Lung tumor recognition system based on deep neural network

Abstract

The present application provides a lung tumor recognition system based on a deep neural network, including a lung original image acquisition module, a lung precise image acquisition module, a deep neural network, a first loss function acquisition module, a second loss function acquisition module, a third loss function acquisition module, a total loss function construction module, and a parameter adjustment module. In the actual recognition stage of lung tumors, the present application can directly input the original lung image obtained by detection into the deep neural network for recognition without performing ROI extraction to obtain a precise lung image, which greatly improves the recognition speed and efficiency on the basis of improving the accuracy of lung tumor recognition.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a lung tumor recognition system based on a deep neural network. Background Art

[0002] Cancer is extremely destructive and can occur in various parts of the human body. It can invade surrounding tissues and organs, causing damage to the functions of these tissues and organs. It is also diffusible and can spread to other parts of the body through the blood or lymphatic system, causing metastatic cancer. Among them, lung tumors are one of the malignant tumors with the highest incidence and mortality rates among cancers, and the incidence and mortality rates are increasing year by year. At present, medical imaging methods can be used to diagnose lung tumors in a timely manner and provide a reference for treatment, including ultrasound, X-ray imaging, electronic computed tomography, magnetic resonance imaging, and positron emission tomography.

[0003] In order to better identify early tumor symptoms, computer-aided diagnosis technology based on deep neural networks has been widely used. In the prior art, deep neural networks usually need to perform ROI extraction on the original lung images detected by medical imaging equipment, remove background noise interference, and obtain accurate lung images to ensure the accuracy of lung tumor recognition results. If ROI extraction is not performed, the exact position of the lungs cannot be determined, and it will be interfered by background noise, resulting in misjudgment of lung tumor recognition results, affecting accuracy.

[0004] Inevitably, since ROI extraction of the original lung image requires a lot of time for image processing, this will lead to low recognition speed and efficiency of the lung tumor recognition system. Therefore, there is an urgent need for a lung tumor recognition system based on a deep neural network that can improve the recognition speed and efficiency while improving the accuracy of lung tumor recognition. Summary of the invention

[0005] In view of the above problems, the present application is proposed to provide a lung tumor recognition system based on a deep neural network. In the actual recognition stage of lung tumors, the original lung image obtained by detection is directly input into the deep neural network for recognition without the need for ROI extraction to obtain an accurate lung image. This greatly improves the recognition speed and efficiency on the basis of improving the accuracy of lung tumor recognition.

[0006] In order to achieve the above-mentioned purpose, the present application provides a lung tumor recognition system based on a deep neural network, including a lung original image acquisition module, a lung precise image acquisition module, a deep neural network, a loss function construction module, and a parameter adjustment module;

[0007] A lung original image acquisition module, used to acquire a first training data set, wherein the first training data set is a collection of multiple lung original images;

[0008] A lung precise image acquisition module receives a first training data set, performs image extraction on the first training data set using a ROI extraction method, and acquires a second training data set, wherein the second training data set is a collection of multiple lung precise images;

[0009] A deep neural network receives a first training data set and a second training data set respectively, recognizes the original lung image of the first training data set to obtain a recognition result of the original lung image, and recognizes the precise lung image of the second training data set to obtain a recognition result of the precise lung image;

[0010] A loss function construction module receives the recognition result of the original lung image, the recognition result of the precise lung image, and the real tumor result of the original lung image, and constructs a total loss function;

[0011] The parameter adjustment module receives the total loss function and adjusts the parameters of the deep neural network through back propagation and stochastic gradient descent according to the total loss function.

[0012] Furthermore, the loss function construction module includes a first loss function acquisition module, a second loss function acquisition module, a third loss function acquisition module and a total loss function acquisition module;

[0013] A first loss function acquisition module receives the recognition result of the original lung image and the recognition result of the precise lung image, and acquires a first loss function according to the recognition result of the original lung image and the recognition result of the precise lung image;

[0014] A second loss function acquisition module receives the recognition result of the original lung image and the actual tumor result of the original lung image, and acquires a second loss function according to the recognition result of the original lung image and the actual tumor result of the original lung image;

[0015] A third loss function acquisition module receives the recognition result of the lung precise image and the real tumor result of the lung original image, and acquires a third loss function according to the recognition result of the lung precise image and the real tumor result of the lung original image;

[0016] The total loss function acquisition module receives the first loss function, the second loss function, and the third loss function to obtain the total loss function.

