Apparatus and method for tubular structure segmentation

Abstract

Systems, methods, and apparatuses associated with image segmentation, such as tubular structure segmentation, are described herein. An artificial neural network is trained based on annotated images of different types of tubular structures, which can have different contrast and / or appearance than a tubular structure of interest, to segment the tubular structure of interest in medical scan images. The training can be performed in multiple stages during which a segmentation model learned from the annotated images during a first stage can be modified to adapt to the tubular structure of interest in a second stage. In examples, the tubular structure of interest can include coronary arteries, catheters, guide wires, etc., and the annotated images used to train the artificial neural network can include blood vessels, such as retinal blood vessels.

Description

Technical Field

[0001] This application relates to the field of medical imaging, specifically to the processing of medical images, and to devices and methods for segmenting tubular structures. Background Technology

[0002] Segmenting tubular structures (such as arteries, veins, catheters, guidewires, etc.) in medical scan images (e.g., X-ray fluorescence fluoroscopy images) can be essential for many downstream image processing tasks, including visibility enhancement, multimodal image registration, path mapping, etc. In recent years, deep learning-based segmentation techniques have been increasingly adopted in medical imaging and have demonstrated superior performance compared to conventional image segmentation techniques. However, due to the scarcity of labeled training data for tubular structures, these deep learning-based techniques have not yet been applied to segmentation tasks involving common tubular structures (such as those described above). Therefore, it is highly desirable to develop deep learning-based systems and devices based on currently available tubular training data, which may be limited to specific types of tubular structures, and to transfer or adapt the knowledge learned from this training data to other types of tubular structures. Summary of the Invention

[0003] This document describes systems, methods, and apparatuses associated with the segmentation of tubular structures. An apparatus configured to perform a segmentation task may include one or more processors configured to: receive a medical image (e.g., an X-ray fluorescence image) depicting a first type of tubular structure (e.g., a catheter, guidewire, etc.) and segment the first type of tubular structure from the medical image using an artificial neural network (ANN). The ANN may be trained to segment the first type of tubular structure through a process that may include: training the ANN during a first phase of the process to segment a second type of tubular structure (e.g., retinal vessels) based on an annotated medical image of a second type of tubular structure; and further training the ANN during a second phase of the process to segment the first type of tubular structure based on a segmentation model learned from the first phase of the training process. The second phase of the training process may include: providing the ANN with a first training image including the first type of tubular structure; causing the ANN to generate a first segmentation of the first type of tubular structure based on the segmentation model learned from the first phase of the training process; correcting the first segmentation generated by the ANN based on one or more features of the first type of tubular structure to derive a corrected segmentation; and adjusting the segmentation model based on the difference between the first segmentation generated by the ANN and the corrected segmentation.

[0004] In the example, correcting a first segmentation based on one or more features of a first type of tubular structure may include: identifying one or more connected regions in the first segmentation corresponding to a first type of tubular structure; determining the corresponding size of each of the one or more connected regions; and indicating, in the corrected segmentation, whether each of the one or more connected regions includes a first type of tubular structure based on the size of the connected regions. For example, if the size of the connected region is higher than a threshold, the connected region may be indicated as including a first type of tubular structure, and if the size of the connected region is lower than a threshold, the connected region may be indicated as not including a first type of tubular structure. In the example, the size of the connected region may be determined based on the number of pixels included in the connected region, and identifying connected regions in the first segmentation that may correspond to a first type of tubular structure may include: identifying a consecutive number of pixels in the first segmentation that are indicated to belong to a first type of tubular structure, and considering the region occupied by the consecutive number of pixels as a connected region.

[0005] In the example, correcting the first segmentation based on one or more features of the first type of tubular structure may include: determining the motion of regions in the first segmentation that can be indicated as including the first type of tubular structure, and indicating whether a region includes the first type of tubular structure in the corrected segmentation based on the motion of the region. In the example, the motion of the region can be determined by calculating the changes (e.g., signal changes, pixel-level displacements, etc.) between a first training image including the first type of tubular structure and a second training image including the first type of tubular structure. The motion of the region can be determined by first registering the first training image with the second training image. Using these techniques, if a region has greater motion than the background region of the first segmentation, the region can continue to be labeled as including the first type of tubular structure. Conversely, if the region has motion substantially similar to the background region of the first segmentation, the region can be relabeled as not including the first type of tubular structure. Attached Figure Description

[0006] The examples disclosed herein can be understood in more detail from the following description, which is given by way of example in conjunction with the accompanying drawings.

