Acute aortic syndrome early warning method, system and device and storage medium

Abstract

The invention discloses an acute aortic syndrome early warning method, system and device and a storage medium, and belongs to the technical field of medical image processing. The method comprises the steps that an aorta CT image is acquired and preprocessed; generating a CTA image by using the cascade generative adversarial network model; and carrying out key information feature extraction on the CTA image by using a deep learning classification model DenseNet, and outputting an aortic syndrome classification result. According to the method, the high-quality CTA image can be generated without using a contrast agent, the risk possibly generated by contrast agent examination of a patient is reduced, and the method is particularly suitable for patients with renal function impairment or contrast agent allergy; compared with the prior art, an economical and efficient alternative scheme is provided, dependence on traditional CTA examination equipment and professionals is reduced, meanwhile, the accuracy rate and the timeliness rate of early warning of the acute aortic syndrome can be effectively increased, and remarkable clinical application value and social benefits are achieved.

Description

Technical Field

[0001] The present invention relates to an early warning method, system, device and storage medium for acute aortic syndrome, and belongs to the technical field of medical image processing. Background Art

[0002] Acute Aortic Syndrome (AAS) is a group of life-threatening aortic emergencies, mainly including aortic dissection, intramural hematoma of the aorta and penetrating aortic ulcer. These diseases have similar pathophysiological mechanisms, that is, the integrity of the aortic wall is damaged, resulting in blood entering the aortic wall layer or wall, which may lead to fatal complications such as aortic rupture.

[0003] At present, Computed Tomography Angiography (CTA) is recognized as the gold standard for diagnosing AAS. CTA can clearly show the aortic lumen, wall and its surrounding structures, and the sensitivity to aortic dissection can reach 90-100%, and the specificity can reach 87-100%. However, CTA examination has some limitations: First, it requires intravenous injection of iodine contrast agent, which may pose a risk of contrast-induced nephropathy to patients with renal insufficiency; Second, some patients may be allergic to iodine contrast agent, and in severe cases, even anaphylactic shock may occur; In addition, CTA examination requires special equipment and trained technical personnel, which may delay the diagnosis time in emergency situations.

[0004] In contrast, non-contrast CT is often used as a preliminary screening method for emergency patients due to its advantages such as simplicity, rapidity, and no need for contrast agent. Non-contrast CT has unique value in the diagnosis of intramural hematoma of the aorta, and can clearly show aortic wall thickening (>5mm) and high-density crescent shadow. However, the overall diagnostic sensitivity of non-contrast CT for AAS is relatively low, only 60-80%, especially in the diagnosis of aortic dissection, false negative results are prone to occur. These limitations may lead to missed diagnosis or misdiagnosis, delaying the timely treatment of patients. Summary of the Invention

[0005] In order to improve the accuracy and timeliness of early warning of acute aortic syndrome, the present invention provides an early warning method, system, device and storage medium for acute aortic syndrome, and the technical solutions are as follows:

[0006] The present invention provides an early warning method for acute aortic syndrome, and the method includes:

[0007] Step 1: Obtain aortic CT images;

[0008] Step 2: Preprocess the aortic CT images and extract the region of interest containing the aorta;

[0009] Step 3: Based on the image obtained in Step 2, use a cascaded generative adversarial network model to generate CTA images;

[0010] The generative adversarial network model includes a generator, a discriminator, and a frequency perception module. The frequency perception module uses a frequency feature extraction unit and a convolution module, and simultaneously integrates a central residual connection mechanism to extract frequency information. The frequency feature extraction unit uses discrete cosine transform to obtain different frequency components, and the calculation formula for each frequency component is expressed as:

[0011]

[0012] where f h,w is the two-dimensional DCT spectrum, h ∈ {0, 1..., h - 1}, w ∈ {0, 1..., w - 1}x i,j is the input signal, and i and j are the corresponding height and width;

[0013] Step 4: Use the deep learning classification model DenseNet to extract key information features from the CTA images generated in Step 3, and output the classification results of aortic syndrome.

