DWI-flair mismatch evaluation method, and apparatus, medium and product

Abstract

The present application relates to the technical field of image recognition. Disclosed are a DWI-FLAIR mismatch evaluation method, and an apparatus, a medium and a product. The method comprises: acquiring a diffusion-weighted image and a fluid-attenuated inversion recovery image of a target object; using a first image segmentation model to segment the diffusion-weighted image to obtain a first image segmentation result; using a second image segmentation model to segment the fluid-attenuated inversion recovery image to obtain a second image segmentation result; registering the second image segmentation result to the first image segmentation result to obtain a registered second image segmentation result; and using a Dice index to evaluate the degree of mismatch between the first image segmentation result and the registered second image segmentation result. The present application can improve the efficiency of evaluating whether DWI matches FLAIR.

Description

A DWI-FLAIR mismatch assessment method, device, medium and product

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on March 22, 2024, with application number 202410329867.5 and invention name “A DWI-FLAIR mismatch assessment method, device, medium and product”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application belongs to the field of image recognition technology, and in particular relates to a DWI-FLAIR mismatch assessment method, device, medium and product. Background Art

[0003] Acute ischemic stroke (AIS) is a common type of cerebrovascular disease, with awakening stroke accounting for approximately 20%. If asymptomatic strokes and patients with a known onset >6 hours are included, the proportion of patients missing the opportunity for reperfusion therapy due to exceeding the time window is even higher. "Mismatch" is an imaging concept introduced based on the pathophysiological mechanisms of the ischemic penumbra and infarct core. Using neuroimaging techniques to assess the infarct core, ischemic penumbra, and the resulting mismatch volume is expected to provide further guidance for intravenous thrombolysis and endovascular treatment of AIS.

[0004] Cytotoxic edema caused by acute cerebral infarction reduces the free diffusion of water, leading to a decrease in the apparent diffusion coefficient and enhanced DWI (diffusion-weighted imaging) signal. DWI can early demonstrate the size and location of ischemic lesions, while the positivity rate of fluid attenuated inversion recovery (FLAIR) increases with increasing intervals between onset and MRI (magnetic resonance imaging). Therefore, patients with positive DWI but negative FLAIR are likely within the safe and effective treatment window for thrombolytic therapy. The WAKE-UP (Efficacy and Safety of MRI-Based Thrombolysis in Wake-Up Stroke) study further confirmed the effectiveness of using DWI-FLAIR mismatch to guide intravenous thrombolysis in patients with AIS whose onset time is unknown. In patients with AIS without significant FLAIR hyperintensity, higher relative FLAIR signal intensity of DWI lesions is associated with reduced efficacy of intravenous thrombolysis, consistent with the inverse correlation between treatment efficacy and time since stroke onset. In addition, according to the literature [1]Thomalla G,Fiebach JB, L,et al.A multicenter,randomized,double-blind,placebo-controlled trial to test efficacy and safety of magnetic resonance imaging-based thrombolysis in wake-up stroke(WAKE-UP)[J].International Journal of Stroke,2014,9(6):829-836. and [2]Thomalla G,Cheng B,Ebinger M,et al.DWI-FLAIR mismatch for the identification of patients with acute ischaemic stroke within4·5h of symptom onset(PRE-FLAIR):a multicentre observational study[J].The Lancet Neurology,2011,10(11):978-986. studies showed that it is feasible to select patients with wake-up stroke or stroke with unknown onset time for thrombolytic therapy based on modified DWI-FLAIR mismatch (defined as the presence of DWI lesions but no corresponding signal changes in the FLAIR sequence, or the FLAIR lesions are significantly smaller than the DWI lesions). In patients with ASI due to LVO (large vessel occlusion) undergoing EVT (endovascular therapy), DWI-FLAIR mismatch is also associated with better outcomes and fewer symptomatic intracranial hemorrhages. Although not based on tissue window mismatch, this mismatch, arising from differences in onset time between imaging sequences, provides theoretical support for the treatment of stroke after awakening.