[0007] Figure 1 This is a diagram illustrating an example of a tubular structure segmentation (TSS) system or device according to one or more embodiments described herein.

[0008] Figure 2A This is a diagram illustrating the training of an artificial neural network using labeled medical images of a specific type of tubular structure, and Figure 2B This is a simplified diagram illustrating how to further train an artificial neural network to handle different types of tubular structures without labeled medical images of different types of tubular structures.

[0009] Figure 3A This is a diagram illustrating an example technique for correcting a segment based on the corresponding size of one or more connected regions of a segment generated by an artificial neural network.

[0010] Figure 3B This is a diagram illustrating an example technique for correcting a segment based on motion information associated with the identified tubular structure in a segment generated by an artificial neural network.

[0011] Figure 4 This is a flowchart illustrating example operations that can be performed to train a neural network according to one or more embodiments described herein.

[0012] Figure 5 This is a block diagram illustrating example components of a device that can be configured to perform the tubular structure segmentation task described herein. Detailed Implementation

[0013] The present disclosure is illustrated by way of example rather than limitation in the figures.

[0014] Figure 1An example of a tubular structure segmentation (TSS) system or device (e.g., TSS 102) is illustrated, which can be configured to receive a medical image 104 comprising multiple tubular structures and obtain segmentation 106 of the tubular structures using a deep learning-based technique. The medical image 104 may include medical scan images (such as X-ray fluorescence images of the human body), and the tubular structures depicted in image 104 may include anatomical tubular structures of the human body (e.g., blood vessels) and / or artificial tubular structures placed in the human body (e.g., catheters, guidewires, etc.). TSS 102 may include an artificial neural network (ANN) pre-trained to segment tubular structures from the medical image 104. In the example, the ANN may include a convolutional neural network (CNN) (e.g., with a U-Net architecture) having multiple convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Following the convolutional layers may be batch normalization layers and / or linear or nonlinear activation functions (e.g., modified linear units or ReLU activation functions). Each convolutional layer may include multiple convolutional kernels or filters with corresponding weights, the values ​​of which can be learned through a training process to extract features from the medical image 104. The features extracted by the convolutional layers may be downsampled by one or more pooling layers to obtain a representation of the features, for example, in the form of a feature map or feature vector. The CNN may also include one or more up-pooling layers and one or more transposed convolutional layers. Through the up-pooling layers, the CNN can upsample the features extracted from the medical image 104, and process the upsampled features through one or more transposed convolutional layers (e.g., via multiple deconvolution operations) to derive an enlarged or dense feature map or feature vector. The dense feature map or vector can then be used to predict regions (e.g., pixels) in the medical image 104 that belong to tubular structures.

[0015] As will be described in more detail below, the execution parameters of the ANN associated with segmenting tubular structures from medical image 104 (e.g., the weights of various filters or kernels of the ANN) can be learned through a training process (e.g., an offline training process) that can be performed using labeled training images of different types of tubular structures. Furthermore, even in Figure 1 Only one input medical image is shown in the image, and those skilled in the art will understand that TSS can be configured to process multiple medical images, for example, sequentially or in parallel.

[0016] Segmentation 106 can be obtained (e.g., generated) in different formats, including, for example, one or more segmentation masks or one or more binary images. For example, the binary image generated by TSS 102 may include pixels corresponding to medical image 104, and pixels in the binary image belonging to the identified tubular structures may be assigned a first value (e.g., 1), while pixels belonging to non-tubular regions (e.g., background pixels) may be assigned a second value (e.g., 0).

