Spinal artery recognition method and device, storage medium and computer program product

Abstract

The invention discloses a spinal artery recognition method and device, a storage medium and a computer program product, and relates to the technical field of image processing. The electronic equipment can obtain the interested area containing the blood vessel from the scanned image and perform segmentation processing on the interested area to obtain the initial segmentation result of the spinal artery, and then perform progressive refining processing on the initial segmentation result to obtain the target segmentation result of the spinal artery, and the target segmentation result comprises the target position of the spinal artery. The interested area containing the blood vessel can be firstly obtained, and the target segmentation result of the spinal artery is obtained based on the interested area, so that the calculation amount of electronic equipment can be reduced, and the identification efficiency of the spinal artery is further improved. Moreover, the initial segmentation result of the spinal artery can be obtained firstly, and then the initial segmentation result is refined progressively, so that the segmentation boundary can be optimized step by step, and the target segmentation result is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a spinal artery recognition method, device, storage medium and computer program product. Background Art

[0002] When doctors perform surgery to treat patients with massive hemoptysis caused by lung disease, they need to identify the patient's spinal arteries in order to perform protective occlusion of the spinal arteries to avoid accidental embolism of the spinal arteries.

[0003] Currently, doctors can identify the spinal artery from the digital subtraction angiography (DSA) image of the patient's chest by observing the DSA image with naked eyes based on their own experience. However, the efficiency of doctors identifying the spinal artery by observing the DSA image with naked eyes is low. Summary of the invention

[0004] The present invention provides a spinal artery identification method, device, storage medium and computer program product, which can solve the problem of low efficiency in the related art that doctors identify spinal arteries by observing DSA images of the chest with naked eyes. The technical solution is as follows:

[0005] In one aspect, a method for identifying a spinal artery is provided, the method comprising:

[0006] Acquire a region of interest containing blood vessels from the scanned image;

[0007] Segment the region of interest to obtain the initial segmentation result of the spinal artery;

[0008] The initial segmentation result is gradually refined to obtain the target segmentation result of the spinal artery.

[0009] Optionally, the initial segmentation result is gradually refined to obtain a target segmentation result of the spinal artery, including:

[0010] The initial segmentation result is refined step by step by sequentially connecting multiple spinal artery thinning networks to obtain the target segmentation result of the spinal artery;

[0011] The output of each spinal artery thinning network among the multiple spinal artery thinning networks serves as the input of the next spinal artery thinning network.

[0012] Optionally, the input of each spinal artery thinning network also includes: a feature map of the region of interest;

[0013] The resolution of the feature map input to each spinal artery thinning network is smaller than the resolution of the feature map input to the subsequent spinal artery thinning network.

[0014] Optionally, the region of interest is segmented to obtain an initial segmentation result of the spinal artery, including:

[0015] Based on the attention heat map of the scanned image, feature extraction is performed on the region of interest to obtain a feature map of the region of interest;

[0016] The feature map is processed by the spinal artery localization network to obtain the initial segmentation result of the spinal artery.

[0017] Optionally, a region of interest containing blood vessels is obtained from the scanned image, including:

[0018] Get the attention heatmap of the scanned image;

[0019] Based on the attention heatmap, the region of interest containing blood vessels is obtained from the scanned image.

[0020] Optionally, the scanned image is a digital subtraction angiography (DSA) image; the method further comprises:

[0021] Acquire a DSA image sequence of the scanned object;

[0022] When the spinal artery exists in multiple consecutive DSA image frames, it is determined that the DSA image sequence includes the spinal artery.

[0023] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the spinal artery identification method described in the above aspect is implemented.

[0024] In yet another aspect, a computer program product is provided. The computer program product includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, the spinal artery identification method described in the above aspect is implemented.

[0025] On the other hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the spinal artery identification method described in the above aspects is implemented.

[0026] In another aspect, a spinal artery identification device is provided, the device comprising:

[0027] An acquisition module, used for acquiring a region of interest including a blood vessel from a scanned image;

[0028] The first segmentation module is used to segment the region of interest to obtain an initial segmentation result of the spinal artery;

[0029] The second segmentation module is used to perform progressive refinement processing on the initial segmentation result to obtain a target segmentation result of the spinal artery.

