A convolutional neural network-based method for reconstructing field of view image processing

Abstract

The present invention relates to the field of fine processing of medical imaging data, and in particular to a method for processing a reconstructed field of view image based on a convolutional neural network, comprising: obtaining image correlation features using a real-time reconstructed field of view image; establishing a multi-level analysis and screening model for a reconstructed field of view image based on a convolutional neural network using the image correlation features; obtaining a reconstructed field of view image processing result using the multi-level analysis and screening model for the reconstructed field of view image, and setting up a reconstructed field of view image error analysis processing to extract various features at all levels under a scanning field of view and a reconstructed field of view, which is of great help to the applicability of a scheme, and can be deployed and implemented for CT scanners with different settings. At the same time, without changing the scanning image and the reconstructed field of view image, a multi-level classification and screening model is established based on a convolutional neural network, and a verification and iteration process is added to the scheme to ensure stable and efficient operation in actual application.

Description

Technical Field

[0001] The present invention relates to the field of medical image data refinement processing, and in particular to a method for reconstructing visual field images based on a convolutional neural network. Background Art

[0002] Reconstructing the visual field image is the reconstruction process of the CT scan image, which converts multiple X-ray image data acquired at different angles into two-dimensional or three-dimensional images to facilitate diagnosis by doctors and radiologists. This process is a key step in modern medical imaging technology. It makes the originally complex internal structure clearly visible, providing an important visual basis for clinical diagnosis and treatment. This process requires not only highly precise hardware equipment, but also advanced software algorithms to ensure the quality and accuracy of the image. However, due to the superposition of various influencing factors during CT scanning, the final output reconstructed image is not stable enough, and it is impossible to fully distinguish the correct content of the scanned target or the error manifestation of the scanning process. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides a reconstructed field of view image processing method based on convolutional neural network, which efficiently and accurately processes multi-feature node data of traditional convolutional network and quickly and accurately outputs the error between CT scan image and reconstructed image.

[0004] To achieve the above object, the present invention provides a method for reconstructing visual field images based on a convolutional neural network, comprising:

[0005] S1, using real-time reconstruction of the visual field image to obtain image correlation features;

[0006] S2. Using the image correlation features to establish a multi-level analysis and screening model for reconstructed visual field images based on a convolutional neural network;

[0007] S3. Obtaining a reconstructed field of view image processing result by using the reconstructed field of view image multi-level analysis screening model.

[0008] Preferably, the method of obtaining image correlation features by reconstructing the visual field image in real time includes:

[0009] S1-1, obtaining a scanning field of view image range of the real-time reconstructed field of view image;

[0010] S1-2, obtaining a reconstructed field of view image range of the real-time reconstructed field of view image;

[0011] S1-3, using the ratio of the scanned field of view image range to the reconstructed field of view image range as the scanned reconstructed image change rate;

[0012] S1-4, obtaining a scanning aperture of the real-time reconstructed field of view image as a hardware feature of the reconstructed field of view image;

[0013] S1-5, acquiring corresponding image pixels according to the reconstructed field of view image range as basic features of the reconstructed field of view image;

[0014] S1-6. Using the scanned field of view image range, the reconstructed field of view image range, the scanned reconstructed image change rate, the reconstructed field of view image hardware features and the reconstructed field of view image basic features as image association features.

[0015] Furthermore, using the image association features to establish a multi-level analysis and screening model for reconstructed visual field images based on a convolutional neural network includes:

[0016] S2-1, using the image association features to establish a scan reconstruction image transformation matrix;

[0017] S2-2, using the image association features to establish a basic reconstruction field of view image analysis model based on a convolutional neural network;

[0018] S2-3, establishing a basic reconstructed visual field image screening model based on a convolutional neural network according to the basic reconstructed visual field image analysis model;

[0019] S2-4, using the scanned reconstructed image transformation matrix, sequentially perform structural verification processing on the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model to obtain a reconstructed field of view image multi-level analysis and screening model.

