CT automatic lung cancer diagnosis and classification system and method based on deep learning

Abstract

The invention relates to the technical field of medical images, and discloses a CT automatic lung cancer diagnosis and classification system and method based on deep learning. According to the system and the method, a lung CT scanning image is acquired from medical imaging equipment through a data acquisition module, and image denoising, translation and cutting operation is performed on the acquired image through a data preprocessing module; the feature extraction module extracts lung contour features, lung boundary features, contrast features, gray features and focus structure features based on the processed image through a convolutional neural network, and the classification model module inputs the extracted features into a full-connection neural network for classification and judges whether the image is a normal lung or has lung cancer. The cancer diagnosis module is used for further classifying the specific types and stages of cancers and generating classified diagnosis results, the output and feedback module is used for displaying the classified diagnosis results in a report form and providing the classified diagnosis results to doctors for further analysis, and the model is continuously iterated and optimized by collecting feedback of the doctors and new case data.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a CT automatic lung cancer diagnosis and classification system and method based on deep learning. Background Art

[0002] Lung cancer is one of the leading causes of cancer-related deaths worldwide, and its impact involves many aspects such as personal health, family well-being and public health. Lung cancer has a direct impact on the patient's body, such as cough, difficulty breathing, chest pain and other symptoms. As the disease progresses, further complications may occur, such as lung infection, airway obstruction and metastasis to other organs. Early detection and timely intervention of lung cancer can significantly improve the survival rate of patients, but most patients are already in the advanced stage when diagnosed, resulting in a low overall survival rate. Through classification and typing, we can better understand the biological characteristics of lung cancer and promote the development of new therapies, such as targeted therapy and immunotherapy.

[0003] Currently, when CT is automatically classified for lung cancer diagnosis, due to subjective biases in doctors' diagnostic experience and techniques, benign lesions or other lesions may be misdiagnosed as malignant tumors when diagnosing patients' lung tumor images, reducing the accuracy of medical diagnosis. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the deficiencies of the prior art, the present invention provides a CT automatic lung cancer diagnosis and classification system and method based on deep learning, which has the following features: a data acquisition module acquires lung CT scan images from a medical imaging device; a data preprocessing module performs image denoising, translation and cropping operations on the acquired images, and marks the tumors in the lung CT images to form labels of a data set; a feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features and lesion structure features based on the processed images through a convolutional neural network; a classification model module inputs the extracted features into a fully connected neural network for classification, determines whether the image is a normal lung or has lung cancer, further classifies the specific type and stage of the cancer and generates a classification diagnosis result; an output and feedback module displays the classification diagnosis result in the form of a report, provides it to a doctor for further analysis, and continuously iterates and optimizes the model by collecting feedback from the doctor and new case data; and by establishing a model, extracts features from the patient's lung images, and generates a detailed diagnosis report through detailed classification, thereby avoiding misjudgment of malignant tumors and improving the accuracy of lung medical diagnosis, thereby solving the above problems.

[0006] (II) Technical solution

[0007] To achieve the above object, the present invention provides the following technical solutions: a CT automatic lung cancer diagnosis and classification system based on deep learning, comprising a data acquisition module, a data preprocessing module, a feature extraction module, a classification model module, and an output and feedback module;

[0008] The data acquisition module obtains lung CT scan images, including normal and different stages of lung cancer images, and obtains the patient's age, gender, smoking history and clinical background information from the medical imaging device, and transmits the image data to the data preprocessing module;

[0009] The data preprocessing module performs image denoising, translation and cropping operations on the collected images, and marks the tumors in the lung CT images to form labels for the data set;

[0010] The feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features, and lesion structure features based on the processed image through a convolutional neural network, and uniformly sends the extracted features to the classification module;

[0011] The classification model module inputs the extracted features into a fully connected neural network for classification, determines whether the image is a normal lung or lung cancer, further classifies the specific type and stage of the cancer, and generates a classification diagnosis result and sends it to the output module;

[0012] The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

[0013] Preferably, the data preprocessing module performs image denoising on the collected image, and the image denoising formula is as follows:

[0014] D(x,y)=med{f(x+i,y+j)|-k<=i,j<=k}

[0015] In the formula, D(x,y) represents the value of the pixel in the denoised image, f(x,y) represents the value of the pixel in the original image, k represents the radius of the filter window, med represents the median calculation of the pixel values ​​in the window, and i,j represent the offset of the filter window.

