An image processing method, system, device, and medium for CTA images

By utilizing lesion information from preoperative CTA images as prior knowledge, combining deep learning techniques and neural network models to remove metal artifacts, and using Mango to adjust contrast, the problem of inaccurate metal information in existing technologies has been solved, achieving accurate diagnosis from CTA images.

CN118781020BActive Publication Date: 2026-11-17SICHUAN UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410897447.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2026-11-17
Estimated Expiration
2044-07-05

Smart Images

  • Figure CN118781020B_ABST
    Figure CN118781020B_ABST
Patent Text Reader

Abstract

This invention discloses an image processing method, system, device, and medium for CTA images, specifically relating to the field of image processing technology. The key technical points are as follows: The method includes: acquiring a first sinusoidal feature map without metal artifacts, a second sinusoidal feature map with metal artifacts, and a metal trajectory feature map; removing metal trajectory values ​​from the second sinusoidal feature map using the metal trajectory feature map to obtain a third sinusoidal feature map without metal trajectory values; using the third sinusoidal feature map as input and the first sinusoidal feature map as the ground truth of the network, and training a neural network model using a loss function; obtaining a neural network model capable of removing metal artifacts from CTA images after training; using the trained neural network model to perform metal artifact removal processing on the postoperative CTA image to be processed, obtaining a CTA image with metal artifacts removed; and adjusting the contrast of the CTA image with metal artifacts removed using Mango to obtain a CTA image displaying blood vessels.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to an image processing method, system, device, and medium for CTA images. Background Technology

[0002] Treatment methods for intracranial aneurysms are mainly divided into two categories: open craniotomy and interventional embolization. Interventional embolization involves puncturing the femoral artery, inserting a microcatheter along the vessel into the aneurysm cavity, and filling the aneurysm with embolic materials such as coils to eliminate blood flow within the aneurysm sac and prevent rupture. Compared to open craniotomy, it has advantages such as minimal invasiveness and rapid recovery, making it the preferred treatment for intracranial aneurysms. However, patients undergoing interventional surgery still face certain risks post-operatively, including aneurysm rupture, intraoperative vasospasm, coil displacement, and aneurysm recanalization. Therefore, close follow-up and management are necessary for patients undergoing embolization. Digital subtraction angiography (DSA) is considered the gold standard for observing vascular anatomy after coil embolization, but DSA is an invasive, high-radiation, time-consuming examination with a certain risk of thromboembolization.

[0003] Computed tomography (CTA), as a non-invasive vascular imaging technique, is increasingly used for the screening and follow-up of intracranial aneurysms. However, due to beam hardening and photon starvation, band-like artifacts can occur around intracranial metallic implants, making it difficult to detect postoperative complications and aneurysm recurrence, thus leading to misdiagnosis and missed diagnosis. Therefore, removing metal artifacts from CTA images, improving CTA readability, and enhancing vascular reconstruction are of great clinical significance.

[0004] In recent years, with the development of technology, the application of spectral CT and various artifact removal post-processing techniques has solved this problem to some extent. However, in existing technologies, simply thresholding the CT data affected by metal artifacts to obtain metal information leads to inaccurate metal information acquisition. Inaccurate metal information can cause errors in the process of removing metal artifacts, making it difficult to achieve satisfactory results for blood vessels surrounding the metal.

[0005] Therefore, the present invention aims to provide an image processing method, system, device, and medium for CTA images to solve the aforementioned problems. Summary of the Invention

[0006] The technical problem this invention aims to solve is that in existing technologies, simply thresholding CT data affected by metal artifacts to obtain metal information leads to inaccurate metal information acquisition. The purpose is to provide an image processing method, system, device, and medium for CTA images. By using preoperative CTA images as guidance and taking the lesion information from the preoperative CTA images as prior knowledge, the inaccurate metal information problem in existing technologies is solved. Simultaneously, deep learning technology is used to acquire metal information and remove metal artifacts from postoperative follow-up CTA images. Furthermore, after removing CTA image artifacts, Mango is used to adjust the contrast of the CTA image after metal artifact removal, resulting in a CTA image displaying blood vessels. This facilitates accurate diagnosis of the patient's postoperative vascular and lesion conditions by physicians.