[0030] Optionally, a second segmentation module is used to perform step-by-step refinement processing on the initial segmentation result by sequentially connecting multiple spinal artery refinement networks to obtain a target segmentation result of the spinal artery;

[0031] The output of each spinal artery thinning network among the multiple spinal artery thinning networks serves as the input of the next spinal artery thinning network.

[0032] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0033] The present invention provides a spinal artery identification method, device, storage medium and computer program product. The electronic device can obtain a region of interest containing blood vessels from a scanned image, and segment the region of interest to obtain an initial segmentation result of the spinal artery, and then gradually refine the initial segmentation result to obtain a target segmentation result of the spinal artery, and the target segmentation result includes the target position of the spinal artery. Since the doctor does not need to observe the chest DSA image with the naked eye to identify the spinal artery, the recognition efficiency and accuracy of the spinal artery are improved. In addition, since the region of interest containing blood vessels can be obtained first, and the target segmentation result of the spinal artery can be obtained based on the region of interest, the calculation amount of the electronic device can be reduced, and the recognition efficiency of the spinal artery can be further improved. Since the initial segmentation result of the spinal artery can be obtained first and then the initial segmentation result can be gradually refined, the segmentation boundary can be gradually optimized, so that the target segmentation result is more accurate. In addition, since the doctor does not need to identify the spinal artery, the doctor's operation can be simplified.

[0034] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of a spinal artery identification method provided by an embodiment of the present invention;

[0036] Figure 2 is a flow chart of another spinal artery identification method provided by an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of the structure of a spinal artery identification network provided by an embodiment of the present invention;

[0038] Figure 4 is a schematic diagram of the structure of an attention network provided by an embodiment of the present invention;

[0039] Figure 5 is a schematic diagram of the structure of a spinal artery thinning network provided by an embodiment of the present invention;

[0040] Figure 6 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention;

[0041] Figure 7 It is a block diagram of a spinal artery identification device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0043] Figure 1 The present invention provides a flowchart of a spinal artery identification method, which is applied to an electronic device. Optionally, the electronic device may be a terminal, a medical imaging device or a server. The server may be a single server, or a server cluster composed of several servers, or a cloud computing service center. Figure 1 , the method comprising:

[0044] Step 101: Acquire a region of interest including a blood vessel from a scanned image.

[0045] The scanned image may be obtained by scanning a scanned part of the scanned object by the electronic device. The scanned part may include: the chest or abdomen of the scanned object. The scanned object may be a human body, a phantom or other animal body.

[0046] In an embodiment of the present invention, the electronic device may first obtain an attention heat map of the scanned image, and then, based on the attention heat map, the electronic device may crop the region of interest containing the blood vessel from the scanned image.

[0047] Alternatively, the medical imaging device may pre-store a region of interest segmentation model. The medical imaging device may input the scanned image into the region of interest segmentation model to obtain the region of interest output by the region of interest segmentation model.

[0048] Specifically, the electronic device may pre-acquire a region of interest segmentation dataset, which may include multiple sample scanned images and label scanned images corresponding to the multiple sample scanned images. Each label scanned image may be obtained by manually marking a region of interest in a sample scanned image. Afterwards, the electronic device may perform model training based on the region of interest segmentation dataset to obtain a region of interest segmentation model.

[0049] Step 102: Segment the region of interest to obtain an initial segmentation result of the spinal artery.

[0050] The initial segmentation result may include an initial position of the spinal artery.

[0051] Step 103: progressively refine the initial segmentation result to obtain a target segmentation result of the spinal artery.

[0052] Optionally, the electronic device may perform step-by-step refinement processing on the initial segmentation result through a spinal artery refinement network to obtain a target segmentation result of the spinal artery. The target segmentation result may include a target position of the spinal artery. The target position is more accurate than the initial position.