[0016] Preferably, the data preprocessing module performs image translation on the collected image, and the image translation formula is as follows:

[0017]

[0018] In the formula, (x, y) represents the coordinates of the midpoint of the original image, (x′, y′) represents the coordinates of the midpoint of the translated image, and t x , t yIndicates the amount of translation in the horizontal and vertical directions.

[0019] Preferably, the data preprocessing module performs image cropping on the collected image, and the image cropping formula is as follows:

[0020] I crop =I[x1:x2,y1:y2]

[0021] In the formula, I crop represents the cropped image, I represents the original image, (x1, y1) and (x2, y2) represent the coordinates of the upper left corner and lower right corner of the cropped area.

[0022] Preferably, the feature extraction module extracts lung contour features based on the processed image through a convolutional neural network, and the feature extraction formula is as follows:

[0023] Z l =I l-1 *W l +b l

[0024] A l =ReLU(Z l )

[0025] F=P(A l )

[0026] In the formula, * represents the convolution operation, I l-1 Represents the output of the previous layer, Z l represents the linear output of the convolutional layer, W l represents the weight of the convolutional layer, b l represents the convolution layer bias, A l ReLU represents the activation function, P represents the pooling operation, and F represents the lung contour feature.

[0027] Preferably, the feature extraction module extracts lung boundary features based on the processed image through a convolutional neural network, and the feature extraction formula is as follows:

[0028]

[0029] In the formula, G represents the lung boundary feature, G x represents the horizontal direction of the lung x boundary, G y Indicates the vertical direction of the y boundary of the lung.

[0030] Preferably, the feature extraction module extracts contrast features based on the processed image through a convolutional neural network, and the feature extraction formula is as follows:

[0031]

[0032] In the formula, C represents the contrast feature, N represents the total number of pixels in the image, μ represents the average grayscale value of the image, and I(i) represents the grayscale value of the i-th pixel in the image.

[0033] Preferably, the feature extraction module extracts grayscale features based on the processed image through a convolutional neural network, and the feature extraction formula is as follows:

[0034]

[0035] In the formula, Fm represents the grayscale feature, N represents the total number of pixels in the image, and the grayscale histogram h(q) is used to understand the frequency of each grayscale level q in the image.

[0036] Preferably, the feature extraction module extracts the lesion structure features based on the processed image through a convolutional neural network, and the feature extraction formula is as follows:

[0037] R s =concat(A l ,A l+1 ,A l+2 )

[0038] In the formula, R s Represents the structural features of the lesion, and concat represents the feature map A output by connecting different layers l ,A l+1 ,A l+2 Extract features from .

[0039] A deep learning-based CT automatic lung cancer diagnosis and classification method comprises the following steps:

[0040] S1. Obtain lung CT scan images, including normal and different stages of lung cancer images, from medical imaging equipment through a data acquisition module, as well as clinical background information such as age, gender, and smoking history of the patient, and transmit the image data to a data preprocessing module;

[0041] S2, the data preprocessing module performs image denoising, translation and cropping operations on the collected images, and marks the tumors in the lung CT images to form the labels of the data set;

[0042] S3, the feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features and lesion structure features based on the processed image through a convolutional neural network, and sends the extracted features to the classification module in a unified manner;

[0043] S4, the classification model module inputs the extracted features into the fully connected neural network for classification, determines whether the image is normal lung or lung cancer, further classifies the specific type and stage of cancer and generates a classification diagnosis result and sends it to the output module;

[0044] S5. The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

[0045] Compared with the prior art, the present invention provides a CT automatic lung cancer diagnosis and classification system and method based on deep learning, which has the following beneficial effects:

[0046] The present invention acquires lung CT scan images from medical imaging equipment through a data acquisition module, performs image denoising, translation and cropping operations on the acquired images through a data preprocessing module, and marks tumors in the lung CT images to form labels for data sets, and extracts lung contour features, lung boundary features, contrast features, grayscale features and lesion structure features based on processed images through a convolutional neural network. The classification model module inputs the extracted features into a fully connected neural network for classification to determine whether the image is a normal lung or has lung cancer, further classifies the specific type and stage of cancer and generates a classification diagnosis result, and the output and feedback module displays the classification diagnosis result in the form of a report, which is provided to doctors for further analysis, and continuously iterates and optimizes the model by collecting feedback from doctors and new case data, and extracts features from the patient's lung images through the establishment of a model, and generates a detailed diagnosis report through detailed classification, thereby avoiding misjudgment of malignant tumors and improving the accuracy of lung medical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the system flow of the present invention.

