An MRI mismatch scoring method, device, equipment, medium and product

Abstract

The present application discloses an MRI mismatch scoring method, device, equipment, medium and product, relating to the field of intelligent healthcare. The method includes: registering the low-field magnetic resonance stroke discrimination sequence images onto the DWI images, applying the lesion segmentation result on the DWI images to the registered low-field magnetic resonance stroke discrimination sequence images to obtain the region corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence images, calculating the mean image intensity of this region, and comparing the mean image intensity of this region with the mean intensity within the whole brain of the registered low-field magnetic resonance stroke discrimination sequence images to obtain the mismatch ratio. The present application can perform DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition based on DWI and the low-field magnetic resonance stroke discrimination sequence, so as to improve the efficiency of MRI mismatch scoring and the accuracy of MRI mismatch scoring results.

Description

Technical Field

[0001] This application relates to the field of intelligent medical technologies, and particularly to an MRI mismatch scoring method, device, equipment, medium, and product. Background Art

[0002] Cerebrovascular diseases, including acute cerebral hemorrhage and cerebral infarction, are one of the main causes of death and disability globally. Accurately and rapidly differentiating acute cerebral hemorrhage and cerebral infarction is crucial for timely treatment and improved prognosis. Acute cerebral hemorrhage refers to a disease in which a cerebral blood vessel ruptures, causing blood to enter the brain tissue and resulting in brain tissue damage. Cerebral infarction, on the other hand, is a condition in which cerebral blood vessels are blocked, leading to ischemia and hypoxia of the brain tissue and causing brain tissue damage. Although both involve problems with cerebral blood circulation, their etiologies and pathophysiological mechanisms are different, and the treatment decisions are completely different: for cerebral hemorrhage, blood pressure needs to be controlled and cerebral edema reduced; for the acute phase of cerebral infarction, thrombolysis, mechanical thrombectomy can be performed, and antiplatelet aggregation and circulation improvement treatment can also be carried out subsequently. Therefore, being able to rapidly differentiate cerebral hemorrhage and cerebral infarction in the ultra-early stage will be very beneficial for the treatment plan decision-making of patients.

[0003] The differential diagnosis of cerebral hemorrhage and cerebral infarction includes symptoms and imaging examination results. The usual differential methods are as follows: Cerebral hemorrhage generally occurs during activity, manifested as sudden headache, nausea, and vomiting, and then combined with the clinical features of hemiplegia, aphasia, and coma, it can basically be determined as cerebral hemorrhage. Cerebral infarction generally occurs in a quiet state and also shows symptoms such as hemiplegia, aphasia, and even coma. The most crucial differentiation is that once this situation occurs, if high density is seen on Computed Tomography (CT), it is determined as cerebral hemorrhage. On the other hand, imaging examination also plays an important role in differential diagnosis. Commonly used imaging examinations include head CT and Magnetic Resonance Imaging (MRI). By observing the cerebral blood vessel conditions and the degree of brain tissue damage, it can help doctors determine the etiology and type. In addition, special examination methods such as angiography can be used to further clarify the diagnosis. Clinical scoring tools are an auxiliary diagnostic means that can help doctors more accurately judge the condition of patients. Commonly used clinical scoring tools include the Glasgow Coma Scale (GCS), the National Institute of Health stroke scale (NIHSS), etc. By evaluating indicators such as the patient's consciousness state and limb movement, it can provide important reference basis for doctors' diagnosis.

[0004] For traditional imaging differential methods, CT has high detection sensitivity for acute cerebral hemorrhage, but low detection sensitivity for acute cerebral infarction. This limitation may lead to misdiagnosis or missed diagnosis, delaying the treatment opportunity and thus affecting the prognosis of patients. MRI has high detection sensitivity for acute cerebral infarction, especially DWI is the most sensitive. Although MRI sequences can provide detailed brain structure information, the differentiation cannot be achieved by observing a certain imaging feature in a certain MRI sequence. Usually, multiple MRI sequences need to be compared, observed and analyzed, and finally a conclusion is drawn, which is highly dependent on the doctor's experience. Therefore, the sensitivity and specificity of MRI sequences in ultra-acute lesions are limited. This deficiency leads to doctors relying on experience and subjective judgment, increasing the risk of misdiagnosis.

[0005] The MRI mismatch score refers to the assessment of the infarct core and ischemic penumbra of patients with ischemic stroke through various imaging techniques in MRI, and then calculating the mismatch volume and mismatch ratio (i.e., mismatch rate) to guide clinical treatment. The MRI mismatch score is mainly based on imaging techniques. For example, based on Diffusion Weighted Imaging (DWI) and Perfusion Weighted Imaging (PWI), DWI-PWI mismatch recognition (i.e., DWI-PWI mismatch score) is performed, or based on DWI and Fluid-attenuated Inversion Recovery (FLAIR), DWI-FLAIR mismatch recognition is performed, or based on DWI and Arterial Spin Labeling (ASL), DWI-ASL mismatch recognition is performed, and so on. In these mismatch recognitions, such as DWI-PWI mismatch recognition, the current mainstream methods are mainly based on traditional imaging analysis and mainly rely on doctors' interpretation of DWI-PWI, resulting in problems of low efficiency of MRI mismatch scoring and low accuracy of MRI mismatch scoring results. There are many MRI scan sequences, including not only PWI, DWI, FLAIR, and ASL sequences, but also low-field magnetic resonance stroke differentiation sequences (i.e., low-field strength magnetic resonance stroke differentiation sequences). Like these sequences, such as DWI, the low-field magnetic resonance stroke differentiation sequences are also scanned by magnetic resonance. DWI and the low-field magnetic resonance stroke differentiation sequences can be quickly scanned and obtained within a few minutes using a low-field strength mobile magnetic resonance device (such as the mobile head and neck magnetic resonance system ACUTA Elfin manufactured by Foshan Regato Medical Technology Co., Ltd.). Compared with sequences such as PWI, FLAIR, and ASL, the low-field magnetic resonance stroke differentiation sequences can quickly and accurately judge hemorrhagic stroke. Based on this, there is an urgent need in this field for an MRI mismatch scoring method based on DWI and low-field magnetic resonance stroke differentiation sequences to perform DWI-low-field magnetic resonance stroke differentiation sequence mismatch recognition, so as to improve the efficiency of MRI mismatch scoring and the accuracy of MRI mismatch scoring results.