[0053] In summary, an embodiment of the present invention provides a method for identifying a spinal artery, wherein an electronic device can obtain a region of interest containing blood vessels from a scanned image, and segment the region of interest to obtain an initial segmentation result of the spinal artery, and then gradually refine the initial segmentation result to obtain a target segmentation result of the spinal artery, wherein the target segmentation result includes a target position of the spinal artery. Since the doctor does not need to observe the chest DSA image with the naked eye to identify the spinal artery, the recognition efficiency and accuracy of the spinal artery are improved. Furthermore, since the region of interest containing blood vessels can be obtained first, and the target segmentation result of the spinal artery can be obtained based on the region of interest, the amount of calculation of the electronic device can be reduced, and the recognition efficiency of the spinal artery can be further improved. Since the initial segmentation result of the spinal artery can be obtained first, and then the initial segmentation result can be gradually refined, the segmentation boundary can be gradually optimized, so that the target segmentation result is more accurate. In addition, since the doctor does not need to identify the spinal artery, the doctor's operation can be simplified.

[0054] Taking the electronic device as a medical imaging device as an example, the spinal artery identification method provided by the embodiment of the present invention is exemplarily described. The method can be applied to the medical imaging device, and the medical imaging device can be a DSA device and a computed tomography (CT) device. Figure 2 , the method may include:

[0055] Step 201: Acquire a scanned image of a scanned part of a scanned object.

[0056] The scanning part may include: a chest of a scanning object. The scanning object may be a human body, a phantom or other animal body.

[0057] In an embodiment of the present invention, the original image may be a DSA image. At this time, the medical imaging device may first obtain a mask image of the scanned part and an angiography image. Afterwards, the medical imaging device may subtract the mask image from the angiography image to obtain a DSA image of the scanned part. The mask image is obtained by the medical imaging device scanning the scanned part before the contrast agent is injected into the scanned object. The angiography image is obtained by the medical imaging device scanning the scanned part after the contrast agent is injected into the scanned object. The DSA image of the scanned part may be a two-dimensional image.

[0058] Alternatively, the medical imaging device can obtain a CT image of the scanned part and segment the CT image (such as threshold segmentation) to obtain a scanned image of the scanned part. The CT image can be an image containing a contrast agent, or an image without a contrast agent, or a subtraction image obtained based on an image containing a contrast agent and an image without a contrast agent.

[0059] Step 202: Obtain an attention heat map of the scanned image.

[0060] The medical imaging device can obtain the vascular tree segmentation result from the scanned image, and generate an attention heat map of the scanned image based on the vascular tree segmentation result. The attention heat map can reflect the importance of different areas in the scanned image. Since the medical imaging device in the embodiment of the present invention needs to identify the spinal artery, the importance of the vascular area in the attention heat map is higher than that of the non-vascular area. In this way, the medical imaging device can focus on the vascular area in the scanned image when processing the scanned image.

[0061] Optionally, the attention heat map may be a probability distribution map representing the importance of different regions in the scanned image.

[0062] In an embodiment of the present invention, a vascular tree segmentation model may be pre-stored in the medical imaging device. After obtaining a scanned image, the medical imaging device may input the scanned image into the vascular tree segmentation model to obtain a vascular tree segmentation result output by the vascular tree segmentation model.

[0063] Before inputting the scanned image into the vascular tree segmentation model, the medical imaging device may first obtain a vascular tree segmentation dataset and perform model training based on the vascular tree segmentation dataset to obtain a vascular tree segmentation model. Optionally, the vascular tree segmentation model may be a U-Net neural network.

[0064] The vascular tree segmentation data set may include: a plurality of sample vascular images, and a plurality of first labeled vascular images corresponding to the plurality of sample vascular images. Each of the plurality of first labeled vascular images is marked with a vascular tree. Each first labeled vascular image may be obtained by manually marking the vascular tree in a sample vascular image.

[0065] In an embodiment of the present invention, the medical imaging device may divide the vascular tree segmentation data set into a first training set and a first test set. Then, based on the first training set, the medical imaging device may adjust the parameters of the first initial model to minimize the first loss function to train the model, and determine the first initial model after training as the vascular tree segmentation model. Afterwards, the medical imaging device may use the first test set to evaluate the performance of the vascular tree segmentation model, and tune the vascular tree segmentation model based on the evaluation result to further improve the performance of the vascular tree segmentation model. The first initial model may be pre-stored in the medical imaging device.

[0066] Optionally, the ratio of the first training set to the first test set may be 4:1. The first loss function may include: a cross entropy loss function and / or a dice loss function. For example, the first loss function may be a weighted sum of a cross entropy loss function and a dice loss function, that is, the first loss value of the vascular tree segmentation model may be a weighted sum of a loss value of a cross entropy loss function and a loss value of a dice loss function. The weight of the cross entropy loss function and the weight of the dice loss function may be pre-stored. For example, the ratio of the weight of the cross entropy loss function to the weight of the dice loss function may be 0.5:1.