[0017] Figure 2A and Figure 2B Examples are given of what can be used to train artificial neural networks (e.g., regarding...). Figure 1 The TSS102 describes an example technique for segmenting tubular structures from medical images using an ANN. Figure 2A This illustrates how an artificial neural network can be trained first using a first training dataset containing labeled medical images of a specific type (e.g., a second type) of tubular structures, and... Figure 2B This illustrates how an artificial neural network trained on a first training dataset can be further trained to process (e.g., segment) other types of tubular structures without the availability of labeled training images of other types (e.g., the first type) of tubular structures. As will be described in more detail below, these example training techniques can allow an artificial neural network to acquire (e.g., learn) the ability to segment multiple types of tubular structures, even with only labeled training data for a specific type of tubular structure.

[0018] refer to Figure 2A An ANN 202 can be initially trained (e.g., during the first phase of the training process) using a medical image 204 of a specific type (e.g., a second type) of tubular structures. The medical image 204 may include, for example, an optical image of the retina, and the type of tubular structure included in the medical image may include retinal vessels depicted in the optical image. The medical image 204 may be obtained from a public database and may include a corresponding labeled (e.g., marked) segment 206 that can be used as a gold standard for training. When referenced herein, a labeled medical image may refer to the medical image 204 and the labeled segment (e.g., segment 206) of the tubular structure associated with the medical image. In the example, the ANN 202 may receive (e.g., be provided) the medical image 204 during training iterations and predict segment 208 based on the received image and the network's current execution parameters (e.g., weights). Once predicted, segment 208 can be compared to segment 206 (e.g., the gold standard), and the loss associated with the prediction can be determined based on this comparison. The loss can be computed using an appropriate loss function, such as a loss function based on cross-entropy loss, mean squared error, L1 norm, L2 norm, etc. The computed loss can then be used to tune the parameters of the ANN 202, for example, by backpropagating the loss through the ANN 202 (e.g., gradient descent based on the loss).

[0019] The above operations can be repeated iteratively until specific criteria for terminating training are met. For example, the criteria for terminating training can be met if the loss falls below a predetermined threshold, or if the change in the loss value between two training iterations (e.g., between consecutive training iterations). Upon completion of training, the execution parameters of ANN 202 (e.g., constituting the segmentation model) can be considered suitable for segmenting tubular structures (e.g., retinal vessels) of the type included in medical image 204. Subsequently, the segmentation capabilities of ANN 202 can be further extended (e.g., the segmentation model learned during the first phase of training) to cover other types of tubular structures, such as blood vessels from other regions of the body, artificially placed catheters and / or guidewires, etc.

[0020] Figure 2B An example is illustrated using unlabeled (e.g., unannotated) training images of other types of tubular structures to train ANN 202 to handle (e.g., segment) other types of tubular structures. During training (e.g., it can be done...), Figure 2A During the second phase of the training process following the first phase, ANN 202 can receive (e.g., be provided) a medical image 224 including a type (e.g., first type) of tubular structure (e.g., coronary artery), and can predict a segmentation 226 based on the received image and the segmentation model (e.g., the network's execution parameters or weights) learned from the first phase. The segmentation 226 can then be corrected by the segmentation correction module 228 (e.g., a pseudo-label generator) based on one or more features (e.g., predetermined constraints) of the first type of tubular structure to derive a segmentation 230, which can be used as a gold standard for how segmentation 226 should be generated for the first type of tubular structure. For example, segmentation 230 can be compared with segmentation 226 to determine the difference between the two segments (e.g., loss), and this difference can be used to guide the tuning of the ANN's execution parameters (e.g., the segmentation model), allowing for further optimization of these parameters (e.g., based on the use of...). Figure 2A The parameters of the medical image 202 are learned to adapt the segmentation generated by ANN 202 to the corrected segmentation 230. As described herein, the difference or loss between segmentation 226 and segmentation 230 can be determined using a loss function based on MSE, L1 norm, L2 norm, etc. And once determined, the loss can be backpropagated through ANN 202 (e.g., based on gradient descent associated with the loss) to adjust the parameters of the neural network.

[0021] The segmentation correction module 228 (e.g., a pseudo-label generator) can be implemented using software and / or hardware components to achieve the above-described functions, and the correction of segmentation 226 can be performed based on the characteristics (e.g., constraints) of the first type of tubular structure, which can be used to distinguish regions that include the first type of tubular structure from regions that do not include the first type of tubular structure. Figure 3A and Figure 3B An example technique is illustrated for segmentation predicted by ANN 202 based on feature correction of a first type of tubular structure.