[0014] Optionally, in Step 3, the calculation formula for using the generator to generate an image is expressed as:

[0015] G(z; θ G ) = σ(W L h L-1 + b L )

[0016] where z is a random noise vector, σ is an activation function, W L is a weight matrix, b L is a bias vector, and h L-1 = σ(W L- 1h L-2 + b L-1 ) is calculated by the neurons in the previous layer;

[0017] The generator G is optimized by minimizing its loss function, and the formula is as follows:

[0018]

[0019] where D(G(z)) is the discriminator's judgment on the generated image, indicating the probability that the generated image G(z) is judged as "real". Optionally, in Step 3, the calculation formula for using the discriminator to discriminate an image is expressed as:

[0020] D(x; θ D ) = σ(U M k M-1 + vM )

[0021] where x represents the input image, U M is the weight matrix, v M is the bias vector, k M-1 = σ(U M-1 k M-2 + v M-1 ) is calculated by the neurons in the previous layer;

[0022] The objective of the discriminator D is to maximize the probability of correctly judging real and generated images, and the formula is as follows:

[0023]

[0024] where p data (x) is the distribution of real data, D(x) is the probability that the discriminator identifies the real image x as "real", and D(G(z)) is the probability that the discriminator identifies the generated image G(z) as "real".

[0025] Optionally, the deep learning classification model DenseNet includes a plurality of densely connected blocks and transition layers connected in sequence. The densely connected block includes a batch normalization layer, an activation function, and a convolutional layer connected in sequence; the transition layer includes a batch normalization layer, an activation function, a convolutional layer, and an average pooling layer connected in sequence.

[0026] The present invention provides an acute aortic syndrome warning system, and the system includes:

[0027] An image acquisition module configured to acquire aortic CT images;

[0028] A preprocessing module configured to preprocess the aortic CT images and extract the region of interest containing the aorta;

[0029] A generative adversarial network module configured to generate CTA images based on the preprocessed images by using a cascaded generative adversarial network model;

[0030] The generative adversarial network model includes a generator, a discriminator, and a frequency perception module. The frequency perception module uses a frequency feature extraction unit and a convolutional module, and simultaneously integrates a central residual connection mechanism to extract frequency information. The frequency feature extraction unit uses discrete cosine transform to obtain different frequency components, and the calculation formula of each frequency component is expressed as:

[0031]

[0032] where f h,w is the two-dimensional DCT spectrum, h ∈ {0, 1..., h - 1}, w ∈ {0, 1..., w - 1}x i,jis the input signal, where i and j are the corresponding height and width;

[0033] An acute aortic syndrome warning module, configured to extract key information features from the CTA images generated in the said step 3 by using the deep learning classification model DenseNet, and output an aortic syndrome classification result.

[0034] Optionally, the calculation formula for the generative adversarial network module to generate an image by using a generator is expressed as:

[0035] G(z; θ G ) = σ(W L h L-1 + b L )

[0036] where z is a random noise vector, σ is an activation function, W L is a weight matrix, b L is a bias vector, and h L-1 = σ(W L- 1h L-2 + b L-1 ) is calculated by the neurons in the previous layer;

[0037] The generator G is optimized by minimizing its loss function, and the formula is as follows:

[0038]

[0039] where D(G(z)) is the discriminator's judgment on the generated image, representing the probability that the generated image G(z) is judged as "real". Optionally, the calculation formula for the generative adversarial network module to discriminate an image by using a discriminator is expressed as:

[0040] D(x; θ D ) = σ(U M k M-1 + v M )

[0041] where x represents the input image, U M is a weight matrix, v M is a bias vector, and k M-1 = σ(U M-1 k M-2 + v M-1 ) is calculated by the neurons in the previous layer;

[0042] The goal of the discriminator D is to maximize the probability of correctly judging real and generated images, and the formula is as follows:

[0043]

[0044] where p dataLet \(p_{data}(x)\) be the distribution of real data, \(D(x)\) be the probability that the discriminator classifies the real image \(x\) as "real", and \(D(G(z))\) be the probability that the discriminator classifies the generated image \(G(z)\) as "real".

