Image detection method and device, computer equipment and computer readable storage medium

Abstract

The invention relates to an image detection method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: performing blood vessel segmentation on a to-be-detected image through a trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the to-be-detected image; and performing type identification on the blood vessel segmentation result through the trained blood vessel type identification model to obtain the blood vessel type of each pixel point of the to-be-detected image, and obtaining the detection information of the to-be-detected image. By adopting the method, the blood vessels of the to-be-detected image and the object type of each blood vessel can be obtained through the segmentation model, a richer data identification basis can be provided for identification of abnormal conditions in the to-be-detected image, and the diversity of data contained in the detection information is increased.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an image detection method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] In recent years, with the continuous improvement of traditional vascular imaging technology, the visualization effect of imaging results has gradually increased. Commonly used cerebrovascular imaging technologies include magnetic resonance imaging (MRI), computed tomography angiography (CTA) and magnetic resonance angiography (MRA). Among them, MRA images can describe vascular structure and reflect vascular function information such as blood flow velocity. The MRA images can be segmented to achieve vascular detection.

[0003] The cerebral blood vessel segmentation method in the related art generally only distinguishes the blood vessel pixels and non-blood vessel pixels in the image, and the segmentation dimension is relatively single, resulting in low accuracy of the blood vessel segmentation results of the image. Summary of the invention

[0004] Based on this, it is necessary to provide an image detection method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of image segmentation and enrich the segmentation dimension in response to the above technical problems.

[0005] In a first aspect, the present application provides an image detection method, comprising:

[0006] Acquire the image to be detected;

[0007] Performing blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected;

[0008] The trained blood vessel type recognition model is used to perform type recognition on the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained.

[0009] In this embodiment, the trained segmentation model can be used to predict the blood vessels and blood vessel types of each pixel point in the image to be detected, and the detection information corresponding to the image to be detected, including the identified blood vessels and the object type of each blood vessel, can be obtained. This can provide a richer data recognition basis for the identification of abnormal situations in the image to be detected, and increase the diversity of data contained in the detection information.

[0010] In one embodiment, the detection information further includes a brain region prediction label of the pixel point, and the method further includes:

[0011] The image to be detected is segmented by using the trained brain region segmentation model to obtain a brain region segmentation result of the image to be detected, wherein the brain region segmentation result includes a brain region prediction label of each pixel point of the image to be detected.

[0012] In this embodiment, the trained brain region segmentation model is used to perform image segmentation processing on the image to be detected in the brain region dimension, so that the brain region of the image to be detected can be accurately segmented, and the content contained in the obtained detection information can be further enriched, providing a more reliable data basis for the subsequent identification of abnormal situations.

[0013] In one embodiment, the detection information further includes a maximum density projection image, the brain region segmentation result is a brain region segmentation result, and the brain region includes a first brain region and a second brain region; the method further includes:

[0014] Based on the brain region segmentation result, determine the brain region of the image to be detected, and divide the brain region based on the blood vessel types corresponding to each pixel point in the brain region, determine the pixel points that meet the first type condition as the first brain region, and determine the pixel points that meet the second type condition as the second brain region;

[0015] Maximum density projection processing is performed on the brain area of ​​the image to be detected, the first brain area and the second brain area to obtain a maximum density projection image of the image to be detected, a first maximum density projection subimage corresponding to the first brain area and a second maximum density projection subimage corresponding to the second brain area.

[0016] In this embodiment, multiple maximum density projection sub-images can be obtained, which can better describe the stenosis information and dilation information of the blood vessel and provide more diverse detection information.

[0017] In one embodiment, the method further comprises:

[0018] An original contrast image is acquired, and preprocessed to obtain an image to be detected, wherein the preprocessing at least includes resampling processing.

[0019] In this embodiment, by preprocessing the original angiography image, various outliers in the image can be eliminated in advance, and the obtained image to be detected can be ensured to meet the size requirements of the model for the input image, thereby improving the prediction accuracy.

[0020] In one embodiment, obtaining the detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point includes:

[0021] Based on the object prediction label and blood vessel type of each pixel point, obtaining initial detection information of the image to be detected;

[0022] The inverse operation of the resampling is performed on the initial detection information to obtain original detection information corresponding to the original angiography image.

[0023] In this embodiment, when the detection information is obtained after preprocessing the original detection image, the inverse operation of the preprocessing can be performed on the obtained detection information to obtain the original detection information that can characterize the characteristics of the original detection image, thereby reconstructing the detection information of the original detection image and improving the accuracy of the detection information.

[0024] In one embodiment, the image to be detected includes a plurality of sub-images; and the blood vessel segmentation is performed on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, including:

[0025] Performing image segmentation on each of the sub-images of the image to be detected by using the trained object segmentation model to obtain blood vessel segmentation sub-results corresponding to each of the sub-images;

[0026] Based on the blood vessel segmentation sub-results corresponding to the sub-images, a blood vessel segmentation result corresponding to the image to be detected is obtained.

[0027] In this embodiment, for images with larger sizes, image segmentation can be performed using a preset sliding window to expand the image range that can be processed by the model.

[0028] In one embodiment, the method further comprises:

[0029] Acquire target training data, wherein the target training data includes a contrast image sequence, wherein the contrast image sequence includes a plurality of contrast images, and label data of each of the contrast images, wherein the label data includes an object sample label of each pixel point included in each of the contrast images;

[0030] Based on the contrast image sequence in the target training data and the object sample label of each pixel in each contrast image, the object segmentation model to be trained is trained to obtain the object prediction label of each pixel, and based on the object prediction label of each pixel and the sample object label, training is performed to obtain the trained object segmentation model.

[0031] In this embodiment, by performing model training using training data including multiple contrast images and label data of each contrast image, the performance of the segmentation model can be improved, and a more accurate object segmentation result of the contrast image can be obtained.

[0032] In one of the embodiments, the object sample label of each pixel point includes a blood vessel sample label and / or a brain region sample label; the object segmentation model includes a blood vessel segmentation model and / or a brain region segmentation model; and the object prediction label includes a blood vessel prediction label and / or a brain region prediction label.

[0033] In this embodiment, by segmenting the blood vessel segmentation results and the brain region segmentation results, the accuracy of identifying the abnormal conditions of the user can be improved.

[0034] In one embodiment, the label data further includes a sample label of a blood vessel type of each pixel; the method further includes:

[0035] Acquire target pixel points whose object sample labels meet the segmentation conditions, and collect input data of each of the target pixel points, wherein the input data at least includes a blood vessel type sample label;

[0036] Based on the vascular type sample label of each target pixel in the angiography image sequence, the vascular type recognition model to be trained is trained to obtain the vascular type prediction label of each target pixel, and based on the vascular type prediction label of each target pixel and the vascular type sample label, training is performed to obtain the trained vascular type recognition model.

[0037] In this embodiment, under the condition of ensuring low computational complexity, the detection range of the angiography image can be further expanded, each vascular structure included in the angiography image can be accurately determined, and the granularity of image detection can be further refined.

[0038] In a second aspect, the present application also provides an image detection device, comprising:

[0039] An acquisition module, used for acquiring an image to be detected;

[0040] a segmentation module, configured to perform vascular segmentation on the image to be detected by using a trained vascular segmentation model to obtain a vascular segmentation result corresponding to the image to be detected, wherein the vascular segmentation result includes a vascular prediction label of each pixel point of the image to be detected;

[0041] The recognition module is used to perform type recognition on the blood vessel segmentation result through the trained blood vessel type recognition model to obtain the blood vessel type of each pixel point of the image to be detected; and obtain the detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point.