[0022] Figure 3A An example is illustrated by correcting the segmentation based on the corresponding dimensions of one or more connecting regions corresponding to the tubular structures identified in the segmentation (e.g., Figure 2B The example shown is segmentation 226). These dimensions can be used to distinguish tubular structures because real tubular structures (e.g., blood vessels, guide wires, etc.) may occupy a larger area (e.g., a larger length) than misidentified tubular structures (e.g., artifacts resembling the target tubular structure). Figure 3A As shown, the segmentation correction module (e.g., segmentation correction module 228) can be configured to detect multiple connected regions (e.g., 302a, 302b, 302c, etc.) in segmentation 302 corresponding to the identified tubular structure by identifying a corresponding number of connected pixels (e.g., connecting pixels) in each region. For example, the segmentation correction module can identify connected region 302a by identifying a first pixel in the region indicated to belong to the tubular structure, and further identifying a chain of consecutive pixels connected to the first pixel that are also indicated to belong to the tubular structure, until the chain breaks (e.g., through pixels indicated not to belong to the tubular structure). The segmentation correction module can then use the region occupied by the pixel chain (including the first pixel) as connected region 302a.

[0023] Using a similar technique, the segmentation correction module can identify other connected regions (e.g., 302b and 302c) and can further determine the size of each connected region (e.g., connected regions 302a, 302b, 302c, etc.), for example, by counting the number of pixels included in each region. The segmentation correction module can then determine whether each connected region should be properly labeled as a tubular structure region (e.g., including a target tubular structure) based on the size of the connected regions. For example, if the size of connected region 302a is above a threshold (e.g., a preset threshold), the segmentation correction module can determine that connected region 302a is a tubular structure region. If the corresponding sizes of connected regions 302b and 302c are below the threshold, the segmentation correction module can further determine that connected regions 302b and 302c are incorrectly identified as tubular structure regions. In response to making such a determination, the segmentation correction module can maintain the labeling of region 302a as a tubular structure region and change the labeling of regions 302b and 302c as non-tubular structure regions in the correction segmentation 304.

[0024] Figure 3B An example is illustrated by correcting segmentation based on motion information (e.g., inter-frame signal difference) associated with the identified tubular structure during segmentation. Figure 2B The example shown is segmentation 226). Motion information can be used to distinguish tubular structures because the tubular structures of interest (e.g., blood vessels, guide wires, etc.) exhibit greater motion from one image frame to the next compared to other parts of the image frame. Figure 3B As shown, the motion of one or more target tubular structures can be determined based on multiple medical images 322a and 322b including the tubular structures. Medical images 322a and 322b can be adjacent image frames, or can be separated by one or more other image frames, and as described herein, medical image 322a can be determined using an ANN 324 (e.g., Figure 2B The ANN 202) is processed to obtain a segmentation 325 (e.g., a binary image) of the tubular structure. Based on the obtained segmentation 325 and / or the input medical image (e.g., 322a and 322b), the segmentation correction module 326 (e.g., ...) is used to correct the segmentation of the tubular structure. Figure 2BThe segmentation correction module 228 can determine changes that may have occurred in one or more regions (e.g., tubular and / or non-tubular regions) of segmentation 325 (e.g., differences or displacements of features that may indicate signal changes, pixel-level displacements, etc.), and the segmentation correction module 326 can also determine motion associated with one or more regions based on these changes. The segmentation correction module 326 can evaluate these changes, for example, by including a motion calculation module 326a configured to calculate the magnitude of a motion field or flow field 326b based on one or more features of medical images 322a and 322b (e.g., image gradients and / or inter-frame pixel intensity differences). In this example, such a motion field or flow field may include vectors, vector grids, vector-valued functions, etc., that can indicate differences or displacements of features representing tubular and non-tubular objects depicted in the medical images from medical image 322a to medical image 322b.

