mouth, the area of the oral opening area will become relatively smaller, while the area of the remaining facial area basically remains unchanged. In this way, the imbalance ratio will decrease. Therefore, a low imbalance ratio can intuitively reflect that the patient's mouth opening amplitude becomes smaller and accurately reflect the actual situation of the patient's mouth opening limitation; arrange the oral opening images of historical head and neck radiotherapy patients in ascending order of the imbalance ratio, and select N images as the final image set.

[0038] Embodiment Three, refer to Figure 1 , this embodiment is based on the above embodiment. The mouth opening image enhancement module specifically includes the following:

[0039] Oral opening image degradation comprehensive simulation unit; considering the oral opening image degradation caused by factors such as soft tissue fibrosis, mucosal damage and muscle stiffness in radiotherapy patients; and introducing a local attenuation compensation term , used to describe the additional attenuation effect brought by fibrosis or tissue stiffness in the local area. The simulated representation of the collected patient's oral opening image is: ; ; where, is the oral opening image obtained from the patient; is the oral structure image that can accurately measure the intraoral opening and the distance between dental arches ideally; is the additional noise, including ambient light interference and sensor noise; is the visibility inside the oral cavity; L is the light reflection intensity of the oral mirror light; is the convolution operation; is the degraded feature map extracted at (x, y);

[0040] Oral opening image rearrangement unit; simplifies the complex degradation process into local multiplication and addition operations, corrects the acquired image using local visibility, compensates for the effects of light reflection and noise; enables the restoration of key tooth edges and soft tissue details in the image; the rearranged oral opening image is expressed as: ; ; ; and are modulation parameters, representing the combined effects of local imaging conditions and occlusive reflection noise respectively;

[0041] Degraded feature extraction unit; takes the patient's oral opening image, as well as gradient and texture information as input. The degraded feature extraction unit consists of multiple convolutional layers, and each layer is followed by a ReLU activation function and a normalization layer to form a series of local feature representations; the overall representation is: ; where is feature fusion; is feature concatenation operation; , and are to extract features using 3×3, 5×5, and 7×7 convolutional kernels;

[0042] Adaptive modulation parameter generation unit; adaptively adjusts for different regions such as the alveolar region, near the joint, and the soft tissue boundary, thereby correcting feature distortion caused by insufficient mouth opening locally; generates modulation parameters in the channel and spatial dimensions using Conv1DNet and the embedding block and ; for the scale differences existing in the global morphology of the overall opening width and the local details of the lip contour and tooth gaps in the oral opening image, three parallel 1D convolutional branches are adopted on the embedded features, with the convolutional kernel sizes being 3, 5, and 7 respectively; used to capture context information at different scales; fuse the outputs at different scales to obtain the final modulation parameters, expressed as: ; ; where , and are local modulation parameters extracted at position (x, y) through 1D convolutional branches with different convolutional kernel sizes ; , and are local modulation parameters extracted at position (x, y) through 1D convolutional branches with different convolutional kernel sizes ; Perform a fusion operation on multi-scale features using a fully connected layer;

[0043] Reconstruction error minimization unit; Define a reconstruction loss function to optimize the modulation parameters by minimizing the error between the reconstructed image and the ideal image, expressed as: ; where is to select the modulation parameters that minimize the loss function L; Adjust the parameters of Conv1DNet based on the loss gradient, and train the mouth opening image enhancement module based on the original image set.

[0044] By performing the above operations, for the mouth opening difficulty assessment system for general head and neck radiotherapy patients, there are problems such as uneven distribution of foreground and background samples, lack of a mechanism for adaptive correction for different facial regions, and inability to adjust the compensation intensity according to local conditions, resulting in poor accuracy of subsequent mouth opening difficulty assessment. This solution adopts a sampling strategy based on sensitivity weights to alleviate the problem that class imbalance affects subsequent assessment. For the complex local degradation phenomenon of radiotherapy patients, a local attenuation compensation term is introduced to simulate local additional attenuation, and correction is performed based on the multi-scale degradation feature extraction unit and the adaptive modulation parameter generation unit, which can correct both global light reflection and noise, and restore the details of tooth edges and soft tissues; thereby improving the accuracy of subsequent mouth opening difficulty assessment.

