A method and system for oral cavity region segmentation based on head and neck CT images

Abstract

This invention discloses a method and system for oral cavity region segmentation based on head and neck CT images, belonging to the field of medical image processing technology. The method includes: acquiring original head and neck radiotherapy CT images; preprocessing the CT images to obtain preprocessed images; constructing a multi-scale convolutional attention block U-Net model; training and parameter tuning the multi-scale convolutional attention block U-Net model to obtain a trained optimal multi-scale convolutional attention block U-Net model; validating the model performance on a validation set using the trained optimal multi-scale convolutional attention block U-Net model to obtain an oral cavity segmentation model; and inputting the preprocessed images into the oral cavity segmentation model to obtain oral cavity region segmentation results. This method improves the segmentation accuracy of the oral cavity region, enhances its resistance to noise and complex structures, and enables more accurate, stable, and comprehensive oral cavity segmentation, providing reliable technical support for the precision treatment of head and neck radiotherapy.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method and system for oral cavity region segmentation based on head and neck CT images. Background Technology

[0002] Head and neck radiotherapy is a common cancer treatment method, in which CT (Computed Tomography) images play a crucial role in diagnosis and treatment. CT images provide detailed information about the patient's anatomy, thus playing a vital role in radiotherapy planning, dose allocation, and treatment efficacy evaluation. Traditional CT image segmentation methods are mainly based on basic image processing techniques such as threshold segmentation, region growing, and edge detection.

[0003] Traditional CT image segmentation methods have the following problems:

[0004] (1) Low segmentation accuracy: Traditional methods based on thresholding and region growing have low segmentation accuracy for oral cavity regions in head and neck radiotherapy CT images, which limits the accuracy of radiotherapy planning and the evaluation of treatment effects.

[0005] (2) Sensitivity to noise and artifact structures: Traditional methods are highly sensitive to noise and complex structures in images, making them susceptible to external interference and resulting in unstable segmentation results, which in turn affects the formulation and execution of radiotherapy plans. Furthermore, most current techniques focus on processing artifact-free images and delineating the oral cavity region based on these images, but relatively little consideration is given to artifact interference. Oral cavity artifacts are one of the common problems in CT images of head and neck radiotherapy, and they can originate from various factors, such as metal prostheses, intraoral implants, and masticatory muscle movements. These artifacts can not only distort the morphology of the oral cavity region but also affect the accuracy of oral cavity segmentation, leading to inaccurate radiotherapy planning and limited evaluation of treatment effects.

[0006] (3) Lack of multi-scale information: Traditional methods lack effective use of multi-scale information in images and cannot fully capture the details and structural information inside the oral cavity, resulting in a lack of completeness and accuracy in the segmentation results, which affects the accuracy of radiotherapy planning and the evaluation of treatment effects. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for oral cavity region segmentation based on head and neck CT images, which effectively solves the technical problem of low oral cavity segmentation accuracy in existing technologies.

[0008] In a first aspect, the present invention provides a method for oral cavity region segmentation based on head and neck CT images, comprising: including:

[0009] Acquire raw head and neck radiotherapy CT images;

[0010] The CT images are preprocessed to obtain preprocessed images;

[0011] Construct a multi-scale convolutional attention block U-Net model, train and tune the multi-scale convolutional attention block U-Net model to obtain the trained optimal multi-scale convolutional attention block U-Net model;

[0012] The oral cavity segmentation model was obtained by validating the model performance on the validation set using the trained optimal multi-scale convolutional attention block U-Net model.

[0013] The preprocessed image is input into the oral cavity segmentation model to obtain the oral cavity region segmentation result.

[0014] Secondly, the present invention provides an oral cavity region segmentation system based on head and neck CT images, comprising: an image acquisition module, a preprocessing module, a model building module, a model verification module, and an image segmentation module;

[0015] The image acquisition module is used to acquire raw head and neck radiotherapy CT images;

[0016] The preprocessing module is used to preprocess CT images to obtain preprocessed images;

[0017] The model building module is used to build a multi-scale convolutional attention block U-Net model, train and tune the multi-scale convolutional attention block U-Net model, and obtain the trained optimal multi-scale convolutional attention block U-Net model.