[0042] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0043] Acquire the image to be detected;

[0044] Performing blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected;

[0045] The trained blood vessel type recognition model is used to perform type recognition on the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0047] Acquire the image to be detected;

[0048] Performing blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected;

[0049] The trained blood vessel type recognition model is used to perform type recognition on the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained.

[0050] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0051] Acquire the image to be detected;

[0052] Performing blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected;

[0053] The trained blood vessel type recognition model is used to perform type recognition on the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained.

[0054] The above-mentioned image detection method, apparatus, computer equipment, computer-readable storage medium and computer program product, wherein the method may include acquiring an image to be detected; performing vascular segmentation on the image to be detected by a trained vascular segmentation model to obtain a vascular segmentation result corresponding to the image to be detected, the vascular segmentation result including a vascular prediction label of each pixel of the image to be detected; performing type recognition on the vascular segmentation result by a trained vascular type recognition model to obtain a vascular type of each pixel of the image to be detected; and based on the vascular prediction label and vascular type of each pixel, obtaining detection information of the image to be detected. By adopting this method, the vascular and vascular type of each pixel in the image to be detected can be predicted by a trained segmentation model, and detection information corresponding to the image to be detected including the identified vascular and the object type of each vascular can be obtained, which can provide a richer data recognition basis for the recognition of abnormal conditions in the image to be detected and increase the diversity of data contained in the detection information.

[0055] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used to illustrate the technical solutions of the present disclosure together with the specification.

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 Schematic diagram of a flow chart of an image detection method in one embodiment;

[0059] Figure 2 It is a flowchart of the steps of dividing the brain region segmentation result in one embodiment;

[0060] Figure 3 A schematic diagram of a process for obtaining original detection information in one embodiment;

[0061] Figure 4 is a flow chart of the step of obtaining a blood vessel segmentation result in an image detection method in one embodiment;

[0062] Figure 5 A schematic diagram of a flow chart of a training step of an object segmentation model of an image detection method in one embodiment;

[0063] Figure 6 is a schematic diagram of a training process of an object type recognition model of an image detection method in an embodiment;

[0064] Figure 7 A schematic diagram of a training process of an object segmentation model of an image detection method in one embodiment;

[0065] Figure 8 A schematic diagram of a training process of an object type recognition model of an image detection method in one embodiment;

[0066] Fig. 9 is a schematic flow chart of an image detection method in another embodiment;

[0067] Fig.10 is a schematic diagram of a segmentation result of an image detection method in another embodiment;

[0068] Fig.11 It is a schematic diagram of the effect obtained after dividing the brain region segmentation result of the image detection method in another embodiment;

[0069] Fig.12 is a structural block diagram of an image detection device in one embodiment;

[0070] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] like Figure 1As shown, an image detection method is provided, and the execution subject of the image detection method can be an image detection device. For example, the image detection method can be executed by a terminal device or a server or other processing device, wherein the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the image detection method can be implemented by a processor calling a computer-readable instruction stored in a memory. It can be understood that the method can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the image detection method includes the following steps:

[0073] Step 102: Acquire an image to be detected.

[0074] The image to be detected may be an angiography image to be detected, which can be used to identify abnormal conditions, and may be a magnetic resonance angiography (MRA) image, which describes the structure of a blood vessel and the flow rate of the contents contained in the blood vessel, etc. The angiography image to be detected is also an angiography image to be tested.

[0075] Specifically, the terminal device can receive the contrast image to be detected input by the external device, and the terminal device can also collect the contrast image of the user's target position through the image acquisition device to obtain the contrast image sequence of the user's target position; based on this, the terminal device can determine the contrast image to be detected in the collected contrast image sequence. In one example, the terminal device can randomly extract the contrast image to be detected from the contrast image sequence, or determine the contrast image to be detected in the contrast image sequence by other methods. This disclosure does not specifically limit this method, and those skilled in the art can determine the specific method to be adopted based on the actual application scenario.

[0076] Optionally, the image to be detected may be an MRA image to be detected, and the flow velocity of the content contained in the blood vessel may be a blood vessel flow velocity or the like.

[0077] Step 104 , performing blood vessel segmentation on the image to be detected using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected.

[0078] Among them, the blood vessel segmentation result includes the blood vessel prediction label of each pixel point of the image to be detected. The blood vessel segmentation model is used to segment the area belonging to the blood vessel and the area not belonging to the blood vessel in the contrast image, that is, to identify the blood vessels in the contrast image; the blood vessel prediction label of each pixel point can be a prediction label used to characterize whether each pixel point is a blood vessel after the trained blood vessel segmentation model performs segmentation processing on the image to be detected. For example, the blood vessel prediction label as the first target value can be a characterization that the trained blood vessel segmentation model predicts that the area corresponding to the blood vessel prediction label is a blood vessel, that is, the pixel point corresponding to the blood vessel prediction label is a blood vessel; the blood vessel prediction label as the second target value can be a characterization that the trained blood vessel segmentation model predicts that the pixel point corresponding to the blood vessel prediction label is not a blood vessel; in an example, the first target value can be 1 and the second target value can be 0.

[0079] Specifically, the specific process of the terminal device performing vascular segmentation processing on the image to be detected through the trained vascular segmentation model is that the terminal device predicts the vascular prediction label of each pixel point in the image to be detected through the trained vascular segmentation model, obtains the vascular prediction label corresponding to each pixel point in the model to be detected, and obtains the vascular segmentation result of the image to be detected based on the vascular prediction label of each pixel point.

[0080] Optionally, the blood vessel prediction label is used to indicate whether a pixel point corresponding to the blood vessel prediction label is a blood vessel.

[0081] Step 106, using the trained blood vessel type recognition model, the blood vessel segmentation result is type recognized to obtain the blood vessel type of each pixel of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel, the detection information of the image to be detected is obtained.

[0082] The blood vessel type recognition model is a model for identifying the type of blood vessel at each pixel point based on the blood vessel segmentation result output by the blood vessel segmentation model.

[0083] Specifically, the terminal device can input the blood vessel segmentation result of the image to be detected into the trained blood vessel type recognition model, and through the blood vessel type recognition model, the blood vessel type of each pixel point belonging to the blood vessel in the detection contrast image is recognized and processed to obtain the blood vessel type of each pixel point belonging to the blood vessel. In other words, the blood vessel type recognition model is used to predict the blood vessel type of each pixel point belonging to the blood vessel to obtain the predicted blood vessel type of each pixel point. The terminal device can obtain the detection information of the image to be detected based on the blood vessel prediction label of each pixel point of the image to be detected and the blood vessel type of each pixel point belonging to the blood vessel.

[0084] In one example, the terminal device may determine the blood vessel prediction label of each pixel of the image to be detected and the blood vessel type of each pixel belonging to the blood vessel as the detection information of the image to be detected. In another example, the blood vessel may be a blood vessel or a cerebral blood vessel, and the type of the blood vessel may include one or more of a plurality of cerebral blood vessel structures such as anterior cerebral artery, middle cerebral artery, posterior cerebral artery and basilar artery.

[0085] The above-mentioned image detection method obtains an image to be detected. The image to be detected is segmented by the trained blood vessel segmentation model to obtain the blood vessel segmentation result corresponding to the image to be detected, and the blood vessel segmentation result includes the blood vessel prediction label of each pixel point of the image to be detected. The blood vessel type recognition model is used to identify the type of the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected, and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained. By adopting this method, the blood vessels and blood vessel types of each pixel point in the image to be detected can be predicted by the trained segmentation model, and the detection information corresponding to the image to be detected, which includes the identified blood vessels and the object type of each blood vessel, can be obtained, which can provide a richer data recognition basis for the recognition of abnormal conditions in the image to be detected and increase the diversity of data contained in the detection information.

