Oral squamous cell carcinoma two-dimensional image segmentation and benign and malignant differentiation method and system

Abstract

This invention discloses a method and system for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma, belonging to the field of two-dimensional image processing technology for oral squamous cell carcinoma. It aims to improve the accuracy and stability of early screening for oral squamous cell carcinoma. The invention acquires oral image data, labels the acquired images according to pre-set classification standards to obtain training and testing sets; preprocesses the oral image data to obtain preprocessed training and testing sets; performs data augmentation on the oral image data in the training set to obtain a data-augmented training set; constructs an image feature extraction model based on a convolutional neural network; and trains the convolutional neural network-based image feature extraction model using supervised learning to obtain a trained convolutional neural network-based image feature extraction model for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma.

Description

Technical Field

[0001] This invention belongs to the field of two-dimensional image processing technology for oral squamous cell carcinoma, specifically relating to a method and system for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma. Background Technology

[0002] Oral squamous cell carcinoma is a common malignant tumor of the head and neck, clinically characterized by ulcerative, exophytic, or infiltrative growth. Microscopically, abnormally proliferating squamous epithelium can be seen breaking through the basement membrane and infiltrating deeper layers. Existing oral cavity image processing methods are mostly based on color features or superficial semantic information for image segmentation; however, in actual images, factors such as saliva reflection, imaging angle, and changes in lighting can make early lesion features unclear, leading to missed clinical diagnoses. Summary of the Invention

[0003] The problem this invention aims to solve is to improve the accuracy and stability of early screening for oral squamous cell carcinoma, and proposes a two-dimensional image segmentation method and system for identifying benign and malignant oral squamous cell carcinoma.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma includes the following steps:

[0006] S1. Acquire oral cavity image data, label the acquired oral cavity images according to pre-set classification standards, and obtain training set and test set;

[0007] S2. Preprocess the oral cavity image data in the training set and test set obtained in step S1 to obtain the preprocessed training set and test set;

[0008] S3. Perform data augmentation on the oral cavity image data in the training set to obtain a data-augmented training set;

[0009] S4. Construct an image feature extraction model based on a convolutional neural network, which includes an input image layer, a convolutional layer, an MBConV module layer, a global average pooling layer, a fully connected layer, and a classification output layer connected in sequence.

[0010] S5. Using supervised learning, the image feature extraction model based on convolutional neural networks obtained in step S4 is trained using the data-enhanced training set obtained in step S3, resulting in a trained image feature extraction model based on convolutional neural networks.

[0011] S6. Use the test set obtained in step S2 to test the trained image feature extraction model based on convolutional neural network, and perform two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma.

[0012] Furthermore, the classification criteria in step S1 include category 1 corresponding to obvious malignant lesion characteristics, category 2 corresponding to suspicious or potential malignant lesions, and category 3 corresponding to benign or normal tissues, forming category labels. The image path is then paired with the corresponding category label to form a data index list.

[0013] Furthermore, the specific implementation method of step S2 includes the following steps:

[0014] S2.1. Convert the RGB image of the oral cavity image data to the LAB color space, and separate the luminance channel L and the color channels A and B;

[0015] S2.2. An adaptive histogram equalization algorithm is used to enhance the luminance channel L, where the contrast limit parameter is set to 2.0 and the grid size is set to 8×8, resulting in the enhanced luminance channel;

[0016] S2.3. Fuse the enhanced luminance channel with the color channels A and B obtained in step S2.1, and convert it into an RGB image;

[0017] S2.4. The RGB images obtained in step S2.3 are scaled to a fixed resolution and normalized to obtain the preprocessed training and test sets.

[0018] Furthermore, the data augmentation process in step S3 includes randomly flipping the input image horizontally and randomly flipping it vertically.

[0019] Furthermore, the operation process of the convolutional layer in step S4 is as follows:

[0020]

[0021] in, For the input feature map at location First Pixel values ​​of each output channel For input channels, These represent the horizontal and vertical spatial positions of the convolution kernel, respectively. This represents the total number of input channels. This represents the total horizontal space of the convolution kernel. The total number of vertical spaces in the convolution kernel; For the input feature map at location First The pixel values ​​of each input channel. For convolution kernel weights, For the first The bias term corresponding to each output channel;

[0022] After the convolutional layer, a non-linear activation function, Swish, is introduced to enhance the model's expressive power.