[0017] Furthermore, the original lung image is generated by ultrasound, X-ray imaging, electronic computed tomography, magnetic resonance imaging, or positron emission tomography.

[0018] Furthermore, the first loss function is expressed as:

[0019]

[0020] Where N is the total number of samples in the first training data set, is the recognition result of the i-th lung original image, is the recognition result of the i-th lung accurate image, L avg is the first loss function.

[0021] Furthermore, the second loss function L org It is expressed as:

[0022]

[0023] in, is the true tumor result of the i-th lung original image, L org is the second loss function.

[0024] Furthermore, the third loss function is expressed as:

[0025]

[0026] Among them, L roi is the third loss function.

[0027] Furthermore, the total loss function L loss It is expressed as:

[0028] L loss =L org +αL avg +βL roi

[0029] Among them, α and β are pre-set weight coefficients, L loss is the total loss function.

[0030] Furthermore, the deep neural network includes a multi-scale feature extraction module and a lung tumor recognition module connected in sequence;

[0031] A multi-scale feature extraction module receives the original lung image and the precise lung image, performs multi-scale feature extraction on the original lung image and the precise lung image respectively, and obtains image features of the original lung image and image features of the precise lung image;

[0032] The lung tumor recognition module receives the image features of the original lung image and the image features of the precise lung image, recognizes the original lung image and the precise lung image according to the image features of the original lung image and the image features of the precise lung image, and generates recognition results of the original lung image and the precise lung image.

[0033] Further, the multi-scale feature extraction module includes a standard convolution block, a first dilated convolution block, a second dilated convolution block, a third dilated convolution block, a fourth dilated convolution block, a fifth dilated convolution block, a first addition module, a second addition module, a third addition module, a fourth addition module, and a splicing module;

[0034] A standard convolution block, used for receiving an original lung image or an accurate lung image, performing a convolution operation on the original lung image or the accurate lung image using a standard convolution kernel, and generating a first standard convolution output;

[0035] The first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are respectively used to receive the first standard convolution output, perform dilated convolution operation on the first standard convolution output, and generate the first dilated convolution output, the second dilated convolution output, the third dilated convolution output, the fourth dilated convolution output, and the fifth dilated convolution output; the dilation rates of the first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are 1, 2, 4, 8, and 16, respectively;

[0036] A first adding module, configured to receive the first dilated convolution output and the second dilated convolution output, and fuse the first dilated convolution output and the second dilated convolution output to generate a first fused output;

[0037] a second adding module, configured to receive the first fused output and the third dilated convolution output, and fuse the first fused output and the third dilated convolution output to generate a second fused output;

[0038] a third adding module, configured to receive the second fused output and the fourth dilated convolution output, and fuse the second fused output and the fourth dilated convolution output to generate a third fused output;

[0039] a fourth adding module, configured to receive the third fused output and the fifth dilated convolution output, and fuse the third fused output with the fifth dilated convolution output to generate a fourth fused output;

[0040] The splicing module is used to receive the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output, and to generate image features after splicing the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output.

[0041] Furthermore, the standard convolution kernel size is 1×1, and the dilated convolution kernel size is 3×3.

[0042] The beneficial effects of this application are:

[0043] (1) The lung tumor recognition system based on a deep neural network uses the original lung image and the precise lung image to train the deep neural network, and uses the first loss function, the second loss function, and the third loss function to construct a total loss function to optimize the deep neural network, so that the tumor recognition result of the original lung image is the same as the tumor recognition result of the precise lung image. In the actual recognition stage of the lung tumor, the original lung image obtained by detection can be directly input into the deep neural network for recognition without performing ROI extraction to obtain the precise lung image. On the basis of improving the accuracy of lung tumor recognition, the recognition speed and efficiency are greatly improved.