[0086] In an exemplary embodiment, the detection information also includes a brain region prediction label of the pixel point. Accordingly, the image detection method also includes:

[0087] The image to be detected is segmented by the trained brain region segmentation model to obtain the brain region segmentation result of the image to be detected.

[0088] Among them, the brain region segmentation result includes the brain region prediction label of each pixel point of the contrast image to be detected, and the brain region segmentation model is used to identify the brain region in the contrast image, specifically to segment the area belonging to the brain region and the area not belonging to the brain region in the contrast image to be detected. The brain region prediction label can be used to characterize whether each pixel point in the contrast image to be detected is a prediction label of the brain region. For example, the brain region prediction label of the pixel point as the first target value can be a representation that the trained brain region segmentation model predicts that the pixel point is a brain region; the brain region prediction label as the second target value can be a representation that the trained brain region segmentation model predicts that the pixel point is not a brain region; in one example, the first target value can be 1 and the second target value can be 0.

[0089] Specifically, the terminal device can input the contrast image to be detected into the trained brain region segmentation model, and predict the brain region prediction label of each pixel point in the contrast image to be detected through the trained brain region segmentation model to obtain the brain region prediction label corresponding to each pixel point in the model to be detected, and obtain the second segmentation result of the contrast image to be detected based on the brain region prediction label of each pixel point.

[0090] In this embodiment, the trained brain region segmentation model is used to perform image segmentation processing in the brain region dimension on the contrast image to be detected, so that the brain region of the contrast image to be detected can be accurately segmented, further enriching the content contained in the obtained detection information, and providing a more reliable data basis for the subsequent identification of abnormal situations.

[0091] In an exemplary embodiment, the detection information further includes a maximum intensity projection map, which may be a MIP (Maximum intensity projection) map; the brain region may be a brain region of the user; the brain region includes a first brain region and a second brain region, and in an example, the first brain region may be an intracranial anterior circulation region, and the second brain region may be an intracranial posterior circulation region. Figure 2 As shown, the image detection method also includes:

[0092] Step 202, based on the brain region segmentation result, the brain region of the image to be detected is determined, and based on the blood vessel types corresponding to each pixel point in the brain region, the blood vessels are divided, and the pixels that meet the first type condition are determined to be the first brain region, and the pixels that meet the second type condition are determined to be the second brain region.

[0093] Specifically, the terminal device can perform brain region segmentation on the image to be detected through the trained brain region segmentation model to obtain the brain region segmentation result of the image to be detected; the terminal device can determine the various pixel points in the image to be detected that belong to the brain region through the brain region segmentation result, that is, determine the brain region in the image to be detected; the terminal device can obtain the first type of blood vessels and the second type of blood vessels of each blood vessel in the brain region based on the blood vessel type in each pixel point, and determine the pixel points closer to the first type of blood vessels as pixel points that meet the first type condition, that is, divide them into the first brain region; correspondingly, determine the pixel points closer to the second type of blood vessels as pixel points that meet the second type condition, that is, divide them into the second brain region.

[0094] In one example, the first type of blood vessel may be a first type of blood vessel, which may include a middle cerebral artery (MCA), an anterior cerebral artery (ACA), an internal carotid artery (ICA), and the like; the second type of blood vessel may be a second type of blood vessel, which may include a posterior cerebral artery (PCA), a vertebral artery (VA), and a basilar artery (BA), and the like; the pixel points in the brain region that are closer to the first type of blood vessel are divided into the first brain region, and the pixel points in the brain region that are closer to the second type of blood vessel are divided into the second brain region, that is, the intracranial anterior circulation region and the intracranial posterior circulation region are obtained. In one possible implementation, the distances of each pixel point in the brain region from the first type of blood vessel and the second type of blood vessel may be calculated respectively, and the pixel points that are closer to the first type of blood vessel are divided into the intracranial anterior circulation region, and the pixel points that are closer to the second type of blood vessel are divided into the intracranial posterior circulation region. In another possible implementation, the first type of blood vessel and the second type of blood vessel in the brain region may be expanded at the same expansion ratio to obtain the boundary line between the intracranial anterior circulation region and the intracranial posterior circulation region in the brain region, and the intracranial anterior circulation region and the intracranial posterior circulation region are obtained based on the boundary line.

[0095] Step 204, performing maximum density projection processing on the brain region of the image to be detected, the first brain region, and the second brain region to obtain a maximum density projection image of the image to be detected, a first maximum density projection subimage corresponding to the first brain region, and a second maximum density projection subimage corresponding to the second brain region.

[0096] Specifically, the terminal device can perform maximum density projection processing on the brain region segmentation result of the image to be detected based on the maximum density projection processing algorithm to obtain a maximum density projection map of the image to be detected. And based on the maximum density projection processing algorithm, the first brain region and the second brain region are respectively subjected to maximum density projection processing to obtain a first maximum density projection sub-map belonging to the first brain region, and a second maximum density projection sub-map belonging to the second brain region. In one example, the first maximum density projection sub-map can be an intracranial anterior circulation mip map, and the second maximum density projection sub-map can be an intracranial posterior circulation mip map.

[0097] In this embodiment, by dividing the brain into multiple brain regions based on the blood vessel type, multiple brain regions can be obtained, as well as the mip map of the MRA image to be detected and multiple maximum density projection sub-images, which can better describe the stenosis and dilation information of the blood vessels and provide more diverse detection information.

[0098] In an exemplary embodiment, the image detection method further includes:

[0099] The original contrast image is obtained, and the original contrast image is preprocessed to obtain the image to be detected.

[0100] Among them, the preprocessing at least includes resampling processing, and the preprocessing can also include threshold clipping processing and normalization processing. Threshold clipping can be based on a preset pixel size and pixel threshold to process the image, and the normalization processing can include maximum value normalization processing and / or minimum value normalization processing, etc.

[0101] Specifically, after acquiring the original contrast image, the terminal may preprocess the original contrast image based on a preset pixel size and pixel threshold to obtain a resampled original contrast image, and determine the image to be detected based on the resampled original contrast image.

[0102] In one example, the terminal device may determine the resampled original contrast image as the image to be detected. In another example, after obtaining the resampled original contrast image, the terminal may perform threshold clipping and normalization processing on the resampled original contrast image to obtain a processed contrast image, and determine the processed contrast image as the image to be detected.

[0103] In this embodiment, by preprocessing the original angiography image, various outliers in the image can be eliminated in advance, and the obtained image to be detected can be ensured to meet the size requirements of the model for the input image, thereby improving the prediction accuracy.

[0104] In an exemplary embodiment, Figure 3 As shown, the specific implementation process of the step of "obtaining detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point" may include:

[0105] Step 302: obtaining initial detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel.

[0106] Specifically, the terminal device may determine the obtained blood vessel prediction label of each pixel in the image to be detected, and the blood vessel type (blood vessel type label) of each pixel belonging to the blood vessel as the initial detection information of the image to be detected.

[0107] Step 304 , performing an inverse operation of resampling on the initial detection information to obtain original detection information corresponding to the original angiography image.

[0108] The image to be detected may be obtained by resampling the original detection image.

[0109] Specifically, the terminal device can determine a preset pixel size and pixel threshold for resampling the original detection image, and after obtaining the preset pixel size and pixel threshold, based on the preset pixel size and pixel threshold, perform an inverse operation of resampling the initial detection information of the image to be detected to obtain the detection information of the original angiography image, that is, the original detection information. The original detection information may be a blood vessel prediction label including each original pixel point in the original angiography image, and a blood vessel type of each original pixel point belonging to the blood vessel.

