Aortic dissection recognition model construction method, aortic dissection recognition method, equipment and related device

Abstract

The invention discloses an aortic dissection recognition model construction method, an aortic dissection recognition method, aortic dissection recognition equipment and a related device, and relates to the technical field of deep learning, and the model construction method comprises the steps: carrying out the dual-mode matching of a CT plain-scan image and a CTA image of the same period marked with an aortic dissection lesion feature part, and carrying out the recognition of the aortic dissection lesion feature part; determining the same fault with a lesion feature part between the two, and positioning and marking the corresponding lesion feature part on the CT plain-scan image of the same fault; and forming a CT plain-scan image data set by the plurality of CT plain-scan images marked with the diseased regions and the plurality of CT plain-scan images of the normal aortic patients, training a deep learning model by adopting the CT plain-scan image data set, and taking the trained deep learning model as an aortic dissection recognition model. And the aortic dissection recognition model is used for performing aortic dissection recognition on the to-be-recognized CT plain-scan image. The aortic dissection recognition accuracy can be maintained, and meanwhile, the recognition cost can be reduced.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a method for constructing an aortic dissection identification model, an identification method, equipment, and related devices. Background Art

[0002] Aortic dissection (AD) is an acute and critical disease that seriously endangers human life and health. It has clinical characteristics such as rapid onset, rapid progression, severe condition, and high mortality rate, and is extremely easy to miss or misdiagnose. The disease is mainly caused by various reasons that lead to damage to the aortic intima. Blood flows into the media layer through the intima tear, causing the aortic intima to tear from the media, and the lumen is separated into the true lumen and the false lumen by the aortic intima sheet. Subsequently, under the influence of hemodynamics and other factors, the scope of the intima tear may further expand. When the branch vessels are involved, it may cause dysfunction of the corresponding organs. If the disease worsens further, it is often life-threatening. Early and active control of heart rate and blood pressure can significantly reduce the patient's mortality rate and inhibit the progression of the disease, so early identification of the disease is very necessary.

[0003] Early identification and disease classification of AD rely more on medical imaging. The purpose of the examination is to conduct a comprehensive evaluation of the entire aorta, including the scope of AD involvement, morphology, diameter of the aorta in different parts, involvement of the aortic valve and its branches, and relationship with surrounding tissues. Aortic intimal patches and intimal ruptures are the main signs for diagnosing AD. Commonly used imaging techniques in clinical practice include: echocardiography, computed tomography (CT) plain scan, CT angiography (CTA), magnetic resonance imaging (MRI), digital subtraction angiography (DSA), and CT angiography. While these imaging technologies each have their own advantages and disadvantages, echocardiography offers advantages such as simplicity and speed, but its diagnostic accuracy is low and it can easily lead to missed diagnoses. CT scans are fast and noninvasive, but their accuracy is relatively low. Only a few typical cases of AD can be identified by CT, and experienced clinical experts are required to increase the disease's detection rate. Computed tomography (CTA) offers the highest accuracy, can precisely locate the aortic intimal patch in most AD patients, and provides guidance for treatment. Therefore, the European Society of Cardiology recommends CTA as the first test for suspected patients. However, it carries risks such as nephrotoxicity and ionizing radiation, and is expensive, requiring high equipment and personnel requirements. MRI offers high accuracy and is free of nephrotoxicity and ionizing radiation hazards, but its scanning time is long and it is contraindicated in patients with life-support devices and metal implants. DSA is an invasive test, typically performed intraoperatively, and its complex procedure and high cost limit its application in early clinical diagnosis. Therefore, a method for early AD detection with high accuracy and low cost is urgently needed. Summary of the Invention

[0004] The purpose of this application is to provide an aortic dissection identification model construction method, identification method, equipment and related devices, which can maintain the accuracy of aortic dissection identification while reducing the identification cost.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for constructing an aortic dissection identification model, comprising:

[0007] By performing bimodal matching on a CT plain scan image and a CTA image of the same period marked with characteristic lesion sites of aortic dissection, it is determined that there is a common section with characteristic lesion sites between the CT plain scan image and the CTA image, and the corresponding characteristic lesion sites are located and marked on the CT plain scan image of the same section;

[0008] The CT plain scan image dataset comprises a plurality of CT plain scan images with lesion sites marked thereon and a plurality of CT plain scan images of patients with normal aortas;

[0009] A deep learning model is trained using a CT plain scan image dataset, and the trained deep learning model is used as an aortic dissection recognition model. The aortic dissection recognition model is used to recognize aortic dissection on the CT plain scan images to be recognized.