[0045] Optionally, the deep learning classification model DenseNet includes a plurality of densely connected blocks and transition layers connected in sequence. The densely connected block includes a batch normalization layer, an activation function, and a convolutional layer connected in sequence; the transition layer includes a batch normalization layer, an activation function, a convolutional layer, and an average pooling layer connected in sequence.

[0046] The present invention provides an acute aortic syndrome warning device, including a memory and a processor;

[0047] The memory is used to store a computer program;

[0048] The processor is used to implement the acute aortic syndrome warning method described in any one of the above when executing the computer program.

[0049] The present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, the acute aortic syndrome warning method described in any one of the above is implemented.

[0050] The beneficial effects of the present invention are:

[0051] The present invention adopts the generative adversarial network (GAN) technology to generate high-quality CTA images from low-dose non-contrast CT images, and further constructs a deep learning model to efficiently warn of acute aortic syndrome (AAS). High-quality CTA images can be generated without using contrast agents, reducing the risks that patients may face during contrast agent examinations, especially suitable for patients with impaired renal function or contrast agent allergies; the verification results show that the model constructed by the present invention can effectively classify the condition, and then achieve rapid and accurate AAS warning, which can effectively improve the sensitivity and specificity of diagnosis.

[0052] In addition, the present invention uses a modular design, which is convenient for direct integration with existing medical imaging systems, improving the flexibility and promotion ability of practical applications, optimizing the entire process of image generation and diagnosis, enhancing the robustness and reliability of the model, and is particularly suitable for high-risk medical applications such as emergency scenarios and preoperative screening; compared with the prior art, the present invention provides an economical and efficient alternative solution, reducing the dependence on traditional CTA examination equipment and professional personnel, and having significant clinical application value and social benefits. Description of the Drawings

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of an acute aortic syndrome early warning method provided in the first embodiment of the present invention.

[0055] Figure 2 It is a schematic structural diagram of a generative adversarial network model provided in the first embodiment of the present invention.

[0056] Figure 3 It is a schematic structural diagram of a cascaded GAN model provided in the first embodiment of the present invention.

[0057] Figure 4 It is a schematic structural diagram of a frequency perception module of a GAN model provided in the first embodiment of the present invention.

[0058] Figure 5 It is a schematic structural diagram of a deep learning DenseNet classification network model provided in the first embodiment of the present invention.

[0059] Figure 6 It is an example diagram of a synthetic aortic CTA of the present invention. Specific Embodiments

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the accompanying drawings.

[0061] Embodiment 1:

[0062] This embodiment provides an acute aortic syndrome early warning method for generating CTA images based on CT plain scan images. Refer to Figure 1 , the method includes:

[0063] Step 1: Obtain aortic CT images.

[0064] The aortic CT images include aortic dissection, intramural hematoma of the aortic wall, and CT images of penetrating aortic ulcer.

[0065] Step 2: Preprocess the aortic CT images and extract the region of interest containing the aorta.

[0066] Data preprocessing includes: window width and window level adjustment, image normalization, image enhancement, and ROI extraction; specifically, for different types of CT images, the window width and window level are adjusted according to different standards to highlight more tissue structures and ensure that the model can recognize more features; image normalization converts the image pixel values to a unified range to avoid increasing the difficulty of model training due to different image ranges; image enhancement methods include rotation, translation, scaling, flipping, etc., which are used to enhance the generalization ability of the model; ROI extraction is to extract the region of interest containing the aorta.

[0067] Step 3: Based on the preprocessed image, use the cascaded generative adversarial network model to generate CTA images.

[0068] Among them, the generative adversarial network (GAN) module includes a generator, a discriminator, and a frequency perception module, and through the game learning between the generator and the discriminator, good outputs are produced.