[0020] Furthermore, establishing a scan reconstruction image transformation matrix using the image association features includes:

[0021] S2-1-1, establishing a digital image matrix of the scanning field of view image using the scanning field of view image range of the image association feature;

[0022] S2-1-2, establishing a digital image matrix of a reconstructed field of view image using the reconstructed field of view image range of the image association feature;

[0023] S2-1-3, obtaining a ratio of the digital image matrix of the scanned field of view image to the digital image matrix of the reconstructed field of view image as a scanned reconstructed image matrix transformation coefficient;

[0024] S2-1-4, determine whether the scanned reconstructed image matrix transformation coefficient is consistent with the scanned reconstructed image change rate of the image-related features. If so, obtain the corresponding matrix vector of the digital image matrix of the scanned field of view image and the corresponding matrix vector of the digital image matrix of the reconstructed field of view image as the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image, and execute S2-1-5. Otherwise, return to S2-1-1;

[0025] S2-1-5. Use the digital image matrix of the scanned field of view image, the digital image matrix of the reconstructed field of view image, the scanned and reconstructed image matrix transformation coefficients, the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image as the scanned and reconstructed image transformation matrix.

[0026] Furthermore, using the image association features to establish a basic reconstruction field of view image analysis model based on a convolutional neural network includes:

[0027] S2-2-1, determining whether the image center of the scanned field of view image range corresponding to the image association feature is consistent with the image center of the reconstructed field of view image range, if so, executing S2-2-2, otherwise, returning to S1-1;

[0028] S2-2-2, respectively acquiring corresponding historical scanned field of view images and historical reconstructed field of view images according to the image association features to establish a first data set and a second data set;

[0029] S2-2-3, dividing the target area corresponding to the first data set into first target area pixels;

[0030] S2-2-4, dividing the target area corresponding to the second data set into second target area pixels;

[0031] S2-2-5, using the first data set and the first target area pixels as input, and the pixel matrix corresponding to the first target area pixels as output, to establish a primary image analysis model based on convolutional neural network training;

[0032] S2-2-6, using the second data set and the second target area pixels as input, and the pixel matrix corresponding to the second target area pixels as output, to establish a secondary image analysis model based on convolutional neural network training;

[0033] S2-2-7, using the primary image analysis model and the secondary image analysis model as a basis to reconstruct a visual field image analysis model;

[0034] Wherein, the target area is the target area of ​​the CT scan image.

[0035] Furthermore, establishing a basic reconstructed visual field image screening model based on a convolutional neural network according to the basic reconstructed visual field image analysis model includes:

[0036] S2-3-1. Obtaining a model output result of a primary image analysis model according to the basic reconstructed field of view image analysis model;

[0037] S2-3-2, obtaining a model output result of a secondary image analysis model according to the basic reconstructed field of view image analysis model;

[0038] S2-3-3, using the model output result of the primary image analysis model and the model output result of the secondary image analysis model as a third data set;

[0039] S2-3-4. Using the third data set as input, the target area contour of the model output result of the third data set corresponding to the primary image analysis model and the target area contour of the model output result of the secondary image analysis model as output, and establishing a basic reconstructed field of view image screening model based on convolutional neural network training.

[0040] Furthermore, the scanning reconstructed image transformation matrix is ​​used to sequentially perform structural verification processing on the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model to obtain the reconstructed field of view image multi-level analysis and screening model, including:

[0041] S2-4-1, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the first-level image analysis model corresponds to the digital image matrix of the scanned field of view image of the scanned reconstructed image transformation matrix. If so, execute S2-4-2, otherwise, use the digital image matrix of the non-corresponding scanned field of view image to update the first data set corresponding to the scanned field of view image, and return to S2-2-3;

[0042] S2-4-2, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the secondary image analysis model corresponds to the digital image matrix of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, execute S2-4-3, otherwise, use the digital image matrix of the non-corresponding reconstructed field of view image to update the second data set corresponding to the reconstructed field of view image, and return to S2-2-4;

[0043] S2-4-3, determine whether the pixel matrix ratio of the model output result of the primary image analysis model and the model output result of the secondary image analysis model corresponding to the same scanning field of view image is consistent with the scanning reconstructed image matrix transformation coefficient of the scanning reconstructed image transformation matrix, if so, execute S2-4-4, otherwise, use the scanning reconstructed image matrix transformation coefficient to update the scanning field of view image corresponding to the reconstructed field of view image, and return to S2-2-2;

[0044] S2-4-4. Determine whether the model output result of the basic reconstructed field of view image screening model corresponds to the target area contour ratio and the scanned reconstructed image matrix transformation coefficient. If so, use the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model as the reconstructed field of view image multi-level analysis screening model. Otherwise, return to S2-1-1.