[0007] This invention is achieved through the following technical solution:

[0008] An image processing method for CTA images, the method comprising:

[0009] Obtain the first sinusoidal feature map without the influence of metal artifacts, the second sinusoidal feature map with the influence of metal artifacts, and the metal trajectory feature map;

[0010] The metal trajectory value in the second sinusoidal feature map is removed by using the metal trajectory feature map to obtain the third sinusoidal feature map without metal trajectory value. The third sinusoidal feature map is used as input and the first sinusoidal feature map is used as the ground truth value of the network. The neural network model is trained by combining the loss function. After training, a neural network model that can remove metal artifacts in CTA images is obtained.

[0011] The trained neural network model was used to remove metal artifacts from the postoperative CTA images to be processed, resulting in CTA images with metal artifacts removed.

[0012] Mango was used to adjust the contrast of the CTA images after removing metal artifacts, resulting in CTA images that visualize blood vessels.

[0013] Furthermore, the following steps are taken to obtain a first sinusoidal feature map without the influence of metal artifacts, a second sinusoidal feature map with the influence of metal artifacts, and a metal trajectory feature map:

[0014] Acquire preoperative and postoperative CTA images of the patient, and use the DeSeg deep learning method to obtain lesion image information in the preoperative CTA images;

[0015] Based on a preset registration reference, the preoperative CTA images and postoperative CTA images are registered and aligned to obtain the aligned preoperative CTA images and generate registration parameters. The lesion image information is then registered using the registration parameters to obtain the metallic image information of the postoperative CTA images.

[0016] The aligned preoperative CTA image, postoperative CTA image, and metal image information are forward-projected to obtain the corresponding first sinusoidal feature map without metal artifacts, the second sinusoidal feature map with metal artifacts, and the metal trajectory feature map.

[0017] Furthermore, the pre-defined registration reference is the patient's skull.

[0018] Furthermore, the loss function is specifically as follows:

[0019]

[0020] Among them, S corr The restored sine feature map; S free The first sinusoidal feature map is free from metal artifacts; S coil This is a feature map of the metal trajectory.

[0021] Furthermore, convolutional neural networks include multi-head attention mechanism models and feedforward propagation network models.

[0022] Furthermore, the multi-head attention mechanism model utilizes multi-channel convolutional layers to perform depthwise separable convolution operations on the input sinusoidal feature map to obtain query vector, key vector, and value vector;

[0023] The query vector and the key vector are transposed and then multiplied by a dot product. The result is then divided by a preset scaling factor and scaled. Finally, the scaled result is normalized using the SoftMax function to obtain the attention weights.

[0024] The attention matrix is ​​obtained by weighted summation of the attention weights and the value vectors.

[0025] The attention matrix is ​​reduced in dimensionality using 1×1 convolution, and the result is concatenated with the input vector residual to obtain the sinusoidal feature map output by the multi-head attention mechanism model.

[0026] Furthermore, the feedforward propagation network model includes a first linear projection layer and a second linear projection layer;

[0027] The first linear projection layer uses the GELU nonlinear activation function to gate the input sinusoidal feature map, resulting in a gated weight map.

[0028] The second linear projection layer performs depthwise convolution on the input sinusoidal feature map;

[0029] The result of the depthwise convolution is multiplied by the gated weight map to obtain the sinusoidal feature map output by the feedforward propagation network model.

[0030] The present invention also provides an image processing system for CTA images, the system comprising:

[0031] The feature map acquisition module is used to acquire a first sinusoidal feature map without the influence of metal artifacts, a second sinusoidal feature map with the influence of metal artifacts, and a metal trajectory feature map.