[0069] The generator is used to learn from the data distribution and generate images as similar as possible to the training data. The generator is usually represented by G. The calculation formula for using the generator to generate images is expressed as:

[0070] G(z; θ G ) = σ(W L h L-1 +b L )

[0071] In the above formula, z is a random noise vector, usually sampled from a simple distribution, such as a standard normal distribution or a uniform distribution, and mapped to the generated image space. θ G represents the set of parameters of the generator model. σ is the activation function. W L is the weight matrix. b L is the bias vector. h L-1 = σ(W L-1 h L-2 +b L-1 ) is the hidden state calculated by the neurons in the previous layer.

[0072] The generator G is optimized by minimizing its loss function. The formula is as follows:

[0073]

[0074] Among them, D(G(z)) is the discriminator's judgment on the generated image, indicating the probability that the generated image G(z) is judged as "real". represents the expectation of the noise vector z following the distribution p z (z). p z (z) represents the prior distribution of the noise vector z.

[0075] The discriminator is used to distinguish whether the generated samples are highly similar to the paired reference images or false ones that are different from the reference images. The discriminator is usually denoted by D, and the formula for using the discriminator to discriminate images is expressed as:

[0076] D(x; θ D ) = σ(U M k M-1 + v M )

[0077] In the above formula, x represents the input image, which comes from the real data distribution or the image generated by the generator, and θ D represents the set of parameters of the discriminator model, U M is the weight matrix, v M is the bias vector, k M-1 = σ(U M-1 k M-2 + v M-1 ) is the hidden state calculated by the neurons in the previous layer.

[0078] The goal of the discriminator D is to maximize the probability of correctly judging real and generated images, and the formula is as follows:

[0079]

[0080] where p data (x) is the distribution of the real data, D(x) is the probability that the discriminator recognizes the real image x as "real", represents the expectation that the real sample x follows the real data distribution, D(G(z)) is the probability that the discriminator recognizes the generated image G(z) as "real", represents the expectation that the noise vector z follows the noise distribution p z (z).

[0081] Through the adversarial optimization of the generator and the discriminator, the generator continuously improves the quality of the generated images, while the discriminator continuously improves the ability to distinguish true and false images. The optimization goal of the entire network is expressed by the following formula:

[0082]

[0083] The frequency perception module converts the input image into a feature map with precise resolution through blocks with residual connections. Each block involves frequency feature extraction and convolutional blocks, and there is a central residual connection. This design of frequency extraction enhances the model's ability to capture complex details and further enhances the perception ability of microvascular generation. The residual connection helps to effectively learn and distinguish local and global features, contributing to the evaluation of image authenticity.

[0084] Among them, frequency feature extraction obtains different frequency components through the discrete cosine transform (DCT), and each global frequency component can be expressed as:

[0085]

[0086] In the above formula, f h,w is the two-dimensional DCT spectrum, H represents the height of the input signal, W represents the width of the input signal, x i,j is the input signal, i and j are the corresponding height and width, h represents the frequency component index in the vertical direction, w represents the frequency component index in the horizontal direction, c(h) and c(w) respectively represent the normalization coefficients of the frequency components in the vertical and horizontal directions, H ∈ {0, 1..., h - 1}, W ∈ {0, 1..., w - 1}.

[0087] Furthermore, the input channel X is divided into multiple parts [X 0 , X 1 ,..., X n-1 , and the corresponding two-dimensional DCT frequency components are assigned to each part to obtain each local frequency component, which is calculated as follows:

[0088]

[0089] Among them, represents the pixel value of channel X i at position (h, w), represents the weight of the DCT basis function at frequency (u, v) and spatial position (h, w), [u, v] is the frequency component index corresponding to the input variable X i , and the cosine calculation formula corresponding to p is as follows:

[0090]

[0091] Finally, the complete frequency can be expressed as:

[0092] F = cat([F 0 , F 1 ,..., F n-1 )

[0093] Among them, cat represents the concatenation operation, and the attention feature vector of the frequency domain channel attention mechanism adopting the DCT mode can be further expressed as:

[0094] Att dct = sigmoid(f c (F))

[0095] In the above formula, the left side represents the attention feature vector, and the right side represents the complete connection layer. f c (·) represents the fully connected layer.