[0045] Furthermore, the reconstructed visual field image processing result obtained by using the reconstructed visual field image multi-level analysis and screening model includes:

[0046] S3-1, acquiring a real-time scanning field of view image and a real-time reconstructed field of view image respectively according to the real-time reconstructed field of view image;

[0047] S3-2, using the real-time scanning field of view image and the real-time reconstructed field of view image to input the reconstructed field of view image multi-level analysis and screening model to obtain the model output result of the reconstructed field of view image multi-level analysis and screening model;

[0048] S3-3. Obtain a reconstructed field of view image processing result according to a model output result of the reconstructed field of view image multi-level analysis and screening model.

[0049] Furthermore, the reconstructed field of view image processing result is obtained according to the model output result of the reconstructed field of view image multi-level analysis and screening model, including:

[0050] S3-3-1, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the scanned field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-3, otherwise, obtain the target area contour of the model output result of the first-level image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the scanned field of view image, and execute S3-3-2;

[0051] S3-3-2, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the scanned field of view image and the pixel coordinates corresponding to the digital image matrix label of the scanned field of view image as the processing result of reconstructing the field of view image;

[0052] S3-3-3, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-5, otherwise, obtain the target area contour of the model output result of the secondary image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the reconstructed field of view image, and execute S3-3-4;

[0053] S3-3-4, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the reconstructed field of view image and the pixel coordinates corresponding to the digital image matrix label of the reconstructed field of view image as the reconstructed field of view image processing result;

[0054] S3-3-5. Determine whether the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image of the image-associated features at the current moment are consistent with the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image at the previous adjacent moment. If so, update the scanning field of view image range and the reconstructed field of view image range corresponding to the real-time reconstructed field of view image, and execute S2-1-1. Otherwise, return to S1.

[0055] Compared with the closest prior art, the present invention has the following beneficial effects:

[0056] This solution is applied to the post-processing field of CT scan images. Therefore, the hardware characteristics and image features are taken into consideration together, and the reconstructed field of view image error analysis processing is set up to extract various types of features at all levels under the scanning field of view and the reconstructed field of view. This is of great help to the applicability of the solution, and it can be deployed and implemented for CT scanners with different settings. At the same time, without changing the scanning image and the reconstructed field of view image, a multi-level classification and screening model is established based on the convolutional neural network, and a verification and iteration process is added within the solution to ensure stable and efficient operation in actual application. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flow chart of a method for reconstructing visual field images based on a convolutional neural network provided by the present invention. DETAILED DESCRIPTION

[0058] The specific implementation modes of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0060] Embodiment 1:

[0061] The present invention provides a method for reconstructing visual field images based on convolutional neural networks, such as Figure 1 As shown, including:

[0062] S1, using real-time reconstruction of the visual field image to obtain image correlation features;

[0063] S2. Using the image correlation features to establish a multi-level analysis and screening model for reconstructed visual field images based on a convolutional neural network;

[0064] S3. Obtaining a reconstructed field of view image processing result by using the reconstructed field of view image multi-level analysis screening model.

[0065] S1 specifically includes:

[0066] S1-1, obtaining a scanning field of view image range of the real-time reconstructed field of view image;

[0067] S1-2, obtaining a reconstructed field of view image range of the real-time reconstructed field of view image;

[0068] S1-3, using the ratio of the scanned field of view image range to the reconstructed field of view image range as the scanned reconstructed image change rate;

[0069] S1-4, obtaining a scanning aperture of the real-time reconstructed field of view image as a hardware feature of the reconstructed field of view image;

[0070] S1-5, acquiring corresponding image pixels according to the reconstructed field of view image range as basic features of the reconstructed field of view image;

[0071] S1-6. Using the scanned field of view image range, the reconstructed field of view image range, the scanned reconstructed image change rate, the reconstructed field of view image hardware features and the reconstructed field of view image basic features as image association features.