[0018] The model validation module is used to validate the model performance on the validation set using the trained optimal multi-scale convolutional attention block U-Net model to obtain the oral cavity segmentation model.

[0019] The image segmentation module inputs the preprocessed image into the oral cavity segmentation model to obtain the oral cavity region segmentation result.

[0020] The beneficial effects of this invention are:

[0021] This invention provides a method and system for oral cavity region segmentation based on head and neck CT images. The constructed multi-scale convolutional attention block U-Net model, combining the U-Net structure and multi-scale convolutional attention mechanism, can segment both artifact-containing and artifact-free oral cavities, effectively solving the gradient vanishing and exploding problems present in traditional U-Net models, thus improving the model's convergence speed and training stability. Simultaneously, the use of multi-scale convolutional layers and dilated convolutions enriches image feature information, improves segmentation accuracy and precision, and significantly enhances the model's anti-artifact performance and segmentation capabilities, providing reliable technical support for precise head and neck radiotherapy. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0023] Figure 1 A flowchart of a method for oral cavity region segmentation based on head and neck CT images provided in the first embodiment of the present invention is shown;

[0024] Figure 2 A model diagram of the multi-scale convolutional attention block U-Net model in the first embodiment of the present invention is shown;

[0025] Figure 3 This diagram illustrates the structure of three sub-models in the multi-scale convolutional attention block U-Net model in the first embodiment of the present invention.

[0026] Figure 4 This diagram illustrates the structure of the attention branch in the multi-scale convolutional attention block U-Net model according to the first embodiment of the present invention.

[0027] Figure 5 A flowchart of the multi-scale convolutional attention block U-Net model in the first embodiment of the present invention is shown;

[0028] Figure 6 A structural block diagram of an oral cavity region segmentation system based on head and neck CT images provided in another embodiment of the present invention is shown. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0031] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0032] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0033] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0034] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by those skilled in the art to which this invention pertains.

[0035] Please see Figure 1-5 The first embodiment of the present invention provides a method for oral cavity region segmentation based on head and neck CT images, comprising:

[0036] Acquire raw head and neck radiotherapy CT images;

[0037] The CT images are preprocessed to obtain preprocessed images;

[0038] Construct a multi-scale convolutional attention block U-Net model, train and tune the multi-scale convolutional attention block U-Net model to obtain the trained optimal multi-scale convolutional attention block U-Net model;

[0039] The oral cavity segmentation model was obtained by validating the model performance on the validation set using the trained optimal multi-scale convolutional attention block U-Net model.

[0040] The preprocessed image is input into the oral cavity segmentation model to obtain the oral cavity region segmentation result.

[0041] Specifically, the raw data containing head and neck radiotherapy CT images with the oral cavity outlined were divided into a training set (52 cases), a validation set (8 cases), and a test set (12 cases). Data augmentation was performed on the training set to prevent overfitting and improve model stability. Data augmentation methods included rotation, flipping, scaling, and cropping techniques.

[0042] Image preprocessing was performed on the divided dataset, primarily including HU value transformation, window width and level adjustment, adaptive histogram equalization, and image normalization. HU value transformation was performed automatically by the Simple ITK software package. Window width and level were used to adjust the range of gray levels in the CT images, making the contrast between different tissues more pronounced and highlighting structures and tissues of interest. Adaptive histogram equalization adjusted local features of the image based on gray levels, effectively enhancing image contrast and detail, and further emphasizing the oral cavity. Image normalization normalized pixel values, making the model more stable during training. The image range was normalized to [0,1].

[0043] The process of Hu value conversion:

[0044] 1. Read the raw head and neck radiotherapy CT image data.

[0045] 2. Use sitk.GetArrayFromImage to convert the raw data to Hounsfield units (HU) and simultaneously to Numpy format.