[0044] (2) The multi-scale feature extraction module of the lung tumor recognition system based on deep neural network adopts multiple dilated convolution blocks with different dilation rates, which increases the receptive field of the deep neural network, prevents feature loss, and realizes the multi-scale feature extraction function. At the same time, multiple addition modules are added to perform trapezoidal fusion on the output of dilated convolution blocks with different receptive fields, improve the spatial resolution of the output feature map, avoid grid effects and loss of important features, maintain the continuity of extracted features, enhance the extraction performance of multi-scale features, and thus improve the accuracy of lung tumor recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 A structural block diagram of a lung tumor recognition system based on a deep neural network provided in this application;

[0047] Figure 2 A structural block diagram of a deep neural network provided in this application;

[0048] Figure 3 A structural block diagram of the multi-scale feature extraction module provided in this application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms of "a", "said", and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two, but does not exclude the inclusion of at least one.

[0051] The present application provides a lung tumor recognition system based on a deep neural network. The lung tumor recognition system uses original lung images and precise lung images to train the deep neural network, and uses a first loss function, a second loss function, and a third loss function to construct a total loss function to optimize the deep neural network, so that the tumor recognition result of the original lung image is the same as the tumor recognition result of the precise lung image. In the actual recognition stage of the lung tumor, the original lung image obtained by detection can be directly input into the deep neural network for recognition without ROI extraction to obtain a precise lung image. On the basis of improving the accuracy of lung tumor recognition, the recognition speed and efficiency are greatly improved.

[0052] At the same time, the present application also provides a multi-scale feature extraction module that uses multiple dilated convolution blocks with different dilation rates to increase the receptive field of the deep neural network, prevent feature loss, and realize multi-scale feature extraction. At the same time, multiple addition modules are added to perform trapezoidal fusion on the output of dilated convolution blocks with different receptive fields, improve the spatial resolution of the output feature map, avoid grid effects and loss of important features, maintain the continuity of extracted features, enhance the extraction performance of multi-scale features, and thus improve the accuracy of lung tumor recognition.

[0053] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0054] First embodiment

[0055] like Figure 1 As shown, in one embodiment, the present application provides a lung tumor recognition system based on a deep neural network. The lung tumor recognition system based on a deep neural network includes a lung original image acquisition module, a lung precise image acquisition module, a deep neural network, a loss function construction module, and a parameter adjustment module.

[0056] The lung original image acquisition module is used to acquire a first training data set, wherein the first training data set is a collection of multiple lung original images, and the lung original images can be generated by detection equipment such as ultrasound, X-ray imaging, electronic computed tomography, magnetic resonance imaging, positron emission tomography, etc.

[0057] The lung precise image acquisition module receives a first training data set, performs image extraction on the first training data set using a ROI (region of interest) extraction method, and acquires a second training data set, wherein the second training data set is a collection of multiple lung precise images.

[0058] Exemplarily, the background noise of the original lung image is removed through the ROI extraction method to obtain a precise lung image, which can accurately indicate the lung contour.

[0059] In the prior art, accurate lung tumor recognition results can be obtained by using a deep neural network to identify precise lung images after background noise interference has been removed.

[0060] The original lung image can be considered as a comprehensive image of the precise lung image and background noise. Due to the interference of background noise, the lung tumor recognition results generated by the recognition of the precise lung image through the deep neural network will be misjudged, affecting the accuracy.

[0061] Therefore, the existing technology usually requires ROI extraction of the original lung image to obtain an accurate lung image and improve the accuracy of lung tumor recognition results. However, it is inevitable that the lung tumor recognition system needs to spend a lot of time to extract the ROI of the original lung image, resulting in low recognition speed and efficiency of the lung tumor recognition system.

[0062] Exemplarily, the ROI extraction method adopts the OSTU threshold segmentation method or the fuzzy C-means clustering segmentation method.

[0063] The deep neural network receives the first training data set and the second training data set respectively, recognizes the original lung image of the first training data set to obtain the recognition result of the original lung image, and recognizes the precise lung image of the second training data set to obtain the recognition result of the precise lung image.

[0064] A loss function construction module receives the recognition result of the original lung image, the recognition result of the precise lung image, and the real tumor result of the original lung image, and constructs a total loss function;

[0065] Further, the loss function construction module includes a first loss function acquisition module, a second loss function acquisition module, a third loss function acquisition module and a total loss function acquisition module;

[0066] The first loss function acquisition module receives the recognition result of the original lung image and the recognition result of the precise lung image, and acquires the first loss function L according to the recognition result of the original lung image and the recognition result of the precise lung image. avg .

