Cerebral microbleed lesion localization and segmentation method, system and terminal

Abstract

The present invention provides a method, system, and terminal for localizing and segmenting cerebral microbleed lesions. A cerebral microbleed lesion localization and segmentation model based on multi-task and multi-view constraints is used to obtain corresponding cerebral microbleed lesion localization and segmentation results based on the brain SWI scan image. The present invention employs a novel multi-task, multi-view constraint learning framework to achieve fast and accurate CMBs localization and segmentation.

Description

Technical Field

[0001] The present invention relates to the field of cerebral microbleeding localization, and in particular to a method, system and terminal for localizing and segmenting cerebral microbleeding lesions. Background Art

[0002] Cerebral microbleeds (CMBs), a common form of cerebral small vessel disease (SVD), can cause a variety of serious conditions, including movement disorders, affective disorders, and dementia. The number of CMBs is a key clinical indicator of SVD severity. However, due to their small size and potential for confusion with cerebral vascular tissue, manual localization is time-consuming and labor-intensive, and prone to missed and misidentified findings. Furthermore, clinical findings suggest that qualitative analysis based solely on the number of CMBs cannot explain the varying cognitive performance of patients with SVD. Patients with the same cerebral bleeds may present with varying levels of cognitive impairment, including normal cognition (NCI), amnestic cognitive impairment (aMCI), and non-amnestic cognitive impairment (naMCI). Furthermore, studies have suggested that the location of CMBs may be a key factor in varying cognitive outcomes. Therefore, developing an automated and accurate CMB localization and segmentation algorithm could help quantify the relationship between SVD and the resulting SVC.

[0003] Existing localization and segmentation algorithms used in medical images have difficulty locating small lesions, resulting in a high incidence of false positives in localized CMBs. Existing CMB localization algorithms can be categorized by implementation method as follows: 1) algorithms that localize CMBs based on manually designed CMB features and classic machine learning classifiers such as support vector machines (SVMs) and random forests; and 2) algorithms that utilize deep learning networks. Qi Dou et al. used a fully convolutional neural network (FCN) to locate possible CMBs and then designed a multi-layer convolutional neural network (CNN) to identify true CMBs from the localized CMBs. Al-Masni et al. first used the localization network YOLO to locate CMB regions and then used a CNN to detect the detected CMB regions to obtain the final CMB localization results.

[0004] Algorithms based on manually designed features suffer from poor robustness and scalability due to the high data specificity. Deep learning-based methods, on the other hand, typically employ a two-stage learning strategy: 1) localizing CMBs in images; 2) determining whether the localized CMBs are genuine CMBs. This two-stage approach is time-consuming and labor-intensive, and has a high false-positive rate. Furthermore, existing methods fail to consider the texture and geometric features of CMBs themselves, making it easy to misidentify tissues that closely resemble CMBs, such as blood vessels and calcified areas, as CMBs. Furthermore, CMBs are relatively small compared to the brain, but their volume varies significantly between patients. Existing methods focus solely on larger CMB lesions, easily missing smaller ones. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a method, system and terminal for localizing and segmenting cerebral microbleeding lesions, which are used to solve the above technical problems in the prior art.

[0006] To achieve the above-mentioned and other related purposes, the present invention provides a method for localizing and segmenting cerebral microbleeding lesions, which comprises: obtaining a brain SWI scan image to be segmented; and obtaining a corresponding cerebral microbleeding lesion localization and segmentation result based on the brain SWI scan image based on a multi-task and multi-view constraint cerebral microbleeding lesion localization and segmentation model.

[0007] In one embodiment of the present invention, the brain microbleed lesion localization and segmentation model training method includes: based on the brain microbleed lesion volume constraint, training according to multiple lesion localization and segmentation training samples to obtain the brain microbleed lesion localization and segmentation model; wherein each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleed lesion localization and segmentation result.

[0008] In one embodiment of the present invention, the brain microbleed lesion localization and segmentation model includes: a center point detection module, which is used to obtain the corresponding center point localization result based on the input brain SWI scan image; a lesion area segmentation module, which is used to obtain the corresponding lesion area segmentation result based on the input brain SWI scan image; a fusion module, which connects the center point localization module and the lesion area segmentation module, and is used to fuse the center point localization result and the lesion area segmentation result to obtain the brain microbleed lesion localization and segmentation result.

[0009] In one embodiment of the present invention, obtaining the corresponding center point positioning result based on the input brain SWI scan image includes: obtaining a center point positioning probability map based on the brain SWI scan image based on the weighted Hausdorff distance; wherein, the center point positioning probability map includes: the position data of each point and its corresponding brain microbleed lesion center point probability value.