[0005] Currently, DWI-FLAIR mismatch is typically determined by comparing DWI and FLAIR images. DWI images typically display areas of high signal intensity due to their sensitivity to the diffusion of water molecules, while FLAIR images suppress the signal from cerebrospinal fluid and typically reveal areas of low signal intensity. Therefore, the two images exhibit complementary characteristics. However, the time and effort required to analyze and interpret medical images makes the assessment of DWI-FLAIR mismatch inefficient. Summary of the Invention

[0006] The purpose of this application is to provide a DWI-FLAIR mismatch assessment method, device, medium and product, which can improve the efficiency of assessing whether DWI-FLAIR is matched.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] A DWI-FLAIR mismatch assessment method comprises: acquiring a diffusion-weighted image and a fluid-attenuated inversion recovery image of a target object; applying a first image segmentation model to segment the diffusion-weighted image to obtain a first image segmentation result; wherein the first image segmentation model is obtained by training a neural network based on a U-Net framework using a diffusion-weighted image dataset of acute ischemic lesions; applying a second image segmentation model to segment the fluid-attenuated inversion recovery image to obtain a second image segmentation result; wherein the second image segmentation model is obtained by training a neural network based on a U-Net framework using a fluid-attenuated inversion recovery image dataset of acute ischemic lesions; registering the second image segmentation result to the first image segmentation result to obtain a registered second image segmentation result; and applying the Dice index to evaluate the degree of mismatch between the first image segmentation result and the registered second image segmentation result.

[0009] In an exemplary embodiment, the neural network based on the U-Net framework includes four downsampling layers connected in sequence, a residual module, four upsampling layers connected in sequence and a first activation function; the four downsampling layers connected in sequence are connected to the residual module; the residual module is connected to the four upsampling layers connected in sequence; and the four upsampling layers connected in sequence are connected to the first activation function.

[0010] In an exemplary embodiment, the downsampling layer includes a maximum pooling layer, a first convolutional layer, a first InstanceNorm layer, a second activation function, a second convolutional layer, a second InstanceNorm layer and a third activation function connected in sequence; the convolution kernel size of the first convolutional layer and the second convolutional layer is 3.

[0011] In an exemplary embodiment, the upsampling layer includes a deconvolution layer, a third convolution layer, a third InstanceNorm layer, a third activation function, a fourth convolution layer, a fourth InstanceNorm layer and a fourth activation function connected in sequence; the convolution kernel sizes of the third convolution layer and the fourth convolution layer are both 3.

[0012] In an exemplary embodiment, the process of constructing the diffusion-weighted image dataset of acute ischemic lesions specifically includes: obtaining an original diffusion-weighted image of the acute ischemic lesion; eliminating abnormal data from the original diffusion-weighted image to obtain an original diffusion-weighted image after elimination; labeling the target area in the original diffusion-weighted image after elimination to obtain an annotated original diffusion-weighted image, and using the original diffusion-weighted image and the annotated original diffusion-weighted image as the diffusion-weighted image dataset of acute ischemic lesions.

[0013] In an exemplary embodiment, the process of constructing the acute ischemic lesion fluid-attenuated inversion recovery image dataset specifically includes: obtaining the original fluid-attenuated inversion recovery image of the acute ischemic lesion; eliminating abnormal data from the original fluid-attenuated inversion recovery image to obtain the eliminated original fluid-attenuated inversion recovery image; marking the target area in the eliminated original fluid-attenuated inversion recovery image to obtain the marked original fluid-attenuated inversion recovery image, and using the original fluid-attenuated inversion recovery image and the marked original fluid-attenuated inversion recovery image as the acute ischemic lesion fluid-attenuated inversion recovery image dataset.

[0014] In an exemplary embodiment, the second image segmentation result is registered to the first image segmentation result to obtain the registered second image segmentation result, specifically including: registering the liquid attenuated inversion recovery image to the diffusion weighted image to obtain an affine transformation matrix; and applying the affine transformation matrix to register the second image segmentation result to the first image segmentation result.

[0015] A computer device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned DWI-FLAIR mismatch assessment methods.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned DWI-FLAIR mismatch assessment methods.

[0017] A computer program product comprises a computer program, which implements the steps of any one of the above-mentioned DWI-FLAIR mismatch assessment methods when executed by a processor.