[0006] However, the existing mismatch recognition algorithms are mainly used for the evaluation between DWI and PWI, DWI and FLAIR, and DWI and ASL. There is no algorithm for the signal mismatch between DWI and the low-field magnetic resonance stroke discrimination sequence. The applicability and effectiveness of the existing algorithms are limited. This blank limitation leads to unsatisfactory results when directly applying current other imaging mismatch models (including DWI-PWI mismatch recognition models and mismatch models of other different modality images, such as DWI-FLAIR, etc.) in the discrimination of hyperacute cerebral hemorrhage and cerebral infarction, because current imaging algorithms can still basically only process images of known modalities during training. For images of the low-field magnetic resonance stroke discrimination sequence, which is an unseen modality, generally, the algorithms need to be redesigned and experimented. Aiming at the blank of the above existing technologies, there is an urgent need in the art for an algorithm that can perform DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition based on DWI and the low-field magnetic resonance stroke discrimination sequence. Summary of the Invention

[0007] The purpose of this application is to provide an MRI mismatch scoring method, device, equipment, medium and product, which can perform DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition based on DWI and the low-field magnetic resonance stroke discrimination sequence, so as to improve the efficiency of MRI mismatch scoring and the accuracy of MRI mismatch scoring results.

[0008] To achieve the above purpose, the following solutions are provided in this application.

[0009] In a first aspect, the present application provides an MRI mismatch scoring method, which includes: obtaining the DWI image and the low-field magnetic resonance stroke discrimination sequence image of a target object; applying a DWI lesion segmentation model to segment the DWI image to obtain a lesion segmentation result on the DWI image; the DWI lesion segmentation model is obtained by training a neural network based on the U-Net framework using a DWI acute ischemic lesion dataset; the lesion segmentation result is a binary mask; performing skull stripping on the DWI image and the low-field magnetic resonance stroke discrimination sequence image respectively to obtain the DWI image after skull stripping and the low-field magnetic resonance stroke discrimination sequence image after skull stripping; registering the low-field magnetic resonance stroke discrimination sequence image after skull stripping to the DWI image after skull stripping to obtain a registered low-field magnetic resonance stroke discrimination sequence image; applying the lesion segmentation result to the registered low-field magnetic resonance stroke discrimination sequence image to extract a local region; the local region is the region corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence image; the DWI high signal is the image of the part with a value of 1 in the lesion segmentation result; calculating the mean image intensity of the local region to obtain a local signal intensity mean; calculating the mean intensity of the entire brain in the registered low-field magnetic resonance stroke discrimination sequence image to obtain a whole-brain signal intensity mean; comparing the local signal intensity mean with the whole-brain signal intensity mean to obtain a mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist doctors in judging the bleeding or ischemic type of the target object.

[0010] Optionally, registering the low-field magnetic resonance stroke discrimination sequence image after skull stripping to the DWI image after skull stripping to obtain a registered low-field magnetic resonance stroke discrimination sequence image specifically includes: using the registration method of ANTs, keeping the DWI image after skull stripping stationary, and transforming the low-field magnetic resonance stroke discrimination sequence image after skull stripping into the same physical space as the DWI image after skull stripping through an Affine transformation to obtain a registered low-field magnetic resonance stroke discrimination sequence image; the Affine transformation includes rotation and translation operations.

[0011] Optionally, applying the lesion segmentation result to the registered low-field magnetic resonance stroke discrimination sequence image to extract a local region specifically includes: calculating the lesion segmentation result and the registered low-field magnetic resonance stroke discrimination sequence image according to the formula to extract a local region; where represents the lesion segmentation result, , represents element-level multiplication of matrices, Indicates the registered low-field magnetic resonance stroke discrimination sequence image, , Indicates a local area, , Indicates the dimension to which it belongs, and the size and shape of the image corresponding to each dimension, , , Respectively indicate Is a three-dimensional vector in a real number space, Is a three-dimensional vector in a real number space, Is a three-dimensional vector in a real number space, and each dimension represents the height of the three-dimensional vector Width And depth .

[0012] Optionally, calculate the average image intensity of the local area to obtain the local signal intensity average value, specifically including: using the formula Calculate the average image intensity of the local area to obtain the local signal intensity average value; where, Represents the local signal intensity average value, Represents taking the expectation. In a three-dimensional space, a point in the space is uniquely determined by the coordinates ( , , ), Represents the position along the height axis, Represents the position along the width axis, Represents the position along the depth axis, Represents the intensity value of a voxel of an image corresponding to the coordinates ( , , ).

[0013] Optionally, calculate the average intensity of the entire brain in the registered low-field magnetic resonance stroke discrimination sequence image to obtain the average signal intensity in the whole brain, specifically including: using the formula Calculate the average intensity of the entire brain in the registered low-field magnetic resonance stroke discrimination sequence image to obtain the average signal intensity in the whole brain; where, Represents the average signal intensity in the whole brain, Represents the image At ( , , ) the intensity value of the voxel.

[0014] Optionally, compare the local signal intensity average value with the average signal intensity in the whole brain to obtain a mismatch ratio, specifically including: using the formula Calculate the mismatch ratio; wherein, represents the mismatch ratio; Using to provide a reference for the doctor and assist the doctor in judging the specific process of the bleeding or ischemia type of the target object includes: comparing with the high signal threshold in terms of size. If , it is determined that a high signal appears in the local area. If , it is determined that no high signal appears in the local area. Combining the situation of DWI high signal, if a high signal appears in the DWI image but no high signal appears in the low-field magnetic resonance stroke discrimination sequence image, it is determined that the DWI image does not match the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is ischemia. If high signals appear in both the DWI image and the low-field magnetic resonance stroke discrimination sequence image, it is determined that the DWI image matches the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is bleeding.