[0025] Since differences or displacements (e.g., their absence) in features can indicate corresponding motion (or lack thereof) of objects depicted in medical images 322a and 322b, and tubular structures of interest in the images can (e.g., inherently) have greater motion than non-tubular structures in the images, segmentation correction module 328 can be able to distinguish true tubular structures in segment 325 from misidentified tubular structures based on a motion field or flow field 326b. For example, segmentation correction module 326 may include a motion averaging module 326c configured to generate a motion map 326d in which each pixel may contain a value depicting the motion (e.g., average motion calculated from multiple images) of the connected regions in segment 325 to which the pixel may belong. Segmentation correction module 326 may also include a motion thresholding module 326e configured to determine that regions of segment 325 with large motion (e.g., above a threshold) are correctly labeled as tubular structures, and therefore, the labeling of these regions should be maintained in the corrected segment 327. The motion thresholding module 326e can also determine that regions of segmentation 325 with small motion (e.g., below a threshold) are incorrectly labeled as tubular structures, and therefore, these regions should be relabeled as non-tubular regions in the corrected segmentation 327. The motion thresholding module 326e can use the motion of the background region as a threshold for correcting the labeling in segmentation 325. For example, the motion thresholding module 326e can determine that if a region exhibits greater motion than the background region, the region is correctly labeled as including the tubular structure of interest, and if a region exhibits the same or less motion as the background region, the region is incorrectly labeled as including the tubular structure of interest.

[0026] Tubular structure. The motion thresholding module 326e can also use a preset (e.g., pre-configured or pre-determined) motion threshold to correct the markers in segmentation 325.

[0027] Various techniques can be applied to improve the accuracy of the motion estimation described above. For example, optical flow-based techniques can be used to register 325 background pixels (e.g., pixels marked as 0 in a binary image indicating segmentation) to stabilize the background frame by frame (e.g., to maintain stationary objects in the background frame by frame).

[0028] (At the same image coordinates), making it easy to determine the motion of the tubular structure in the foreground. Furthermore, labeled medical images of the target tubular structure (if available) can be compared with the disclosures described herein.

[0029] Available labeled training images (e.g., retinal vessel images) can be used together to train the segmentation neural network described herein, which can accelerate the transfer learning process described herein and / or improve the segmentation capabilities of the neural network.

[0030] Figure 4 Examples are shown that can be used with training neural networks (e.g., Figure 1 ANN 102 and / or Figure 2A and Figure 2B The ANN202) is used to perform example operations associated with the segmentation task described in this paper. (See figure)

[0031] As shown, the training operation may include initializing the parameters of the neural network (e.g., weights associated with the individual filters or kernels of the neural network) at step 402. The parameters may be initialized, for example, based on samples collected from one or more probability distributions or parameter values ​​from another neural network with a similar architecture.

[0032] The training operation may also include providing training data to the neural network at 404 (e.g., publicly available labeled medical images including a type 0 tubular structure), and having the neural network predict at 406.

[0033] Segmentation. In 408, the predicted segmentation can be compared with a gold standard (such as labeled gold standard segmentation, e.g., Figure 2A The segmentation shown is 206) or a segmentation generated using the techniques described herein (e.g., Figure 2B The corrected segment 230) is compared to determine the loss associated with the prediction. The loss can be used...

[0034] A suitable loss function is determined, such as a loss function based on mean squared error (MSE), L1 norm, L2 norm, etc. Once the loss is determined, it can be evaluated at 410 to determine if it has been...

[0035] One or more training termination criteria are met. For example, if the loss is below a predetermined threshold, or if the change in loss between two training iterations (e.g., between consecutive training iterations) falls below a predetermined threshold, then the training termination criteria can be considered met. If it is determined at 410 that the training termination criteria have been met, then training can end. Otherwise, before training returns to 406, the loss can be backpropagated through the neural network at 412 (e.g., based on gradient descent associated with the loss).

[0036] For the sake of simplicity, the training steps are depicted and described in a specific order herein. However, it should be understood that training operations can occur in various orders, simultaneously, and / or with other operations not presented or described herein. Furthermore, it should be noted that not all operations that may be included in the training process are depicted and described herein, and not all exemplified operations need to be performed.