[0067] Step 203: Based on the attention heat map, obtain a region of interest containing blood vessels from the scanned image.

[0068] Taking the attention heat map as a probability distribution map as an example, an exemplary description is given of how a medical imaging device obtains a region of interest containing blood vessels from a scanned image based on the attention heat map:

[0069] The medical imaging device can determine the position of the auxiliary pixel with the largest probability value in the attention heat map in the attention heat map, and determine the target pixel from the scanned image based on the position. The position of the target pixel in the scanned image is the same as the position of the auxiliary pixel in the attention heat map. Then, the medical imaging device can crop the scanned image according to a preset size with the target pixel as the center to obtain a region of interest containing blood vessels from the scanned image. The region of interest is the area where the spinal artery may be located.

[0070] The preset size may be pre-stored in the medical imaging device. For example, the preset size may be 512×512. The preset size is the size of the region of interest.

[0071] Since the size of the scanned image is large (such as generally 1024×1024), and there are large non-vascular areas, while the spinal artery is very small and the signal is relatively weak, based on the attention heat map, the region of interest containing the blood vessels is obtained from the scanned image, so that the areas in the scanned image that are not related to the blood vessels can be directly excluded, so that the medical imaging equipment can directly process the area where the spinal artery may be located (i.e., the region of interest). In this way, the recognition efficiency of the spinal artery can be improved, and the influence of other tissues on the recognition process can be avoided, thereby improving the recognition accuracy of the spinal artery.

[0072] In an embodiment of the present invention, a spinal artery recognition network may be pre-stored in the medical imaging device, and the medical imaging device may process the attention heat map and the region of interest through the spinal artery recognition network to obtain a target segmentation result of the spinal artery. The specific process is described in the following steps 204 to 205.

[0073] Step 204: Based on the attention heat map of the scanned image, feature extraction is performed on the region of interest to obtain a feature map of the region of interest.

[0074] refer to Figure 3 The spinal artery recognition network may include a feature extraction network 30. The medical imaging device may input both the attention heat map and the region of interest into the feature extraction network 30, so that the feature extraction network 30 can extract features of the region of interest based on the attention heat map to obtain a feature map.

[0075] Since feature extraction can be performed based on the attention heat map, it can ensure that the feature extraction network can better perceive the contextual information of the vascular tree topology, so that the extracted feature map can be more comprehensive and representative. The contextual information can include: anatomical features, spatial relationships, and interactions with other tissues.

[0076] In an embodiment of the present invention, the number of the feature maps may be one or more, for example, multiple. The number of channels of the multiple feature maps increases successively, so that the multiple feature maps have different resolutions and semantic information. That is, the feature extraction network can extract multi-level features of different spatial scales from the region of interest. Specifically, feature maps with lower levels have higher spatial resolution and can capture detailed information in the region of interest, such as edges, etc.; while feature maps with higher levels have lower resolution, but have richer semantic information and can capture the overall structure and objects in the region of interest.

[0077] The total number of the plurality of feature maps may depend on the total number of spinal artery localization networks and spinal artery thinning networks in the spinal artery recognition network. For example, assuming that the number of spinal artery localization networks is 1 and the number of spinal artery thinning networks is 4, the number of the plurality of feature maps may be 5.

[0078] Optionally, the spinal artery recognition network may be a progressive refinement learning (PRL) network. The feature extraction network may be a ResNet-50 convolutional neural network.

[0079] Step 205: Process the feature map through the spinal artery localization network to obtain an initial segmentation result of the spinal artery.

[0080] In the embodiment of the present invention, reference Figure 3 The spinal artery identification network 10 may further include a spinal artery positioning network 40 connected to the feature extraction network 30. The medical imaging device may input the feature map output by the feature extraction network 30 into the spinal artery positioning network 40 to obtain an initial segmentation result of the spinal artery output by the spinal artery positioning network 40. The initial segmentation result may include an initial position of the spinal artery.