[0096] Step 4: Use the deep learning classification model DenseNet to extract key information features from the CTA images generated in Step 3, and output the classification results of aortic syndrome.

[0097] In this embodiment, a classification experiment will be conducted on the synthetic CTA images of the aorta, and the classification model used is DenseNet121. As Figure 5 shown, the DenseNet network has 4 stages, and each stage contains a dense block and a transition layer (the last stage only has a dense block). The initial input image is x input , and the output after the first dense block and transition layer is z1, and the output after the second dense block and transition layer is z2, and so on.

[0098] The dense block consists of a batch normalization layer, a ReLU activation function, and a convolutional layer, and the specific formula is expressed as:

[0099] x l = Conv l (ReLU(BN l (x l-1 )))

[0100] where x l-1 is the input of the l-th layer. When l = 1, x l-1 = x0.

[0101] The transition layer consists of a batch normalization layer, a ReLU activation function, a convolutional layer, and an average pooling layer, and the specific formula is expressed as:

[0102] z = AvgPool(Conv 1×1 (ReLU(BN trans (y L ))))

[0103] where y L is the output of the last layer of the dense block, BN trans is the batch normalization layer, Conv 1×1 is a convolutional layer with a size of 1×1, AvgPool is the average pooling layer, the pooling kernel size is 2×2, and the stride is 2.

[0104] In this embodiment, CTA images are generated based on non-contrast-enhanced CT images, and a deep learning model is further constructed to warn of acute aortic syndrome (AAS). High-quality CTA images can be generated without using contrast agents, reducing the risks that patients may suffer from contrast agent examinations. It is especially suitable for patients with impaired renal function or contrast agent allergies; using deep learning technology, it provides fast and accurate diagnosis of AAS, effectively improving the sensitivity and specificity of diagnosis.

[0105] This embodiment uses a modular design, which is convenient for direct integration with existing medical imaging systems, improving the flexibility and promotion ability of practical applications. It can optimize the entire process of image generation and diagnosis, enhancing the robustness and reliability of the model. It is particularly suitable for high-risk medical applications such as emergency scenarios and preoperative screening. Compared with the prior art, the present invention provides an economical and efficient alternative solution, reducing the dependence on traditional CTA examination equipment and professional personnel, and having significant clinical application value and social benefits.

[0106] Embodiment Two:

[0107] This embodiment provides an acute aortic syndrome warning system, which is characterized in that the system includes:

[0108] An image acquisition module, configured to acquire aortic CT images.

[0109] A preprocessing module, configured to preprocess the aortic CT images and extract the region of interest containing the aorta.

[0110] A generative adversarial network module, configured to generate CTA images using a cascaded generative adversarial network model based on the preprocessed images.

[0111] The generative adversarial network model includes a generator, a discriminator, and a frequency perception module. The frequency perception module uses a frequency feature extraction unit and a convolutional module, and simultaneously integrates a central residual connection mechanism to extract frequency information. The frequency feature extraction unit uses the discrete cosine transform to obtain different frequency components, and the calculation formula for each frequency component is expressed as:

[0112]

[0113] where f h,w is the two-dimensional DCT spectrum, h ∈ {0, 1..., h - 1}, w ∈ {0, 1..., w - 1}x i,j is the input signal, and i and j are the corresponding height and width.

[0114] An acute aortic syndrome warning module, configured to extract key information features from the CTA images generated in step 3 using the deep learning classification model DenseNet, and output the classification result of aortic syndrome.

[0115] Embodiment Three:

[0116] This embodiment verifies and explains the technical effect of the present invention in warning aortic syndrome based on synthetic CTA images. Other image synthesis methods are selected for comparison with the synthesis method of the present invention to verify the superiority of the synthesis method proposed by the present invention.