[0018] The beneficial effects of this application are:

[0019] This application applies a first image segmentation model to segment the diffusion-weighted image of the target object to obtain a first image segmentation result, and applies a second image segmentation model to segment the fluid-attenuated inversion recovery image of the target object to obtain a second image segmentation result. The application can automatically identify the target area of ​​the patient displayed on the DWI and FLAIR images and quickly segment it; then the second image segmentation result is aligned with the first image segmentation result to obtain a registered second image segmentation result; finally, the Dice indicator is applied to evaluate the degree of mismatch between the first image segmentation result and the registered second image segmentation result, thereby realizing automatic evaluation of whether DWI-FLAIR matches, and improving the efficiency of evaluating whether DWI-FLAIR matches.

[0020] Figures in the specification

[0021] FIG1 is a schematic diagram of the image segmentation model architecture provided in Example 1 of the present application;

[0022] FIG2 is a schematic diagram of the process of registering FLAIR to DWI using Affine transform provided in Example 1 of the present application;

[0023] FIG3 is a schematic diagram of a process for aligning segmentation results using a transformation matrix according to Example 1 of the present application;

[0024] FIG4 is a flowchart of the practical application of the DWI-FLAIR mismatch assessment method based on the deep learning model provided in Example 1 of the present application;

[0025] FIG5 is a schematic flow chart of the DWI-FLAIR mismatch assessment method provided in Example 1 of the present application;

[0026] FIG6 is a diagram showing the internal structure of a computer device. DETAILED DESCRIPTION

[0027] The present application will be further described below with reference to the following examples. The following examples are provided only to facilitate understanding of the present application. It should be noted that, without departing from the principles of the present application, it is possible for a person skilled in the art to make various modifications to the present application, and such improvements and modifications are intended to fall within the scope of the claims of the present application.

[0028] The purpose of this application is to provide a DWI-FLAIR mismatch assessment method, device, medium and product, which can improve the efficiency of assessing whether DWI-FLAIR is matched.

[0029] This application provides a DWI-FLAIR mismatch assessment model based on a deep learning model, which can automatically identify the target area shown on the patient's DWI and FLAIR images and quickly segment it; then, through the mismatch assessment module designed in this application, a score for the mismatch degree of the current target area is obtained.

[0030] This application uses the capabilities of deep models to achieve image recognition of acute ischemic lesions in DWI and FLAIR images, and through the designed lesion mismatch evaluation algorithm, calculates a score used to describe the degree of mismatch of acute ischemic lesions in DWI and FLAIR images.

[0031] The process is briefly described as follows: (1) Using the DWI acute ischemic lesion dataset, the DWI-UNet segmentation network constructed in this application is trained to obtain a DWI image segmentation model; (2) Using the FLAIR acute ischemic lesion dataset, the FLAIR-UNet segmentation network constructed in this application is trained to obtain a FLAIR image segmentation model; (3) For a pair of DWI and FLAIR images, the FLAIR image is aligned to the DWI image using the FSL tool to obtain an affine transformation matrix; (4) The image is segmented using the DWI image segmentation model and the FLAIR segmentation model respectively to obtain two image segmentation results; (5) The two segmentation results are aligned using the affine transformation matrix; (6) The Dice index of the two segmentation results after alignment is calculated, and the mismatch degree is obtained using the 1-Dice value, which is converted into a score.

[0032] 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.

[0033] Example 1

[0034] As shown in FIG5 , a DWI-FLAIR mismatch assessment method in this embodiment includes:

[0035] Step S1: Acquire a diffusion-weighted image and a fluid-attenuated inversion recovery image of the target object.

[0036] Step S2: Applying a first image segmentation model to segment the diffusion-weighted image to obtain a first image segmentation result; wherein the first image segmentation model is obtained by training a neural network based on a U-Net framework using a diffusion-weighted image dataset of acute ischemic lesions.

[0037] The neural network based on the U-Net framework includes four sequentially connected downsampling layers, a residual module, four sequentially connected upsampling layers and a first activation function.

[0038] The four downsampling layers connected in sequence are connected to the residual module; the residual module is connected to the four upsampling layers connected in sequence; and the four upsampling layers connected in sequence are connected to the first activation function.

[0039] Specifically, the downsampling layer includes a maximum pooling layer, a first convolutional layer, a first InstanceNorm layer, a second activation function, a second convolutional layer, a second InstanceNorm layer and a third activation function connected in sequence; the sizes of the convolution kernels of the first convolutional layer and the second convolutional layer are both 3.