[0015] In a second aspect, the present application provides an MRI mismatch scoring device, which includes: an image acquisition module for acquiring DWI images and low-field magnetic resonance stroke discrimination sequence images of a target object; a lesion segmentation module for segmenting the DWI images using a DWI lesion segmentation model to obtain a lesion segmentation result on the DWI images; the DWI lesion segmentation model is obtained by training a neural network based on the U-Net framework using a DWI acute ischemic lesion dataset; the lesion segmentation result is a binary mask; an image skull stripping module for respectively stripping the skulls of the DWI images and the low-field magnetic resonance stroke discrimination sequence images to obtain the DWI images after skull stripping and the low-field magnetic resonance stroke discrimination sequence images after skull stripping; an image registration module for registering the low-field magnetic resonance stroke discrimination sequence images after skull stripping onto the DWI images after skull stripping to obtain the registered low-field magnetic resonance stroke discrimination sequence images; a local region extraction module for applying the lesion segmentation result to the registered low-field magnetic resonance stroke discrimination sequence images to extract a local region; the local region is the region corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence images; the DWI high signal is the image of the part with a value of 1 in the lesion segmentation result; a local signal intensity mean calculation module for calculating the mean image intensity of the local region to obtain a local signal intensity mean; a whole-brain signal intensity mean calculation module for calculating the mean intensity of the entire brain of the registered low-field magnetic resonance stroke discrimination sequence images to obtain a whole-brain signal intensity mean; a mismatch ratio calculation module for comparing the local signal intensity mean with the whole-brain signal intensity mean to obtain a mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist doctors in judging the bleeding or ischemia type of the target object.

[0016] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the MRI mismatch scoring method described in any one of the above.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the MRI mismatch scoring method described in any one of the above.

[0018] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the MRI mismatch scoring method described in any one of the above.

[0019] According to the specific embodiments provided in the present application, the present application has the following technical effects.

[0020] The present application provides an MRI mismatch scoring method, device, equipment, medium and product. By registering the low-field magnetic resonance stroke discrimination sequence images to the DWI images, and applying the lesion segmentation result on the DWI images to the registered low-field magnetic resonance stroke discrimination sequence images, the area corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence images is obtained. By calculating the mean value of the image intensity in this area and comparing it with the mean value of the intensity in the whole brain of the registered low-field magnetic resonance stroke discrimination sequence images, the mismatch ratio is obtained, thereby realizing the DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition based on DWI and the low-field magnetic resonance stroke discrimination sequence that can quickly and accurately judge hemorrhagic stroke. Compared with traditional imaging analysis and relying on doctor interpretation, the DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition method can improve the efficiency of MRI mismatch scoring and the accuracy of MRI mismatch scoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flow chart of an MRI mismatch scoring method provided by an embodiment of the present application.

[0023] Figure 2 It is a diagram of the UNet segmentation network model of the present application.

[0024] Figure 3 It is a schematic flow chart of the paired DWI-low-field magnetic resonance stroke discrimination sequence data processing of the present application.

[0025] Figure 4 It is a schematic diagram of the steps of the method for quickly discriminating hemorrhage and ischemia of the DWI-low-field magnetic resonance stroke discrimination sequence based on the deep learning model of the present application.

[0026] Figure 5 It is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0028] The purpose of the present application is to provide an MRI mismatch scoring method, device, equipment, medium and product, which can perform DWI-low field magnetic resonance stroke discrimination sequence mismatch recognition based on DWI and low field magnetic resonance stroke discrimination sequences, so as to improve the efficiency of MRI mismatch scoring and the accuracy of MRI mismatch scoring results.

[0029] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0030] As Figure 1 shown, an MRI mismatch scoring method provided by the present application includes steps 101 to 108.

[0031] Step 101: Obtain the DWI image and the low field magnetic resonance stroke discrimination sequence image of the target object.

[0032] Step 102: Apply the DWI lesion segmentation model to segment the DWI image to obtain the lesion segmentation result on the DWI image; the DWI lesion segmentation model is obtained by training a neural network based on the U-Net framework using the DWI acute ischemic lesion dataset; the lesion segmentation result is a binary mask.

[0033] Step 103: Perform skull stripping on the DWI image and the low field magnetic resonance stroke discrimination sequence image respectively to obtain the DWI image after skull stripping and the low field magnetic resonance stroke discrimination sequence image after skull stripping.

[0034] Step 104: Register the low field magnetic resonance stroke discrimination sequence image after skull stripping to the DWI image after skull stripping to obtain the registered low field magnetic resonance stroke discrimination sequence image.

[0035] This step 104 specifically includes: using the registration method of ANTs, keeping the DWI image after skull stripping stationary, and transforming the low field magnetic resonance stroke discrimination sequence image after skull stripping into the same physical space as the DWI image after skull stripping through Affine transformation to obtain the registered low field magnetic resonance stroke discrimination sequence image; the Affine transformation includes rotation and translation operations.

[0036] Step 105: Apply the lesion segmentation result to the registered low-field magnetic resonance stroke discrimination sequence image, and extract the local region; the local region is the region corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence image; the DWI high signal is the image of the part with a value of 1 in the lesion segmentation result.

[0037] This step 105 specifically includes: Calculate the lesion segmentation result and the registered low-field magnetic resonance stroke discrimination sequence image according to the formula to extract the local region; where represents the lesion segmentation result, , represents element-wise multiplication of matrices, represents the registered low-field magnetic resonance stroke discrimination sequence image, , represents the local region, , represents the dimension and the size and shape of the image corresponding to each dimension, , , respectively represent is a three-dimensional vector in the real number space, is a three-dimensional vector in the real number space, is a three-dimensional vector in the real number space, and each dimension represents the height , width and depth of the three-dimensional vector.

[0038] Step 106: Calculate the mean image intensity of the local region to obtain the local signal intensity mean.

[0039] This step 106 specifically includes: Use the formula to calculate the mean image intensity of the local region to obtain the local signal intensity mean; where represents the local signal intensity mean, represents taking the expectation. In three-dimensional space, a point in the space is uniquely determined by the coordinates ( , , ), represents the position along the height axis, represents the position along the width axis, represents the position along the depth axis, represents the intensity value of a voxel of the image corresponding to the coordinates ( , , ).

[0040] Step 107: Calculate the intensity mean value of the entire brain in the registered low-field magnetic resonance stroke discrimination sequence image to obtain the signal intensity mean value of the whole brain.