[0010] Optionally, the characteristic lesion site includes at least one of an aortic intimal patch and an aortic intimal rupture.

[0011] Optionally, the shooting interval between the CT plain scan image in the CT plain scan image data set and the CTA image of the same period as the CT plain scan image does not exceed 12 hours.

[0012] Optionally, the deep learning model is a deep aggregation network structure based on a two-dimensional convolutional neural network.

[0013] Optionally, a deep learning model is trained using a CT plain scan image dataset, and the trained deep learning model is used as an aortic dissection recognition model, specifically including:

[0014] The back propagation algorithm and stochastic gradient descent method are used to update and iterate the deep learning model to obtain a trained deep learning model.

[0015] Optionally, the loss function when training a deep learning model is expressed as:

[0016] L=-∑ k y k logf(x k );

[0017] Among them, L is the loss function value, x k is the kth input of the deep learning model, f(x k ) represents the deep learning model, y k For the deep learning model x k The predicted output.

[0018] In a second aspect, the present application provides a device for constructing an aortic dissection identification model, the device comprising:

[0019] a dual-modality matching module, configured to perform dual-modality matching between a plain CT image and a contemporaneous CTA image marked with characteristic lesion sites of aortic dissection, to determine whether the CT plain image and the CTA image have the same slice with the characteristic lesion site, and to locate and mark the corresponding characteristic lesion site on the plain CT image of the same slice;

[0020] a CT plain scan image data set determination module, configured to form the CT plain scan image data set from a plurality of CT plain scan images with lesion sites marked thereon and a plurality of CT plain scan images of patients with normal aortas;

[0021] The model training module is used to train the deep learning model using a CT plain scan image dataset, and use the trained deep learning model as an aortic dissection recognition model, which is used to identify aortic dissection on the CT plain scan images to be identified.

[0022] In a third aspect, the present application provides an aortic dissection identification device, wherein the aortic dissection identification device includes a host computer, and the host computer includes an aortic dissection identification model constructed according to the aortic dissection identification model construction method.

[0023] In a fourth aspect, the present application provides a method for identifying aortic dissection, the method comprising:

[0024] Obtaining a CT plain scan image to be identified;

[0025] The CT plain scan image to be identified is input into the aortic dissection identification model constructed according to the aortic dissection identification model construction method to obtain an aortic dissection identification result, which is the presence of aortic dissection or the absence of aortic dissection.

[0026] In a fifth aspect, the present application provides an aortic dissection identification device, the aortic dissection identification device comprising:

[0027] A CT plain scan image acquisition module is used to obtain a CT plain scan image to be identified;

[0028] The recognition module is used to input the CT plain scan image to be identified into the aortic dissection recognition model constructed according to the above-mentioned aortic dissection recognition model construction method to obtain the aortic dissection recognition result, and the aortic dissection recognition result is the presence of aortic dissection or the absence of aortic dissection.

[0029] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0030] The present application provides a method for constructing an aortic dissection recognition model, a recognition method, equipment and related devices, which accurately locate the characteristics of aortic dissection based on CTA images that are synchronized with CT plain scan images, construct a CT plain scan image dataset, use the CT plain scan image dataset to train a deep learning model, and use the trained deep learning model as an aortic dissection recognition model. The accuracy of aortic dissection recognition using CTA images is improved, and when the aortic dissection recognition model is actually applied, there is no need to obtain CTA images, and only CT plain scan images are used for aortic dissection recognition, thereby reducing the recognition cost. The present application combines the advantages of CTA and CT examinations, and improves the accuracy of CT recognition of aortic dissection through CTA-weighted CT images. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 A schematic flow chart of a method for constructing an aortic dissection identification model provided in one embodiment of the present application;

[0033] Figure 2 A schematic diagram of the deep learning model training process provided in one embodiment of the present application;

[0034] Figure 3 A schematic diagram of a CTA image and CT plain scan results provided in one embodiment of the present application;

[0035] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0037] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0038] This application provides a method for constructing an aortic dissection identification model, such as Figure 1 As shown, the aortic dissection identification model construction method includes:

[0039] Step 101: By performing bimodal matching on a CT plain scan image and a CTA image of the same period marked with characteristic lesion sites of aortic dissection, it is determined that there is an identical section with characteristic lesion sites between the CT plain scan image and the CTA image, and the corresponding characteristic lesion sites are located and marked on the CT plain scan image of the same section.