[0023] Furthermore, in step S5, the cross-entropy loss function is used during the training process, and the Adam optimization algorithm is used to update the model parameters.

[0024] Furthermore, in step S6, the images to be tested in the test set are input into the image feature extraction model based on the convolutional neural network for forward calculation. The probability value corresponding to each category is obtained through the Softmax function, and the category with the highest probability is selected as the final classification result. The classification results include three types: malignant risk, suspicious risk, and benign risk.

[0025] Then, the probability value corresponding to the final classification result is used as the confidence level to obtain the prediction confidence level. ,in To predict confidence levels, For predicting categories, This represents the probability of the corresponding category.

[0026] A system for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed, it implements the steps of the two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma.

[0027] The beneficial effects of this invention are:

[0028] The present invention provides a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma, which improves the accuracy and stability of early screening for oral cancer. The present invention introduces an automatic recognition model based on deep learning to perform unified image processing and feature extraction, thereby reducing the influence of subjective human factors and improving the consistency and stability of screening results.

[0029] This invention discloses a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma. It optimizes the preprocessing workflow based on the characteristics of oral images, improving lesion identifiability. Oral images often suffer from uneven illumination, localized reflections, and significant color differences, directly impacting recognition performance. This invention converts the image to the LAB color space and performs adaptive histogram equalization enhancement on the brightness channel, effectively improving local contrast and making lesion boundaries clearer, thereby enhancing the effectiveness of subsequent feature extraction.

[0030] The present invention provides a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma, which ensures effective model training under small sample conditions. To address the problems of high difficulty in acquiring medical image data and high annotation costs, the present invention adopts a transfer learning method based on a pre-trained model, enabling the model to utilize the general features learned from existing large-scale data, and still have good recognition ability under limited sample conditions, thereby improving the model training efficiency and generalization performance.

[0031] This invention discloses a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma, enabling risk grading output and enhancing clinical decision support capabilities. By outputting different risk levels and their corresponding probability values ​​through a model, this invention provides a quantitative basis for diagnostic results, helping physicians to stratify patient management and improve the standardization of clinical decision-making. This invention possesses excellent system integration capabilities and engineering feasibility; through modular design of the recognition process and structured encapsulation of the output results, it can be easily integrated into existing medical information systems or independent software platforms, supporting rapid deployment and application, and lowering the barrier to practical use. Attached Figure Description

[0032] Figure 1 This is a flowchart of a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma according to the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely a part of the embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0034] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0035] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 Detailed explanation is as follows:

[0036] Example 1:

[0037] A method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma includes the following steps:

[0038] S1. Acquire oral cavity image data, label the acquired oral cavity images according to pre-set classification standards, and obtain training set and test set;

[0039] Furthermore, the classification criteria in step S1 include category 1 corresponding to obvious malignant lesion characteristics, category 2 corresponding to suspicious or potential malignant lesions, and category 3 corresponding to benign or normal tissues, forming category labels. The image path is then paired with the corresponding category label to form a data index list.

[0040] Furthermore, the oral cavity image data is a two-dimensional color image of the oral cavity; the image data is loaded by reading the image file path and then uniformly converted into RGB three-channel format;

[0041] S2. Preprocess the oral cavity image data in the training set and test set obtained in step S1 to obtain the preprocessed training set and test set;

[0042] Furthermore, the specific implementation method of step S2 includes the following steps:

[0043] S2.1. Convert the RGB image of the oral cavity image data to the LAB color space, and separate the luminance channel L and the color channels A and B;

[0044] S2.2. An adaptive histogram equalization algorithm is used to enhance the luminance channel L, where the contrast limit parameter is set to 2.0 and the grid size is set to 8×8, resulting in the enhanced luminance channel;

[0045] S2.3. Fuse the enhanced luminance channel with the color channels A and B obtained in step S2.1, and convert it into an RGB image;

[0046] S2.4. The RGB images obtained in step S2.3 are scaled to a fixed resolution and normalized to obtain the preprocessed training and test sets.