[0110] In the case where the initial detection information of the image to be detected also includes the brain region, the original detection information of the original detection image obtained by the terminal device may also include the brain region prediction label of each original pixel point in the original detection image, etc.

[0111] In this embodiment, when the detection information is obtained after preprocessing the original detection image, the inverse operation of the preprocessing can be performed on the obtained detection information to obtain the original detection information that can characterize the characteristics of the original detection image, thereby reconstructing the detection information of the original detection image and improving the accuracy of the detection information.

[0112] In an exemplary embodiment, the image to be detected includes multiple sub-images. Specifically, each sub-image represents a different region of the image, and each region can be obtained by dividing the image to be detected by the terminal device according to a preset sliding window. For example, the terminal can extract multiple sub-images by presetting a window in the image to be detected, and input each sub-image into the trained object segmentation model.

[0113] Accordingly, if Figure 4 As shown in the figure, the specific implementation process of the step of "performing image segmentation on the image to be detected by using the trained object segmentation model to obtain the object segmentation result corresponding to the image to be detected" may include:

[0114] Step 402 , using the trained blood vessel segmentation model to perform image segmentation on each sub-image of the image to be detected, and obtaining blood vessel segmentation sub-results corresponding to each sub-image.

[0115] Specifically, the terminal device can read different areas of the image to be detected in sequence according to a preset sliding window method, that is, obtain multiple sub-images of the image to be detected; based on this, the terminal device can input each sub-image read into the trained object segmentation model respectively to obtain the blood vessel segmentation sub-result of each sub-image.

[0116] Step 404: based on the blood vessel segmentation sub-results corresponding to the sub-images, obtain the blood vessel segmentation result corresponding to the image to be detected.

[0117] Specifically, the terminal device can obtain the positional relationship between the object segmentation sub-results corresponding to each sub-image according to the position information of each sub-image, and adjust the position of the object segmentation sub-result corresponding to each sub-image based on the position information to obtain the object segmentation result corresponding to the image to be detected.

[0118] In one example, the terminal device can input the multiple sub-images of each of the images to be detected into the blood vessel segmentation model to obtain the blood vessel segmentation sub-results corresponding to each of the sub-images. Based on the blood vessel segmentation sub-results corresponding to the multiple sub-images of the images to be detected, the blood vessel segmentation result of the images to be detected is obtained.

[0119] In one example, the terminal device can input the multiple sub-images of each of the images to be detected into the brain region segmentation model to obtain the brain region segmentation sub-results corresponding to each of the sub-images. Based on the brain region segmentation sub-results corresponding to the multiple sub-images of the image to be detected, the brain region segmentation result of the image to be detected is obtained.

[0120] In this embodiment, for images with larger sizes, image segmentation can be performed using a preset sliding window to expand the image range that can be processed by the model.

[0121] In an exemplary embodiment, Figure 5 As shown, the image detection method also includes:

[0122] Step 502: Obtain target training data.

[0123] The target training data includes a contrast image sequence, the contrast image sequence includes multiple contrast images, and label data of each contrast image, the label data includes the object sample label of each pixel point contained in each contrast image. The contrast image sequence can be an MRA image sequence of each target object, and the MRA image sequence refers to a 3D image of MRA, which includes multiple 2D images of MRA.

[0124] Specifically, the terminal can obtain multi-sequence data of multiple users, and filter the multi-sequence data to obtain a contrast image sequence of the user's target position, and the contrast image sequence can be an MRA image sequence; the contrast image sequence can contain multiple contrast images that can be MRA images, and the terminal device can annotate each contrast image to obtain an object sample label for each contrast image. For example, the terminal device can determine the position information of each pixel point for each of the multiple pixel points contained in each contrast image, and can also determine the object sample label of each pixel point. For example, the label value of the object sample label of the pixel point belonging to the object can be set to the first target value, and the label value of the object sample label of the pixel point that does not belong to the object can be set to the second target value. In one example, the first target value can be 1, the second target value can be 0, and so on.

[0125] In one example, after acquiring a contrast image sequence and obtaining label data for each contrast image in the contrast image sequence, the terminal device can sequentially resample, threshold clip, and normalize each contrast image in the contrast image sequence and the label data for each contrast image based on a preset pixel size and pixel threshold to obtain target training data.

[0126] Step 504: Based on the contrast image sequence in the target training data and the object sample label of each pixel in each contrast image, the object segmentation model to be trained is trained to obtain the object prediction label of each pixel, and based on the object prediction label of each pixel and the sample object label, training is performed to obtain the trained object segmentation model.

[0127] Specifically, the terminal device can input multiple contrast images contained in the contrast image sequence in the target training data into the object segmentation model to be trained to obtain the output result of the object segmentation model to be trained, and the output result can be the object segmentation result; the terminal device can calculate the error spacing value between the object segmentation result and the object sample label of each pixel point in each contrast image contained in the target training data, and obtain the loss function based on the error spacing value; the terminal device can update the model parameters of the object segmentation model to be trained based on the parameter update algorithm and the loss value corresponding to the loss function, update the object segmentation model to be trained based on the obtained updated model parameters, and re-execute the above-mentioned step in which the terminal device can input multiple contrast images contained in the contrast image sequence in the target training data into the object segmentation model to be trained through the updated object segmentation model, until the obtained loss function meets the preset training completion condition, at this time, the trained object segmentation model is obtained.

[0128] In one example, the parameter update algorithm may be a gradient descent algorithm, and the loss function may be a segmentation loss function, which may include a DICE loss function (a metric function used to measure set similarity) and a cross entropy loss function. The preset training completion condition may include that the current loss function no longer decreases, that is, the loss value corresponding to the loss function has converged, etc. The object segmentation model to be trained may be any one of Fully Convolutional Networks (FCN), UNet, and VNet, and may also include other types of models, which will not be described here.

[0129] In this embodiment, by performing model training using training data including multiple contrast images and label data of each contrast image, the performance of the segmentation model can be improved, and a more accurate object segmentation result of the contrast image can be obtained.

[0130] In an exemplary embodiment, the object sample label of each pixel includes a blood vessel sample label and / or a brain region sample label. The object segmentation model includes a blood vessel segmentation model and / or a brain region segmentation model. The object prediction label includes a blood vessel prediction label and / or a brain region prediction label, the brain region is a brain region, and the blood vessel is a blood vessel.

[0131] In one example, the object sample label of each pixel point may include a blood vessel sample label and a brain region sample label, the object segmentation model may include a blood vessel object segmentation model and a brain region object segmentation model; the object prediction label may be a blood vessel prediction label and a brain region prediction label.

[0132] In this embodiment, the accuracy of identifying abnormal conditions of the user can be improved through the segmented blood vessel segmentation results and the brain region segmentation structure.

[0133] In an exemplary embodiment, the label data also includes an object type sample label for each pixel; specifically, the object in the object type sample label may be a blood vessel, the object type sample label may be a blood vessel type sample label, and the label data may include an object type sample label for each target pixel belonging to the blood vessel, that is, a blood vessel type sample label for each target pixel belonging to the blood vessel.

[0134] Accordingly, if Figure 6 As shown, the image detection method also includes:

[0135] Step 602, obtaining target pixels whose object sample labels meet the segmentation conditions, and collecting input data of each target pixel, the input data at least including a blood vessel type sample label;

[0136] Among them, the object sample label can be a blood vessel sample label; the segmentation condition can be to determine that the pixel point belongs to the blood vessel based on the blood vessel sample label; the target pixel point is the pixel point belonging to the blood vessel determined based on the blood vessel sample label; in one example, the input data can also include point cloud data of each pixel point, and the point cloud data can be, for example, three-dimensional position coordinate information of each pixel point, etc.; the blood vessel type sample label can be a blood vessel type sample label.