[0072] In this embodiment, a method for reconstructing a field of view image based on a convolutional neural network is provided. The reconstructed field of view image is processed based on an overall proportional transformation of a scanned field of view image. Therefore, there are relevant stability coefficients in both image pixels and target contours.

[0073] S2 specifically includes:

[0074] S2-1, using the image association features to establish a scan reconstruction image transformation matrix;

[0075] S2-2, using the image association features to establish a basic reconstruction field of view image analysis model based on a convolutional neural network;

[0076] S2-3, establishing a basic reconstructed visual field image screening model based on a convolutional neural network according to the basic reconstructed visual field image analysis model;

[0077] S2-4, using the scanned reconstructed image transformation matrix, sequentially perform structural verification processing on the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model to obtain a reconstructed field of view image multi-level analysis and screening model.

[0078] S2-1 specifically includes:

[0079] S2-1-1, establishing a digital image matrix of the scanning field of view image using the scanning field of view image range of the image association feature;

[0080] S2-1-2, establishing a digital image matrix of a reconstructed field of view image using the reconstructed field of view image range of the image association feature;

[0081] S2-1-3, obtaining a ratio of the digital image matrix of the scanned field of view image to the digital image matrix of the reconstructed field of view image as a scanned reconstructed image matrix transformation coefficient;

[0082] S2-1-4, determine whether the scanned reconstructed image matrix transformation coefficient is consistent with the scanned reconstructed image change rate of the image-related features. If so, obtain the corresponding matrix vector of the digital image matrix of the scanned field of view image and the corresponding matrix vector of the digital image matrix of the reconstructed field of view image as the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image, and execute S2-1-5. Otherwise, return to S2-1-1;

[0083] S2-1-5. Use the digital image matrix of the scanned field of view image, the digital image matrix of the reconstructed field of view image, the scanned and reconstructed image matrix transformation coefficients, the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image as the scanned and reconstructed image transformation matrix.

[0084] S2-2 specifically includes:

[0085] S2-2-1, determining whether the image center of the scanned field of view image range corresponding to the image association feature is consistent with the image center of the reconstructed field of view image range, if so, executing S2-2-2, otherwise, returning to S1-1;

[0086] S2-2-2, respectively acquiring corresponding historical scanned field of view images and historical reconstructed field of view images according to the image association features to establish a first data set and a second data set;

[0087] S2-2-3, dividing the target area corresponding to the first data set into first target area pixels;

[0088] S2-2-4, dividing the target area corresponding to the second data set into second target area pixels;

[0089] S2-2-5, using the first data set and the first target area pixels as input, and the pixel matrix corresponding to the first target area pixels as output, to establish a primary image analysis model based on convolutional neural network training;

[0090] S2-2-6, using the second data set and the second target area pixels as input, and the pixel matrix corresponding to the second target area pixels as output, to establish a secondary image analysis model based on convolutional neural network training;

[0091] S2-2-7, using the primary image analysis model and the secondary image analysis model as a basis to reconstruct a visual field image analysis model;

[0092] Wherein, the target area is the target area of ​​the CT scan image.

[0093] S2-3 specifically includes:

[0094] S2-3-1. Obtaining a model output result of a primary image analysis model according to the basic reconstructed field of view image analysis model;

[0095] S2-3-2, obtaining a model output result of a secondary image analysis model according to the basic reconstructed field of view image analysis model;

[0096] S2-3-3, using the model output result of the primary image analysis model and the model output result of the secondary image analysis model as a third data set;

[0097] S2-3-4. Using the third data set as input, the target area contour of the model output result of the third data set corresponding to the primary image analysis model and the target area contour of the model output result of the secondary image analysis model as output, and establishing a basic reconstructed field of view image screening model based on convolutional neural network training.