[0081] Specifically, when there are multiple feature maps, the medical imaging device can input the highest-level feature map output by the feature extraction network 30, that is, the feature map containing the richest semantic information, into the spinal artery localization network 40, so that the spinal artery localization network 40 can preliminarily determine the position of the spinal artery from the global image to obtain the initial segmentation result of the spinal artery.

[0082] Optionally, the spinal artery localization network can introduce an attention network, which can assign different importance weights to different parts of the input data, so that the spinal artery localization network can pay more attention to key information related to the spinal artery.

[0083] Figure 4 is a schematic diagram of the structure of an attention network provided by an embodiment of the present invention, from Figure 4 It can be seen that the attention network may include: a channel attention network 4011 and a spatial attention network 4012 connected in sequence. The channel attention network 4011 and the spatial attention network 4012 act on the channel and spatial dimensions respectively to capture the long-range dependency relationship between the spinal artery and other blood vessels in the vascular tree, promote the enhancement of the high-level semantic information of the spinal artery, and thus obtain the initial segmentation result of the spinal artery at a high level.

[0084] Specifically, the channel attention network 4011 can first determine the channel weight vector of the feature map at the highest level, and perform weighted processing on the feature map based on the channel weight vector to obtain a first weighted feature map. Afterwards, the medical imaging device can perform weighted processing on the first weighted feature map based on the spatial weight vector to obtain a second weighted feature map. The second weighted feature map is used for the spinal artery positioning network to determine the initial segmentation result of the spinal artery.

[0085] Optionally, taking the size of the highest-level feature map as C×H×W as an example, combined with Figure 4 , the process of the attention network outputting the second weighted feature map is exemplarily explained.

[0086] See also Figure 4 , the channel attention network 4011 includes: multiple first pooling layers 01. The medical imaging device can perform pooling processing on the feature map of the highest level through the multiple first pooling layers 01, so as to first adjust the size of the feature image to obtain the first feature map, the second feature map and the third feature map after the size adjustment. Then, the medical imaging device can fuse the first feature map with the second feature map (such as by multiplication) to obtain a first fusion result. The size of the first fusion result can be C×C. Subsequently, the medical imaging device can fuse the first fusion result with the third feature map to obtain a channel weight vector. Afterwards, the medical imaging device can fuse the channel weight vector with the feature map of the highest level (such as by element-by-element addition) to obtain a first weighted feature map of size C×H×W.

[0087] The spatial attention network 4012 includes: multiple second pooling layers 02. The medical imaging device can perform pooling operations on the first weighted feature map through the multiple second pooling layers 02 to obtain the fourth feature map, the fifth feature map and the sixth feature map after resizing. Then, the medical imaging device can fuse the fourth feature map and the fifth feature map (such as by multiplication) to obtain a second fusion result. The size of the second fusion result can be (H×W)×(H×W). Subsequently, the medical imaging device can fuse the second fusion result with the sixth feature map to obtain a spatial weight vector, and fuse the spatial weight vector with the first weighted feature map to obtain a second weighted feature map of size C×H×W.

[0088] Step 205 , the initial segmentation result is refined step by step by sequentially connecting multiple spinal artery thinning networks to obtain a target segmentation result of the spinal artery.

[0089] In the embodiment of the present invention, please continue to refer to Figure 3The spinal artery recognition network 10 further includes: a plurality of spinal artery thinning networks 50 connected in sequence, and the output of each spinal artery thinning network 50 can be used as the input of the next spinal artery thinning network 50. Thus, by gradually thinning the initial segmentation result by the plurality of spinal artery thinning networks 50, a relatively accurate target segmentation result of the spinal artery is obtained.

[0090] like Figure 3 As shown, the first spinal artery thinning network 50 among the multiple spinal artery thinning networks is also connected to the spinal artery positioning network 40. The output of the spinal artery positioning network 40 can be used as the input of the first spinal artery thinning network 50. In addition, each spinal artery thinning network 50 is also connected to the feature extraction network 30, and the input of each spinal artery identification network 50 can also include: a feature map output by the feature extraction network 30. Specifically, assuming that the last spinal artery thinning network 50 among the multiple spinal artery thinning networks 50 connected in sequence outputs the target segmentation result of the spinal artery, the level of the feature map input to the multiple spinal artery thinning networks 50 from front to back gradually decreases.