[0046] Window width and window position adjustment process:

[0047] 1. Select the window width and window center parameters.

[0048] 2. Adjust the display range of the image according to the selected window width and window level.

[0049] Window width and window level adjustments are achieved using the following formula:

[0050]

[0051] In the formula, window_width represents the window width, window_center represents the window position, and pixel represents the number of pixels.

[0052] Adaptive histogram equalization process:

[0053] 1. Input the image obtained after HU value conversion and window width and level adjustment for format determination;

[0054] 2. The cv2.createCLAHE algorithm is used to perform histogram equalization on the image;

[0055] 3. Input the image of the equalization result.

[0056] The original image is processed using the above preprocessing method to further enhance the contrast between the region of interest and the surrounding tissue, thereby reducing noise in the image and highlighting the features of the oral cavity region.

[0057] The model was constructed using TensorFlow and Keras. The resulting multi-scale convolutional attention block U-Net model consists of 9 convolutional layers, 9 normalization layers, and 6 attention branch modules. The convolutional layers employ different kernel sizes (e.g., 1×1, 3×3, and 5×5 fully convolutions) and 3×3 dilated convolutions for feature extraction at different scales. The residual connection part comprises a 1×1 fully convolutional layer and a normalization layer, connected to the original input. After processing, the processed input is added to subsequent feature maps to preserve key information. The model performs three different convolutions on the original input, dividing each convolution into three feature extraction channels, named the first, second, and third feature extraction sub-models. The first and second sub-models extract basic oral cavity features, resulting in the first and second feature maps respectively. The third sub-model extracts detailed oral cavity features, resulting in the third feature map. The first, second, and third feature maps are then combined to obtain the relevant features of the entire oral cavity structure.

[0058] The first feature extraction sub-model comprises two convolutional layers to extract basic oral cavity features. The convolutional kernels are 3×3. After each convolutional layer, a batch normalization (BN) layer and a ReLU activation layer are added. An attention branch module is added after each convolutional layer. This attention branch module acquires key features related to the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the first feature map. Through parallel convolution operations, the model can extract multi-scale features of oral cavity structures such as teeth and gums, including tooth edges, oral cavity location, and the microstructure of periodontal tissues. These multi-scale features are then input into the subsequent attention branch module for further processing. The attention branch module receives multi-scale features from the parallel convolutional branch as input.

[0059] The second feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. The convolutional kernel of each layer is 5×5. Each convolutional layer is followed by a normalization layer and a ReLU activation layer. An attention branch module is added after each convolutional layer. The attention branch module obtains the key features related to the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the second feature map.

[0060] The third feature extraction sub-model includes two dilated convolutional layers to extract oral cavity detail features. The convolutional kernel of the convolutional layer is 3×3. Each dilated convolutional layer is followed by a normalization layer and a ReLU activation layer. An attention branch module is added after each dilated convolutional layer. The attention branch module obtains the relevant key features of the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the third feature map.

[0061] The main part of the U-Net model is the multi-scale convolutional attention block, which has three core ideas:

[0062] (1) A normalization layer is added after each convolution, and residual connections are introduced at each resolution level for feature extraction. This effectively solves the gradient explosion problem that occurs during training in the traditional U-Net model due to simple convolutional feature extraction. Furthermore, as the model depth increases, simple convolution stacking may also cause gradient vanishing or exploding during training. Residual connections can effectively solve or mitigate the information loss and gradient vanishing or exploding problems in deep learning models during training.

[0063] The specific process of a convolutional neural network with normalization layers and residual connections:

[0064] Step 1: Input layer: Passes the raw input data to the network.

[0065] Step 2: Convolutional Layer: Feature extraction is performed using the Conv2d package of Tensflow and Keras, with a convolutional kernel size of 3×3.