[0098] S2-4 specifically includes:

[0099] S2-4-1, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the first-level image analysis model corresponds to the digital image matrix of the scanned field of view image of the scanned reconstructed image transformation matrix. If so, execute S2-4-2, otherwise, use the digital image matrix of the non-corresponding scanned field of view image to update the first data set corresponding to the scanned field of view image, and return to S2-2-3;

[0100] S2-4-2, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the secondary image analysis model corresponds to the digital image matrix of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, execute S2-4-3, otherwise, use the digital image matrix of the non-corresponding reconstructed field of view image to update the second data set corresponding to the reconstructed field of view image, and return to S2-2-4;

[0101] S2-4-3, determine whether the pixel matrix ratio of the model output result of the primary image analysis model and the model output result of the secondary image analysis model corresponding to the same scanning field of view image is consistent with the scanning reconstructed image matrix transformation coefficient of the scanning reconstructed image transformation matrix, if so, execute S2-4-4, otherwise, use the scanning reconstructed image matrix transformation coefficient to update the scanning field of view image corresponding to the reconstructed field of view image, and return to S2-2-2;

[0102] S2-4-4. Determine whether the model output result of the basic reconstructed field of view image screening model corresponds to the target area contour ratio and the scanned reconstructed image matrix transformation coefficient. If so, use the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model as the reconstructed field of view image multi-level analysis screening model. Otherwise, return to S2-1-1.

[0103] In this embodiment, a reconstructed visual field image processing method based on a convolutional neural network, step S2-4-4 involves corresponding judgment and screening of the target area contour ratio and the scanned reconstructed image matrix transformation coefficient, wherein the target area contour ratio can be calculated by the target area contour vector of the model output result of the primary image analysis model and the model output result of the secondary image analysis model, and at the same time, the target area is the target area of ​​the CT scan image. When performing a computed tomography scan, doctors and radiologists will pay special attention to specific parts that require detailed examination, which may require special attention due to disease, injury or other medical reasons. By accurately imaging these target areas, doctors can more accurately diagnose the condition, formulate treatment plans, or monitor disease progression. When processing CT scan images, it is crucial to identify and locate these key areas.

[0104] S3 specifically includes:

[0105] S3-1, acquiring a real-time scanning field of view image and a real-time reconstructed field of view image respectively according to the real-time reconstructed field of view image;

[0106] S3-2, using the real-time scanning field of view image and the real-time reconstructed field of view image to input the reconstructed field of view image multi-level analysis and screening model to obtain the model output result of the reconstructed field of view image multi-level analysis and screening model;

[0107] S3-3. Obtain a reconstructed field of view image processing result according to a model output result of the reconstructed field of view image multi-level analysis and screening model.

[0108] S3-3 specifically includes:

[0109] S3-3-1, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the scanned field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-3, otherwise, obtain the target area contour of the model output result of the first-level image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the scanned field of view image, and execute S3-3-2;

[0110] S3-3-2, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the scanned field of view image and the pixel coordinates corresponding to the digital image matrix label of the scanned field of view image as the processing result of reconstructing the field of view image;

[0111] S3-3-3, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-5, otherwise, obtain the target area contour of the model output result of the secondary image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the reconstructed field of view image, and execute S3-3-4;

[0112] S3-3-4, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the reconstructed field of view image and the pixel coordinates corresponding to the digital image matrix label of the reconstructed field of view image as the reconstructed field of view image processing result;

[0113] S3-3-5. Determine whether the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image of the image-associated features at the current moment are consistent with the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image at the previous adjacent moment. If so, update the scanning field of view image range and the reconstructed field of view image range corresponding to the real-time reconstructed field of view image, and execute S2-1-1. Otherwise, return to S1.

[0114] In this embodiment, a method for processing reconstructed field of view images based on a convolutional neural network is provided. Since this scheme is a step between the CT machine and the final image output, the scheme is set to automatically process the next reconstructed image when there is no abnormality or error in the image. At the same time, the scheme has built-in self-check verification iteration, internal and external loops, to ensure stable and efficient operation during actual application deployment.