[0045] Example 4, refer to Figure 1 and Figure 2 , based on the above example, the mouth opening image segmentation module performs pixel-level segmentation on the image output by the mouth opening image enhancement module, aiming to locate the oral opening area of head and neck radiotherapy patients; the mouth opening image segmentation module includes a feature extraction and encoding unit B(·), a decoding and upsampling unit D(·), and a dual loss optimization unit; the image processed by the mouth opening image enhancement module is F used as the input of the segmentation network; the output is a pixel-level probability map P ; specifically including the following content:

[0046] Feature extraction and encoding unit; Extract low-level features from the input image to obtain a feature map : expressed as: ; To enable the mouth opening image segmentation module to perceive the information of each spatial position of the oral opening image, a position encoding P(x,y) is added, and the position encoding is used as a learnable parameter, initialized to a random value, and automatically adjusted through backpropagation during training to obtain the embedded feature : expressed as: ; Considering the scale differences between the global and local information in the oral cavity opening images, three parallel 1D convolutional branches are adopted , and , expressed as: ; ; ; Each branch focuses on different receptive fields, capturing the details of the lip margin and dental gaps and the features of the overall oral cavity contour; The outputs of the multi-scale branches are fused to generate local modulation features , expressed as: ; Adding Transformer encoding to capture long-range dependencies to obtain a high-dimensional feature map , expressed as: ; And taking as the final output of the high-dimensional feature map; Among them, is the initial convolution operation; is the low-level feature map; is a one-dimensional convolution operation with a convolution kernel size of k, where k = 3, 5, 7 are the convolution kernel sizes; is the ReLU activation function; , and are the fusion weights; is the Transformer encoding, which globally reorganizes the fused features using the self-attention mechanism to highlight key structures and edge information;

[0047] Decoding upsampling unit; The decoder converts the high-dimensional feature map into a pixel-level prediction map of the same size as the input image; Gradual upsampling is used to restore the spatial resolution, and at the same time, skip connections are combined to retain low-level details, expressed as: ; Finally, a pixel-level prediction probability is obtained through a layer of convolution and activation function , expressed as: ; Among them, is the high-dimensional feature at the l-th layer in the encoding stage; is the upsampling operation; is the low-level feature from the corresponding layer of the encoder; is the decoding upsampling output at the l-th layer; is the Sigmoid function; is the last layer of convolution operation, which converts the upsampled features into prediction values; is the total decoding upsampling output;

[0048] Dual loss optimization unit; To improve the segmentation accuracy, especially in dealing with the edge blur and local morphological changes in the oral cavity opening images after radiotherapy, two loss functions are designed;

[0049] Define the regional penalty; to accurately segment the oral cavity opening area at the pixel level, design a regional penalty loss to impose a higher penalty on the mis-segmented area; define the pixel-level label Y ∈ {0, 1}, where 0 represents the background and 1 represents the oral cavity opening area; the predicted probability is P F (x), and perform power modulation on the predicted probability to reduce the loss contribution of easily classified pixels; the regional penalty loss is expressed as: ; where and are weight factors used to balance the penalty strengths of positive and negative samples; is the modulation index;

[0050] Define the boundary numerical supervision loss ; used to optimize the complex boundary of the oral cavity opening area, which is more sensitive to the oral cavity opening area under radiotherapy; and introduce boundary distance weighting , assign higher weights to pixels closer to the boundary, which is more sensitive to the complex boundary of the oral cavity area after radiotherapy; the boundary numerical supervision loss is expressed as: ; ; where is the matching probability between the predicted boundary numerical label and the true boundary label; is the image area; and are adjustment parameters; is the distance from the pixel to the true boundary;

[0051] The final loss function L is expressed as: ; where is the weight factor of the boundary loss.

[0052] By performing the above operations, there is a problem that the existing oral cavity opening difficulty degree evaluation system for general head and neck radiotherapy patients lacks a dedicated processing mechanism for the edge region. After radiotherapy, due to obvious tissue degeneration and local morphological changes, the edge information is easily blurred, resulting in unclear edges of the segmentation result, which in turn affects the overall evaluation effect. In this solution, a learnable position encoding is added to the feature extraction and encoding unit to capture the spatial distribution information, so as to more accurately locate the local fine regions of oral cavity images; based on multi-scale feature extraction, it is ensured that in the case of dealing with edge blurring and local degeneration, both details can be captured and the global contour can be maintained, adapting to the complex situation of the morphological changes in the oral cavity region after radiotherapy; through the skip connection of the decoding and upsampling unit, low-level details are effectively retained, which is particularly crucial for the problem of local detail loss caused by radiotherapy, so that the final segmentation result is more accurate in terms of details and overall morphology; finally, based on constructing a regional penalty loss and introducing a boundary numerical supervision loss weighted by the boundary distance, both global and local information are taken into account, adapting to the characteristics of edge blurring and inconsistent local morphology commonly existing in the images of radiotherapy patients; thereby improving the evaluation effect of the oral cavity opening difficulty degree of subsequent head and neck radiotherapy patients.