[0040] The upsampling layer includes a deconvolution layer, a third convolution layer, a third InstanceNorm layer, a third activation function, a fourth convolution layer, a fourth InstanceNorm layer and a fourth activation function connected in sequence; the sizes of the convolution kernels of the third convolution layer and the fourth convolution layer are both 3.

[0041] Furthermore, the process of constructing the acute ischemic lesion diffusion-weighted imaging dataset specifically includes:

[0042] Step 11: Obtain original diffusion-weighted images of the acute ischemic lesion.

[0043] Step 12: Eliminate abnormal data from the original diffusion-weighted image to obtain an original diffusion-weighted image after elimination.

[0044] Step 13: annotating the target area in the original diffusion-weighted image after elimination to obtain an annotated original diffusion-weighted image, and using the original diffusion-weighted image and the annotated original diffusion-weighted image as the acute ischemic lesion diffusion-weighted image dataset.

[0045] Step S3: Apply a second image segmentation model to segment the fluid-attenuated inversion recovery image to obtain a second image segmentation result; wherein the second image segmentation model is obtained by training a neural network based on a U-Net framework using an acute ischemic lesion fluid-attenuated inversion recovery image dataset.

[0046] The process of constructing the acute ischemic lesion fluid-attenuated inversion recovery imaging dataset specifically includes:

[0047] Step 21: Obtain original fluid-attenuated inversion recovery images of the acute ischemic lesion.

[0048] Step 22: Eliminate abnormal data from the original LAIRI image to obtain the eliminated original LAIRI image.

[0049] Step 23: Annotate the target area in the original fluid-attenuated inversion recovery image after elimination to obtain an annotated original fluid-attenuated inversion recovery image, and use the original fluid-attenuated inversion recovery image and the annotated original fluid-attenuated inversion recovery image as the acute ischemic lesion fluid-attenuated inversion recovery image dataset.

[0050] In practical applications, the two image segmentation models used in this application: 1) DWI-UNet image segmentation model (i.e., the first image segmentation model); 2) FLAIR-UNet image segmentation model (i.e., the second image segmentation model) are both obtained by training an improved network based on UNet.

[0051] The two image segmentation models constructed in this application have the same architecture, as shown in Figure 1. Based on the traditional U-Net framework, the image segmentation network of this application is mainly composed of 4 downsampling and upsampling stages. Among them, each downsampling layer includes a maximum pooling layer, two convolutional layers (convolution kernel size is 3, step size is 1, padding is 1), two InstanceNorm layers and two ReLU activation functions, and is constructed in the order of maximum pooling layer, convolution layer-1 (first convolution layer), InstanceNorm-1 (first InstanceNorm layer), ReLU-1 (second activation function), convolution layer-2 (second convolution layer), InstanceNorm-2 (second InstanceNorm layer), ReLU-2 (third activation function); each upsampling layer includes a maximum pooling layer, two convolutional layers (convolution kernel size is 3, step size is 1, padding is 1), two InstanceNorm layers and two ReLU activation functions. The network consists of one deconvolution layer (kernel size 2, stride 2, and padding 0), two convolution layers (kernel size 3, stride 1, and padding 1), two InstanceNorm layers, and two ReLU activation functions. The network is constructed in the following order: deconvolution layer, Convolution-1 (third convolution layer), InstanceNorm-1 (third InstanceNorm layer), ReLU-1 (third activation function), Convolution-2 (fourth convolution layer), InstanceNorm-2 (fourth InstanceNorm layer), and ReLU-2 (fourth activation function). To improve the network's feature extraction capabilities for medical images, dense connections are used in the downsampling phase, and a residual module is added at the bottleneck layer to ensure that the segmentation model can extract sufficient information for segmentation. Finally, the network output is passed through a Sigmoid activation function, limiting the output range to [0, 1]. This value is considered the probability result of the target region prediction.

[0052] When training the DWI-UNet image segmentation model and the FLAIR-UNet image segmentation model, the DWI and FLAIR acute ischemic lesion datasets were first constructed. The original DWI and FLAIR images were selected and abnormal data were removed. Then, doctors manually outlined the lesion area as the target area annotation of the corresponding image, which together with the original image data formed the image segmentation dataset.