[0040] Step 102: A plurality of CT plain scan images with lesion sites marked and a plurality of CT plain scan images of patients with normal aortas constitute the CT plain scan image dataset.

[0041] Step 103: Use the CT plain scan image dataset to train the deep learning model, and use the trained deep learning model as an aortic dissection recognition model, which is used to perform aortic dissection recognition on the CT plain scan image to be recognized.

[0042] This application constructs a CT plain scan image dataset based on CTA images that are acquired at the same time as the CT plain scan images. Figure 2 As shown in the figure, a deep learning model is trained using a CT plain scan image dataset, and the trained deep learning model is used as an aortic dissection recognition model. The accuracy of CTA images in recognizing aortic dissection is utilized to improve the recognition accuracy of the aortic dissection recognition model. When the aortic dissection recognition model is actually applied, there is no need to obtain CTA images, and only CT plain scan images are used for aortic dissection recognition, thereby reducing the recognition cost.

[0043] In an exemplary embodiment, the lesion site includes the site of the aortic intimal patch and / or the site of the aortic intimal rupture, that is, the characteristic lesion site includes at least one of the sites of the aortic intimal patch and the aortic intimal rupture, and if the lesion site exists, aortic dissection exists.

[0044] In an exemplary embodiment, target image data is retrieved from the hospital information department's image database, including corresponding CTA and CT plain scan image sequences of AD patients and normal aorta patients during the same period. A CT plain scan image dataset is constructed through a dual-modality matching method, specifically including:

[0045] Step a. For CTA images showing suspected aortic intimal patches (with or without intimal ruptures), the slices and specific areas showing suspected AD characteristic images are selected and marked by expert manual identification.

[0046] Step b. For the CT plain scan image, dual-modality matching is performed between it and a CTA image from the same source that shows a characteristic image of a suspected aortic intimal patch (including / excluding intimal rupture), and the slice and specific area containing the characteristic image of the aortic intimal patch (including / excluding intimal rupture) in the CT image matched with the CTA image are automatically annotated.

[0047] Step c. For the image data of the slices and specific parts marked in steps a and b, crop several slice sequences at the same slice interval to obtain 1250*1250 vectors. Then, normalize the 1250*1250 vectors and use the normalized results as the CT plain scan image dataset.

[0048] The specific process of normalization is: mapping the 1250*1250 vector value to [0,1].

[0049] Step d. Divide the CT plain scan image dataset into a training set, a validation set, and a test set, with the data volume ratio of the training set, validation set, and test set being 8:1:1.

[0050] The deep learning model is a deep layer aggregation (DLA) network structure based on a two-dimensional convolutional neural network.

[0051] In the DLA network structure, the basic operation blocks and aggregation nodes are composed of multiple two-dimensional convolutional neural layers with a convolution kernel of 1 and a stride of 1, a batch normalization layer and a relu layer. The input of the deep learning model is the training set obtained above. The output layer of the deep learning model is the softmax layer, and the output is the aortic dissection recognition result. The aortic dissection recognition result is divided into two categories: the absence of aortic dissection (Class I aorta) or the presence of aortic dissection (Class II aorta). Class I aorta is the aorta with normal physiological structure, and Class II aorta is the aorta with AD lesion characteristics.

[0052] The basic operation blocks of the DLA structure are usually convolution blocks or residual blocks, and the aggregation nodes connect these operation blocks through two aggregation methods: iterative aggregation (IDA) and hierarchical aggregation (HAD), forming a deep neural network with powerful feature fusion capabilities.

[0053] The effect of the DLA structure on two-dimensional convolutional neural networks is mainly reflected in improving model performance, enhancing feature fusion capabilities, reducing the number of parameters and video memory overhead, etc.