[0032] The model training module is used to remove the metal trajectory values ​​in the second sinusoidal feature map using the metal trajectory feature map, so as to obtain the third sinusoidal feature map without metal trajectory values. The third sinusoidal feature map is used as input and combined with the loss function to train the neural network model. After training, a neural network model that can remove metal artifacts in CTA images is obtained.

[0033] The artifact removal module is used to remove metal artifacts from the postoperative CTA image to be processed using the trained neural network model, so as to obtain a CTA image with metal artifacts removed.

[0034] The vessel display module is used to adjust the contrast of the CTA image after removing metal artifacts using Mango, so as to obtain a CTA image displaying blood vessels.

[0035] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] In this invention, by using preoperative CTA images as guidance and taking the lesion information in the preoperative CTA images as prior knowledge, the technical problem of inaccurate metal information in the prior art is solved. At the same time, metal information is obtained through deep learning technology and metal artifacts in postoperative follow-up CTA images are removed. After removing CTA image artifacts, Mango is used to adjust the contrast of the CTA image after removing metal artifacts to obtain a CTA image showing blood vessels, which facilitates doctors to make accurate diagnoses of the patient's postoperative vascular condition and lesion condition. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0040] Figure 1 This is a flowchart of an image processing method for CTA images in this embodiment;

[0041] Figure 2 This is a flowchart illustrating the deep learning technique DeSeg in an image processing method for CTA images in this embodiment.

[0042] Figure 3 This is a schematic diagram of the DeSeg image processing technology, a deep learning technique used in image processing for CTA images, in this embodiment.

[0043] Figure 4 This is a schematic diagram of two registrations in an image processing method for CTA images in this embodiment;

[0044] Figure 5 This is a schematic diagram of a forward projection image in an image processing method for CTA images in this embodiment;

[0045] Figure 6 This is a model structure diagram of a neural network model in an image processing method for CTA images in this embodiment;

[0046] Figure 7 This is a schematic diagram of a blood vessel display in an image processing method for CTA images according to this embodiment;

[0047] Figure 8 This is a schematic diagram of the system structure of an image processing system for CTA images in this embodiment;

[0048] Figure 9 This is a schematic diagram of the structure of a computer device in this embodiment. Detailed Implementation

[0049] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0050] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to define the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0051] The terminology used in the description of the various examples in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. Furthermore, the term "and / or" as used in this disclosure covers any one of the listed items and all possible combinations thereof.

[0052] As mentioned in the background, in recent years, with the development of technology, the application of spectral CT and various artifact removal post-processing techniques has solved this problem to some extent. However, in existing technologies, simply thresholding the CT data affected by metal artifacts to obtain metal information leads to inaccurate metal information acquisition. Inaccurate metal information makes it difficult to achieve satisfactory results in the process of removing metal artifacts, especially for blood vessels surrounding the metal.

[0053] Therefore, this invention provides an image processing method, system, device, and medium for CTA images. By using preoperative CTA images as guidance and taking the lesion information in the preoperative CTA images as prior knowledge, it solves the technical problem of inaccurate metal information in the prior art. At the same time, it uses deep learning technology to obtain metal information and remove metal artifacts in postoperative follow-up CTA images. After removing CTA image artifacts, Mango is used to adjust the contrast of the CTA image after removing metal artifacts to obtain a CTA image showing blood vessels, which facilitates doctors to make accurate diagnoses of the patient's postoperative vascular condition and lesion condition.

[0054] Example 1

[0055] An image processing method for CTA images, the method comprising:

[0056] S1: Obtain the first sinusoidal feature map without the influence of metal artifacts, the second sinusoidal feature map with the influence of metal artifacts, and the metal trajectory feature map;

[0057] S101: Acquire preoperative and postoperative CTA images of the patient, and use the DeSeg deep learning method to obtain lesion image information in the preoperative CTA images;

[0058] Specifically, in this embodiment, the raw data required to implement the image processing method is divided into two categories. The first is the patient's preoperative CTA data (preoperative CTA images), referred to as CTA. pre The first is used to guide subsequent acquisition of metal information and removal of metal artifacts; the second is the patient's postoperative follow-up CTA data (postoperative CTA images), referred to as CTA. re ;