[0041] This step 107 specifically includes: Using the formula to calculate the intensity mean value of the entire brain in the registered low-field magnetic resonance stroke discrimination sequence image to obtain the signal intensity mean value of the whole brain; where represents the signal intensity mean value of the whole brain, represents the image at the voxel intensity value at ( , , ).

[0042] Step 108: Compare the local signal intensity mean value with the signal intensity mean value of the whole brain to obtain a mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist doctors in judging the bleeding or ischemia type of the target object.

[0043] This step 108 specifically includes: Using the formula to calculate the mismatch ratio; where represents the mismatch ratio; the specific process of using to provide a reference for doctors to assist doctors in judging the bleeding or ischemia type of the target object includes: comparing with the high signal threshold . If , it is judged that a high signal appears in the local area. If , it is judged that no high signal appears in the local area. Combining the situation of high signal in DWI, if a high signal appears in the DWI image but no high signal appears in the low-field magnetic resonance stroke discrimination sequence image, it is judged that the DWI image does not match the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is ischemia. If high signals appear in both the DWI image and the low-field magnetic resonance stroke discrimination sequence image, it is judged that the DWI image matches the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is bleeding.

[0044] The following uses a specific embodiment to illustrate the technical solution of the present application.

[0045] The MRI mismatch scoring method provided by this application is a DWI-low-field magnetic resonance stroke discrimination sequence signal mismatch scoring algorithm based on a deep learning model (i.e., a DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition algorithm based on a deep learning model). By applying the deep learning method to the recognition of DWI-low-field magnetic resonance stroke discrimination sequence mismatches, the deep learning model is used to automatically extract the image features of DWI and low-field magnetic resonance stroke discrimination sequences, enabling rapid analysis and evaluation, providing real-time diagnostic results, shortening the diagnostic time, overcoming the limitations of traditional imaging that rely on doctors' subjective judgments, establishing an objective and quantifiable DWI-low-field magnetic resonance stroke discrimination sequence signal mismatch evaluation system, providing a reliable diagnostic basis, assisting clinicians in making decisions, improving the accuracy and efficiency of differentiating ultra-early cerebral hemorrhage and cerebral infarction, reducing the risk of misdiagnosis and missed diagnosis, and providing a brand-new technical means for the rapid and accurate differentiation of ultra-early cerebral hemorrhage and cerebral infarction.

[0046] This application constructs an index for calculating the high-signal mismatch between Diffusion Weighted Imaging (DWI) and low-field magnetic resonance stroke discrimination sequences, and provides a DWI-low-field magnetic resonance stroke discrimination sequence mismatch scoring model based on a deep learning model to rapidly differentiate hemorrhage and ischemia in the ultra-early stage of stroke and assist doctors in diagnosis. This method can automatically segment the lesions shown on the DWI image of the patient and perform rapid segmentation. Then, through the mismatch scoring module designed in this application, the score of the signal mismatch degree between the lesions shown on the current DWI and the signals on the corresponding low-field magnetic resonance stroke discrimination sequence is calculated.

[0047] This application realizes the segmentation of DWI stroke lesions by leveraging the capabilities of the deep model, and calculates the score used to describe the mismatch degree through the designed signal mismatch scoring algorithm to prompt doctors about the hemorrhage and ischemia types of patients in the ultra-early stage. The process of the DWI-low-field magnetic resonance stroke discrimination sequence mismatch score rapid hemorrhage and ischemia differentiation method is briefly described as follows.

[0048] (1) Use the DWI acute ischemic lesion dataset to train the DWI-UNet segmentation network (i.e., a neural network based on the U-Net framework) constructed in this application as shown in Figure 2 to obtain a DWI lesion segmentation model.

[0049] (2) Use the DWI lesion segmentation model to segment the DWI image to obtain the lesion segmentation result on the DWI.

[0050] (3) For a pair of DWI and low-field magnetic resonance stroke discrimination sequence images (the DWI image and the low-field magnetic resonance stroke discrimination sequence image are corresponding), using the registration library in ANTs (Advanced Normalization Tools), register the low-field magnetic resonance stroke discrimination sequence image onto the DWI image. The DWI sequence and the low-field magnetic resonance stroke discrimination sequence can be quickly scanned within a few minutes using a low-field mobile magnetic resonance device. Since when the same patient undergoes a nuclear magnetic resonance scan, multiple sequences are scanned for diagnosis, the DWI and the low-field magnetic resonance stroke discrimination sequence obtained by scanning the same patient are corresponding.

[0051] (4) Apply the obtained segmentation result (the binarized mask) to the registered low-field magnetic resonance stroke discrimination sequence image, calculate the mean value of the image intensity of the part with a value of 1 within the mask, and then calculate the mean value of the intensity within the entire brain of the low-field magnetic resonance stroke discrimination sequence. Divide the two mean values to obtain a ratio that can represent the signal difference between the inside and outside of the region. Among them, the process of obtaining the image intensity of the part with a value of 1 is as follows: The segmentation mask is binarized (only 0 and 1), and each pixel value in the computer represents the voxel intensity value of the image at that place. Therefore, multiply the segmentation mask by the image (a matrix with the same shape and size). The intensity values corresponding to 0 in the segmentation mask in the image become 0, and the parts with 1 remain unchanged, that is: the part corresponding to 1 in the segmentation mask is taken out. For example, to calculate the mean value of the intensity within the entire brain of the low-field magnetic resonance stroke discrimination sequence, for a low-field magnetic resonance stroke discrimination sequence image represented by A, its shape is a (13, 512, 512) matrix, and each element represents the image intensity value. For example, A[0, 0, 0] = 725. First, perform skull stripping on A, as Figure 3 shown, to obtain an image with only the brain, and calculate the overall mean value, that is: add up each element of the matrix and then divide by the total number of elements.

[0052] (5) Compare the size with the ratio obtained in (4) according to the set threshold ( ), and obtain the discrimination result.

[0053] The lesion segmentation model in this application will be described below.