[0137] Specifically, the terminal device can determine each target pixel belonging to a blood vessel in each angiography image based on the output result of the blood vessel segmentation model, and collect blood vessel type sample labels annotating each target pixel. That is, the terminal device can also collect blood vessel type sample labels corresponding to each target pixel, that is, the terminal device needs to determine the blood vessel type label of each target pixel.

[0138] Step 604: Based on the sample blood vessel type labels of each target pixel in the angiography image sequence, the blood vessel type recognition model to be trained is trained to obtain the predicted blood vessel type labels of each target pixel, and based on the predicted blood vessel type labels of each target pixel and the sample blood vessel type labels, training is performed to obtain a trained blood vessel type recognition model.

[0139] Specifically, the terminal device can input each target pixel point in the angiography image sequence into the vascular type recognition model to be trained to obtain the output result of the vascular type recognition model to be trained, and the output result can be the vascular type prediction label corresponding to each pixel point; the terminal device can calculate the loss function between the vascular type sample label of each pixel point and , and update the model parameters of the vascular type recognition model to be trained through the parameter updating algorithm and the loss value corresponding to the loss function, update the vascular type recognition model to be trained based on the obtained updated model parameters, and re-execute the above-mentioned step in which the terminal device can input each target pixel point in the angiography image sequence into the vascular type recognition model to be trained through the updated vascular type recognition model, until the obtained loss function meets the preset training completion condition, at which time, the trained vascular type recognition model is obtained.

[0140] In another example, the terminal device can input each target pixel point in the angiography image sequence into the vascular type recognition model to be trained to obtain the output result of the vascular type recognition model to be trained, and the output result can be the vascular type prediction label corresponding to each position data, that is, the vascular type prediction label of each pixel point; the terminal device can calculate the loss function between the vascular type sample label of each position data and , and update the model parameters of the vascular type recognition model to be trained through the parameter updating algorithm and the loss value corresponding to the loss function, update the vascular type recognition model to be trained based on the obtained updated model parameters, and re-execute the above-mentioned step that the terminal device can input the position data of each target pixel point in the angiography image sequence into the vascular type recognition model to be trained through the updated vascular type recognition model, until the obtained loss function meets the preset training completion condition, at which time, the trained vascular type recognition model is obtained.

[0141] In one example, the parameter update algorithm may be a back propagation algorithm, and the loss function may be a segmentation loss function, which may include a DICE loss function and a cross entropy loss function. The preset training completion condition may include that the current loss function no longer decreases, that is, the loss value corresponding to the loss function has converged, etc.

[0142] In one example, the terminal device can extract the point cloud data of the blood vessel from the segmentation label, and then assign corresponding naming labels to the point cloud data at the corresponding position according to the blood vessel segmentation label. Since the number of blood vessel point cloud data of different patients is different, it is necessary to randomly sample the point cloud data to the same number. The point cloud data can be used for model training after z-score regularization. Similar to the training process of blood vessel segmentation, the training will go through multiple epochs. In one epoch, the point cloud data in the training set is first input into the blood vessel naming model according to a certain batch size, and the output is the predicted value corresponding to each point cloud. The loss value between it and the naming label is calculated, and then the naming model is optimized and updated through back propagation.

[0143] In this embodiment, under the condition of ensuring low computational complexity, the detection range of the angiography image can be further expanded, each vascular structure included in the angiography image can be accurately determined, and the granularity of image detection can be further refined.

[0144] In the following, in conjunction with a specific embodiment, a detailed description of the specific implementation process of the above-mentioned image detection method may include:

[0145] Cerebrovascular abnormalities are relatively serious abnormalities that users may experience, including intracranial vascular abnormalities, cerebral vascular stenosis, and stroke. The probability of cerebrovascular abnormalities has been on the rise in recent years. Cerebrovascular status is an important indicator for determining whether cerebrovascular abnormalities occur. At present, fully understanding the structure and morphology of cerebrovascular information can further detect and know the abnormalities of users' cerebrovascular conditions early, which is of great significance for the determination and improvement of abnormal conditions.

[0146] Medical imaging technology is the basis for assisting in the identification of abnormal cerebrovascular conditions. By observing the imaging information, the cerebrovascular structure can be analyzed and abnormal changes in cerebrovascular morphology can be found, so as to judge and prevent abnormal cerebrovascular conditions. In recent years, with the continuous improvement of traditional vascular imaging technology, the visualization effect of imaging results has gradually increased. Commonly used cerebrovascular imaging technologies include magnetic resonance imaging (MRI), computed tomography angiography (CTA) and magnetic resonance angiography (MRA). Among them, MRA has the advantages of simple imaging and no radiation. MRA images can not only describe the vascular structure, but also reflect vascular function information such as blood flow velocity. It is a common choice for cerebrovascular examination and is widely used.

[0147] Since cerebral blood vessels have multiple branches, some small structures and complex and changeable shapes, their imaging quality is easily affected by factors such as hemodynamics, noise and bias fields. Direct analysis of MRA images depends on the anatomical basis and film reading experience in the empirical database. A more intuitive and clear observation method is to segment and reconstruct the three-dimensional structure of cerebral blood vessels. With the help of the results of vascular reconstruction, the user's vascular information can be directly observed in a non-invasive way, and various abnormalities of cerebral blood vessels such as arterial stenosis and aneurysm can be detected and found.

[0148] Accurate cerebral vascular segmentation is an important prerequisite for three-dimensional reconstruction, and its purpose is to distinguish vascular and non-vascular pixels in the image. However, due to the complex structure of blood vessels, the most direct manual segmentation requires a lot of time and manpower, and the segmentation results often depend on the experience of clinical workers, lack repeatability and reproducibility, and are also inefficient. Semi-automatic segmentation methods including threshold segmentation and region growing also require manual participation in the selection of algorithm seed points, which are highly subjective, and such methods do not perform well in the segmentation of small blood vessels. A fully automatic vascular segmentation method is needed.

[0149] The fully automatic blood vessel segmentation method in the related technology is a deep neural network represented by UNet, which has shown good performance in various medical image segmentation fields. The head MRA blood vessel reconstruction method based on deep neural network has important practical significance in reducing the workload of abnormal condition identifiers and providing accurate and reliable cerebrovascular information. However, most cerebrovascular segmentation methods only distinguish between vascular pixels and non-vascular pixels, the division dimension is relatively single, and the segmentation accuracy is poor.

[0150] With the continuous improvement of computer technology and cerebrovascular imaging databases, deep learning algorithms are gradually applied from natural images to medical images. The cerebrovascular segmentation method based on deep neural networks automatically mines and learns the semantic information of cerebrovascular images in an end-to-end manner to drive the model, which has the advantages of high accuracy and strong generalization performance. The image detection method in the present disclosure can realize the recognition of different types of each blood vessel based on the unique structure of various types of cerebrovascular itself, that is, to realize the naming of different segments of blood vessels. The image detection method in the present disclosure can be a cerebrovascular reconstruction method based on deep learning. On the basis of completing cerebrovascular segmentation, different blood vessels are also segmented and named. At the same time, the intracranial vascular MIP map (including blood vessels and the brain area they supply blood) is given in combination with the brain region segmentation results, which provides more abundant information for the subsequent user's identification and determination of cerebrovascular abnormalities. The occurrence of different types of abnormalities may be related to the morphology of different types of cerebrovascular vessels.