[0091] It can be seen that multiple spinal artery thinning networks can exchange and fuse information between feature maps of different scales to achieve cross-scale information interaction, thereby enhancing the spinal artery thinning network's contextual semantic perception of fuzzy areas, enabling the spinal artery thinning network to accurately infer the location of the spinal artery and achieve step-by-step refinement of the initial segmentation results of the spinal artery.

[0092] In an embodiment of the present invention, each spinal artery refinement network can fuse the output of the previous level network with the input feature map, split the fusion result into four-channel features, and perform convolution processing with different void rates before splicing and fusion to promote full interaction of cross-scale context information of adjacent levels, thereby achieving refinement of the spinal artery segmentation result.

[0093] Figure 5 is a schematic diagram of a spinal artery thinning network provided by an embodiment of the present invention. Figure 5 The refinement network may include: an upsampling layer 03, a grouped convolution layer 04, multiple hole convolution layers 05 and multiple ordinary convolution layers 06.

[0094] For each refined network, the medical imaging device can upsample the output result of the previous network (such as the refined network or the positioning network) of the refined network through the upsampling layer 03, and splice the obtained sampling result with the feature map input to the refined network to obtain a first splicing result. Then, the medical imaging device can split the first splicing result through the grouped convolution layer 04 to obtain new feature maps of four independent channels.

[0095] For the first new feature map among the four feature maps, the medical imaging device fuses the first new feature map with the first splicing result (such as fusion by element-by-element addition) to obtain a third fusion result. Then, the medical imaging device can perform a dilated convolution process on the third fusion result through the dilated convolution layer 05, and fuse the obtained first processing result with the second new feature map, and then perform a dilated convolution process through the dilated convolution layer 05. By analogy, the first processing result of each new feature map after the dilated convolution layer 05 performs the dilated convolution process is obtained.

[0096] Subsequently, the medical imaging device may splice the obtained multiple first processing results to obtain a second splicing result, and perform convolution processing on the second splicing result through a common convolution layer 06 to obtain a second processing result. Afterwards, the medical imaging device may fuse the second processing result with the first splicing result, and perform convolution processing on the fused fourth fusion result through a common convolution layer 06 to obtain a refined segmentation result.

[0097] Optionally, the medical imaging device obtains the target segmentation result of the spinal artery and can also display the target segmentation result of the spinal artery.

[0098] Optionally, the medical imaging device obtains the target segmentation result of the spinal artery and can also display the target segmentation result of the spinal artery.

[0099] Optionally, the process of the medical imaging device acquiring the scanned image of the scanned part of the scanned object may include: acquiring a DSA image sequence of the scanned part, the DSA image sequence including multiple frames of DSA images. The medical imaging device can determine whether there is a spinal artery in each frame of the DSA image. In the case where the spinal artery exists in multiple consecutive frames of DSA images, the medical imaging device can determine that the spinal artery exists in the DSA image sequence and output a target segmentation result of the spinal artery. Each frame of the target image is obtained based on a frame of DSA image with a spinal artery. In this way, the probability of misjudgment of the medical imaging device can be effectively reduced, the accuracy of the medical imaging device in identifying the spinal artery can be improved, and the reliability of the medical imaging device can be improved.

[0100] It should be noted that, in an embodiment of the present invention, before the medical imaging device inputs the region of interest into the spinal artery recognition network, it can first obtain a spinal artery recognition dataset and perform model training based on the spinal artery recognition dataset to obtain a spinal artery recognition network.

[0101] The spinal artery identification dataset may include: a plurality of sample blood vessel images, and a plurality of second labeled blood vessel images corresponding to the plurality of sample blood vessel images. Each of the plurality of second labeled blood vessel images is marked with a spinal artery. Each second labeled blood vessel image may be obtained by manually marking a spinal artery in a sample blood vessel image. Furthermore, after the medical imaging device acquires the spinal artery identification dataset, the data of the spinal artery identification dataset may be expanded by rotation, scaling, etc., so as to enrich the data in the spinal artery identification dataset.