[0059] Meanwhile, the DeSeg deep learning method is a conventional technique, and its specific technical means are disclosed in the invention patent with publication number CN115170802A. These details will not be elaborated upon here; please refer to [link to relevant documentation]. Figures 2-3 As shown, using the existing deep learning technology DeSeg, CTA is located and delineated. pre To obtain prior information about the location and size of the metal in aneurysms, the process involves: using a center point detector (SSD) to locate the central region of the lesion in preoperative CTA images; then using a multi-head U-Net segmentation model (MHUM) to refine the lesion contour; and finally using a data cascade unit to smoothly connect these two tasks.

[0060] M IA =DeSeg(CTA) pre )

[0061] Among them, M IA It is the lesion image information delineated by the DeSeg deep learning method.

[0062] S102: Based on the patient's skull, the preoperative CTA images and postoperative CTA images are registered and aligned to obtain the aligned preoperative CTA images and generate registration parameters. The lesion image information is registered using the registration parameters to obtain the metal image information of the postoperative CTA images.

[0063] Specifically, in this embodiment, when performing interventional embolization surgery, doctors always use coils to fill the entire aneurysm cavity; therefore, the lesion information M... IA After registration, metal image information can be obtained, which involves two registration processes. See [link to documentation]. Figure 4 As shown; First registration: Using the edge contour of the patient's skull, the preoperative CTA image and the postoperative CTA image are registered and aligned to obtain the aligned preoperative CTA image. reg And generate registration parameters, specifically:

[0064] x′=x+t x y′=y+t y

[0065] x′=x·cos(θ)-y·sin(θ), y′=x·sin(θ)-y·cos(θ)

[0066] Where (x, y) is CTA pre The pixel coordinates (x′, y′) in the CTA are re Pixel coordinates in (t) x , t y ) is the translation offset, and θ is the rotation angle. Using the above formulas, the registration parameters (t) can be obtained. x , t y ) and θ, based on the obtained parameters, for CTA pre By performing translation and rotation operations, the registered image can be obtained from the CTA. re Aligned image CTA reg ;

[0067] Second registration: The above registration parameters are applied to the obtained lesion image information M. IA Above, because this registration parameter can enable CTA pre To CTA re Alignment, so it can be considered that CTA pre The lesion image information M obtained from IA After registration, it can be used with CTA. re Align the metal image information in the image to obtain metal image information M. coil .

[0068] S103: Project the aligned preoperative CTA image, postoperative CTA image and metal image information forward to obtain the corresponding first sinusoidal feature map without metal artifacts, the second sinusoidal feature map with metal artifacts and the metal trajectory feature map.

[0069] Specifically, in this embodiment, see Figure 5 As shown, the aligned preoperative CTA images, postoperative CTA images, and metal images are forward-projected (FP) to obtain their corresponding projection domain data, specifically:

[0070] S free =FP(CTA) reg ), S raw =FP(CTA) re ), S coil =FP(M coil )

[0071] Among them, S free The first sinusoidal feature map without the influence of metal artifacts (a clean sinusoidal map unaffected by metal), S raw The second sinusoidal feature map affected by metal artifacts (the sinusoidal map affected by metal), S coil Refers to the feature map of metal trajectory.

[0072] S2: Remove the metal trajectory values ​​from the second sinusoidal feature map using the metal trajectory feature map to obtain a third sinusoidal feature map without metal trajectory values. Use the third sinusoidal feature map as input and the first sinusoidal feature map as the ground truth value of the network. Combine this with a loss function to train the neural network model. After training, a neural network model capable of removing metal artifacts from CTA images is obtained, specifically:

[0073] S201: Treat the values ​​at the metal trajectory in the second sinusoidal feature map as missing, and obtain the third sinusoidal feature map S without metal trajectory values. ma Specifically:

[0074] S ma =S raw -S raw ⊙S coil

[0075] Here, ⊙ represents element-wise multiplication;