[0151] The image detection device used in the image detection method (i.e., the head MRA vascular reconstruction method) disclosed in the present invention may include the following modules:

[0152] Data preprocessing module. Filter out the human brain MRA image sequence from the user's multi-sequence data, annotate and obtain the corresponding brain region label, vascular segmentation label and vascular segmentation label, and store them in NIfTI format. Use SimpleITK to read MRA images, and perform preprocessing operations such as resampling, threshold clipping and normalization on different images according to the preset pixel size and pixel threshold to suppress outliers. For training data, various labels corresponding to the image will also be preprocessed through pixel resampling and clipping to achieve matching between images and labels.

[0153] Specifically, the user's medical data usually contains sequences of multiple modalities, and it is necessary to screen out suitable MRA sequence data. There may be differences in the pixel size and image size between different MRA data. The terminal device can perform pixel resampling, pixel threshold clipping, and image normalization on the MRA data. The preset pixel size can be 0.5mm*0.5mm*0.5mm. In addition, since MRA images usually have a high resolution and a large image space dimension, it is difficult to directly use the complete image for model training due to the limitation of the video memory size. Therefore, it is necessary to cut the image into blocks (for example, into 192*192*192 3D image blocks), and a random cutting strategy is implemented for the image during the training process.

[0154] The segmentation model training module includes cerebrovascular segmentation model training and brain region segmentation model training. Cerebrovascular segmentation uses preprocessed MRA images and cerebrovascular segmentation labels to participate in training. Model training usually includes multiple processes (epochs) in which the entire training data set is completely traversed by the neural network. In one epoch, the images in the training set are divided into multiple batches according to a certain batch size (the number of images contained in the batch). The MRA images in each batch are subjected to Gaussian noise enhancement, scaling, rotation, and flipping operations according to a certain probability to generate more perturbation samples and improve the generalization ability of the model. The perturbed data will be directly input into the vascular segmentation model, and the output result will be obtained after the forward propagation of the network model. The error distance between the output result and the segmentation label is calculated to obtain the loss value, and finally the model parameters are optimized and updated through back propagation. During training, AdamW is selected as the optimizer, and the learning rate is 0.001 until the model loss converges. Brain region segmentation uses preprocessed MRA images and brain region labels to participate in training. Its segmentation model and training process are similar to those of cerebrovascular segmentation.

[0155] like Figure 7The figure may be a specific training process of the segmentation model, wherein the input of the model is a three-dimensional MRA image with a dimension of 1×d×h×w. The segmentation model as a whole is an encoder-decoder architecture, including but not limited to segmentation models such as UNet, VNet and their variants, including; an input part, a downsampling part, an upsampling part and an output part. The segmentation model receives the preprocessed MRA image as input, and can perform encoding processing corresponding to stage1, encoding processing corresponding to stage2, encoding processing corresponding to stage3, and encoding processing corresponding to stage4 on the input data to obtain the encoding result, and then perform decoding processing corresponding to stage4, decoding processing corresponding to stage3, decoding processing corresponding to stage2, and decoding processing corresponding to stage1 on the encoding result in turn to obtain the output result, that is, output the blood vessel segmentation result or the brain region segmentation result at the output end, calculate the error distance between the output result and the corresponding label, and then obtain the segmentation loss (segmentation loss, Ls). The segmentation loss can be measured by dice or cross entropy. The purpose of optimizing the loss is to improve the accuracy of segmentation. Formulas (1) and (2) of dice and cross entropy are as follows:

[0156] (1)

[0157] (2)

[0158] Among them, Y represents the segmentation label label, that is, the sample label, X represents the segmentation prediction result, that is, the predicted label; m represents the number of categories, A preset constant used to avoid the denominator being 0.

[0159] Blood vessel naming is based on blood vessel segmentation. Before training begins, the blood vessel point cloud data is first extracted from the segmentation label, and then the corresponding point cloud data at the corresponding position is assigned a corresponding naming label based on the blood vessel segmentation label. Since the number of blood vessel point cloud data of different patients is different, the terminal device can randomly sample the point cloud data to the same number. The training will go through multiple epochs. In one epoch, the point cloud data in the training set is first input into the blood vessel naming model according to a certain batch size, and the output is the predicted value corresponding to each point cloud. The loss value between it and the naming label is calculated, and then the naming model is optimized and updated through back propagation. Figure 8As shown, it can be a structural diagram of the naming model (object type recognition model) training module. Among them, the input data is the point cloud data of the blood vessel, the dimension is n*3, n is the number of point clouds after sampling, 3 represents the three-dimensional coordinates of each point in the point cloud, and the dimension of the output result is n*m, m represents the number of blood vessel naming types. The blood vessel naming model includes but is not limited to point cloud analysis models such as PointNet and PointNet++. The PointNet uses a multi-layer perceptron (MLP) to extract features from each point, and then aggregates the features of the points into global features through a maximum pooling layer, and finally uses these features to perform tasks such as classification or regression; the PointNet++ can be a model that divides the point cloud data into multiple local areas through a hierarchical feature learning mechanism and extracts features in each area; the naming model is mainly composed of a fully connected layer and a maxpool layer, which receives point cloud data as input, outputs the naming result of the point cloud data, and calculates the error distance between it and the naming label of the point cloud data to obtain the naming loss (naming loss, Ln), which can be measured by cross entropy.

[0160] like Fig. 9 As shown in the figure, during data inference, the MRA image to be tested will first be processed according to the same resampling and threshold clipping operations in the data preprocessing module, and then directly input into the vascular segmentation model to obtain the segmentation results of the intracranial arteries, and then extract the three-dimensional coordinates of the blood vessels from the segmentation results to obtain point cloud data. The point cloud data is randomly sampled and regularized and then input into the naming model (vascular type recognition model) to obtain the naming results. After that, the K nearest neighbor algorithm can be used to assign corresponding naming results to all point cloud data before sampling, so as to obtain the names of all blood vessels. Finally, the inverse operation of pixel resampling can be used to reconstruct the cerebral vascular segmentation and naming results of the original image. When segmenting blood vessels, if the hardware performance does not meet the requirements of inputting the entire image at one time, different areas of the image can be input multiple times in a sliding window manner to obtain segmentation results of different areas. These segmentation results can be merged to obtain the cerebral vascular segmentation results.

[0161] In one example, the terminal device can sample the point cloud data of each pixel belonging to a blood vessel in the MRA image to be tested, for example, by random sampling or sampling based on a preset sampling interval, to obtain target point cloud data, and input the target point cloud data into a naming model to obtain the blood vessel type of each blood vessel pixel corresponding to the target point cloud data, and fit the blood vessel type of each blood vessel pixel corresponding to the target point cloud data through a K-nearest neighbor algorithm to obtain the blood vessel type of all pixel points belonging to the blood vessel in the MRA image to be tested.

[0162] like Fig.10As shown, it can be the blood vessel segmentation result, the brain region segmentation result and the blood vessel naming result; Fig.11 As shown, the MRA image to be tested will also be input into the brain region segmentation model to obtain a brain region segmentation result. The brain region in the MRA image to be tested is determined based on the brain region segmentation result, and the brain region is divided based on the blood vessel type of each pixel in the brain region to obtain a first brain region and a second brain region. The first brain region may be an intracranial anterior circulation region, and the second brain region may be an intracranial posterior circulation region. Based on this, the brain region, the intracranial anterior circulation region, and the intracranial posterior circulation region may be projected with maximum density, respectively, to obtain a mip map of the brain region (i.e., the mip map of the MRA image to be tested), an intracranial anterior circulation mip map, and an intracranial posterior circulation mip map, so as to better display information such as stenosis and dilation of blood vessels.