[0048] Figure 2 It is a schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] In view of the current problem that when CT is used for automatic lung cancer diagnosis and classification, due to the subjective bias of doctors' diagnostic experience and technology, benign lesions or other lesions may be misjudged as malignant tumors when diagnosing patients' lung tumor images, which reduces the accuracy of medical diagnosis, a CT automatic lung cancer diagnosis and classification system based on deep learning is proposed. Please refer to Figure 1 ,The system includes a data acquisition module, a data preprocessing module, a feature extraction module, a classification model module, and an output and feedback module;

[0051] The data acquisition module is connected to the CT scanner through an interface and uses the DICOM digital imaging and communication protocol to acquire lung CT images. The DICOM format contains not only image data, but also image metadata, including patient information, scanning parameters, etc. The acquired CT images include normal lung images and lung cancer images at different stages, such as early, locally advanced, and metastatic lung cancer, to ensure the diversity and representativeness of the data set for subsequent analysis and model training. After the image is acquired, the module converts the image data in DICOM format into a JPEG format suitable for subsequent processing for processing by the data preprocessing module;

[0052] The data preprocessing module performs image denoising, translation, and cropping operations on the acquired images, and marks the tumors in the lung CT images to form the labels of the data set, where:

[0053] The image denoising formula is as follows:

[0054] D(x,y)=med{f(x+i,y+j)|-k<=i,j<=k}

[0055] The deep learning model performs classification through feature extraction. If there is a lot of noise in the input image, it may interfere with the model's feature learning. Denoising can help the model capture tumor-related structural features more accurately. In the formula, D(x, y) represents the value of the pixel in the denoised image, f(x, y) represents the value of the pixel in the original image, k represents the radius of the filter window, med represents the median calculation of the pixel value in the window, and i, j represent the offset of the filter window. After denoising, especially in the detection of early diseases, the model's ability to identify lesions can be significantly improved, thereby improving the overall diagnostic accuracy.

[0056] The image translation formula is as follows:

[0057]

[0058] By translating the CT image, multiple images with different displacements can be generated to expand the data set. Data enhancement can help the model better cope with image changes in different situations and improve the robustness of the model. In the formula, (x, y) represents the coordinates of the midpoint of the original image, (x′, y′) represents the coordinates of the midpoint of the translated image, and t x , t y Represents the amount of horizontal and vertical translation. The translation operation enables the model to be exposed to features from multiple perspectives and positions during training, thereby improving its performance on unseen images and reducing its reliance on specific position information.

[0059] The image cropping formula is as follows:

[0060] I crop =I[x1:x2,y1:y2]

[0061] By cropping, we can remove the background information irrelevant to diagnosis in the image and highlight the lesion area, so that the model learning can focus more on key features and improve the analysis efficiency. In the formula, I crop represents the cropped image, I represents the original image, (x1, y1) and (x2, y2) represent the coordinates of the upper left corner and lower right corner of the cropped area. The cropped image usually reduces its size, thereby reducing the computational requirements in subsequent processing and improving the processing speed, especially in the case of large-scale data sets;

[0062] The feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features, and lesion structure features based on the processed image through a convolutional neural network, where:

[0063] The formula for extracting lung contour features is as follows:

[0064] Z l =I l-1 *W l +b l

[0065] A l =ReLU(Z l )

[0066] F=P(A l )

[0067] Accurate lung contour extraction helps to remove the influence of tissues or organs that are not related to lung health, thereby improving the focus of subsequent feature learning. In the formula, * represents the convolution operation, I l-1 Represents the output of the previous layer, Z l represents the linear output of the convolutional layer, W l represents the weight of the convolutional layer, b l represents the convolution layer bias, Al ReLU represents the output after activation, P represents the pooling operation, and F represents the lung contour feature. By extracting the lung contour, the model can better understand the anatomical structure of the lung and provide a basis for subsequent functional analysis.

[0068] The formula for extracting lung boundary features is as follows:

[0069]

[0070] The lung boundary characteristics can reflect the shape and size changes of tissues, which is very important for distinguishing normal and pathological states. In the formula, G represents the lung boundary characteristics, G x represents the horizontal direction of the lung x boundary, G y Indicates the vertical direction of the lung y boundary. By analyzing boundary changes, information on the development and outcome of lung lesions can be obtained, thus helping clinicians provide a more comprehensive diagnosis.