[0037] The systems, methods, and / or apparatuses described herein may be implemented using one or more processors, one or more storage devices, and / or other suitable auxiliary devices (such as display devices, communication devices, input / output devices, etc.). Figure 5 This is a block diagram illustrating an example device 500 that can be configured to perform the tubular structure segmentation task described herein. As shown, device 500 may include a processor (e.g., one or more processors) 502, which may be a central processing unit (CPU), graphics processing unit (GPU), microcontroller, reduced instruction set computer (RISC) processor, application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), physical processing unit (PPU), digital signal processor (DSP), field-programmable gate array (FPGA), or any other circuitry or processor capable of performing the functions described herein. Device 500 may also include communication circuitry 504, memory 506, mass storage device 508, input device 510, and / or communication link 512 (e.g., communication bus) through which one or more components shown in the figure exchange information.

[0038] Communication circuitry 504 can be configured to send and receive information using one or more communication protocols (e.g., TCP / IP) and one or more communication networks, including local area networks (LANs), wide area networks (WANs), the Internet, and wireless data networks (e.g., Wi-Fi, 3G, 4G / LTE, or 5G networks). Memory 606 may include a storage medium (e.g., a non-transitory storage medium) configured to store machine-readable instructions that, when executed, cause processor 502 to perform one or more functions described herein. Examples of machine-readable media may include volatile or non-volatile memory, including but not limited to semiconductor memory (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)), flash memory, etc.). Mass storage device 508 may include one or more disks, such as one or more internal hard disks, one or more removable disks, one or more magneto-optical disks, one or more CD-ROMs or DVD-ROMs, etc., on which instructions and / or data may be stored for operation of processor 502. Input device 510 may include a keyboard, mouse, voice-controlled input device, touch-sensitive input device (e.g., touch screen), etc., for receiving user input from device 500.

[0039] It should be noted that device 500 can operate as a standalone device or can be connected to other computing devices (e.g., networked or grouped) to perform the functions described herein. And even in Figure 5 Only one example of each component is shown in the figure, and those skilled in the art will understand that the device 500 may include multiple instances of one or more components shown in the figure.

[0040] Although this disclosure has been described according to certain embodiments and generally associated methods, changes and variations of the embodiments and methods will be apparent to those skilled in the art. Therefore, the above description of exemplary embodiments does not limit this disclosure. Other changes, substitutions, and modifications are possible without departing from the spirit and scope of this disclosure. Furthermore, unless specifically stated otherwise, discussions using terms such as “analyze,” “determine,” “enable,” “identify,” and “modify” refer to the actions and processes of a computer system or similar electronic computing device that manipulate and transform data representing physical (e.g., electronic) quantities within the registers and memories of the computer system into other data representing physical quantities within the computer system's memory or other such information storage, transmission, or display devices.

[0041] It should be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will become apparent to those skilled in the art upon reading and understanding the above description. Therefore, the scope of this disclosure should be determined by reference to the appended claims and the full scope of their equivalents.

Claims

1. An apparatus for dividing a tubular structure, comprising: One or more processors, which are configured as follows: Receive a medical image, wherein the medical image depicts a tubular structure of a first type; and The first type of tubular structure is segmented from the medical image using an artificial neural network (ANN), wherein the ANN is trained to segment the first type of tubular structure through a training process including: During the first phase of the training process, the ANN is trained to learn a segmentation model for segmenting tubular structures of the second type based on labeled medical images of the second type of tubular structures; and During the second phase of the training process, the ANN is further trained to segment the first type of tubular structure based on the segmentation model learned from the first phase of the training process, wherein the second phase of the training process includes: The ANN is provided with a first training image including the first type of tubular structure, wherein the first type of tubular structure is different from the second type of tubular structure; The ANN generates a first segmentation of the first type of tubular structure based on the first training image and the segmentation model learned from the first stage of the training process. The first segment generated by the ANN is corrected based on one or more features of the first type of tubular structure to derive the corrected segment; and The ANN adjusts the segmentation model based on the difference between the first segment generated by the ANN and the corrected segment.