[0163] In one example, the specific process of dividing the brain region based on the blood vessel type may be to determine the first type of blood vessels and the second type of blood vessels based on the blood vessel type of each pixel point, the first type of blood vessels includes the middle cerebral artery (MCA), the anterior cerebral artery (ACA) and the internal carotid artery (ICA) and other blood vessels, the second type of blood vessels includes the posterior cerebral artery (PCA), the vertebral artery (VA) and the basilar artery (BA) and other blood vessels, and the pixel points in the brain region that are closer to the first type of blood vessels are divided into the first brain region, and the pixel points in the brain region that are closer to the second type of blood vessels are divided into the second brain region, that is, the intracranial anterior circulation region and the intracranial posterior circulation region are obtained. In a possible implementation method, the distance of each pixel point in the brain region from the first type of blood vessels and the second type of blood vessels can be calculated respectively, and the pixel points closer to the first type of blood vessels are divided into the intracranial anterior circulation region, and the pixel points closer to the second type of blood vessels are divided into the intracranial posterior circulation region. In another possible implementation, the first type of blood vessels and the second type of blood vessels in the brain region can be expanded at the same expansion ratio to obtain a boundary line between the anterior intracranial circulation region and the posterior intracranial circulation region in the brain region, and the anterior intracranial circulation region and the posterior intracranial circulation region can be obtained based on the boundary line.

[0164] Fig.10 What is displayed is the blood vessel segmentation results of the contrast image to be detected and the brain region segmentation results, as well as the blood vessel type of each pixel, that is, the blood vessel naming; Fig.11 Displayed are the MIP maps of MRA images, the MIP maps of the anterior intracranial circulation, and the MIP maps of the posterior intracranial circulation. The vascular naming types include various cerebrovascular structures such as the anterior cerebral artery, middle cerebral artery, posterior cerebral artery, and basilar artery.

[0165] The image detection device corresponding to the image detection method in the present disclosure may include the following contents:

[0166] The data preprocessing module mainly includes operations such as data screening and data regularization, the segmentation model training module: using the existing segmentation data to train the segmentation network, the naming model training module: using the existing segmentation and naming data to train the naming network, the model inference module: using the trained neural network model to segment and name the cerebral blood vessels in the new MRA image, and also extract the mip map.

[0167] The segmentation model training module includes cerebral vascular segmentation and brain region segmentation. The vascular segmentation model receives the preprocessed MRA image as input, and outputs the vascular segmentation result or the brain region segmentation result at the output end. The error distance between it and the corresponding label is calculated to obtain the segmentation loss (Ls). The segmentation loss function includes conventional segmentation losses (such as DICE and cross entropy). The gradient descent algorithm is used to optimize the loss function, and the network parameters are continuously iterated until the loss function no longer decreases, the model converges, and the training ends.

[0168] The segmentation model can be composed of an input part, a downsampling part, an upsampling part, and an output part. The input part consists of a 3D convolution layer, a BN layer (atch Normalization layer), and an activation layer, which is used to improve the feature expression ability of the input image. Both the downsampling part and the upsampling part include multiple stages, each of which consists of a 3D convolution layer, a BN layer, an activation layer, and a Dropout layer. At the same time, in each corresponding stage, the image features are also transferred from the downsampling part to the upsampling part for feature merging. The output part consists of a 3D convolution layer and a BN layer, which is used to output the segmentation result.

[0169] The training module of the object type recognition model uses point cloud data for training. The point cloud data represents the three-dimensional coordinates of the blood vessel, which is extracted from the blood vessel segmentation label. The label corresponding to each point in the point cloud data is the named type of the blood vessel pixel at the coordinate, which is extracted from the blood vessel segmentation label. The training process of the naming model is similar to that of the segmentation model. The network structure used includes but is not limited to point cloud analysis models such as PointNet and PointNet++. The object type recognition model includes structures such as inputtransform (input module), feature transform (feature transformation), mlp (multi-layer perceptron) and maxpool (maximum pooling layer). Among them, input transform and feature transform are both composed of 2D convolutional layer, BN layer, activation layer and maxpool layer, which are used to aggregate point cloud features. After maxpool combines global features and local features, the predicted type of the point cloud is output.

[0170] Specifically, the process of obtaining the blood vessel naming result may be to process the blood vessel segmentation result of the MRA image to obtain the blood vessel naming result of the MRA image. Another implementation method may be to first extract the point cloud data of each target pixel belonging to the blood vessel, and perform blood vessel naming for the pixel corresponding to each point cloud data, and then name all blood vessel pixels (i.e., each target pixel) in the MRA image according to the nearest neighbor algorithm to obtain the type of each blood vessel contained in the MRA image.

[0171] The MRA image to be tested will first be processed by the data preprocessing module, and then input into the vascular segmentation model and the point cloud naming model to obtain the vascular segmentation and naming results, and then input into the brain region segmentation model to obtain the brain region segmentation result. Unlike the training phase, which can use only part of the image data by cutting, the inference phase needs to output the result of the complete image. If the hardware performance does not meet the requirements of inputting the entire image at one time, different areas of the image can be input multiple times in a sliding window manner, and the results of different areas can be merged to obtain the overall result. The input of the naming model is the sampled point cloud data, and the output is also the naming result of some vascular pixels. After fitting this part of the data using the K nearest neighbor algorithm, all vascular pixels can be assigned naming results. The image detection method disclosed in the present invention is a cerebrovascular reconstruction method based on deep learning. On the basis of completing the cerebrovascular segmentation, different blood vessels are also segmented and named, and a vascular MIP map is given at the same time, providing richer information for detecting and diagnosing cerebrovascular abnormalities. The cerebrovascular vessels can be automatically extracted and named by this method. The cerebrovascular vessels of patients can be better observed and abnormal conditions can be identified through the MIP map, thereby improving the recognition efficiency and accuracy. Moreover, the abnormal condition improvement and prevention can be formulated according to the three-dimensional structure of the cerebrovascular vessels, thereby improving the improvement success rate.

[0172] It should be noted that the term 'and / or' in this article is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent the three situations of: A exists alone, A and B exist at the same time, and B exists alone. In addition, 'at least one of the terms in this article' represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent any one or more elements selected from the set consisting of A, B, and C.

[0173] It should be noted that those skilled in the art can understand that in the above method of the specific implementation mode, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0174] It should be noted that the above description of the various embodiments tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. For the sake of brevity, they will not be repeated herein.

[0175] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0176] Based on the same inventive concept, the embodiment of the present application also provides an image detection device for implementing the above-mentioned image detection method. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above-mentioned method, so the specific limitations in the one or more image detection device embodiments provided below can refer to the limitations on the image detection method above, and for the sake of brevity, they will not be repeated here.

[0177] In an exemplary embodiment, Fig.12 As shown, an image detection device 1200 is provided, comprising:

[0178] An acquisition module 1202 is used to acquire an image to be detected;

[0179] The segmentation module 1204 is used to perform blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected;

[0180] The recognition module 1206 is used to perform type recognition on the blood vessel segmentation results through the trained blood vessel type recognition model to obtain the blood vessel type of each pixel point of the image to be detected; and obtain detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point.

[0181] In one embodiment, the detection information further includes a brain region prediction label of the pixel point, and the image detection device further includes:

[0182] The image processing module is used to perform image segmentation on the image to be detected by using the trained brain region segmentation model to obtain a brain region segmentation result of the image to be detected, wherein the brain region segmentation result includes a brain region prediction label of each pixel point of the image to be detected.

[0183] In one embodiment, the detection information further includes a maximum density projection image, and the brain region includes a first brain region and a second brain region; the image detection device further includes:

[0184] The division module is used to determine the brain region of the image to be detected based on the brain region segmentation result, and to divide the brain region based on the blood vessel types corresponding to each pixel point in the brain region, so as to determine that the pixel points meeting the first type condition are the first brain region, and to determine that the pixel points meeting the second type condition are the second brain region.