[0066] Step 3: Add a normalization (BN) layer after each convolutional layer. The BN layer code is implemented using keras.layers.BatchNormalization, and its calculation formula is as follows:

[0067] Calculate the mean:

[0068] Calculate the variance:

[0069] standardization:

[0070] Scale and displacement:

[0071] Where μ is the sample mean, σ 2 Let γ be the sample variance, and β be the hyperparameters. xi This represents the i-th sample value in the input data, where i is an integer greater than or equal to 1. yi This represents the i-th sample value after scaling and displacement processing, m represents the number of samples in the current batch, and ε represents a very small positive number used to prevent division by zero during the standardization process.

[0072] Step 4: Residual Connections: Residual connections are introduced at each resolution level for feature extraction. Specifically, after convolution, the original input and the convolution result are summed. This step ensures the network can learn residuals, thereby better preserving and propagating information. The calculation formula is:

[0073] z = F(x) input )+x input (6)

[0074] In the formula, x input Let z represent the input of the residual block, z represent the output of the residual block, and F(x) represent the mapping function of the residual block, i.e., convolution and other operations.

[0075] Step 5: Activation Function: Activate each feature map using the ReLU loss function. The specific formula is as follows:

[0076]

[0077] (2) Multi-scale convolutional layers, combining regular convolution and dilated convolution, are used for feature extraction. Multi-scale convolutional layers can enrich image features by adjusting the kernel size without changing the size of the original feature map, and can encode and decode the features of interest in the image from a global perspective, thereby improving segmentation performance. In the traditional U-Net model, downsampling or stride convolution is usually used to increase the receptive field, which leads to a decrease in spatial resolution. Dilated convolution can expand the receptive field without losing resolution, thereby improving the accuracy and detail of segmentation. In addition, dilated convolution can effectively capture multi-scale contextual information, making the model pay more attention to global features during training. However, the calculation method of dilated convolution may lead to the loss of local information. Therefore, dilated convolution is used on one path of the multi-scale convolutional layer, while the remaining paths are trained using regular convolution to supplement the local information lost during training.

[0078] Specifically, the model performs three different convolutions on the original input data: a 3×3 regular convolution, a 5×5 regular convolution, and a 3×3 dilated convolution. These three convolutions are then divided into three feature extraction pathways. A first, second, and third feature extraction sub-model are used to extract features at different scales. The first and second pathways extract basic oral cavity features, including a batch normalization (BN) layer and two attention mechanisms. The third pathway, the dilated convolution pathway, effectively captures multi-scale contextual information, allowing the model to focus more on detailed oral cavity features during training, thereby improving the model's segmentation ability. Finally, the extraction results from the three pathways are combined by summing them to summarize the relevant features of the entire oral cavity structure.

[0079] The specific steps of the dilated convolution algorithm are as follows:

[0080] 1. Input:

[0081] Feature map size: H×W×C, where H is the height of the feature map, W is the width of the feature map, and C is the number of channels of the feature map.

[0082] Kernel size: F×F×C×K, where F is the size of the convolution kernel (usually square, hence F×F), C is the same as the number of channels in the input feature map, and K is the number of convolution kernels, which determines the number of channels in the output feature map.

[0083] Stride: s, where s is the stride of the convolution operation, which determines the step size by which the convolution kernel moves on the feature map.

[0084] Dilation Rate: d. The dilation rate represents the number of zeros inserted between convolution kernels. The dilation rate is used to increase the receptive field of the convolution kernel.

[0085] 2. Padding: Zero padding is applied to the input feature map I so that the size of the feature map after dilated convolution is the same as the size of the input feature map.

[0086] 3. Calculate the output size: The formula for calculating the size of the output feature map is:

[0087]

[0088]

[0089] Among them, H in and W in These are the height and width of the input feature map, H. out and W out These are the height and width of the output feature map, respectively, and padding is the amount of zero padding.

[0090] 4. The formula for calculating dilated convolution is:

[0091]

[0092] Among them, O i,j,k These are the values ​​of the output feature map, where m, n, j, and l are all integers greater than or equal to 1.