[0102] In an embodiment of the present invention, the medical imaging device may divide the spinal artery recognition data set into a second training set and a second test set. Then, the medical imaging device may adjust the parameters of the second initial model based on the second training set to minimize the second loss function to train the model, and determine the second initial model after training as the spinal artery recognition network. Afterwards, the medical imaging device may use the second test set to evaluate the performance of the spinal artery recognition network, and tune the spinal artery recognition network based on the evaluation results and the optimizer to further improve the performance of the spinal artery recognition network. The second initial model may be pre-stored in the medical imaging device, and the optimizer may be an adaptive moment estimation (Adam) optimizer with a weight decay of 0.95.

[0103] Optionally, the ratio of the second training set to the second test set may be 4:1. The second loss function may include: a cross entropy loss function and / or a dice loss function. For example, the second loss function may be a weighted sum of a cross entropy loss function and a dice loss function, that is, the second loss value of the spinal artery recognition network may be a weighted sum of a loss value of a cross entropy loss function and a loss value of a dice loss function. The weight of the cross entropy loss function and the weight of the dice loss function may be pre-stored. For example, the ratio of the weight of the cross entropy loss function to the weight of the dice loss function may be 0.5:1.

[0104] It is understandable that the order of the steps of the spinal artery identification method provided by the embodiment of the present invention can be appropriately adjusted, and the steps can be increased or decreased accordingly according to the situation. For example, step 202 can be deleted according to the situation. Any method that can be easily thought of by a person skilled in the art of the present invention within the technical scope disclosed by the present invention should be included in the protection scope of the present invention, so it will not be repeated.

[0105] In summary, an embodiment of the present invention provides a method for identifying a spinal artery, wherein an electronic device can obtain a region of interest containing blood vessels from a scanned image, and segment the region of interest to obtain an initial segmentation result of the spinal artery, and then gradually refine the initial segmentation result to obtain a target segmentation result of the spinal artery, wherein the target segmentation result includes a target position of the spinal artery. Since the doctor does not need to observe the chest DSA image with the naked eye to identify the spinal artery, the recognition efficiency and accuracy of the spinal artery are improved. Furthermore, since the region of interest containing blood vessels can be obtained first, and the target segmentation result of the spinal artery can be obtained based on the region of interest, the amount of calculation of the electronic device can be reduced, and the recognition efficiency of the spinal artery can be further improved. Since the initial segmentation result of the spinal artery can be obtained first, and then the initial segmentation result can be gradually refined, the segmentation boundary can be gradually optimized, so that the target segmentation result is more accurate. In addition, since the doctor does not need to identify the spinal artery, the doctor's operation can be simplified.

[0106] An embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the spinal artery identification method described above is implemented. Figure 1 or Figure 2 Method for identification of the spinal artery shown.

[0107] The embodiment of the present invention provides a computer program product, which includes a computer program or a computer instruction, and when the computer program or the computer instruction is executed by a processor, the spinal artery identification method described above is implemented. Figure 1 or Figure 2 Method for identification of the spinal artery shown.

[0108] Figure 6 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 6 As shown, the electronic device 60 may include a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program, the spinal artery identification method shown in the above embodiment is implemented. Figure 1 or Figure 2 Method for identification of the spinal artery shown.

[0109] Figure 7 is a block diagram of a spinal artery identification device provided by an embodiment of the present invention, such as Figure 7 As shown, the device 70 includes:

[0110] The acquisition module 701 is used to acquire a region of interest including a blood vessel from a scanned image.

[0111] The first segmentation module 702 is used to segment the region of interest to obtain an initial segmentation result of the spinal artery.

[0112] The second segmentation module 703 is used to perform progressive refinement processing on the initial segmentation result to obtain a target segmentation result of the spinal artery.

[0113] Optionally, the second segmentation module 703 may be used to: sequentially connect multiple spinal artery thinning networks to perform step-by-step thinning processing on the initial segmentation result to obtain a target segmentation result of the spinal artery;

[0114] The output of each spinal artery thinning network among the multiple spinal artery thinning networks serves as the input of the next spinal artery thinning network.

[0115] Optionally, the input of each spinal artery thinning network also includes: a feature map of the region of interest; and a resolution of the feature map input to each spinal artery thinning network is smaller than a resolution of the feature map input to a subsequent spinal artery thinning network.

[0116] Optionally, the first segmentation module 702 may be used to: extract features of the region of interest based on the attention heat map of the scanned image to obtain a feature map of the region of interest;

[0117] The feature map is processed by the spinal artery localization network to obtain the initial segmentation result of the spinal artery.