[0115] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0117] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for reconstructing visual field images based on convolutional neural networks, characterized in that: include: S1, using real-time reconstruction of the visual field image to obtain image correlation features; S1-1, obtaining a scanning field of view image range of the real-time reconstructed field of view image; S1-2, obtaining a reconstructed field of view image range of the real-time reconstructed field of view image; S1-3, using the ratio of the scanned field of view image range to the reconstructed field of view image range as the scanned reconstructed image change rate; S1-4, obtaining a scanning aperture of the real-time reconstructed field of view image as a hardware feature of the reconstructed field of view image; S1-5, acquiring corresponding image pixels according to the reconstructed field of view image range as basic features of the reconstructed field of view image; S1-6, using the scanned field of view image range, the reconstructed field of view image range, the scanned reconstructed image change rate, the reconstructed field of view image hardware features and the reconstructed field of view image basic features as image association features; S2. Using the image correlation features to establish a multi-level analysis and screening model for reconstructed visual field images based on a convolutional neural network; S2-1, using the image association features to establish a scan reconstruction image transformation matrix; S2-1-1, establishing a digital image matrix of the scanning field of view image using the scanning field of view image range of the image association feature; S2-1-2, establishing a digital image matrix of a reconstructed field of view image using the reconstructed field of view image range of the image association feature; S2-1-3, obtaining a ratio of the digital image matrix of the scanned field of view image to the digital image matrix of the reconstructed field of view image as a scanned reconstructed image matrix transformation coefficient; S2-1-4, determine whether the scanned reconstructed image matrix transformation coefficient is consistent with the scanned reconstructed image change rate of the image-related features. If so, obtain the corresponding matrix vector of the digital image matrix of the scanned field of view image and the corresponding matrix vector of the digital image matrix of the reconstructed field of view image as the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image, and execute S2-1-5. Otherwise, return to S2-1-1; S2-1-5, using the digital image matrix of the scanned field of view image, the digital image matrix of the reconstructed field of view image, the scanned and reconstructed image matrix transformation coefficients, the digital image matrix label of the scanned field of view image and the digital image matrix label of the reconstructed field of view image as the scanned and reconstructed image transformation matrix; S2-2, using the image association features to establish a basic reconstruction field of view image analysis model based on a convolutional neural network; S2-2-1, determining whether the image center of the scanned field of view image range corresponding to the image association feature is consistent with the image center of the reconstructed field of view image range, if so, executing S2-2-2, otherwise, returning to S1-1; S2-2-2, respectively acquiring corresponding historical scanned field of view images and historical reconstructed field of view images according to the image association features to establish a first data set and a second data set; S2-2-3, dividing the target area corresponding to the first data set into first target area pixels; S2-2-4, dividing the target area corresponding to the second data set into second target area pixels; S2-2-5, using the first data set and the first target area pixels as input, and the pixel matrix corresponding to the first target area pixels as output, to establish a primary image analysis model based on convolutional neural network training; S2-2-6, using the second data set and the second target area pixels as input, and the pixel matrix corresponding to the second target area pixels as output, to establish a secondary image analysis model based on convolutional neural network training; S2-2-7, using the primary image analysis model and the secondary image analysis model as a basis to reconstruct a visual field image analysis model; Wherein, the target area is the target area of ​​the CT scan image; S2-3, establishing a basic reconstructed visual field image screening model based on a convolutional neural network according to the basic reconstructed visual field image analysis model; S2-3-1. Obtaining a model output result of a primary image analysis model according to the basic reconstructed field of view image analysis model; S2-3-2, obtaining a model output result of a secondary image analysis model according to the basic reconstructed field of view image analysis model; S2-3-3, using the model output result of the primary image analysis model and the model output result of the secondary image analysis model as a third data set; S2-3-4, using the third data set as input, the target area contour of the model output result of the first-level image analysis model and the target area contour of the model output result of the second-level image analysis model corresponding to the third data set as output, and establishing a basic reconstructed field of view image screening model based on convolutional neural network training; S2-4, using the scanned reconstructed image transformation matrix to sequentially perform structural verification processing on the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model to obtain a reconstructed field of view image multi-level analysis and screening model; S2-4-1, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the first-level image analysis model corresponds to the digital image matrix of the scanned field of view image of the scanned reconstructed image transformation matrix. If so, execute S2-4-2, otherwise, use the digital image matrix of the non-corresponding scanned field of view image to update the first data set corresponding to the scanned field of view image, and return to S2-2-3; S2-4-2, determine whether the model output result of the basic reconstructed field of view image analysis model corresponding to the secondary image analysis model corresponds to the digital image matrix of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, execute S2-4-3, otherwise, use the digital image matrix of the non-corresponding reconstructed field of view image to update the second data set corresponding to the reconstructed field of view image, and return to S2-2-4; S2-4-3, determine whether the pixel matrix ratio of the model output result of the primary image analysis model and the model output result of the secondary image analysis model corresponding to the same scanning field of view image is consistent with the scanning reconstructed image matrix transformation coefficient of the scanning reconstructed image transformation matrix, if so, execute S2-4-4, otherwise, use the scanning reconstructed image matrix transformation coefficient to update the scanning field of view image corresponding to the reconstructed field of view image, and return to S2-2-2; S2-4-4, determine whether the model output result of the basic reconstructed field of view image screening model corresponds to the target area contour ratio and the scanned reconstructed image matrix transformation coefficient. If so, use the basic reconstructed field of view image analysis model and the basic reconstructed field of view image screening model as the reconstructed field of view image multi-level analysis screening model. Otherwise, return to S2-1-1; S3. Obtaining a reconstructed field of view image processing result by using the reconstructed field of view image multi-level analysis screening model.