[0093] The core design feature of this model lies in its modular structure, which incorporates the architecture of U-Net. The overall structure of the model starts with the input image, processes it through multiple modules, and finally outputs a segmentation result of the oral cavity.

[0094] (3) A cSE attention mechanism (channel attention mechanism) was introduced, and the specific steps are as follows:

[0095] 1. Add a cSE attention mechanism after each convolution + BN layer to obtain key features related to oral structure at each step.

[0096] 2. After two cSE attention mechanisms, a residual connection is used to add and merge the original input data and the key feature maps obtained after the attention mechanism processing to obtain a feature map, so as to ensure the stability of the model.

[0097] The attention branch module receives multi-scale features from the parallel convolutional branches as input. The specific methods for obtaining key features related to the oral structure at each step include: spatially compressing the input features using a global average pooling layer to extract a global feature representation; extracting key features of the oral structure through a linear layer and a ReLU activation function; generating a weight vector through another linear layer and a Sigmoid activation function; and weighting the input features using the weight vector to generate an enhanced feature map. The attention branch module can selectively focus on features, allowing the model to pay more attention to primary channel features during training, while suppressing secondary channel features through compression, activation, and scaling. This helps to highlight key oral features and suppress irrelevant information, thereby improving segmentation accuracy.

[0098] An attention branch module is added after each downsampling convolution operation to obtain weights for different feature channels. This attention branch module is used to extract key feature channels for oral segmentation during training. Finally, to prevent overfitting and reduce the number of model parameters, a dropout layer with a parameter of 0.1 is added to each multi-scale attention branch module. We also replace max pooling with stride convolution, because in the traditional U-Net architecture, max pooling is mainly used to compress feature maps and resolution. While this effectively reduces parameters and resolution, it also leads to a loss in feature extraction.

[0099] The pre-constructed multi-scale convolutional attention block U-Net model was trained and its parameters tuned using the training set to obtain the optimal multi-scale convolutional attention block U-Net model. The model performance was then validated using the validation set, and its performance was evaluated using metrics such as the Dessie similarity coefficient, resulting in the final oral cavity segmentation model. Finally, the model was tested using the test set to obtain the model's prediction results and the resulting oral cavity segmentation, which clearly labels the teeth and oral cavity regions.

[0100] This invention provides a method for oral cavity region segmentation based on head and neck CT images. It introduces a novel attention and multi-scale fusion model, constructing a multi-scale convolutional attention block U-Net model. Combining the U-Net structure and multi-scale convolutional attention mechanism, this method can segment both artifact-containing and artifact-free oral cavities, effectively solving the gradient vanishing and exploding problems present in traditional U-Net models, thus improving the model's convergence speed and training stability. Simultaneously, the use of multi-scale convolutional layers and dilated convolutions enriches image feature information, improving segmentation accuracy and precision, significantly enhancing the model's anti-artifact performance and segmentation capabilities, and providing reliable technical support for precise head and neck radiotherapy.

[0101] In the first embodiment described above, a method for oral cavity region segmentation based on head and neck CT images is provided. Correspondingly, this application also provides an oral cavity region segmentation system based on head and neck CT images. Please refer to... Figure 6 This is a structural block diagram of an oral cavity region segmentation system based on head and neck CT images provided in the second embodiment of the present invention. Since the device embodiment is basically similar to the method embodiment, it is described simply; relevant details can be found in the description of the method embodiment. The device embodiment described below is merely illustrative.

[0102] like Figure 6 As shown, another embodiment of the present invention provides an oral cavity region segmentation system based on head and neck CT images, comprising: an image acquisition module, a preprocessing module, a model building module, a model validation module, and an image segmentation module; the image acquisition module is used to acquire original head and neck radiotherapy CT images; the preprocessing module is used to preprocess the CT images to obtain preprocessed images; the model building module is used to construct a multi-scale convolutional attention block U-Net model, train and tune the multi-scale convolutional attention block U-Net model to obtain a trained optimal multi-scale convolutional attention block U-Net model; the model validation module is used to use the trained optimal multi-scale convolutional attention block U-Net model to perform model performance validation on a validation set to obtain an oral cavity segmentation model; the image segmentation module inputs the preprocessed images into the oral cavity segmentation model to obtain oral cavity region segmentation results.