[0010] In one embodiment of the present invention, obtaining the corresponding lesion area segmentation result based on the input brain SWI scan image includes: obtaining the lesion area segmentation result based on the brain SWI scan image based on the weight set by the lesion area size; wherein, the lesion area segmentation result includes: one or more brain microbleeding lesion areas.

[0011] In one embodiment of the present invention, the smaller the size of the lesion area is, the greater the corresponding weight is.

[0012] To achieve the above-mentioned and other related purposes, the present invention provides a system for localizing and segmenting cerebral microbleeding lesions, which includes: an image acquisition module for acquiring a SWI scan image of the brain to be located; a positioning and segmentation module, connected to the image acquisition module, for obtaining a corresponding cerebral microbleeding lesion localization and segmentation result based on a multi-task and multi-view constraint cerebral microbleeding lesion localization and segmentation model according to the brain SWI scan image.

[0013] In one embodiment of the present invention, the brain microbleed lesion localization and segmentation model training method includes: based on the brain microbleed lesion volume constraint, training according to multiple lesion localization and segmentation training samples to obtain the brain microbleed lesion localization and segmentation model; wherein each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleed lesion localization and segmentation result.

[0014] In one embodiment of the present invention, the cerebral microbleeding lesion localization and segmentation model includes: a center point detection module for obtaining a corresponding center point localization result based on an input brain SWI scan image; a lesion area segmentation module for obtaining a corresponding lesion area segmentation result based on an input brain SWI scan image; a fusion module for connecting the center point localization module and the lesion area segmentation module for fusing the center point localization result and the lesion area segmentation result to obtain a cerebral microbleeding lesion localization and segmentation result.

[0015] To achieve the above-mentioned objectives and other related objectives, the present invention provides a terminal for localizing and segmenting cerebral microbleeding lesions, comprising: one or more memories and one or more processors; the one or more memories are used to store computer programs; the one or more processors are connected to the memories and are used to run the computer programs to execute the cerebral microbleeding lesion localization and segmentation method.

[0016] As described above, the present invention provides a method, system, and terminal for locating and segmenting cerebral microbleeding lesions, which have the following beneficial effects: the present invention obtains corresponding cerebral microbleeding lesion localization and segmentation results based on the brain SWI scan image through a cerebral microbleeding lesion localization and segmentation model based on multi-task and multi-view constraints; the present invention adopts a novel multi-task, multi-view constrained learning framework to achieve fast and accurate CMBs localization and segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Shown is a flow chart of a method for localizing and segmenting cerebral microbleeding lesions in one embodiment of the present invention.

[0018] Figure 2 Shown is a schematic structural diagram of a cerebral microbleed lesion localization and segmentation model in one embodiment of the present invention.

[0019] Figure 3 Shown is a schematic structural diagram of a cerebral microbleed lesion localization and segmentation model in one embodiment of the present invention.

[0020] Figure 4 Shown is a schematic structural diagram of a system for localizing and segmenting cerebral microbleeding lesions in one embodiment of the present invention.

[0021] Figure 5 Shown is a schematic structural diagram of a cerebral microbleed lesion localization and segmentation terminal in one embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0023] It should be noted that in the following description, reference is made to the accompanying drawings, which describe several embodiments of the present invention. It should be understood that other embodiments may be used and that mechanical, structural, electrical and operational changes may be made without departing from the spirit and scope of the present invention. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present invention is limited only by the claims of the published patents. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. Spatially related terms, such as "upper", "lower", "left", "right", "below", "below", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between one element or feature shown in the figure and another element or feature.

[0024] Throughout this specification, when a part is said to be "connected" to another part, this includes not only "direct connection" but also "indirect connection" with other elements interposed therebetween. Furthermore, when a part is said to "include" a certain component, unless otherwise stated, this does not exclude the other component but rather implies that the other component may be included.

[0025] The terms "first," "second," and "third" are used to describe various parts, components, regions, layers, and / or segments, but are not intended to be limiting. These terms are used solely to distinguish one part, component, region, layer, or segment from another. Therefore, a reference to a first part, component, region, layer, or segment below may also refer to a second part, component, region, layer, or segment without departing from the scope of the present invention.

[0026] Furthermore, as used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Thus, "A, B, or C" or "A, B, and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B, and C". Exceptions to this definition occur only when the combination of elements, functions, or operations is inherently mutually exclusive in some way.