2. A method for reconstructing visual field images based on a convolutional neural network as claimed in claim 1, characterized in that: The reconstructed visual field image processing result obtained by using the reconstructed visual field image multi-level analysis and screening model includes: S3-1, acquiring a real-time scanning field of view image and a real-time reconstructed field of view image respectively according to the real-time reconstructed field of view image; S3-2, using the real-time scanning field of view image and the real-time reconstructed field of view image to input the reconstructed field of view image multi-level analysis and screening model to obtain the model output result of the reconstructed field of view image multi-level analysis and screening model; S3-3. Obtain a reconstructed field of view image processing result according to a model output result of the reconstructed field of view image multi-level analysis and screening model.

3. A method for reconstructing visual field images based on a convolutional neural network as claimed in claim 2, characterized in that: The reconstructed field of view image processing result obtained according to the model output result of the reconstructed field of view image multi-level analysis and screening model includes: S3-3-1, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the scanned field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-3, otherwise, obtain the target area contour of the model output result of the first-level image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the scanned field of view image, and execute S3-3-2; S3-3-2, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the scanned field of view image and the pixel coordinates corresponding to the digital image matrix label of the scanned field of view image as the processing result of reconstructing the field of view image; S3-3-3, determine whether the model output result of the reconstructed field of view image multi-level analysis and screening model corresponds to the digital image matrix label of the reconstructed field of view image of the scanned reconstructed image transformation matrix, if so, directly execute S3-3-5, otherwise, obtain the target area contour of the model output result of the secondary image analysis model corresponding to the model output result of the reconstructed field of view image multi-level analysis and screening model as the abnormal contour of the reconstructed field of view image, and execute S3-3-4; S3-3-4, obtaining the different pixels between the pixel coordinates corresponding to the abnormal contour of the reconstructed field of view image and the pixel coordinates corresponding to the digital image matrix label of the reconstructed field of view image as the reconstructed field of view image processing result; S3-3-5. Determine whether the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image of the image-associated features at the current moment are consistent with the hardware features of the reconstructed field of view image and the basic features of the reconstructed field of view image at the previous adjacent moment. If so, update the scanning field of view image range and the reconstructed field of view image range corresponding to the real-time reconstructed field of view image, and execute S2-1-1. Otherwise, return to S1.

Images