[0071] The contrast feature extraction formula is as follows:

[0072]

[0073] Contrast features help to enhance the contrast between different tissues in CT images, making the lesions more prominent in the image and improving the visualization effect. In the formula, C represents the contrast feature, N represents the total number of pixels in the image, μ represents the average gray value of the image, and I(i) represents the gray value of the i-th pixel in the image. Significant contrast features enable the model to learn features related to lesions more quickly, thereby improving the convergence speed and performance of the model.

[0074] The grayscale feature extraction formula is as follows:

[0075]

[0076] Grayscale features are one of the basic image features. They help improve the effectiveness of feature extraction in the feature mapping process of the subsequent deep learning model. In the formula, Fm represents the grayscale feature, N represents the total number of pixels in the image, and the grayscale histogram h(q) is used to understand the frequency of each grayscale level q in the image. Through deep learning, grayscale features are adaptively extracted, allowing the model to better understand the subtle changes in lesions, thereby improving the overall diagnostic effect.

[0077] The formula for extracting lesion structure features is as follows:

[0078] R s =concat(A l ,A l+1 ,A l+2 )

[0079] Different lesion structural features such as shape, density and boundary smoothness can be used to predict pathological types, thus providing a basis for clinical treatment. In the formula, R s Represents the structural features of the lesion, and concat represents the feature map A output by connecting different layers l ,A l+1 ,A l+2 Extracting features from the lesion structure can help doctors observe the dynamic characteristics of the lesion over time, thereby effectively evaluating the treatment effect and disease progression;

[0080] The classification model module is responsible for inputting the extracted multi-level features into the fully connected neural network for further analysis and classification. First, the extracted features include information such as lung contours, boundaries, contrast, grayscale, and lesion structure. These features are used as the input layer feature vectors of the network. The fully connected neural network performs nonlinear transformations on these features through multiple hidden layers to learn the potential complex patterns and relationships in the data. During the forward propagation of the network, the activation function introduces nonlinearity, allowing the model to capture more complex feature interactions. Through a series of weight adjustments and feature combinations, the network finally outputs a probability distribution, which represents the possibility that the input image belongs to different categories. Specifically, first determine whether the image is a normal lung or has lung cancer. If the model predicts that the probability of lung cancer exceeds the set threshold, the system will further enter the classification stage. Based on the previously learned features, the specific types of cancer, such as adenocarcinoma, squamous cell carcinoma, etc., and their stages, such as early, middle or late stages, can be subdivided. This can be achieved through multiple layers of output nodes. Once the classification is completed, the generated diagnosis results will depend on the labels in the output layer, corresponding to the classification output of the model. These results will be formatted and sent to the output module for use by doctors or clinical decision support systems;

[0081] The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

[0082] See also Figure 2 , a CT automatic lung cancer diagnosis and classification method based on deep learning, comprising the following steps:

[0083] S1. Obtain lung CT scan images, including normal and different stages of lung cancer images, from medical imaging equipment through a data acquisition module, as well as clinical background information such as age, gender, and smoking history of the patient, and transmit the image data to a data preprocessing module;

[0084] S2, the data preprocessing module performs image denoising, translation and cropping operations on the collected images, and marks the tumors in the lung CT images to form the labels of the data set;

[0085] S3, the feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features and lesion structure features based on the processed image through a convolutional neural network, and sends the extracted features to the classification module in a unified manner;

[0086] S4, the classification model module inputs the extracted features into the fully connected neural network for classification, determines whether the image is normal lung or lung cancer, further classifies the specific type and stage of cancer and generates a classification diagnosis result and sends it to the output module;

[0087] S5. The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

[0088] By building a model to extract features from the patient's lung images and classify them in detail to generate detailed diagnostic reports, we can avoid misdiagnosis of malignant tumors and improve the accuracy of lung medical diagnosis.