[0076] S202: The third sine feature map is used as input, and the first sine feature map is used as the ground truth of the network. The neural network model is trained using a loss function to predict the third sine feature map S. ma The value at the missing metal trajectory in the repaired feature map S corr The data distribution conforms to the first sinusoidal feature map;

[0077] Specifically, in this embodiment, see Figure 6 As shown, the neural network model is a hybrid architecture of CNN and transformer, mainly composed of two parts: a multi-head attention mechanism model (Multi-DconvHeadTransposedAttention (MDTA)) and a feedforward network model (Gated-DconvFeed-ForwardNetwork (GDFN)); the loss function specifically adopts the L1 loss function, as follows:

[0078]

[0079] Among them, S corr The restored sine feature map; S free The first sinusoidal feature map is free from metal artifacts; S coil This is a feature map of the metal trajectory;

[0080] S203: Finally, S is reconstructed using the Filtered Back Projection (FBP) reconstruction algorithm. corr Obtain the corrected CTA image. MAR .

[0081] S3: Use the trained neural network model to remove metal artifacts from the postoperative CTA image to be processed, and obtain a CTA image with metal artifacts removed.

[0082] S4: Use Mango to adjust the contrast of the CTA image after removing metal artifacts to obtain a CTA image showing blood vessels.

[0083] Specifically, in this embodiment, see Figure 7 As shown, in Mango, by adjusting the contrast between soft tissue and skull, a CTA image of blood vessels can be obtained. A threshold greater than 2000 is set for skull, and less than 2000 is set for soft tissue.

[0084] Furthermore, the multi-head attention mechanism can apply self-attention along the channel dimension, rather than the spatial dimension, thus eliminating the need for explicit pixel-level interaction computation. Therefore, the multi-head attention model utilizes 1×1 convolutions to achieve pixel-level aggregation across channels and efficient 3×3 depthwise convolutions to achieve channel-level aggregation of local context. A depthwise separable convolution operation is performed on the input sinusoidal feature map to obtain the query vector Q, key vector K, and value vector V, specifically:

[0085]

[0086] Where X is the input to the multi-head attention mechanism model, and LN(·) represents the layer normalization operation. and These represent 3×3 depthwise convolution and 1×1 pointwise convolution, respectively.

[0087] Perform a dot product of the transposes of the query vector Q and the key vector K to obtain QK. T Then process the result QK T Divide by the preset scaling factor The scaling process is performed, and then the scaling result is normalized using the SoftMax function to obtain the attention weights.

[0088] The attention matrix is ​​obtained by weighted summation of the attention weights and the value vector, as follows:

[0089]

[0090] The attention matrix is ​​reduced in dimensionality using a 1×1 convolution. The processed result is then concatenated with the input vector residual to obtain the sinusoidal feature map output by the multi-head attention mechanism model, specifically:

[0091]

[0092] Specifically, in this embodiment, the multi-head attention mechanism (MDTA) used does not compute self-attention spatially (between pixels), but rather on a channel-by-channel basis. This means that it computes cross-channel covariance to generate an attention map that implicitly encodes the global context; in computing self-attention, MDTA uses depthwise separable convolution operations to generate Q (query), K (key), and V (value), which helps to emphasize local information.

[0093] Furthermore, the feedforward propagation network model includes a first linear projection layer and a second linear projection layer;

[0094] The first linear projection layer uses the GELU nonlinear activation function to gate the input sinusoidal feature map, controlling which complementary features should be forwarded, allowing the network to focus on finer image attributes, and obtaining a gated weight map. The second linear projection layer performs depthwise convolution on the input sinusoidal feature map. The result of the depthwise convolution is multiplied by the gated weight map to obtain the sinusoidal feature map output by the feedforward propagation network model, specifically:

[0095]

[0096] Where ⊙ represents element-wise multiplication, LN(·) represents layer normalization, and GELU(·) represents a nonlinear activation function. This represents the 3×3 depthwise convolution and the 1×1 dimension-expanded convolution of the first projection layer. This represents the 3×3 depthwise convolution and the 1×1 expanded dimension convolution of the second projection layer, where X and These are the input and output of GDFN, respectively.