[0054] The DWI-UNet lesion segmentation model (i.e., the DWI lesion segmentation model) used in this application is trained using an improved network based on UNet. First, a DWI acute ischemic lesion dataset is constructed. The original DWI images are selected, and abnormal data is excluded. Then, doctors manually outline the lesion areas as the lesion annotations for the corresponding images, which together with the original image data form the lesion segmentation dataset. The model constructed in this application is based on the traditional U-Net framework. The lesion segmentation network mainly consists of 4 downsampling and upsampling stages. Among them, each downsampling layer includes a max pooling layer, two convolutional layers (with a convolutional kernel size of 3, a stride of 1, and a padding of 1), two InstanceNorm layers, and two ReLU activation functions, and is constructed in the order of max pooling layer, convolutional layer-1, InstanceNorm-1, ReLU-1, convolutional layer-2, InstanceNorm-2, ReLU-2; each upsampling layer includes a transposed convolutional layer (with a convolutional kernel size of 2, a stride of 2, and a padding of 0), two convolutional layers (with a convolutional kernel size of 3, a stride of 1, and a padding of 1), two InstanceNorm layers, and two ReLU activation functions, and is constructed in the order of transposed convolutional layer, convolutional layer-1, InstanceNorm-1, ReLU-1, convolutional layer-2, InstanceNorm-2, ReLU-2. To improve the network's feature extraction ability for medical images, a dense connection method is also adopted in the downsampling stage, and a residual module is added at the Bottleneck to ensure that the segmentation model can extract sufficient information for segmentation. Finally, the output of the network passes through a Sigmoid activation function to limit the output value range to [0, 1], and this value is considered the probability result of the lesion prediction. The reason for the model to output a probability value during prediction is as follows: The annotation is a binary mask of 0 or 1, where 0 indicates normal and 1 indicates a lesion. During model training, it is necessary to predict whether a certain location is a lesion, and the output value is a number between 0 and 1. The larger the value, the higher the confidence of the model that there is a lesion at that location. Therefore, the model outputs a probability value during prediction. The final output generates a segmentation mask based on a confidence threshold. For example, if it is set to 0.5, then when the model predicts that the probability of a lesion at that location is 0.6 (>0.5), it is considered that there is a lesion here, and the mask value finally output for that location is set to 1, otherwise it is set to 0.

[0055] During training, the weighted cross-entropy loss (Weighted Binary Cross Entropy Loss) and DICE loss are used, and the total loss function for training is the sum of the two, and their calculation formulas are as follows respectively.

[0056] 。

[0057] 。

[0058] Among them, represents the weighted cross-entropy loss, represents the DICE loss, represents the weight parameter, represents the th pixel point in an image, represents the true label value at this pixel point, represents the predicted value of the network at this pixel point, and N represents the total number of pixel points in an image. Finally, the network is trained separately by the backpropagation of gradients and the gradient descent algorithm until the network converges. In the way of cross-validation, on the test set, by evaluating the average DICE value and the average Hausdorff Distance value of the model segmentation results, the one with the best evaluation index, that is, the model with the highest segmentation accuracy, is selected for parameter saving, and finally the DWI-UNet lesion segmentation model is obtained.

[0059] The UNet lesion segmentation network is only a step in the scoring method. First, it is necessary to use it to identify the lesion part in the DWI image, and then with the help of the lesion mask map, and then through the following calculation method, corresponding calculations are made on the low-field magnetic resonance stroke discrimination sequence image to obtain the corresponding indicators to achieve the final judgment.

[0060] Next, the DWI-low-field magnetic resonance stroke discrimination sequence mismatch scoring method in this application will be elaborated.

[0061] For the DWI image and the low-field magnetic resonance stroke discrimination sequence image of the same patient, it is considered that if there is a high signal shown on the DWI but no high signal shown on the low-field magnetic resonance stroke discrimination sequence, it means a mismatch, and in this case, the current patient can be directly identified as ischemic; on the contrary, if both the DWI and the low-field magnetic resonance stroke discrimination sequence show high signals, it is a match, and in this case, the current patient can be directly identified as hemorrhagic. The traditional deep learning-based modality signal mismatch algorithm usually calculates based on the segmentation results of different modalities, and it is very challenging to use the lesion segmentation method for the low-field magnetic resonance stroke discrimination sequence: First, the high-signal area in the low-field magnetic resonance stroke discrimination sequence is not as significant as that in the DWI sequence, which will increase the difficulty of model learning and cannot guarantee the segmentation accuracy; Second, if the way of increasing the model scale and complexity is adopted to enhance its learning ability, it will lead to a large increase in training resources and inference time, losing the original goal of this application which aims to quickly and economically identify ischemia; Finally, the construction of the low-field magnetic resonance stroke discrimination sequence lesion dataset will be more time-consuming and laborious compared with other modalities.

[0062] Therefore, this application constructs an index to infer whether there is a high signal in the corresponding low-field magnetic resonance stroke discrimination sequence area based on the DWI high-signal lesion area, and then combines the DWI signal situation (in this application, for the condition, DWI will definitely show a high signal. If there is no high signal in DWI, it means the patient is healthy and no bleeding-ischemia discrimination is required) for comprehensive evaluation to achieve rapid bleeding-ischemia discrimination. The following describes the specific index calculation method: Usually, the images of DWI and the low-field magnetic resonance stroke discrimination sequence of the same patient are not completely aligned, that is, for the same coordinate, the two may correspond to different positions. Therefore, it is necessary to align the same anatomical structures in the two images. When represented by coordinates, it is hoped that they are on the same coordinates. As Figure 3 shown, first use the Synthstrip model to perform skull stripping on DWI and the low-field magnetic resonance stroke discrimination sequence respectively. This not only facilitates the accuracy of subsequent registration, but also is a necessary step for calculating the mean ratio of high signal intensity on the low-field magnetic resonance stroke discrimination sequence later, because the skull part usually has a higher intensity signal than other tissues of the brain; then use the registration method in the ANTs library and adopt the Affine transformation to register the low-field magnetic resonance stroke discrimination sequence image to the DWI image to obtain the registered low-field magnetic resonance stroke discrimination sequence image, and record the image shape as ( , , ), which represent height, width, and the number of slices (i.e., depth) respectively. Among them, registering the low-field magnetic resonance stroke discrimination sequence to DWI means that DWI remains stationary, and through operations such as rotation and translation, the low-field magnetic resonance stroke discrimination sequence is transformed into the same physical space as DWI, and the new image obtained is the registered low-field magnetic resonance stroke discrimination sequence image.