[0089] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A deep learning-based CT automatic lung cancer diagnosis and classification system, characterized by: It includes data acquisition module, data preprocessing module, feature extraction module, classification model module and output and feedback module; The data acquisition module obtains lung CT scan images, including normal and different stages of lung cancer images, and obtains the patient's age, gender, smoking history and clinical background information from the medical imaging device, and transmits the image data to the data preprocessing module; The data preprocessing module performs image denoising, translation and cropping operations on the collected images, and marks the tumors in the lung CT images to form labels for the data set; The feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features, and lesion structure features based on the processed image through a convolutional neural network, and uniformly sends the extracted features to the classification model module; The classification model module inputs the extracted features into a fully connected neural network for classification, determines whether the image is a normal lung or lung cancer, further classifies the specific type and stage of the cancer, and generates a classification diagnosis result and sends it to the output and feedback module; The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

2. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 1, characterized in that: The data preprocessing module performs image denoising on the collected image. The image denoising formula is as follows: D(x,y)=med{f(x+i,y+j)|-k<=i,j<=k} In the formula, D(x,y) represents the value of the pixel in the denoised image, f(x,y) represents the value of the pixel in the original image, k represents the radius of the filter window, med represents the median calculation of the pixel values ​​in the window, and i,j represent the offset of the filter window.

3. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 2, characterized in that: The data preprocessing module performs image translation on the collected image, and the image translation formula is as follows: In the formula, (x, y) represents the coordinates of the midpoint of the original image, (x′, y′) represents the coordinates of the midpoint of the translated image, and t x , t y Indicates the amount of translation in the horizontal and vertical directions.

4. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 3, characterized in that: The data preprocessing module performs image cropping on the collected image, and the image cropping formula is as follows: I crop =I[x1:x2,y1:y2] In the formula, I crop represents the cropped image, I represents the original image, (x1, y1) and (x2, y2) represent the coordinates of the upper left corner and lower right corner of the cropped area.

5. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 4, characterized in that: The feature extraction module extracts lung contour features based on the processed image through a convolutional neural network. The formula is as follows: Z l =I l-1 *W l +b l AND l =ReLU(Z l ) F=P(A l ) In the formula, * represents the convolution operation, I l-1 Represents the output of the previous layer, Z l represents the linear output of the convolutional layer, W l represents the weight of the convolutional layer, b l represents the convolution layer bias, A l ReLU represents the activation function, P represents the pooling operation, and F represents the lung contour feature.

6. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 5, characterized in that: The feature extraction module extracts lung boundary features based on the processed image through a convolutional neural network. The feature extraction formula is as follows: In the formula, G represents the lung boundary feature, G x represents the horizontal direction of the lung x boundary, G y Indicates the vertical direction of the y boundary of the lung.

7. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 6, characterized in that: The feature extraction module extracts contrast features based on the processed image through a convolutional neural network. The feature extraction formula is as follows: In the formula, C represents the contrast feature, N represents the total number of pixels in the image, μ represents the average grayscale value of the image, and I(i) represents the grayscale value of the i-th pixel in the image.

8. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 7, characterized in that: The feature extraction module extracts grayscale features based on the processed image through a convolutional neural network. The feature extraction formula is as follows: In the formula, Fm represents the grayscale feature, N represents the total number of pixels in the image, and the grayscale histogram h(q) is used to understand the frequency of each grayscale level q in the image.

9. The deep learning-based CT automatic lung cancer diagnosis and classification system according to claim 8, characterized in that: The feature extraction module extracts the lesion structure features based on the processed image through a convolutional neural network. The feature extraction formula is as follows: R s =concat(A l ,A l+1 ,A l+2 ) In the formula, R s Represents the structural features of the lesion, and concat represents the feature map A output by connecting different layers l ,A l+1 ,A l+2 Extract features from .

10. A deep learning-based CT automatic lung cancer diagnosis and classification method, characterized in that: The following steps are involved: S1. Obtain lung CT scan images, including normal and different stages of lung cancer images, from medical imaging equipment through a data acquisition module, as well as clinical background information such as age, gender, and smoking history of the patient, and transmit the image data to a data preprocessing module; S2, the data preprocessing module performs image denoising, translation and cropping operations on the collected images, and marks the tumors in the lung CT images to form the labels of the data set; S3, the feature extraction module extracts lung contour features, lung boundary features, contrast features, grayscale features and lesion structure features based on the processed image through a convolutional neural network, and sends the extracted features to the classification module in a unified manner; S4, the classification model module inputs the extracted features into the fully connected neural network for classification, determines whether the image is normal lung or lung cancer, further classifies the specific type and stage of cancer and generates a classification diagnosis result and sends it to the output module; S5. The output and feedback module presents the classified diagnosis results in the form of a report and provides it to doctors for further analysis. It also continuously iterates and optimizes the model by collecting doctors' feedback and new case data.

Images