[0047] S3. Perform data augmentation on the oral cavity image data in the training set to obtain a data-augmented training set;

[0048] Furthermore, the data augmentation process in step S3 includes randomly flipping the input image horizontally and randomly flipping it vertically.

[0049] S4. Construct an image feature extraction model based on a convolutional neural network, which includes an input image layer, a convolutional layer, an MBConV module layer, a global average pooling layer, a fully connected layer, and a classification output layer connected in sequence.

[0050] Furthermore, the operation process of the convolutional layer in step S4 is as follows:

[0051]

[0052] in, For the input feature map at location First Pixel values ​​of each output channel For input channels, These represent the horizontal and vertical spatial positions of the convolution kernel, respectively. This represents the total number of input channels. This represents the total horizontal space of the convolution kernel. The total number of vertical spaces in the convolution kernel; For the input feature map at location First The pixel values ​​of each input channel. For convolution kernel weights, For the first The bias term corresponding to each output channel;

[0053] After the convolutional layer, a non-linear activation function, Swish, is introduced to enhance the model's expressive power.

[0054] Furthermore, the expression for the nonlinear activation function Swish is:

[0055]

[0056] in, It is the input value. It is the output of the activation function. It is the Sigmoid function;

[0057] ;

[0058] Furthermore, global average pooling is used at the end of the feature extraction stage to reduce the dimensionality of the feature map. Its calculation method is as follows:

[0059]

[0060] in, For the first Global eigenvalues ​​of each channel For the input feature map, The height and width of the feature map are respectively;

[0061] Fully connected layers are used to map the extracted features to the classification space. The calculation formula is as follows:

[0062]

[0063] in, It is the first Output values ​​for each category, It is the first of the input feature vectors One portion, These are weight parameters. It is a bias term. It is the dimension of the feature vector;

[0064] Then, the Softmax classification function normalizes the output of the fully connected layer to obtain the probabilities of each class:

[0065]

[0066] in, It belongs to the first Predicted probability of class It is the output of the fully connected layer. It is the total number of categories. It is the summation index variable;

[0067] Furthermore, transfer learning is implemented during the model initialization phase. Pre-trained model parameters based on the ImageNet dataset are loaded, and the last layer of the network is adjusted according to the actual classification task to ensure the number of output nodes matches the number of classification categories. This approach can improve the model's convergence speed and recognition accuracy even with a limited number of samples.

[0068] S5. Using supervised learning, the image feature extraction model based on convolutional neural networks obtained in step S4 is trained using the data-enhanced training set obtained in step S3, resulting in a trained image feature extraction model based on convolutional neural networks.

[0069] Furthermore, image paths are paired with corresponding labels to form a data index list; data is read in batches during training. Perform preprocessing and data augmentation operations on each batch of data; convert images into tensor form and input them into the model.

[0070] Furthermore, in step S5, the cross-entropy loss function is used during the training process, and the Adam optimization algorithm is used to update the model parameters.

[0071] The expression for the cross-entropy loss function is:

[0072]

[0073] in, It is the value of the loss function. It is the unique hot encoding of the real label;

[0074] Furthermore, in each training round, forward propagation, loss calculation, backpropagation, and parameter update operations are performed sequentially; the model performance is evaluated, and the optimal model parameters are saved based on the evaluation results.

[0075] S6. Use the test set obtained in step S2 to test the trained image feature extraction model based on convolutional neural network, and perform two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma.

[0076] Furthermore, in step S6, the images to be tested in the test set are input into the image feature extraction model based on the convolutional neural network for forward calculation. The probability value corresponding to each category is obtained through the Softmax function, and the category with the highest probability is selected as the final classification result. The classification results include three types: malignant risk, suspicious risk, and benign risk.