[0063] At this time, the lesion mask segmented from DWI and the registered low-field magnetic resonance stroke discrimination sequence image are calculated according to the following formula to extract the low-field magnetic resonance stroke discrimination sequence area corresponding to the DWI high signal

[0064] .

[0065] Among them, represents the dimension and the size and shape of the image corresponding to each dimension. For example means: is a three-dimensional vector in the real number space, and each dimension represents its height (Height), width (Width), and depth (Depth) respectively. DWI high signal refers to the image of the part with a value of 1 in the lesion segmentation result on DWI.

[0066] Calculate the mean value of the local signal intensity and the mean value of the signal intensity within the whole brain .

[0067] .

[0068] .

[0069] Among them, represents taking the expectation. As described in the meaning of , in three-dimensional space, a point in the space is uniquely determined by the coordinates ( , , ). represents the position along the height axis upward, represents the position along the width axis upward, represents the position along the depth axis upward, represents the intensity value of a voxel of an image corresponding to the coordinates ( , , ). For a 2D image, a position can be determined by ( , ), and the point corresponding to the image position is usually called a pixel point; for a 3D image, a position can be determined by ( , , ), and the point corresponding to the image position is usually called a voxel point. The mean value of the local signal intensity refers to the partial local mean corresponding to the part with a DWI value of 1 in the registered low-field magnetic resonance stroke discrimination sequence, and the mean value of the signal intensity within the whole brain refers to the intensity mean within the whole brain of the low-field magnetic resonance stroke discrimination sequence.

[0070] Calculate the ratio of the local intensity to the overall intensity .

[0071] .

[0072] By comparing with a high-signal threshold that can be set according to the imaging data , if indicates that the intensity signal value of the local area is significantly higher than the intensity mean of the whole brain, it can be considered that there is a high signal in this area; on the contrary, there is no high signal in the local area. Thus, combined with the situation of the DWI high signal, the discrimination of ischemia can be achieved. Among them, is recommended by experienced doctor experts according to the imaging characteristics and can be adjusted according to the data.

[0073] Such as Figure 4As shown in the figure, a method for quickly differentiating hemorrhage and ischemia in a DWI-low-field magnetic resonance stroke differentiation sequence based on a deep learning model proposed in this application is as follows.

[0074] S1: Train a lesion segmentation model.

[0075] Using the constructed segmentation dataset, train the DWI-UNet lesion segmentation network to obtain an accurate lesion segmentation model.

[0076] S2: Segment the lesions in the DWI image.

[0077] Use the trained DWI-UNet segmentation network to segment the lesions in the DWI image (the DWI image without skull stripping). Among them, the input data accepted by the trained UNet segmentation network is the DWI image without skull stripping. Using the trained model to segment the DWI image does not require additional operations on the DWI. Only need to send it to the UNet segmentation network to obtain the segmentation result.

[0078] S3: Preprocess the low-field magnetic resonance stroke differentiation sequence image and the DWI image.

[0079] Use Synthstrip for skull stripping, and use the registration method of ANTs to register the low-field magnetic resonance stroke differentiation sequence with the DWI using Affine transformation to obtain the registered low-field magnetic resonance stroke differentiation sequence brain image and the corresponding DWI brain image. Among them, the preprocessing of the low-field magnetic resonance stroke differentiation sequence and the DWI is for the subsequent calculation steps.

[0080] S4: Calculate the signal intensity of the low-field magnetic resonance stroke differentiation sequence.

[0081] Through the lesion segmentation mask obtained in step S2, calculate on the low-field magnetic resonance stroke differentiation sequence image according to the above-mentioned constructed image signal intensity mean ratio calculation formula to obtain .

[0082] S5: Hemorrhage and ischemia differentiation prompt.

[0083] Compare with the set high-signal threshold and combine the DWI high-signal situation to obtain the final judgment result of hemorrhage or ischemia type. The DWI-low-field magnetic resonance stroke differentiation sequence signal mismatch scoring algorithm based on the deep learning model in this application finally obtains a mismatch ratio, and realizes the judgment of ischemia according to the size relationship between the ratio and the threshold, which is used to assist the doctor's judgment.

[0084] This application constructs an index for the visibility mismatch of lesions between DWI and low-field magnetic resonance stroke discrimination sequences, and develops a DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition algorithm based on a deep learning model to quickly distinguish hyperacute cerebral hemorrhage and cerebral infarction to assist clinicians in making decisions.

[0085] The key points of this application are as follows: 1. This application constructs a model for discriminating hemorrhage and ischemia based on the signal mismatch of DWI-low-field magnetic resonance stroke discrimination sequences using a deep learning model. This model has the ability of automatic segmentation and signal mismatch discrimination, which can help doctors quickly distinguish the types of hemorrhage and ischemia in hyperacute patients and assist in treatment decisions. 2. This application constructs a method for identifying the high signal in the corresponding low-field magnetic resonance stroke discrimination sequence area based on the high signal area of DWI to quantify the matching degree of the high signal of the lesions shown on the DWI image and the low-field magnetic resonance stroke discrimination sequence image.

[0086] The advantages of this application are as follows: 1. The DWI-low-field magnetic resonance stroke discrimination sequence mismatch recognition model based on a deep learning model in this application helps doctors quickly distinguish the types of hemorrhage and ischemia in hyperacute patients, realizes the automatic analysis of medical image data without manual intervention, overcomes the limitation of traditional methods that rely only on doctors' subjective judgment, shortens the diagnosis time, and helps reduce the risk of misdiagnosis and missed diagnosis. 2. The mismatch recognition model in this application is based on deep learning, can accurately identify the phenomenon of high signal mismatch between DWI and low-field magnetic resonance stroke discrimination sequence images, establishes an objective mismatch evaluation system, provides a reliable diagnosis basis, and assists doctors in making better treatment strategies according to the discrimination type. 3. With the help of this application, the discrimination accuracy and efficiency of hyperacute cerebral hemorrhage and cerebral infarction can be improved, further assisting doctors in quickly making treatment decisions, which is expected to have a positive clinical impact on the treatment results of patients, increase the opportunity for patients to obtain timely intervention, and help improve the survival rate and recovery rate of patients.