[0117] In this embodiment, an image synthesis experiment was performed on aortic dissection images. 3D computed tomography images of available aortic dissections were collected, with 80 used as samples and the remaining 20 used as test samples.

[0118] The synthetic aortic CTA images of the present invention were compared with traditional methods, and the comparison results are shown in Table 1 below:

[0119] Table 1: Comparison results of the synthetic image method of the present invention with other synthetic image methods

[0120] model PSNR (dB) metric SSIM metric Pix2pix 30.24 0.9793 Pix2pixHD 32.09 0.9823 the present invention 33.94 0.9912

[0121] In this embodiment, the proposed method was compared and analyzed using the Pix2pix and Pix2pixHD synthetic image models. All benchmark models were from the source codes provided by the relevant authors. The dataset of aortic dissection was used for training and testing, and the corresponding objective evaluation index values were measured. From the above comparison data, it can be seen that the comprehensive effect of the synthetic image method of the present invention on the evaluation indexes is better than other methods.

[0122] Example 4:

[0123] In this embodiment, the synthetic CTA images were used to construct the pre-training data queue of the deep learning model. At the same time, public CTA images were also collected to construct the training and validation queues. In order to verify that the synthetic images of the present invention can achieve a training effect similar to real images, this embodiment was compared with the model that did not participate in pre-training, and 56 patient samples available publicly were used as the comparative experiment data.

[0124] The present invention generates CTA images based on non-contrast-enhanced CT images and real CTA images for early warning of acute aortic syndrome diagnosis, and the comparison results are shown in Table 2 below:

[0125] Table 2 Comparison of early warning effects between generated CTA images and real CTA images

[0126] comparison method ACC AUC sensitivity specificity real image training 0.942 0.981 0.902 0.963 synthetic image pre-training 0.975 0.993 0.917 0.989

[0127] From the above comparison data, it can be seen that the indexes of the model after pre-training with synthetic CTA images are higher than those without pre-training, and it performs well in the verification comparison. It proves that the artificial intelligence model of the present invention that generates CTA images based on non-contrast-enhanced CT images and is used for early warning of acute aortic syndrome can improve the diagnostic rate and accuracy of the existing acute aortic syndrome diagnosis methods, is applicable to patients without the need for contrast agent examination, and has broad clinical application prospects.

[0128] Some steps in the embodiments of the present invention can be implemented by software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk, etc.

[0129] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An early warning method for acute aortic syndrome, characterized in that, The method includes: Step 1: Obtain an aortic CT image; Step 2: Preprocess the aortic CT image to extract the region of interest containing the aorta; Step 3: Based on the image obtained in Step 2, use a cascaded generative adversarial network model to generate a CTA image; The generative adversarial network model includes a generator, a discriminator, and a frequency perception module. The frequency perception module uses a frequency feature extraction unit and a convolutional module, and at the same time fuses a central residual connection mechanism to extract frequency information. The frequency feature extraction unit uses a discrete cosine transform to obtain different frequency components, and the calculation formula for each frequency component is expressed as: where f h,w is the two-dimensional DCT spectrum, h ∈ {0, 1..., h - 1}, w ∈ {0, 1..., w - 1}x i,j is the input signal, and i and j are the corresponding height and width; Step 4: Use the deep learning classification model DenseNet to extract key information features from the CTA image generated in Step 3 and output the classification result of aortic syndrome.

2. The acute aortic syndrome early warning method according to claim 1, wherein, In Step 3, the calculation formula for using the generator to generate an image is expressed as: G(z; θ G ) = σ(W L h L-1 + b L ) where, z is a random noise vector, σ is the activation function, W L is the weight matrix, b L is the bias vector, h L-1 = σ(W L-1 h L-2 + b L-1 ) is calculated by the neurons in the previous layer; The generator G is optimized by minimizing its loss function, and the formula is as follows: where D(G(z)) is the discriminator's judgment on the generated image, indicating the probability that the generated image G(z) is judged to be "real".