[0067] Exemplarily, the first loss function L avg It is expressed as:

[0068]

[0069] Where N is the total number of samples in the first training data set, is the recognition result of the i-th lung original image, is the recognition result of the i-th lung accurate image.

[0070] The first loss function is used to indicate the degree of difference between the recognition result of the original lung image and the recognition result of the precise lung image.

[0071] The second loss function acquisition module receives the recognition result of the original lung image and the actual tumor result of the original lung image, and acquires the second loss function L according to the recognition result of the original lung image and the actual tumor result of the original lung image. org .

[0072] Exemplarily, the second loss function L org It is expressed as:

[0073]

[0074] in, is the true tumor result of the i-th lung original image.

[0075] The second loss function is used to indicate the degree of difference between the recognition result of the original lung image and the actual tumor result of the original lung image.

[0076] The third loss function acquisition module receives the recognition result of the lung precise image and the real tumor result of the lung original image, and acquires the third loss function L according to the recognition result of the lung precise image and the real tumor result of the lung original image. org .

[0077] Exemplarily, the third loss function L roi It is expressed as:

[0078]

[0079] The second loss function is used to indicate the degree of difference between the recognition result of the precise lung image and the true tumor result of the original lung image.

[0080] The total loss function construction module receives the first loss function L avg , the second loss function L org , the third loss function L roi , we get the total loss function L loss .

[0081] For example, the total loss function L loss It is expressed as:

[0082] L loss =L org +αL avg +βL roi

[0083] Among them, α and β are pre-set weight coefficients.

[0084] Parameter adjustment module, receiving the total loss function L loss , according to the total loss function L loss The parameters of the deep neural network are adjusted through back propagation and stochastic gradient descent.

[0085] The deep neural network parameters are adjusted by the first loss function in the total loss function so that the recognition results of the deep neural network for the original lung image and the precise lung image are similar. The deep neural network parameters are adjusted by the second loss function in the total loss function so that the recognition results of the deep neural network for the original lung image are similar to the real tumor results. The deep neural network parameters are adjusted by the third loss function in the total loss function so that the recognition results of the deep neural network for the precise lung image are similar to the real tumor results.

[0086] In the prior art, in order to improve the accuracy of lung tumor recognition results, the lung tumor recognition system needs to perform ROI extraction on the original lung image to obtain an accurate lung image for lung tumor recognition, but this takes a lot of time, resulting in low recognition speed and efficiency of the lung tumor recognition system.

[0087] In this embodiment, the lung tumor recognition system uses the original lung image and the precise lung image to train the deep neural network, and uses the first loss function, the second loss function, and the third loss function to construct a total loss function to optimize the deep neural network, so that the tumor recognition result of the original lung image is the same as the tumor recognition result of the precise lung image. In the actual recognition stage of the lung tumor, the detected original lung image can be directly input into the deep neural network for recognition without ROI extraction to obtain the precise lung image. On the basis of improving the accuracy of lung tumor recognition, the recognition speed and efficiency are greatly improved.

[0088] Furthermore, after the deep neural network-based lung tumor recognition system is trained, the actual original lung image is recognized to obtain the lung tumor recognition result.

[0089] Second embodiment

[0090] In another embodiment, the present application also provides a lung tumor recognition system based on a deep neural network, such as Figure 2 As shown, the deep neural network includes a multi-scale feature extraction module and a lung tumor recognition module which are connected in sequence.

[0091] The multi-scale feature extraction module receives the original lung image and the precise lung image, performs multi-scale feature extraction on the original lung image and the precise lung image respectively, and obtains the image features of the original lung image and the image features of the precise lung image.

[0092] The lung tumor recognition module receives the image features of the original lung image and the image features of the precise lung image, recognizes the original lung image and the precise lung image according to the image features of the original lung image and the image features of the precise lung image, and generates recognition results of the original lung image and the precise lung image.