[0053] Example VI. Refer to Figure 1 , based on the above example, the oral cavity opening difficulty degree evaluation module continuously collects the oral cavity opening images of head and neck radiotherapy patients. After being processed by the oral cavity opening image enhancement module and the oral cavity opening image segmentation module, the oral cavity opening boundary and regional contour are extracted by using morphological processing; the oral cavity opening width is obtained by calculating the maximum distance between the two sides of the segmentation region; pixel counting is performed on the segmentation region and converted into the actual area to obtain the oral cavity opening area; and the edge smoothness and curvature are calculated; the above indexes are input into a pre-trained regression model to generate a quantitative oral cavity opening difficulty score for the evaluation of the oral cavity opening difficulty degree; if the evaluation result is oral cavity opening difficulty, a warning is given to relevant personnel.

[0054] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0055] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0056] The above description of the present invention and its embodiments is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and, without departing from the spirit of the present invention, design similar structural forms and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. An evaluation system for the degree of trismus in head and neck radiotherapy patients based on image enhancement, characterized in that: The system includes an image acquisition module, an open - mouth image enhancement module, an open - mouth image segmentation module, and an evaluation module for the degree of trismus; The image acquisition module acquires oral opening images of historical head - and - neck radiotherapy patients and screens the images using the imbalance ratio and sensitivity weight; The open - mouth image enhancement module is generated by degradation feature extraction and adaptive modulation parameters, and corrects the acquired oral opening images to achieve open - mouth image enhancement; The open - mouth image segmentation module realizes fine pixel - level segmentation of the oral opening area after radiotherapy through multi - scale feature extraction, position encoding, and Transformer global modeling, combined with progressive upsampling and dual - loss optimization; The evaluation module for the degree of trismus uses morphological processing to extract the oral opening boundary and contour, and generates a quantitative score using a pre - trained regression model to evaluate the degree of trismus; 2. The system for evaluating the degree of trismus in head and neck radiotherapy patients based on image enhancement according to claim 1, wherein: The image acquisition module annotates the oral opening area in the oral opening images of historical head - and - neck radiotherapy patients; Define the imbalance ratio and introduce a sampling method based on the imbalance ratio; consider the oral opening area as the foreground and the remaining facial area as the background; introduce the sensitivity weight , and finally the imbalance ratio IR is expressed as: ; ; ; where x and y are pixel coordinate indices; is the weighted area of the oral opening area; is the area of the oral opening area; is the area of the remaining facial area; is the adjustment factor; and are the gradient values in the horizontal and vertical directions at the pixel (x, y), respectively; Arrange the oral opening images of historical head and neck radiotherapy patients in ascending order of the imbalance ratio, and select N images as the final image set.