[0077] Furthermore, the image to be tested is input into the model for forward computation. The probability values ​​corresponding to each category are obtained through the Softmax function, and the category with the highest probability is selected as the final classification result. The classification results include three types: high risk, medium risk, and low risk. The classification determination rules are as follows:

[0078]

[0079] in, The final predicted category has the following correspondence: Low risk (benign); Medium risk (suspicious); High risk (malignant);

[0080] Then, the probability value corresponding to the final classification result is used as the confidence level to obtain the prediction confidence level. ,in To predict confidence levels, For predicting categories, This represents the probability of the corresponding category.

[0081] Example 2:

[0082] A system for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed, it implements the steps of the two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma as described in Example 1.

[0083] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0084] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma, characterized in that, Includes the following steps: S1. Acquire oral cavity image data, label the acquired oral cavity images according to pre-set classification standards, and obtain training set and test set; S2. Preprocess the oral cavity image data in the training set and test set obtained in step S1 to obtain the preprocessed training set and test set; S3. Perform data augmentation on the oral cavity image data in the training set to obtain a data-augmented training set; S4. Construct an image feature extraction model based on a convolutional neural network, which includes an input image layer, a convolutional layer, an MBConV module layer, a global average pooling layer, a fully connected layer, and a classification output layer connected in sequence. S5. Using supervised learning, the image feature extraction model based on convolutional neural networks obtained in step S4 is trained using the data-enhanced training set obtained in step S3, resulting in a trained image feature extraction model based on convolutional neural networks. S6. Use the test set obtained in step S2 to test the trained image feature extraction model based on convolutional neural network, and perform two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma.

2. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 1, characterized in that, The classification criteria in step S1 include category 1 corresponding to obvious malignant lesion characteristics, category 2 corresponding to suspicious or potential malignant lesions, and category 3 corresponding to benign or normal tissues, forming category labels. The image path is then paired with the corresponding category label to form a data index list.

3. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 2, characterized in that, The specific implementation method of step S2 includes the following steps: S2.

1. Convert the RGB image of the oral cavity image data to the LAB color space, and separate the luminance channel L and the color channels A and B; S2.

2. An adaptive histogram equalization algorithm is used to enhance the luminance channel L, where the contrast limit parameter is set to 2.0 and the grid size is set to 8×8, resulting in the enhanced luminance channel; S2.

3. Fuse the enhanced luminance channel with the color channels A and B obtained in step S2.1, and convert it into an RGB image; S2.

4. The RGB images obtained in step S2.3 are scaled to a fixed resolution and normalized to obtain the preprocessed training and test sets.

4. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 3, characterized in that, The data augmentation process in step S3 includes randomly flipping the input image horizontally and randomly flipping it vertically.

5. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 4, characterized in that, The operation process of the convolutional layer in step S4 is as follows: in, For the input feature map at location First Pixel values ​​of each output channel For input channels, These represent the horizontal and vertical spatial positions of the convolution kernel, respectively. This represents the total number of input channels. This represents the total horizontal space of the convolution kernel. The total number of vertical spaces in the convolution kernel; For the input feature map at location First The pixel values ​​of each input channel. For convolution kernel weights, For the first The bias term corresponding to each output channel; After the convolutional layer, a non-linear activation function, Swish, is introduced to enhance the model's expressive power.

6. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 5, characterized in that, In step S5, the cross-entropy loss function is used during the training process, and the Adam optimization algorithm is used to update the model parameters.

7. The method for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma according to claim 6, characterized in that, Step S6 inputs the images to be tested in the test set into the image feature extraction model based on convolutional neural network for forward calculation, obtains the probability value corresponding to each category through the Softmax function, and selects the category with the highest probability as the final classification result; The classification results include three types: malignant risk, suspected risk, and benign risk; Then, the probability value corresponding to the final classification result is used as the confidence level to obtain the prediction confidence level. ,in To predict confidence levels, For predicting categories, This represents the probability of the corresponding category.

8. A system for two-dimensional image segmentation and benign / malignant differentiation of oral squamous cell carcinoma, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed, it implements the steps of a two-dimensional image segmentation and benign / malignant differentiation method for oral squamous cell carcinoma as described in any one of claims 1-7.

Images