[0087] Based on the same inventive concept, the embodiment of this application also provides an MRI mismatch scoring device for implementing the above-mentioned MRI mismatch scoring method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the MRI mismatch scoring device provided below can refer to the limitations on the MRI mismatch scoring method in the above text and will not be repeated here.

[0088] In an exemplary embodiment, an MRI mismatch scoring device is provided, which includes the following modules.

[0089] An image acquisition module for acquiring the DWI image and the low-field magnetic resonance stroke discrimination sequence image of the target object.

[0090] A lesion segmentation module, which is used to segment the DWI image by applying a DWI lesion segmentation model to obtain a lesion segmentation result on the DWI image; the DWI lesion segmentation model is obtained by training a neural network based on the U-Net framework using a DWI acute ischemic lesion dataset; the lesion segmentation result is a binary mask.

[0091] An image skull stripping module, which is used to perform skull stripping on the DWI image and the low-field magnetic resonance stroke discrimination sequence image respectively to obtain the DWI image after skull stripping and the low-field magnetic resonance stroke discrimination sequence image after skull stripping.

[0092] An image registration module, which is used to register the low-field magnetic resonance stroke discrimination sequence image after skull stripping onto the DWI image after skull stripping to obtain the registered low-field magnetic resonance stroke discrimination sequence image.

[0093] A local area extraction module, which is used to apply the lesion segmentation result to the registered low-field magnetic resonance stroke discrimination sequence image to extract the local area; the local area is the area corresponding to the DWI high signal on the registered low-field magnetic resonance stroke discrimination sequence image; the DWI high signal is the image of the part with a value of 1 in the lesion segmentation result.

[0094] A local signal intensity mean calculation module, which is used to calculate the mean value of the image intensity of the local area to obtain the local signal intensity mean.

[0095] A whole-brain signal intensity mean calculation module, which is used to calculate the mean value of the intensity within the whole brain of the registered low-field magnetic resonance stroke discrimination sequence image to obtain the whole-brain signal intensity mean.

[0096] A mismatch ratio calculation module, which is used to compare the local signal intensity mean with the whole-brain signal intensity mean to obtain the mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist doctors in judging the bleeding or ischemic type of the target object.

[0097] Among them, the image registration module specifically includes: an image registration unit, which is used to use the registration method of ANTs, keep the DWI image after skull stripping unchanged, and transform the low-field magnetic resonance stroke discrimination sequence image after skull stripping into the same physical space as the DWI image after skull stripping through Affine transformation to obtain the registered low-field magnetic resonance stroke discrimination sequence image; the Affine transformation includes rotation and translation operations.

[0098] The local area extraction module specifically includes: a local area extraction unit, which is used to calculate the lesion segmentation result and the registered low-field magnetic resonance stroke discrimination sequence image according to the formula for calculation to extract the local area; where represents the lesion segmentation result, , represents element - level multiplication of a matrix, represents the registered low - field magnetic resonance stroke discrimination sequence image, , represents a local region, , represents the dimension to which it belongs, and the size and shape of the image corresponding to each dimension, 、 、 respectively represent is a three - dimensional vector in a real - number space, is a three - dimensional vector in a real - number space, is a three - dimensional vector in a real - number space, and each dimension represents the height of the three - dimensional vector 、width and depth .

[0099] The local signal intensity mean calculation module specifically includes: a local signal intensity mean calculation unit, which is used to calculate the image intensity mean of the local region by using the formula to obtain the local signal intensity mean; where, represents the local signal intensity mean, represents taking the expectation. In a three - dimensional space, a point in the space is uniquely determined by the coordinates ( , , ). represents the position along the height axis, represents the position along the width axis, represents the position along the depth axis, represents corresponding to ( , , ) coordinates, the intensity value of a voxel of an image.

[0100] The whole - brain signal intensity mean calculation module specifically includes: a whole - brain signal intensity mean calculation unit, which is used to calculate the intensity mean of the entire brain of the registered low - field magnetic resonance stroke discrimination sequence image by using the formula to obtain the whole - brain signal intensity mean; where, represents the whole - brain signal intensity mean, represents the image at ( , , ) the intensity value of the voxel.

[0101] The mismatch ratio calculation module specifically includes: a mismatch ratio calculation unit, which is used to use the formula Calculate the mismatch ratio; wherein, represents the mismatch ratio; using To provide a reference for the doctor and assist the doctor in judging the specific process of the bleeding or ischemia type of the target object includes: comparing with the high signal threshold in terms of size. If , it is determined that a high signal appears in the local area. If , it is determined that no high signal appears in the local area. Combining the situation of the DWI high signal, if a high signal appears in the DWI image but not in the low-field magnetic resonance stroke discrimination sequence image, it is determined that the DWI image does not match the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is ischemia. If high signals appear in both the DWI image and the low-field magnetic resonance stroke discrimination sequence image, it is determined that the DWI image matches the low-field magnetic resonance stroke discrimination sequence image, and the bleeding or ischemia type of the target object is bleeding.

[0102] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 5 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, 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 MRI mismatch score data. The input / output interface of the computer device is used for the processor to exchange information with external devices. 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, it implements an MRI mismatch scoring method.

[0103] Those skilled in the art can understand that Figure 5 The structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0104] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0105] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0106] 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 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 need to comply with relevant regulations.

[0107] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing 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 method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories 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), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can 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), etc.

[0108] In each of the embodiments provided in the present application, the database involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on a blockchain, etc., and is not limited thereto. In each of the embodiments provided in the present application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., and is not limited thereto.

[0109] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, 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, it should be considered as the scope recorded in this specification.