[0054] The DLA structure's two-dimensional convolutional neural network is connected using two mechanisms: iterative aggregation (IDA) and hierarchical aggregation (HDA). The IDA mechanism connects features at different levels to fuse shallow underlying information with deep semantic information. In the DLA structure, IDA uses skip connections to bring features from lower layers to higher layers. These skip connections add connection and aggregation nodes outside of multiple stages, enabling effective cross-layer information fusion.

[0055] The HDA mechanism is used to fuse features from multiple blocks within the same hierarchy (i.e., within the same stage). In the DLA architecture, HDA preserves and combines feature channels across blocks and stages in a tree-like structure. HDA nodes combine shallower and deeper layers to learn richer combinations across more feature hierarchies. To improve computational efficiency, aggregation nodes at the same depth can be fused.

[0056] In a two-dimensional convolutional neural network with a DLA architecture, the IDA and HDA mechanisms complement each other. IDA connects multiple stages, while HDA fuses features within each stage. IDA and HDA share aggregation nodes, forming a cohesive network structure. The entire network achieves efficient information transfer and feature utilization by extracting features layer by layer, fusing features across layers, and fusing features within the same layer.

[0057] The CT plain-contrast image with the aortic dissection area marked is the CT plain-contrast image with the intimal sheet of the dissection.

[0058] In an exemplary embodiment, Figure 3 As shown, the CT scan results to be identified are as follows Figure 3 As shown in parts (b) and (d), Figure 3 Part (a) is the corresponding CTA image of (b). Figure 3 Part (c) is the corresponding CTA image of the same period as (d). The characteristic structure of aortic dissection lesions in the corresponding CTA image shows that the CT plain scan image recognition results can accurately identify whether there is aortic dissection, and its accuracy in identifying aortic dissection is comparable to that of CTA.

[0059] This application uses a two-dimensional convolutional neural network to directly process the obtained CT image data. After marking the basic image data required for the training set, it can eliminate the need for manpower to identify the aortic intimal patches and intimal ruptures in CT or CTA images, thereby enhancing the accuracy of CT image recognition and saving social resources.

[0060] This application is based on a two-dimensional convolutional neural network and uses the DLA network architecture to train a deep learning model. Using the deep learning model as a toolbox, it can automatically identify CT image features of the lesion without manually specifying them. The computer learns the recognition process on its own, greatly improving the recognition speed while cleverly combining the accuracy of CTA with the simplicity of CT scans. In actual application, the CT scan images of all layers of the patient are input into the aortic dissection recognition model one by one, and the recognition results are obtained separately, reducing information loss.

[0061] The main purpose of AD imaging examination is to identify the aortic intimal sheet (including intimal rupture). CTA is the primary examination recommended by the European Society of Cardiology for suspected patients. It can accurately show the aortic intimal sheet and intimal rupture site of most AD patients during acute onset. Therefore, CTA is currently used as a diagnostic measure for suspected AD patients in clinical practice. However, due to its high price, high equipment and personnel requirements, its application in grassroots hospitals is limited. Therefore, compared with CTA, CT plain scan is more popular in clinical practice. This application cleverly combines the advantages of the two and avoids the disadvantages, thereby improving the accuracy of CT in identifying intimal sheets.

[0062] In an exemplary embodiment, step 103 specifically includes: using a back propagation algorithm and a stochastic gradient descent method to iteratively update the deep learning model until the deep learning model converges to obtain a trained deep learning model.

[0063] The deep learning model is trained repeatedly to obtain multiple trained deep learning models. The model with the highest overlap between the recognition results and the known annotation results on the validation set and the test set is selected as the aortic dissection recognition model.

[0064] The loss function when training a deep learning model is expressed as:

[0065] L=-∑ k y k logf(x k );

[0066] Among them, L is the loss function value, x k is the kth input of the deep learning model, specifically a 12*1250 vector, f(x k ) represents the deep learning model, y k For the deep learning model x k The predicted output is a vector of dimension n, where n is the number of categories to be classified, representing the input data x k Category, y k =(y0,y1,...,y i ,...,y n ),y i∈{0,1}.y k The elements in x k y corresponding to the category i is 1, and the rest are 0. After data post-processing, y k Converted into aortic dissection recognition results, in this embodiment, k=1.