[0093] Further, such as Figure 3 As shown, the multi-scale feature extraction module includes a standard convolution block, a first dilated convolution block, a second dilated convolution block, a third dilated convolution block, a fourth dilated convolution block, a fifth dilated convolution block, a first addition module, a second addition module, a third addition module, a fourth addition module, and a splicing module.

[0094] Specifically, the standard convolution block is used to receive the original lung image or the precise lung image, perform a convolution operation on the original lung image or the precise lung image using a standard convolution kernel, perform dimensionality reduction, and generate a first standard convolution output.

[0095] Exemplarily, the standard convolution kernel size is 1×1.

[0096] The first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are respectively used to receive the first standard convolution output, perform dilated convolution operations on the first standard convolution output, and generate the first dilated convolution output, the second dilated convolution output, the third dilated convolution output, the fourth dilated convolution output, and the fifth dilated convolution output. The dilation rates of the first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are 1, 2, 4, 8, and 16, respectively, which increases the receptive field of the deep neural network, prevents feature loss, and realizes the multi-scale feature extraction function.

[0097] Exemplarily, the dilated convolution kernel size is 3×3.

[0098] The first adding module is used to receive the first dilated convolution output and the second dilated convolution output, fuse the first dilated convolution output and the second dilated convolution output, and generate a first fused output.

[0099] The second adding module is used to receive the first fused output and the third dilated convolution output, fuse the first fused output and the third dilated convolution output, and generate a second fused output.

[0100] The third adding module is used to receive the second fused output and the fourth dilated convolution output, fuse the second fused output and the fourth dilated convolution output, and generate a third fused output.

[0101] The fourth adding module is used to receive the third fused output and the fifth dilated convolution output, fuse the third fused output and the fifth dilated convolution output, and generate a fourth fused output.

[0102] In the prior art, in order to increase the receptive field, the outputs of different dilated convolutions are usually directly spliced ​​to obtain image features. However, during the dilated convolution operation, as the dilation rate continues to increase, the interval between the convolution kernels also increases, and the spatial resolution of the output feature map decreases, resulting in a grid effect, which in turn will lose important features and destroy the continuity of the features.

[0103] In this embodiment of the present application, the outputs of the dilated convolution blocks with five different receptive fields are trapezoidally fused through the first addition module, the second addition module, the third addition module, and the fourth addition module, and the output of the dilated convolution block with a certain dilation rate is fused with the outputs of all the dilated convolution blocks with smaller dilation rates to improve the spatial resolution of the output feature map, avoid grid effects and loss of important features, maintain the continuity of features, and enhance the extraction performance of multi-scale features.

[0104] The splicing module is used to receive the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output, and to generate image features after splicing the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output.

[0105] In this embodiment, the multi-scale feature extraction module uses multiple dilated convolution blocks with different dilation rates to increase the receptive field of the deep neural network, prevent feature loss, and realize the multi-scale feature extraction function. At the same time, multiple addition modules are added to perform trapezoidal fusion on the output of the dilated convolution blocks with different receptive fields, improve the spatial resolution of the output feature map, avoid grid effects and loss of important features, maintain the continuity of the extracted features, enhance the extraction performance of multi-scale features, and thus improve the accuracy of lung tumor recognition.

[0106] The above description shows and describes several preferred embodiments of the present application, but as mentioned above, it should be understood that the present application is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present application, and should be within the scope of protection of the claims attached to the present application.

Claims

1. A lung tumor recognition system based on deep neural network, characterized in that: It includes lung original image acquisition module, lung precise image acquisition module, deep neural network, loss function construction module, and parameter adjustment module; A lung original image acquisition module, used to acquire a first training data set, wherein the first training data set is a collection of multiple lung original images; A lung precise image acquisition module receives a first training data set, performs image extraction on the first training data set using a ROI extraction method, and acquires a second training data set, wherein the second training data set is a collection of multiple lung precise images; A deep neural network receives a first training data set and a second training data set respectively, recognizes the original lung image of the first training data set to obtain a recognition result of the original lung image, and recognizes the precise lung image of the second training data set to obtain a recognition result of the precise lung image; A loss function construction module receives the recognition result of the original lung image, the recognition result of the precise lung image, and the real tumor result of the original lung image, and constructs a total loss function; The parameter adjustment module receives the total loss function and adjusts the parameters of the deep neural network through back propagation and stochastic gradient descent according to the total loss function; The loss function construction module includes a first loss function acquisition module, a second loss function acquisition module, a third loss function acquisition module and a total loss function acquisition module; A first loss function acquisition module receives the recognition result of the original lung image and the recognition result of the precise lung image, and acquires a first loss function according to the recognition result of the original lung image and the recognition result of the precise lung image; A second loss function acquisition module receives the recognition result of the original lung image and the actual tumor result of the original lung image, and acquires a second loss function according to the recognition result of the original lung image and the actual tumor result of the original lung image; A third loss function acquisition module receives the recognition result of the lung precise image and the real tumor result of the lung original image, and acquires a third loss function according to the recognition result of the lung precise image and the real tumor result of the lung original image; The total loss function acquisition module receives the first loss function, the second loss function, and the third loss function to obtain the total loss function.