[0103] The preprocessing module performs preprocessing on CT images, including: HU value conversion, window width and window level adjustment, adaptive histogram equalization, and image normalization.

[0104] The multi-scale convolutional attention block U-Net model includes a first feature extraction sub-model, a second feature extraction sub-model, and a third feature extraction sub-model. The first and second feature extraction sub-models are used to extract basic features of the oral cavity to obtain a first feature map and a second feature map, respectively. The third feature extraction sub-model is used to extract detailed features of the oral cavity to obtain a third feature map. The first feature map, the second feature map, and the third feature map are added and merged to obtain the relevant features of the entire oral cavity structure.

[0105] The first feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. The convolutional kernel of each layer is 3×3. Each convolutional layer is followed by a normalization layer and a ReLU activation layer. An attention branch module is added after each convolutional layer. The attention branch module obtains the relevant key features of the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the first feature map.

[0106] The second feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. The convolutional kernel of each layer is 5×5. Each convolutional layer is followed by a normalization layer and a ReLU activation layer. An attention branch module is added after each convolutional layer. The attention branch module obtains the key features related to the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the second feature map.

[0107] The third feature extraction sub-model includes two dilated convolutional layers to extract oral cavity detail features. The convolutional kernel of the convolutional layer is 3×3. Each dilated convolutional layer is followed by a normalization layer and a ReLU activation layer. An attention branch module is added after each dilated convolutional layer. The attention branch module obtains the relevant key features of the oral cavity structure at each step. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the third feature map.

[0108] The specific methods used by the attention branch module to obtain key features relevant to the oral structure at each step include:

[0109] The attention branch module uses a global average pooling layer to spatially compress the input features and extract global feature representations;

[0110] Key features of the oral cavity structure were extracted using a linear layer and the ReLU activation function.

[0111] A weight vector is generated using another linear layer and a Sigmoid activation function;

[0112] The input features are weighted using a weight vector to generate an enhanced feature map.

[0113] This invention provides an oral cavity region segmentation system based on head and neck CT images. It introduces a novel attention and multi-scale fusion model, constructing a multi-scale convolutional attention block U-Net model. Combining the U-Net structure and multi-scale convolutional attention mechanism, it can segment both artifact-containing and artifact-free oral cavities, effectively solving the gradient vanishing and exploding problems present in traditional U-Net models, thus improving the model's convergence speed and training stability. Simultaneously, the use of multi-scale convolutional layers and dilated convolutions enriches image feature information, improving segmentation accuracy and precision, significantly enhancing the model's anti-artifact performance and segmentation capabilities, providing reliable technical support for precise head and neck radiotherapy.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for oral cavity region segmentation based on head and neck CT images, characterized in that, include: Acquire raw head and neck radiotherapy CT images; The CT images are preprocessed to obtain preprocessed images; A multi-scale convolutional attention block U-Net model was constructed using Tensorflow and Keras. The multi-scale convolutional attention block U-Net model was trained and its parameters were tuned to obtain the trained optimal multi-scale convolutional attention block U-Net model. The oral cavity segmentation model was obtained by validating the model performance on the validation set using the trained optimal multi-scale convolutional attention block U-Net model. The preprocessed image is input into the oral cavity segmentation model to obtain the oral cavity region segmentation result; The multi-scale convolutional attention block U-Net model includes 9 convolutional layers, 9 normalization layers, and 6 attention branch modules. The convolutional layers use convolutional kernels of different sizes and dilated convolutions for feature extraction at different scales. The attention branch module acquires key features related to the oral structure at each step, specifically: The attention branch module uses a global average pooling layer to spatially compress the input features and extract global feature representations; Key features of the oral cavity structure were extracted using a linear layer and the ReLU activation function. A weight vector is generated using another linear layer and a Sigmoid activation function; The input features are weighted using a weight vector to generate an enhanced feature map; The preprocessing of the CT images includes: HU value conversion: Use sitk.GetArrayFromImage to convert the original head and neck radiotherapy CT image to Hounsfield units, and simultaneously to NumPy format; Window width and window position adjustment: using formulas Implementation; In the formula, window_width represents the window width, window_center represents the window position, and pixel represents the number of pixels; Adaptive histogram equalization and image normalization processing; The multi-scale convolutional attention block U-Net model exhibits anti-artifact performance, including: The first feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. After each convolutional layer, a normalization layer and a ReLU activation layer are added. After each convolutional layer, an attention branch module is added. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the first feature map. The second feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. After each convolutional layer, a normalization layer and a ReLU activation layer are added. After each convolutional layer, an attention branch module is added. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the second feature map. The third feature extraction sub-model uses two dilated convolutional layers to extract oral cavity detail features. After each dilated convolutional layer, a normalization layer and a ReLU activation layer are added sequentially. An attention branch module is added after each dilated convolutional layer. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the third feature map.