[0118] Optionally, the acquisition module 701 may be used to:

[0119] Get the attention heatmap of the scanned image;

[0120] Based on the attention heatmap, the region of interest containing blood vessels is obtained from the scanned image.

[0121] Optionally, the scanned image is a digital subtraction angiography (DSA) image; Figure 7 , the apparatus 70 may further include a determination module 704, and the determination module 704 may be used to: obtain a DSA image sequence of the scanned object;

[0122] When the spinal artery exists in multiple consecutive DSA image frames, it is determined that the DSA image sequence includes the spinal artery.

[0123] In summary, the present invention provides a spinal artery identification device. The electronic device can obtain a region of interest containing blood vessels from a scanned image, and segment the region of interest to obtain an initial segmentation result of the spinal artery, and then gradually refine the initial segmentation result to obtain a target segmentation result of the spinal artery, and the target segmentation result includes the target position of the spinal artery. Since the doctor does not need to observe the chest DSA image with the naked eye to identify the spinal artery, the recognition efficiency and accuracy of the spinal artery are improved. In addition, since the region of interest containing blood vessels can be obtained first, and the target segmentation result of the spinal artery can be obtained based on the region of interest, the calculation amount of the electronic device can be reduced, and the recognition efficiency of the spinal artery can be further improved. Since the initial segmentation result of the spinal artery can be obtained first, and then the initial segmentation result can be gradually refined, the segmentation boundary can be gradually optimized, so that the target segmentation result is more accurate. In addition, since there is no need for a doctor to identify the spinal artery, the doctor's operation can be simplified.

[0124] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0125] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0126] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0127] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0128] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0129] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for identifying a spinal artery, characterized in that: The method comprises: Acquire a region of interest containing blood vessels from the scanned image; Segmenting the region of interest to obtain an initial segmentation result of the spinal artery; The initial segmentation result is gradually refined to obtain a target segmentation result of the spinal artery.

2. The method according to claim 1, characterized in that: The step of performing progressive refinement processing on the initial segmentation result to obtain a target segmentation result of the spinal artery includes: The initial segmentation result is refined step by step by sequentially connecting a plurality of spinal artery thinning networks to obtain a target segmentation result of the spinal artery; The output of each of the multiple spinal artery thinning networks serves as the input of the next spinal artery thinning network.

3. The method according to claim 2, characterized in that The input of each of the spinal artery thinning networks also includes: a feature map of the region of interest; The resolution of the feature map input to each spinal artery thinning network is smaller than the resolution of the feature map input to the subsequent spinal artery thinning network.

4. The method according to any one of claims 1 to 3, characterized in that: The segmenting of the region of interest to obtain an initial segmentation result of the spinal artery includes: Based on the attention heat map of the scanned image, extract features of the region of interest to obtain a feature map of the region of interest; The feature map is processed by a spinal artery localization network to obtain an initial segmentation result of the spinal artery.

5. The method according to any one of claims 1 to 3, characterized in that: The step of acquiring a region of interest containing a blood vessel from a scanned image comprises: Get the attention heatmap of the scanned image; Based on the attention heat map, a region of interest including a blood vessel is acquired from the scanned image.

6. The method according to any one of claims 1 to 3, characterized in that: The scanned image is a digital subtraction angiography (DSA) image; the method further comprises: Acquire a DSA image sequence of the scanned object; When a spinal artery exists in a plurality of consecutive frames of the DSA images, it is determined that the DSA image sequence includes the spinal artery.

7. A spinal artery identification device, characterized in that: The device comprises: An acquisition module, used for acquiring a region of interest including a blood vessel from a scanned image; A first segmentation module is used to segment the region of interest to obtain an initial segmentation result of the spinal artery; The second segmentation module is used to perform progressive refinement processing on the initial segmentation result to obtain a target segmentation result of the spinal artery.

8. The device according to claim 7, characterized in that A second segmentation module is used to perform step-by-step refinement processing on the initial segmentation result by sequentially connecting multiple spinal artery refinement networks to obtain a target segmentation result of the spinal artery; The output of each of the multiple spinal artery thinning networks serves as the input of the next spinal artery thinning network.

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

10. A computer program product, characterized in that The computer program product comprises a computer program or a computer instruction, and when the computer program or the computer instruction is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images