[0110] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for scoring MRI mismatch, characterized in that: The MRI mismatch scoring method includes: Acquire DWI images and low-field magnetic resonance stroke identification sequence images of the target object; The DWI image is segmented by applying a DWI lesion segmentation model to obtain a lesion segmentation result on the DWI image; the DWI lesion segmentation model is obtained by training a neural network based on a U-Net framework using a DWI acute ischemic lesion data set; the lesion segmentation result is a binary mask; Performing skull stripping on the DWI image and the low-field magnetic resonance stroke identification sequence image respectively to obtain the DWI image after skull stripping and the low-field magnetic resonance stroke identification sequence image after skull stripping; Registering the low-field magnetic resonance stroke identification sequence images after skull stripping to the DWI images after skull stripping to obtain registered low-field magnetic resonance stroke identification sequence images; Apply the lesion segmentation result to the registered low-field magnetic resonance stroke identification sequence image to extract a local area; the local area is an area corresponding to a DWI high signal on the registered low-field magnetic resonance stroke identification sequence image; the DWI high signal is an image of a portion of which the value is 1 in the lesion segmentation result; Calculating the image intensity mean of the local area to obtain the local signal intensity mean; Calculating the intensity mean of the registered low-field magnetic resonance stroke identification sequence images throughout the brain to obtain the signal intensity mean of the whole brain; The local signal intensity mean is compared with the whole brain signal intensity mean to obtain a mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist them in determining the type of hemorrhage or ischemia of the target object.

2. The MRI mismatch scoring method according to claim 1, characterized in that: The low-field magnetic resonance stroke identification sequence image after skull stripping is registered with the DWI image after skull stripping to obtain the registered low-field magnetic resonance stroke identification sequence image, specifically comprising: The ANTs registration method is used to keep the DWI image after skull stripping still, and the low-field magnetic resonance stroke identification sequence image after skull stripping is transformed into the same physical space as the DWI image after skull stripping through Affine transformation to obtain the registered low-field magnetic resonance stroke identification sequence image; the Affine transformation includes rotation and translation operations.

3. The MRI mismatch scoring method according to claim 1, characterized in that: Applying the lesion segmentation result to the registered low-field magnetic resonance stroke identification sequence image to extract the local area specifically includes: The lesion segmentation result and the registered low-field magnetic resonance stroke identification sequence image are combined according to the formula Calculate and extract the local area; among them, represents the lesion segmentation result, represents matrix element-wise multiplication, represents the registered low-field magnetic resonance stroke identification sequence image, Represents a local area, , , Respectively , , It belongs to a three-dimensional vector in the real number space, and each dimension represents the height of the three-dimensional vector ,width and depth .

4. The MRI mismatch scoring method according to claim 3, characterized in that: Calculating the image intensity mean of the local area to obtain the local signal intensity mean, specifically including: Using the formula Calculate the image intensity mean of the local area to obtain the local signal intensity mean; wherein, represents the local signal intensity mean, Indicates the expectation. In three-dimensional space, through the coordinates ( , , ) uniquely identifies a point in space, represents the position along the height axis, Indicates the position along the width axis, represents the position along the depth axis, Indicates that it corresponds to ( , , ) coordinates, the intensity value of an image voxel.

5. The MRI mismatch scoring method according to claim 4, characterized in that: Calculating the intensity mean of the registered low-field magnetic resonance stroke identification sequence image in the whole brain to obtain the signal intensity mean of the whole brain, specifically including: Using the formula Calculate the intensity mean of the registered low-field magnetic resonance stroke identification sequence image in the whole brain to obtain the signal intensity mean of the whole brain; wherein, represents the mean signal intensity in the whole brain, Representing images exist( , , ) is the intensity value of the voxel at .

6. The MRI mismatch scoring method according to claim 5, characterized in that: Comparing the local signal intensity mean with the whole brain signal intensity mean to obtain a mismatch ratio, specifically including: Using the formula Calculate the mismatch ratio; where, Represents the mismatch ratio; using The specific process of providing a reference for doctors and assisting them in determining the type of bleeding or ischemia of the target object includes: With high signal threshold For size comparison, if , it is determined that a high signal appears in the local area. If , it is determined that no high signal appears in the local area, combined with the situation of DWI high signal, if a high signal appears on the DWI image, but no high signal appears on the low-field magnetic resonance stroke identification sequence image, it is determined that the DWI image and the low-field magnetic resonance stroke identification sequence image do not match, and the hemorrhage or ischemia type of the target object is ischemia; if high signals appear on both the DWI image and the low-field magnetic resonance stroke identification sequence image, it is determined that the DWI image and the low-field magnetic resonance stroke identification sequence image match, and the hemorrhage or ischemia type of the target object is hemorrhage.

7. An MRI mismatch scoring device, characterized in that: The MRI mismatch scoring device comprises: An image acquisition module, used to acquire DWI images and low-field magnetic resonance stroke identification sequence images of the target object; A lesion segmentation module is used to segment the DWI image using a DWI lesion segmentation model to obtain a lesion segmentation result on the DWI image; the DWI lesion segmentation model is obtained by training a neural network based on a U-Net framework using a DWI acute ischemic lesion data set; the lesion segmentation result is a binary mask; An image skull stripping module is used to perform skull stripping on the DWI image and the low-field magnetic resonance stroke identification sequence image, respectively, to obtain the DWI image after skull stripping and the low-field magnetic resonance stroke identification sequence image after skull stripping; An image registration module, used for registering the low-field magnetic resonance stroke identification sequence image after skull stripping to the DWI image after skull stripping to obtain the registered low-field magnetic resonance stroke identification sequence image; A local area extraction module is used to apply the lesion segmentation result to the registered low-field magnetic resonance stroke identification sequence image to extract a local area; the local area is the area corresponding to the DWI high signal on the registered low-field magnetic resonance stroke identification sequence image; the DWI high signal is the image of the part with a value of 1 in the lesion segmentation result; A local signal intensity mean value calculation module is used to calculate the image intensity mean value of the local area to obtain the local signal intensity mean value; A whole-brain signal intensity mean calculation module is used to calculate the intensity mean of the registered low-field magnetic resonance stroke identification sequence image in the whole brain to obtain the whole-brain signal intensity mean; The mismatch ratio calculation module is used to compare the local signal intensity mean with the whole brain signal intensity mean to obtain a mismatch ratio; the mismatch ratio is used to provide a reference for doctors to assist doctors in determining the type of bleeding or ischemia of the target object.

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 MRI mismatch scoring method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the MRI mismatch scoring method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the MRI mismatch scoring method according to any one of claims 1 to 6 is implemented.

Images