[0067] This application adopts a deep learning method that has the ability to self-learn effective information from CT plain scan images of lesions, and combines it with the accurate positioning ability of CTA diagnosis to mine complex correlation patterns in the data. Without using any hard-coded features, more efficient image feature extraction and more accurate lesion identification are achieved. The deep learning method used in this application is highly scalable and does not require the design of complex features for disease images based on professional medical knowledge. It only needs to provide a large amount of labeled data to input into the convolutional neural network. As the amount of data increases, the accuracy of the network in identifying lesions increases. The deep learning model used in this application uses a two-dimensional convolutional neural network as the basic network and selects DLA as the overall architecture, which has a good recognition effect on the image features of the lesion site. This deep learning model extracts CT image features corresponding to the CTA lesion site of patients with aortic dissection through two-dimensional convolution, obtains the spatial and geometric features of the aortic intimal patch and intimal rupture on the patient's CT plain scan images of the same period, completes morphological analysis, and obtains the mutual correlation between morphologies. It can be applied to the automatic recognition of the image features of the aortic intimal patch and intimal rupture based on CT images.

[0068] This application helps improve the detection rate of aortic dissection in primary hospitals by improving the accuracy of CT identification of aortic intimal sheets (including intimal ruptures), avoids the occurrence of zero-treatment transfers, reduces the risk of transfer of patients with aortic dissection, and reduces medical disputes. This application is mainly used for the early and rapid identification of aortic intimal sheets and intimal ruptures, in order to simplify the diagnosis and treatment process of aortic dissection, reduce missed diagnosis and misdiagnosis of the disease, and avoid further progression of the disease through early treatment, thereby greatly alleviating the patient's pain and reducing the socioeconomic burden.

[0069] This application uses the patient's CTA lesion site to automatically identify the corresponding site in the CT plain scan image, and uses this as the region of interest data to train the DLA model based on CTA-weighted CT image deep learning for the identification of aortic intimal patches, thereby improving the accuracy of CT plain scan identification of aortic intimal patches, in order to simplify the diagnosis and treatment process of aortic dissection, reduce the missed diagnosis rate and misdiagnosis rate, and guide early treatment.

[0070] Based on the same inventive concept, the present application also provides an aortic dissection identification model construction device for implementing the aforementioned aortic dissection identification model construction method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more of the following embodiments of the aortic dissection identification model construction device can be found in the above-mentioned limitations of the aortic dissection identification model construction method, and will not be repeated here.

[0071] In an exemplary embodiment, the present application provides a device for constructing an aortic dissection identification model, the device comprising:

[0072] The dual-modality matching module is used to perform dual-modality matching between a CT plain scan image and a CTA image of the same period marked with characteristic lesion sites of aortic dissection, determine whether there is an identical fault section with characteristic lesion sites between the CT plain scan image and the CTA image, and locate and mark the corresponding characteristic lesion sites on the CT plain scan image of the same fault section.

[0073] The CT plain scan image data set determination module is used to form the CT plain scan image data set from a plurality of CT plain scan images with lesion sites marked and a plurality of CT plain scan images of patients with normal aortas.

[0074] The model training module is used to train the deep learning model using a CT plain scan image dataset, and use the trained deep learning model as an aortic dissection recognition model, which is used to identify aortic dissection on the CT plain scan images to be identified.

[0075] In an exemplary embodiment, the present application provides an aortic dissection identification device, which includes a host computer, and the host computer includes an aortic dissection identification model constructed according to the aortic dissection identification model construction method.

[0076] In an exemplary embodiment, the present application provides a method for identifying aortic dissection, the method comprising:

[0077] Step 201: Obtain a CT plain scan image to be identified.

[0078] Step 202: Input the CT plain scan image to be identified into the aortic dissection identification model constructed according to the aortic dissection identification model construction method to obtain an aortic dissection identification result, wherein the aortic dissection identification result is the presence of aortic dissection or the absence of aortic dissection.

[0079] In an exemplary embodiment, the present application provides an aortic dissection identification device, the aortic dissection identification device comprising:

[0080] The CT plain scan image acquisition module is used to obtain the CT plain scan image to be identified.

[0081] The recognition module is used to input the CT plain scan image to be identified into the aortic dissection recognition model constructed according to the aortic dissection recognition model construction method to obtain the aortic dissection recognition result, and the aortic dissection recognition result is the presence of aortic dissection or the absence of aortic dissection.