2. The device according to claim 1, wherein, Correcting the first segmentation based on one or more features of the first type of tubular structure includes: Identify one or more connecting regions in the first segment that correspond to the first type of tubular structure; Determine the corresponding size of each of the one or more connecting regions; and in the corrected segmentation, indicate whether each of the one or more connecting regions includes a tubular structure of the first type based on the size of the connecting region, wherein the connecting region is indicated to include a tubular structure of the first type when the size of the connecting region is higher than a threshold, and wherein the connecting region is indicated to not include a tubular structure of the first type when the size of the connecting region is lower than the threshold.

3. The device according to claim 2, wherein, The corresponding size of each of the one or more connected regions is determined based on the corresponding number of pixels included in each of the one or more connected regions, wherein identifying the one or more connected regions in the first segment that correspond to the tubular structure of the first type includes: identifying a consecutive number of pixels in the first segment that are indicated to belong to the tubular structure of the first type, and considering the area occupied by the consecutive number of pixels as a connected region.

4. The device according to claim 1, wherein, Correcting the first segmentation based on one or more features of the first type of tubular structure includes: Determine the movement of the region indicated by the first segmentation as including the tubular structure of the first type; and In the corrected segmentation, the movement of the region is used to indicate whether the region includes the tubular structure of the first type.

5. The device according to claim 4, wherein, Determining the motion of the region includes: calculating the change between the first training image and a second training image including the first type of tubular structure, wherein determining the motion of the region further includes: registering the first training image with the second training image.

6. The device according to claim 4, wherein, Correcting the first segmentation based on one or more features of the first type of tubular structure includes: determining that the region of the first segmentation indicated to include the first type of tubular structure has greater motion than the background region of the first segmentation, and marking the region as including the first type of tubular structure in the corrected segmentation based on the determination.

7. The device according to claim 4, wherein, Correcting the first segmentation based on one or more features of the first type of tubular structure includes: determining that the region of the first segmentation indicated to include the first type of tubular structure has a motion substantially similar to the background region of the first segmentation, and marking the region in the corrected segmentation as not including the first type of tubular structure based on the determination.

8. The device according to claim 1, wherein, The first type of tubular structure includes coronary vessels, catheters placed in the human body, or guidewires placed in the human body, and wherein the second type of tubular structure includes vessels located in a region of the human body different from the coronary vessels.

9. A method for segmenting a tubular structure, the method comprising: Receive a medical image, wherein the medical image depicts a tubular structure of a first type; and The first type of tubular structure is segmented from the medical image using an artificial neural network (ANN), wherein the ANN is trained to segment the first type of tubular structure through a training process including: During the first phase of the training process, the ANN is trained to learn a segmentation model for segmenting tubular structures of the second type based on labeled medical images of the second type of tubular structures; and During the second phase of the training process, the ANN is further trained to segment the first type of tubular structure based on the segmentation model learned from the first phase of the training process, wherein the second phase of the training process includes: The ANN is provided with a first training image including the first type of tubular structure, wherein the first type of tubular structure is different from the second type of tubular structure; The ANN generates a first segmentation of the first type of tubular structure based on the first training image and the segmentation model learned from the first stage of the training process. The first segment generated by the ANN is corrected based on one or more features of the first type of tubular structure to derive the corrected segment; and The ANN adjusts the segmentation model based on the difference between the first segment generated by the ANN and the corrected segment.

10. A method for training an artificial neural network (ANN) to segment a first type of tubular structure, the method comprising: The ANN is trained during the first phase of the training process to learn a segmentation model for segmenting the second type of tubular structures based on labeled medical images of the second type of tubular structures. as well as During the second phase of the training process, the ANN is further trained to segment the first type of tubular structure based on the segmentation model learned from the first phase of the training process, wherein the second phase of the training process includes: The ANN is provided with a first training image including the first type of tubular structure, wherein the first type of tubular structure is different from the second type of tubular structure; The ANN generates a first segmentation of the first type of tubular structure based on the first training image and the segmentation model learned from the first stage of the training process. The first segment generated by the ANN is corrected based on one or more features of the first type of tubular structure to derive the corrected segment; and The ANN adjusts the segmentation model based on the difference between the first segment generated by the ANN and the corrected segment.

Images