2. The method as described in claim 1, characterized in that, The residual connection consists of a fully convolutional layer with a 1×1 kernel and a normalized layer.

3. A system for segmenting the oral cavity region based on head and neck CT images, characterized in that, include: Image acquisition module, preprocessing module, model building module, model validation module, and image segmentation module; The image acquisition module is used to acquire raw head and neck radiotherapy CT images; The preprocessing module is used to preprocess CT images to obtain preprocessed images; The model building module is used to build a multi-scale convolutional attention block U-Net model using Tensorflow and Keras, train and tune the multi-scale convolutional attention block U-Net model, and obtain the trained optimal multi-scale convolutional attention block U-Net model. The model validation module is used to validate the model performance on the validation set using the trained optimal multi-scale convolutional attention block U-Net model to obtain the oral cavity segmentation model. The image segmentation module inputs the preprocessed image into the oral cavity segmentation model to obtain the oral cavity region segmentation result; The multi-scale convolutional attention block U-Net model includes convolutional layers, normalization layers, and attention branch modules. The convolutional layers use convolutional kernels of different sizes and dilated convolutions for feature extraction at different scales. The attention branch module acquires key features related to the oral structure at each step, specifically: The attention branch module uses a global average pooling layer to spatially compress the input features and extract global feature representations; Key features of the oral cavity structure were extracted using a linear layer and the ReLU activation function. A weight vector is generated using another linear layer and a Sigmoid activation function; The input features are weighted using a weight vector to generate an enhanced feature map; The preprocessing of the CT images includes: HU value conversion: Use sitk.GetArrayFromImage to convert the original head and neck radiotherapy CT image to Hounsfield units, and simultaneously to NumPy format; Window width and window position adjustment: using formulas Implementation; In the formula, window_width represents the window width, window_center represents the window position, and pixel represents the number of pixels; Adaptive histogram equalization and image normalization processing; The multi-scale convolutional attention block U-Net model exhibits anti-artifact performance, including: The first feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. After each convolutional layer, a normalization layer and a ReLU activation layer are added. After each convolutional layer, an attention branch module is added. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained through the attention mechanism to obtain the first feature map. The second feature extraction sub-model includes two convolutional layers to extract basic oral cavity features. After each convolutional layer, a normalization layer and a ReLU activation layer are added. After each convolutional layer, an attention branch module is added. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained through the attention mechanism to obtain the second feature map. The third feature extraction sub-model uses two dilated convolutional layers to extract oral cavity detail features. After each dilated convolutional layer, a normalization layer and a ReLU activation layer are added sequentially. An attention branch module is added after each dilated convolutional layer. After the two attention mechanisms, a residual connection is used to add the original input and the key features obtained after the attention mechanism to obtain the third feature map.

Images