[0185] The projection processing module is used to perform maximum density projection processing on the brain area of ​​the image to be detected, the first brain area and the second brain area to obtain a maximum density projection image of the image to be detected, a first maximum density projection sub-image corresponding to the first brain area and a second maximum density projection sub-image corresponding to the second brain area.

[0186] In one embodiment, the image detection device further includes:

[0187] The image acquisition module is used to acquire the original contrast image and preprocess the original contrast image to obtain the image to be detected, wherein the preprocessing at least includes resampling processing.

[0188] In one embodiment, the identification module is specifically used for:

[0189] Based on the object prediction label and blood vessel type of each pixel point, obtaining initial detection information of the image to be detected;

[0190] The inverse operation of the resampling is performed on the initial detection information to obtain original detection information corresponding to the original angiography image.

[0191] In one embodiment, the image to be detected includes a plurality of sub-images; and the segmentation module is specifically used for:

[0192] Performing image segmentation on each of the sub-images of the image to be detected by using the trained object segmentation model to obtain blood vessel segmentation sub-results corresponding to each of the sub-images;

[0193] Based on the blood vessel segmentation sub-results corresponding to the sub-images, a blood vessel segmentation result corresponding to the image to be detected is obtained.

[0194] In one embodiment, the image detection device further includes:

[0195] A data acquisition module, used to acquire target training data, wherein the target training data includes a contrast image sequence, wherein the contrast image sequence includes a plurality of contrast images, and label data of each of the contrast images, wherein the label data includes an object sample label of each pixel point included in each of the contrast images;

[0196] A segmentation model training module is used to train the object segmentation model to be trained based on the contrast image sequence in the target training data and the object sample label of each pixel in each contrast image to obtain the object prediction label of each pixel, and to train based on the object prediction label of each pixel and the sample object label to obtain a trained object segmentation model.

[0197] In one of the embodiments, the object sample label of each pixel point includes a blood vessel sample label and / or a brain region sample label; the object segmentation model includes a blood vessel segmentation model and / or a brain region segmentation model; and the object prediction label includes a blood vessel prediction label and / or a brain region prediction label.

[0198] In one embodiment, the label data further includes an object type sample label for each pixel point; the image detection device further includes:

[0199] A pixel acquisition module, used to acquire target pixels whose object sample labels meet the segmentation conditions, and to collect input data of each of the target pixels, wherein the input data at least includes an object type sample label;

[0200] The recognition model training module is used to train the object type recognition model to be trained based on the object type sample label of each target pixel in the contrast image sequence to obtain the object type prediction label of each target pixel, and to train based on the object type prediction label and object type sample label of each target pixel to obtain the trained object type recognition model.

[0201] Each module in the above-mentioned image detection device can be implemented in whole or in part by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0202] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.13As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. 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. 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 data related to contrast images. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an image detection method is implemented.

[0203] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure 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 certain components, or have a different arrangement of components.

[0204] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps described in the above embodiment when executing the computer program.

[0205] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps described in the above embodiment are implemented.

[0206] In one embodiment, a computer program product is provided, including a computer program, which implements the steps described in the above embodiment when executed by a processor.

[0207] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0208] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

[0209] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed 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 the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an 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 each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0210] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0211] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An image detection method, characterized in that: The method comprises: Acquire the image to be detected; Performing blood vessel segmentation on the image to be detected by using the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected; The trained blood vessel type recognition model is used to perform type recognition on the blood vessel segmentation result to obtain the blood vessel type of each pixel point of the image to be detected; and based on the blood vessel prediction label and blood vessel type of each pixel point, the detection information of the image to be detected is obtained.

2. The method according to claim 1, characterized in that The detection information also includes a brain region prediction label of the pixel point, and the method further includes: The trained brain region segmentation model is used to perform brain region segmentation on the image to be detected to obtain a brain region segmentation result of the image to be detected, wherein the brain region segmentation result includes a brain region prediction label of each pixel point of the image to be detected.

3. The method according to claim 2, characterized in that The detection information further includes a maximum density projection map, and the brain region includes a first brain region and a second brain region; the method further includes: Based on the brain region segmentation result, determine the brain region of the image to be detected, and divide the brain region based on the blood vessel types corresponding to each pixel point in the brain region, determine the pixel points that meet the first type condition as the first brain region, and determine the pixel points that meet the second type condition as the second brain region; Maximum density projection processing is performed on the brain area of ​​the image to be detected, the first brain area and the second brain area to obtain a maximum density projection image of the image to be detected, a first maximum density projection subimage corresponding to the first brain area and a second maximum density projection subimage corresponding to the second brain area.

4. The method according to claim 1, characterized in that: The method further comprises: An original contrast image is acquired, and preprocessed to obtain an image to be detected, wherein the preprocessing at least includes resampling processing.

5. The method according to claim 4, characterized in that The obtaining the detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point includes: Based on the object prediction label and blood vessel type of each pixel point, obtaining initial detection information of the image to be detected; The inverse operation of the resampling is performed on the initial detection information to obtain original detection information corresponding to the original angiography image.

6. The method according to claim 1, characterized in that The image to be detected includes a plurality of sub-images; the image to be detected is segmented by the trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, including: Performing image segmentation on each of the sub-images of the image to be detected by using the trained object segmentation model to obtain blood vessel segmentation sub-results corresponding to each of the sub-images; Based on the blood vessel segmentation sub-results corresponding to the sub-images, a blood vessel segmentation result corresponding to the image to be detected is obtained.

7. The method according to claim 1, characterized in that The method further comprises: Acquire target training data, wherein the target training data includes a contrast image sequence, wherein the contrast image sequence includes a plurality of contrast images, and label data of each of the contrast images, wherein the label data includes an object sample label of each pixel point included in each of the contrast images; Based on the contrast image sequence in the target training data and the object sample label of each pixel in each contrast image, the object segmentation model to be trained is trained to obtain the object prediction label of each pixel, and based on the object prediction label of each pixel and the sample object label, training is performed to obtain the trained object segmentation model.

8. The method according to claim 7, characterized in that The object sample label of each pixel point includes a blood vessel sample label and / or a brain region sample label; the object segmentation model includes a blood vessel segmentation model and / or a brain region segmentation model; the object prediction label includes a blood vessel prediction label and / or a brain region prediction label.

9. The method according to claim 7, characterized in that: The label data also includes an object type sample label for each pixel point; the method further includes: Acquire target pixel points whose object sample labels meet the segmentation conditions, and collect input data of each of the target pixel points, wherein the input data at least includes a blood vessel type sample label; Based on the vascular type sample label of each target pixel in the angiography image sequence, the object type recognition model to be trained is trained to obtain the vascular type prediction label of each target pixel, and based on the vascular type prediction label of each target pixel and the vascular type sample label, training is performed to obtain the trained vascular type recognition model.

10. An image detection device, characterized in that: The device comprises: An acquisition module, used for acquiring an image to be detected; a segmentation module, configured to perform image segmentation on the image to be detected by using a trained blood vessel segmentation model to obtain a blood vessel segmentation result corresponding to the image to be detected, wherein the blood vessel segmentation result includes a blood vessel prediction label of each pixel point of the image to be detected; The recognition module is used to perform type recognition on the blood vessel segmentation result through the trained blood vessel type recognition model to obtain the blood vessel type of each pixel point of the image to be detected; and obtain the detection information of the image to be detected based on the blood vessel prediction label and blood vessel type of each pixel point.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

12. 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 method according to any one of claims 1 to 9 are implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images