3. The acute aortic syndrome early warning method according to claim 1, wherein, In Step 3, the calculation formula for using the discriminator to discriminate an image is expressed as: D(x; θ D ) = σ(U M k M-1 + v M ) where x represents the input image, U M is the weight matrix, v M is the bias vector, k M-1 = σ(U M-1 k M-2 + v M-1 ) is calculated by the neurons in the previous layer; The goal of the discriminator D is to maximize the probability of correctly judging real and generated images, and the formula is as follows: where p data (x) is the distribution of real data, D(x) is the probability that the discriminator recognizes the real image x as "real", and D(G(z)) is the probability that the discriminator recognizes the generated image G(z) as "real".

4. The acute aortic syndrome early warning method according to claim 1, characterized in that, The deep learning classification model DenseNet includes a plurality of densely connected blocks and transition layers connected in sequence. The densely connected block includes a batch normalization layer, an activation function, and a convolutional layer connected in sequence; the transition layer includes a batch normalization layer, an activation function, a convolutional layer, and an average pooling layer connected in sequence.

5. An acute aortic syndrome early warning system, characterized in that, The system includes: An image acquisition module configured to obtain an aortic CT image; A preprocessing module configured to preprocess the aortic CT image and extract the region of interest containing the aorta; A generative adversarial network module configured to generate a CTA image based on the preprocessed image using a cascaded generative adversarial network model; The generative adversarial network model includes a generator, a discriminator, and a frequency perception module. The frequency perception module uses a frequency feature extraction unit and a convolutional module, and at the same time fuses a central residual connection mechanism to extract frequency information. The frequency feature extraction unit uses a discrete cosine transform to obtain different frequency components, and the calculation formula for each frequency component is expressed as: where f h,w is the two-dimensional DCT spectrum, h ∈ {0, 1..., h - 1}, w ∈ {0, 1..., w - 1} x i,j is the input signal, and i and j are the corresponding height and width; An acute aortic syndrome warning module configured to use the deep learning classification model DenseNet to extract key information features from the CTA image generated in Step 3 and output the classification result of aortic syndrome.

6. The acute aortic syndrome early warning system according to claim 5, wherein The calculation formula for the generative adversarial network module to use the generator to generate an image is expressed as: G(z; θ G ) = σ(W L h L-1 + b L ) where, z is a random noise vector, σ is the activation function, W L is the weight matrix, b L is the bias vector, h L-1 = σ(W L-1 h L-2 + b L-1 ) is calculated by the neurons in the previous layer; The generator G is optimized by minimizing its loss function, and the formula is as follows: where D(G(z)) is the discriminator's judgment on the generated image, indicating the probability that the generated image G(z) is judged to be "real".

7. The acute aortic syndrome warning system according to claim 5, characterized in that, The calculation formula for the generative adversarial network module to use the discriminator to discriminate an image is expressed as: D(x; θ D ) = σ(U M k M-1 + v M ) where x represents the input image, U M is the weight matrix, v M is the bias vector, k M-1 = σ(U M-1 k M-2 + v M-1 ) is calculated by the neurons in the previous layer; The goal of the discriminator D is to maximize the probability of correctly judging real and generated images, and the formula is as follows: where p data (x) is the distribution of real data, D(x) is the probability that the discriminator identifies the real image x as "real", and D(G(z)) is the probability that the discriminator identifies the generated image G(z) as "real".

8. The acute aortic syndrome early warning system according to claim 5, wherein, The deep learning classification model DenseNet includes a plurality of densely connected blocks and transition layers connected in sequence. The densely connected block includes a batch normalization layer, an activation function, and a convolutional layer connected in sequence. The transition layer includes a batch normalization layer, an activation function, a convolutional layer, and an average pooling layer connected in sequence.

9. An acute aortic syndrome warning device, characterized in that, It includes a memory and a processor; The memory is used for storing a computer program; The processor is used for implementing the acute aortic syndrome early warning method according to any one of claims 1 to 4 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by the processor, the acute aortic syndrome early warning method according to any one of claims 1 to 4 is implemented.

Images