[0082] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data for constructing an aortic dissection identification model. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for constructing an aortic dissection identification model is implemented.

[0083] Those skilled in the art will understand that Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the above-mentioned aortic dissection identification model construction method embodiments are implemented.

[0084] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps in the above-mentioned embodiments of the method for constructing an aortic dissection identification model.

[0085] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0086] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of the aortic dissection identification model construction method. Among them, any reference to memory, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0087] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, data processing logic of programmable logic devices, and the like.

[0088] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for constructing an aortic dissection identification model, characterized in that: The aortic dissection identification model construction method comprises: By performing bimodal matching on a CT plain scan image and a CTA image of the same period marked with characteristic lesion sites of aortic dissection, it is determined that there is a common section with characteristic lesion sites between the CT plain scan image and the CTA image, and the corresponding characteristic lesion sites are located and marked on the CT plain scan image of the same section; The CT plain scan image dataset comprises a plurality of CT plain scan images with lesion sites marked thereon and a plurality of CT plain scan images of patients with normal aortas; A deep learning model is trained using a CT plain scan image dataset, and the trained deep learning model is used as an aortic dissection recognition model. The aortic dissection recognition model is used to recognize aortic dissection on the CT plain scan images to be recognized.

2. The method for constructing an aortic dissection identification model according to claim 1, wherein: The characteristic lesion site includes at least one of an aortic intimal patch and an aortic intimal rupture.

3. The method for constructing an aortic dissection identification model according to claim 1, wherein: The shooting interval between the CT plain scan image in the CT plain scan image data set and the CTA image of the same period as the CT plain scan image does not exceed 12 hours.

4. The method for constructing an aortic dissection identification model according to claim 1, wherein: The deep learning model is a deep aggregation network structure based on a two-dimensional convolutional neural network.

5. The method for constructing an aortic dissection identification model according to claim 1, wherein: A deep learning model was trained using a CT plain scan image dataset and used as an aortic dissection recognition model. Specifically, the model includes: The back propagation algorithm and stochastic gradient descent method are used to update and iterate the deep learning model to obtain a trained deep learning model.

6. The method for constructing an aortic dissection identification model according to claim 1, wherein: The loss function when training a deep learning model is expressed as: L=-∑ k y k logf(x k ); Among them, L is the loss function value, x k is the kth input of the deep learning model, f(x k ) represents the deep learning model, y k For the deep learning model x k The predicted output.

7. A device for constructing an aortic dissection identification model, characterized in that: The aortic dissection identification model construction device comprises: a dual-modality matching module, configured to perform dual-modality matching between a plain CT image and a contemporaneous CTA image marked with characteristic lesion sites of aortic dissection, to determine whether the CT plain image and the CTA image have the same slice with the characteristic lesion site, and to locate and mark the corresponding characteristic lesion site on the plain CT image of the same slice; a CT plain scan image data set determination module, configured to form the CT plain scan image data set from a plurality of CT plain scan images with lesion sites marked thereon and a plurality of CT plain scan images of patients with normal aortas; The model training module is used to train the deep learning model using a CT plain scan image dataset, and use the trained deep learning model as an aortic dissection recognition model, which is used to identify aortic dissection on the CT plain scan images to be identified.

8. An aortic dissection identification device, characterized in that: The aortic dissection identification device includes a host computer, and the host computer includes an aortic dissection identification model constructed according to the aortic dissection identification model construction method according to any one of claims 1-6.

9. A method for identifying aortic dissection, characterized in that: The aortic dissection identification method comprises: Obtaining a CT plain scan image to be identified; The CT plain scan image to be identified is input into the aortic dissection identification model constructed according to the aortic dissection identification model construction method according to any one of claims 1-6 to obtain an aortic dissection identification result, wherein the aortic dissection identification result is the presence of aortic dissection or the absence of aortic dissection.

10. An aortic dissection identification device, characterized in that: The aortic dissection identification device comprises: A CT plain scan image acquisition module is used to obtain a CT plain scan image to be identified; An identification module is used to input the CT plain scan image to be identified into the aortic dissection identification model constructed according to the aortic dissection identification model construction method according to any one of claims 1 to 6, to obtain an aortic dissection identification result, wherein the aortic dissection identification result is the presence of aortic dissection or the absence of aortic dissection.

Images