[0097] Specifically, in this embodiment, GDFN introduces a gating mechanism to control the flow of information. The gating mechanism uses the GELU nonlinear activation function to generate a weight between 0 and 1. This weight can be applied to the output of the feedforward network to control which information should be retained or discarded. A weight map is generated using the gating mechanism, which is multiplied with the output of the depthwise separable convolution to control the flow of information. This allows the model to suppress irrelevant or redundant information while preserving important features.

[0098] Finally, features from the encoder and decoder are cascaded via skip connections. For each stage of the decoder, the input features are first upsampled by 2x, and then fused features from the encoder at the same resolution are cascaded using skip connections. The concatenated features are then sent to the decoder, enabling contextual interaction and the creation of long-distance dependencies.

[0099] Example 2

[0100] The present invention also provides an image processing system for CTA images, the system comprising:

[0101] The feature map acquisition module 100 is used to acquire a first sinusoidal feature map without the influence of metal artifacts, a second sinusoidal feature map with the influence of metal artifacts, and a metal trajectory feature map.

[0102] The model training module 200 is used to remove the metal trajectory values ​​in the second sinusoidal feature map using the metal trajectory feature map to obtain the third sinusoidal feature map without metal trajectory values. The third sinusoidal feature map is used as input and combined with the loss function to train the neural network model. After training, a neural network model that can remove metal artifacts in CTA images is obtained.

[0103] The artifact removal module 300 is used to perform metal artifact removal processing on the postoperative CTA image to be processed using the trained neural network model, so as to obtain a CTA image with metal artifacts removed.

[0104] The vessel display module 400 is used to adjust the contrast of the CTA image after removing metal artifacts using Mango to obtain a CTA image displaying blood vessels.

[0105] It should be noted that the modules in the system of Embodiment 2 correspond to the steps in the method of Embodiment 1. The steps in the method of Embodiment 1 have been described in detail in Embodiment 1, and the module content in the system will not be described in detail in this Embodiment 2.

[0106] Example 3

[0107] This embodiment also provides a computer device, including a memory 1005 and a processor 1001. The memory 1005 stores a computer program, and the processor 1001 executes the computer program to implement the steps of any of the methods described above.

[0108] It should be noted that the processor 1001 is used to execute the steps in the above method embodiments according to the instructions in the program code. Alternatively, when the processor 1001 executes the computer program, it implements the functions of each module / unit in the above system / device embodiments.

[0109] Specifically, in this embodiment, the computer program can be divided into one or more modules / units. One or more modules / units are stored in the memory 1005 and executed by the processor 1001 to complete this application. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0110] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 1001 and a memory 1005. Those skilled in the art will understand that this does not constitute a limitation on the terminal device, which may include more or fewer components than shown in the figures, or a combination of certain components, or different components. For example, the terminal device may also include an input / output device 1003, a network access device 1002, a bus 1006, etc.

[0111] The processor 1001 can be a Central Processing Unit (CPU), or other general-purpose processors 1001, digital signal processors 1001 (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor 1001 can be a microprocessor or any conventional processor, etc.

[0112] The memory 1005 can be an internal storage unit of the terminal device, such as a hard drive or RAM. The memory 1005 can also be an external storage device 1004 of the terminal device, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 1005 can include both internal storage units and external storage devices 1004. The memory 1005 is used to store computer programs and other programs and data required by the terminal device. The memory 1005 can also be used to temporarily store data that has been output or will be output.

[0113] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0114] Example 4

[0115] This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0116] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), registers, hard disks, optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing, or any other form of computer-readable storage medium in the art.

[0117] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside within an application-specific integrated circuit (ASIC). In embodiments of the invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device.