[0053] Then, the corresponding image segmentation datasets are applied to train the DWI-UNet image segmentation model and the FLAIR-UNet image segmentation model respectively. The weighted cross entropy loss (Weighted Binary Cross Entropy Loss) and Dice loss are used in the training. The total loss function of the training is the sum of the two, and the calculation formulas are as follows:

[0054] Among them, w represents the weight parameter, i represents the i-th pixel in an image, and y i Represents the true label value at the pixel point, Represents the predicted value of the network at the pixel point, and * indicates the product operation of the corresponding elements. represents the weighted cross entropy loss, represents Dice loss, and N represents the number of pixels in an image.

[0055] Finally, the network was trained using both back-gradient propagation and gradient descent algorithms until convergence. Using cross-validation, the average DICE and Hausdorff Distance values ​​of the model segmentation results were evaluated on the test set. The model with the best evaluation metrics, i.e., the highest segmentation accuracy, was selected and its parameters saved, ultimately yielding the trained DWI-UNet and FLAIR-UNet image segmentation models.

[0056] Step S4: registering the second image segmentation result with the first image segmentation result to obtain a registered second image segmentation result.

[0057] S4 specifically includes:

[0058] Step S41: registering the LAIRA image to the DWRA image to obtain an affine transformation matrix.

[0059] Step S42: applying an affine transformation matrix to register the second image segmentation result to the first image segmentation result.

[0060] Step S5: applying the Dice index to evaluate the mismatch between the first image segmentation result and the registered second image segmentation result.

[0061] The DWI-FLAIR mismatch assessment method provided in this application can be constructed as a DWI-FLAIR mismatch assessment model. The DWI-FLAIR mismatch assessment model in this application is described below:

[0062] First, we construct a mismatch evaluation metric: for DWI and FLAIR images from the same user, a mismatch is considered when the target region is absent on the FLAIR image but present on the DWI image. The Dice metric is a set similarity metric, typically used to calculate the similarity between two samples, with a threshold of [0, 1]. It is often used for image segmentation in medical imaging, with the best segmentation result being 1 and the worst being 0. Its calculation formula is as follows:

[0063] In this application, A and B represent the DWI image segmentation results and the FLAIR image segmentation results, respectively. Therefore, after obtaining the image segmentation results of the two images through the deep learning model, calculating the Dice value of the two can reflect the degree of matching of the target areas displayed on the DWI image and the FLAIR image. Among them, the target area is determined by the user according to needs. For example, the target area can be an acute ischemic stroke lesion. And 1-Dice reflects the degree of mismatch between the two segmented target areas. The calculation formula is as follows, where Score missmatch Score for mismatch: missmatch (A,B)=1-dice(A,B);

[0064] Applying the aforementioned evaluation metrics, the specific evaluation process is described. Often, DWI and FLAIR images from the same user are not perfectly aligned. Directly using segmentation results is not sufficient for mismatch assessment because the same coordinates may correspond to different locations. Therefore, it is necessary to align the same anatomical structures in the two images. When expressed in coordinates, they should be located at the same coordinates. As shown in Figure 2, the FSL tool is first used to register the FLAIR image (with its width, height, and depth dimensions as H', W', and D') to the DWI image (with its width, height, and depth dimensions as H, W, and D). Affine transformations with 12 degrees of freedom are used (the calculation method used is an affine transformation with 12 degrees of freedom because there are three coordinate axes in Euclidean space: x, y, and z. Each axis corresponds to four transformations: translation, rotation, scaling, and tilt, for a total of 12 degrees of freedom). Although the types of transformations are numerous, they can be expressed using the algebraic equation: Y = X·M, X,Y∈R 4×1 ,M∈R 4×4 ;

[0065] Where X and Y represent the augmented original and transformed images, their first three elements are the actual positions in Euclidean space, and their last element is 1, which is used to achieve scaling; M represents the transformation matrix. With the help of FSL tools, after the registration is completed, the affine transformation matrix M can be obtained. The form of this matrix is ​​shown in Figure 2. This affine transformation matrix describes the matrix transformation required to register the FLAIR image with the DWI image, where Represents the 12 degrees of freedom mentioned above. Therefore, the same matrix transformation operation is applied to the FLAIR-UNet segmentation result to achieve alignment with the DWI-UNet segmentation result.