2. The lung tumor recognition system according to claim 1, characterized in that: The original lung image is generated by ultrasound, X-ray imaging, electronic computer tomography, magnetic resonance imaging, and positron emission tomography.

3. The lung tumor recognition system according to claim 1, characterized in that: The first loss function is expressed as: Where N is the total number of samples in the first training data set, is the recognition result of the i-th lung original image, is the recognition result of the i-th lung accurate image, L avg is the first loss function.

4. The lung tumor recognition system according to claim 3, characterized in that: The second loss function L org It is expressed as: in, is the true tumor result of the i-th lung original image, L org is the second loss function.

5. The lung tumor recognition system according to claim 4, characterized in that: The third loss function is expressed as: Among them, L roi is the third loss function.

6. The lung tumor recognition system according to claim 5, characterized in that: The total loss function L loss It is expressed as: L loss =L org +αL avg +βL roi Among them, α and β are pre-set weight coefficients, L loss is the total loss function.

7. The lung tumor recognition system according to any one of claims 1 to 6, characterized in that: The deep neural network includes a multi-scale feature extraction module and a lung tumor recognition module connected in sequence; A multi-scale feature extraction module receives the original lung image and the precise lung image, performs multi-scale feature extraction on the original lung image and the precise lung image respectively, and obtains image features of the original lung image and image features of the precise lung image; The lung tumor recognition module receives the image features of the original lung image and the image features of the precise lung image, recognizes the original lung image and the precise lung image according to the image features of the original lung image and the image features of the precise lung image, and generates recognition results of the original lung image and the precise lung image.

8. The lung tumor recognition system according to claim 7, characterized in that: The multi-scale feature extraction module includes a standard convolution block, a first dilated convolution block, a second dilated convolution block, a third dilated convolution block, a fourth dilated convolution block, a fifth dilated convolution block, a first addition module, a second addition module, a third addition module, a fourth addition module, and a splicing module; A standard convolution block, used for receiving an original lung image or an accurate lung image, performing a convolution operation on the original lung image or the accurate lung image using a standard convolution kernel, and generating a first standard convolution output; The first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are respectively used to receive the first standard convolution output, perform dilated convolution operation on the first standard convolution output, and generate the first dilated convolution output, the second dilated convolution output, the third dilated convolution output, the fourth dilated convolution output, and the fifth dilated convolution output; the dilation rates of the first dilated convolution block, the second dilated convolution block, the third dilated convolution block, the fourth dilated convolution block, and the fifth dilated convolution block are 1, 2, 4, 8, and 16, respectively; A first adding module, configured to receive the first dilated convolution output and the second dilated convolution output, and fuse the first dilated convolution output and the second dilated convolution output to generate a first fused output; a second adding module, configured to receive the first fused output and the third dilated convolution output, and fuse the first fused output and the third dilated convolution output to generate a second fused output; a third adding module, configured to receive the second fused output and the fourth dilated convolution output, and fuse the second fused output and the fourth dilated convolution output to generate a third fused output; a fourth adding module, configured to receive the third fused output and the fifth dilated convolution output, and fuse the third fused output with the fifth dilated convolution output to generate a fourth fused output; The splicing module is used to receive the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output, and to generate image features after splicing the first dilated convolution output, the first fusion output, the second fusion output, the third fusion output, and the fourth fusion output.

Images