3. The head and neck radiotherapy patient trismus degree evaluation system based on image enhancement according to claim 2, wherein: The open - mouth image enhancement module specifically includes the following: Oral opening image degradation comprehensive simulation unit; introducing a local attenuation compensation term , the acquired oral opening image of the patient is simulated as: ; ; where is the oral opening image obtained from the patient; is the ideal oral structure image; is the additional noise; is the visibility inside the oral cavity; L is the intensity of the oral mirror light reflection; is the convolution operation; is the degraded feature map extracted at (x, y); Oral cavity opening image rearrangement unit; rearranged oral cavity opening image Expressed as: ； ; ; and are modulation parameters; Degradation feature extraction unit; taking the patient's oral opening image, as well as gradient and texture information as input, the degradation feature extraction unit consists of T convolutional layers, and after each convolution, a ReLU activation function and a normalization layer are connected to form a series of local feature representations; the overall representation is: ; where is feature fusion; is the feature concatenation operation; , and are to extract features using 3×3, 5×5 and 7×7 convolutional kernels; Adaptive modulation parameter generation unit; generates modulation parameters in the channel and spatial dimensions using Conv1DNet and the embedding block and ; adopts three parallel 1D convolutional branches on the embedded features, with convolutional kernel sizes of 3, 5, and 7 respectively; used to capture context information at different scales; fuses the outputs at different scales to obtain the final modulation parameters, expressed as: ; ; where , and are local modulation parameters extracted at position (x, y) through 1D convolutional branches with different convolutional kernel sizes ; , and are local modulation parameters extracted at position (x, y) through 1D convolutional branches with different convolutional kernel sizes ; uses a fully connected layer to perform a fusion operation on multi-scale features; Reconstruction error minimization unit; define a reconstruction loss function, and optimize the modulation parameters by minimizing the error between the reconstructed image and the ideal image, expressed as: ; where is the modulation parameter that selects the one that minimizes the loss function L; adjust the parameters of Conv1DNet based on the loss gradient, and train the open-mouth image enhancement module based on the original image set.

4. The head and neck radiotherapy patient trismus degree evaluation system based on image enhancement according to claim 3, wherein: The open - mouth image segmentation module performs pixel - level segmentation on the image output by the open - mouth image enhancement module. The open - mouth image segmentation module includes a feature extraction and encoding unit B(·), a decoding and up - sampling unit D(·), and a dual - loss optimization unit. The image processed by the open - mouth image enhancement module is used as the input of the segmentation network; the output is a pixel - level probability map P F (x), where the value of each pixel represents the probability that it belongs to the oral opening area, and finally the segmentation result is obtained. The entire open - mouth image segmentation module is expressed as: ; specifically Includes the following: Feature extraction and encoding unit; extract low-level features from the input image to obtain a feature map, expressed as: ; Add the position encoding P(x, y), and use the position encoding as a learnable parameter, which is initialized to a random value and automatically adjusted through backpropagation during training to obtain the embedded features , which is expressed as: ; For the scale difference between the global and local information in the oral opening image, three parallel 1D convolutional branches are adopted , and , which is expressed as: ; ; ; Each branch focuses on different receptive fields, capturing the details of the lip edge and tooth gap and the features of the overall oral cavity contour; The outputs of the multi-scale branches are fused to generate local modulation features , which is expressed as: ; Add the Transformer encoding to capture long-range dependencies and obtain a high-dimensional feature map , which is expressed as: ; and take as the final output of the high-dimensional feature map; where is the initial convolution operation; is the low-level feature map; is a one-dimensional convolution operation with a convolution kernel size of k, where k = 3, 5, 7 are the convolution kernel sizes; is the ReLU activation function; , and are the fusion weights; is the Transformer encoding, which uses the self-attention mechanism to globally reorganize the fused features and highlight key structures and edge information; Decoder upsampling unit; the decoder converts the high-dimensional feature map into a pixel-level prediction map with the same size as the input image; gradually upsampling is used to restore the spatial resolution, and at the same time, skip connections are combined to retain low-level details, expressed as: ; Finally, pixel-level prediction probabilities are obtained through a layer of convolution and activation function , expressed as: ; where is the high-dimensional feature at the l-th layer in the encoding stage; is the upsampling operation; is the low-level feature from the corresponding layer of the encoder; is the output of the decoder upsampling at the l-th layer; is the Sigmoid function; is the last layer of convolution operation, which converts the upsampled features into prediction values; is the total output of the decoder upsampling; Dual - loss optimization unit.

5. The system for evaluating the degree of trismus in head and neck radiotherapy patients based on image enhancement according to claim 4, characterized in that: The dual - loss optimization unit includes: Define the region penalty; design the region penalty loss; define the pixel-level label Y ∈ {0, 1}, where 0 represents the background and 1 represents the oral opening region; the predicted probability is P F (x), and perform power modulation on the predicted probability; the region penalty loss is expressed as: ; where and are weight factors; is the modulation index; Define the boundary numerical supervision loss ; Introduce boundary distance weighting , and the boundary numerical supervision loss is expressed as: ; ; Among them, is the matching probability between the predicted boundary numerical label and the true boundary label; is the image region; and are adjustment parameters; is the distance from the pixel to the true boundary; The final loss function L is expressed as: ; where is the weight factor of the boundary loss.

Images