[0118] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An image processing method for CTA images, characterized in that the method... include: Acquire preoperative and postoperative CTA images of the patient, and use the DeSeg deep learning method to obtain lesion image information in the preoperative CTA images; Based on a preset registration reference, the preoperative CTA images and postoperative CTA images are registered and aligned to obtain the aligned preoperative CTA images and generate registration parameters. The lesion image information is then registered using the registration parameters to obtain the metallic image information of the postoperative CTA images. The aligned preoperative CTA image, postoperative CTA image, and metal image information are forward-projected to obtain the corresponding first sinusoidal feature map without metal artifacts, the second sinusoidal feature map with metal artifacts, and the metal trajectory feature map. The metal trajectory feature map is used to remove the metal trajectory values ​​in the second sinusoidal feature map to obtain the third sinusoidal feature map without metal trajectory values. The third sinusoidal feature map is used as input, and the first sinusoidal feature map is used as the ground truth value of the network. The neural network model is trained by combining the loss function. After training, a neural network model that can remove metal artifacts in CTA images is obtained. The trained neural network model was used to remove metal artifacts from the postoperative CTA images to be processed, resulting in CTA images with metal artifacts removed. Mango was used to adjust the contrast of the CTA images after removing metal artifacts, resulting in CTA images that visualize blood vessels.

2. The image processing method for CTA images according to claim 1, characterized in that, The default registration reference is the patient's skull.

3. The image processing method for CTA images according to claim 1, characterized in that, The loss function is as follows: ; in, The restored sine feature map; The first sinusoidal feature map is unaffected by metal artifacts. This is a feature map of the metal trajectory.

4. The image processing method for CTA images according to claim 1, characterized in that, Convolutional neural networks include multi-head attention mechanism models and feedforward propagation network models.

5. The image processing method for CTA images according to claim 4, characterized in that, The multi-head attention mechanism model uses multi-channel convolutional layers to perform depthwise separable convolution operations on the input sinusoidal feature map to obtain query vector, key vector, and value vector; The query vector and the key vector are transposed and then multiplied by a dot product. The result is then divided by a preset scaling factor and scaled. Finally, the scaled result is normalized using the SoftMax function to obtain the attention weights. The attention matrix is ​​obtained by weighted summation of the attention weights and the value vectors. The attention matrix is ​​reduced in dimensionality using 1×1 convolution, and the result is concatenated with the input vector residual to obtain the sinusoidal feature map output by the multi-head attention mechanism model.

6. The image processing method for CTA images according to claim 5, characterized in that, The feedforward propagation network model includes a first linear projection layer and a second linear projection layer; The first linear projection layer uses the GELU nonlinear activation function to gate the input sinusoidal feature map, resulting in a gated weight map. The second linear projection layer performs depthwise convolution on the input sinusoidal feature map; The result of the depthwise convolution is multiplied by the gated weight map to obtain the sinusoidal feature map output by the feedforward propagation network model.

7. An image processing system for CTA images, characterized in that, This system is used in the image processing method for CTA images according to any one of claims 1-6, the system comprising: The feature map acquisition module is used to acquire a first sinusoidal feature map without the influence of metal artifacts, a second sinusoidal feature map with the influence of metal artifacts, and a metal trajectory feature map. The model training module is used to remove the metal trajectory values ​​in the second sinusoidal feature map using the metal trajectory feature map, so as to obtain the third sinusoidal feature map without metal trajectory values. The third sinusoidal feature map is used as input and combined with the loss function to train the neural network model. After training, a neural network model that can remove metal artifacts in CTA images is obtained. The artifact removal module is used to remove metal artifacts from the postoperative CTA image to be processed using the trained neural network model, so as to obtain a CTA image with metal artifacts removed. The vessel display module is used to adjust the contrast of the CTA image after removing metal artifacts using Mango, so as to obtain a CTA image displaying blood vessels.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Segmentation method and system of tiny solid allogenic tissue, electronic equipment and medium

    CN115170802A

  • Method and device for reducing metal artifacts in tomography image

    CN113592968A

  • Image processing method and device for artifact removal

    CN115131452A