[0066] As shown in Figure 3, using the apply function provided by the FSL tool, the affine transformation matrix obtained above is applied to the FLAIR image segmentation network results (size H', W', D'). The transformation is then applied via nearest neighbor interpolation, applying the registration transformation to the FLAIR target region. This results in a registered segmentation result (size H, W, D), which is aligned with the DWI image segmentation result (size H, W, D). A mismatch evaluation metric is then calculated using the two registered segmentation results to obtain a final DWI-FLAIR mismatch score. A score above a certain threshold indicates a DWI-FLAIR mismatch for the user. Research has shown that a mismatch between acute ischemic lesions visible on diffusion-weighted imaging (DWI) but not on fluid-attenuated inversion recovery (FLAIR) imaging indicates that the patient's stroke onset occurred within 4.5 hours (≤4.5 hours from symptom onset). Therefore, the DWI-FLAIR mismatch can provide physicians with additional information to confirm the onset of stroke.

[0067] As shown in FIG4 , the present application proposes a DWI-FLAIR mismatch assessment method based on a deep learning model, and the steps are as follows:

[0068] Step 1: Train the image segmentation model.

[0069] Using the constructed segmentation dataset, the DWI-UNet image segmentation network and the FLAIR-UNet image segmentation network were trained respectively to obtain two accurate image segmentation models.

[0070] Step 2: Construct a DWI-FLAIR mismatch assessment model. Construct a target region mismatch score based on the Dice metric.

[0071] Step 3: Segment the DWI image. Use the trained DWI-UNet segmentation network to segment the DWI image.

[0072] Step 4: Segment the FLAIR image. Use the trained FLAIR-UNet segmentation network to segment the FLAIR image.

[0073] Step 5: Register the FLAIR image with the DWI image. Use the FSL tool to register the FLAIR image with the DWI image using the Affine transformation to obtain the affine matrix followed by the transformation.

[0074] Step 6: Align the FLAIR segmentation results with the DWI segmentation results. Use the affine matrix obtained in step 5 and apply it to the FLAIR image segmentation results to obtain a FLAIR image segmentation result that is aligned with the DWI segmentation result.

[0075] Step 7: Calculate the mismatch score. Use the constructed DWI-FLAIR mismatch assessment model to calculate the mismatch score of the registered FLAIR and DWI segmentation results.

[0076] The advantages of this application are as follows:

[0077] 1. This application uses a deep learning-based DWI-FLAIR mismatch assessment model to inform doctors of stroke onset time. This automated analysis of medical imaging data eliminates the need for manual intervention by doctors, improving work efficiency and reducing workload. It can analyze imaging data from a large number of users in a short period of time, accelerating stroke diagnosis and intervention.

[0078] 2. The mismatch assessment model in this application is based on deep learning and can accurately assess the degree of mismatch between DWI and FLAIR images. It provides a timeline for stroke onset, helping doctors determine the time window for treatment and enabling them to better select treatment strategies.

[0079] 3. By providing a reminder of the time of stroke occurrence, this application is expected to have a positive clinical impact on the treatment outcomes of stroke patients, increase the chances of stroke patients receiving timely intervention, and help improve the survival and recovery rates of stroke patients.

[0080] 4. This application constructs a DWI-FLAIR mismatch assessment model based on a deep learning model. This model has the ability to automatically segment and evaluate, and can provide auxiliary information for doctors to identify acute ischemic stroke lesions.

[0081] 5. This application constructs a DWI-FLAIR mismatch evaluation index based on the calculation of the Dice index, which is used to quantify the degree of match between the target area on the DWI image and the FLAIR image.

[0082] 6. This application uses the FSL tool to align the FLAIR image segmentation results with the DWI image segmentation results, and realizes automated DWI-FLAIR mismatch interpretation and calculation.

[0083] Example 2

[0084] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the DWI-FLAIR mismatch assessment method in embodiment 1.

[0085] Example 3

[0086] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the DWI-FLAIR mismatch assessment method in embodiment 1.

[0087] Example 4

[0088] A computer program product includes a computer program, which implements the steps of the DWI-FLAIR mismatch assessment method in embodiment 1 when executed by a processor.

[0089] Example 5

[0090] A computer device, which can be a database, and its internal structure diagram can be shown in Figure 6. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Wherein, the processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. Wherein, 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 the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. 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 via a network connection. When the computer program is executed by the processor, the DWI-FLAIR mismatch assessment method in Example 1 is implemented.

[0091] It should be noted that the object information (including but not limited to object device information, object 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 object or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0092] 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, and 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 embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may 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 may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database 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 processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0093] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

Claims

1. A DWI-FLAIR mismatch assessment method, characterized in that: The method comprises: Acquire diffusion-weighted images and fluid-attenuated inversion recovery images of the target object; Applying a first image segmentation model to segment the diffusion-weighted image to obtain a first image segmentation result; wherein the first image segmentation model is obtained by training a neural network based on a U-Net framework using a diffusion-weighted image dataset of acute ischemic lesions; Applying a second image segmentation model to segment the fluid-attenuated inversion recovery image to obtain a second image segmentation result; wherein the second image segmentation model is obtained by training a neural network based on a U-Net framework using a dataset of fluid-attenuated inversion recovery images of acute ischemic lesions; registering the second image segmentation result with the first image segmentation result to obtain a registered second image segmentation result; The Dice index is applied to evaluate the mismatch between the first image segmentation result and the registered second image segmentation result.

2. The DWI-FLAIR mismatch assessment method according to claim 1, wherein: The neural network based on the U-Net framework includes four sequentially connected downsampling layers, a residual module, four sequentially connected upsampling layers and a first activation function; The four sequentially connected downsampling layers are connected to the residual module; the residual module is connected to the four sequentially connected upsampling layers; The four upsampling layers connected in sequence are connected to the first activation function.

3. The DWI-FLAIR mismatch assessment method according to claim 2, characterized in that: The downsampling layer includes a maximum pooling layer, a first convolutional layer, a first InstanceNorm layer, a second activation function, a second convolutional layer, a second InstanceNorm layer and a third activation function connected in sequence; The sizes of the convolution kernels of the first convolution layer and the second convolution layer are both 3.

4. The DWI-FLAIR mismatch assessment method according to claim 2, wherein: The upsampling layer includes a deconvolution layer, a third convolution layer, a third InstanceNorm layer, a third activation function, a fourth convolution layer, a fourth InstanceNorm layer and a fourth activation function connected in sequence; The sizes of the convolution kernels of the third convolution layer and the fourth convolution layer are both 3.

5. The DWI-FLAIR mismatch assessment method according to claim 1, wherein: The construction process of the acute ischemic lesion diffusion-weighted imaging dataset includes: Obtain original diffusion-weighted images of acute ischemic lesions; Eliminating abnormal data from the original diffusion-weighted image to obtain an original diffusion-weighted image after elimination; The target area in the eliminated original diffusion-weighted image is labeled to obtain a labeled original diffusion-weighted image, and the original diffusion-weighted image and the labeled original diffusion-weighted image are used as the acute ischemic lesion diffusion-weighted image dataset.

6. The DWI-FLAIR mismatch assessment method according to claim 1, wherein: The process of constructing the acute ischemic lesion fluid-attenuated inversion recovery imaging dataset includes: Acquire raw fluid-attenuated inversion recovery images of acute ischemic lesions; Eliminating abnormal data from the original liquid attenuated inversion recovery image to obtain an original liquid attenuated inversion recovery image after elimination; The target area in the removed original fluid-attenuated inversion recovery image is labeled to obtain a labeled original fluid-attenuated inversion recovery image, and the original fluid-attenuated inversion recovery image and the labeled original fluid-attenuated inversion recovery image are used as the acute ischemic lesion fluid-attenuated inversion recovery image dataset.

7. The DWI-FLAIR mismatch assessment method according to claim 1, wherein: Registering the second image segmentation result to the first image segmentation result to obtain a registered second image segmentation result, comprising: registering the liquid-attenuated inversion recovery image to the diffusion-weighted image to obtain an affine transformation matrix; An affine transformation matrix is ​​applied to register the second image segmentation result to the first image segmentation result.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the DWI-FLAIR mismatch assessment method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the DWI-FLAIR mismatch assessment method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the DWI-FLAIR mismatch assessment method according to any one of claims 1 to 7 are implemented.

Images