[0027] The present invention provides a method, system, and terminal for locating and segmenting cerebral microbleed lesions. By using a cerebral microbleed lesion localization and segmentation model based on multi-task and multi-view constraints, corresponding cerebral microbleed lesion localization and segmentation results are obtained according to the brain SWI scan image. The present invention adopts a novel multi-task, multi-view constraint learning framework to achieve fast and accurate CMBs localization and segmentation.

[0028] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0029] like Figure 1 A flowchart illustrating a method for localizing and segmenting cerebral microbleeding lesions according to an embodiment of the present invention is shown.

[0030] The method comprises:

[0031] Step S11: Acquire a brain SWI scan image to be segmented.

[0032] Optionally, the acquired susceptibility-weighted imaging (SWI) images are subjected to preprocessing operations such as normalization, image enhancement, and denoising to remove interference factors as much as possible.

[0033] Step S12: Based on the multi-task and multi-view constraint cerebral microbleeding lesion localization and segmentation model, the corresponding cerebral microbleeding lesion localization and segmentation results are obtained according to the brain SWI scan image.

[0034] Alternatively, because existing methods do not consider the texture and geometric features of CMBs themselves, it is easy to misidentify tissues that are very similar to CMBs, such as blood vessels and calcification areas, as CMBs. Considering the morphological characteristics of microbleeds themselves, multi-perspective constraints such as the volume, shape, and position of CMBs themselves are introduced into the brain microbleed lesion localization and segmentation model.

[0035] Optionally, the training method of the cerebral microbleed lesion localization and segmentation model includes:

[0036] Based on the volume constraint of cerebral microbleed lesions, a cerebral microbleed lesion localization and segmentation model was obtained by training multiple lesion localization and segmentation training samples. Specifically, for cerebral microbleed lesions of different volumes, the segmentation and localization effects are different. Therefore, we use the volume of CMBs as a constraint to train the cerebral microbleed lesion localization and segmentation model, and introduce the volume of CMBs themselves into the model training to achieve accurate CMBs localization.

[0037] Each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleed lesion localization and segmentation result.

[0038] Optional, such as Figure 2 As shown, the structure of the cerebral microbleed lesion localization and segmentation model includes:

[0039] The center point detection module 21 is used to perform the CMBs positioning task to obtain the coordinates of the CMBs center point; this module obtains the corresponding center point positioning result based on the input brain SWI scan image;

[0040] The lesion region segmentation module 22 is used to perform the lesion region segmentation task to obtain the lesion region of CMBs; the module inputs the brain SWI scan image to obtain the corresponding lesion region segmentation result;

[0041] A fusion module is connected to the center point positioning module and the lesion area segmentation module, and is used to fuse the center point positioning result and the lesion area segmentation result to obtain the cerebral microbleeding lesion positioning and segmentation result.

[0042] This solution employs an end-to-end strategy, simultaneously localizing and segmenting CMBs through a center point detection module 21 and a lesion segmentation module 22. These two parallel tasks utilize geometric information from multiple viewpoints to constrain the CMB center point coordinates and lesion shape. Finally, a fusion module 23 combines the results of these two branches into the final CMB localization and segmentation results, providing the localization and segmentation results for cerebral microbleed lesions.

[0043] We transform the task of localizing CMBs into detecting their center points, rather than using CMB bounding boxes like Al-Masni et al., significantly reducing computational effort. Furthermore, to better describe the geometric information of CMBs, we supervise the CMB lesion regions while locating their center points, enabling our network to learn more semantic information about CMBs.

[0044] Optional, such as Figure 3 As shown, the center point detection module 21 adopts a center positioning network, which is a V-Net-like structure composed of a series of downsampling and upsampling layers. Skip connections are used between downsampling and upsampling layers with the same feature size to fuse low-order and high-order information. Each downsampling and upsampling layer consists of two residual modules, which enables the network to learn more image features and increase the stability of the network.

[0045] Optionally, obtaining the corresponding center point location result based on the input brain SWI scan image includes: obtaining a center point location probability map based on the brain SWI scan image based on a weighted Hausdorff distance; wherein the center point location probability map includes: location data of each point and its corresponding probability value of the cerebral microbleed lesion center. That is, the center point location network ultimately outputs a probability map of the same size as the input image, with the value of each position in the probability map corresponding to the probability of that position being the center point of a CMB.

[0046] Unlike conventional center point detection, our model directly outputs the probability of each point belonging to the center point. It also uses a weighted Hausdorff distance modified for the task to suppress the probability of locations far from the center point and enhance the probability around the center point. The network outputs a probability map to obtain the final center point detection result, ensuring that the detection network has the highest probability of detecting the center of the CMB.

[0047] Optional, such as Figure 3 As shown, the lesion area segmentation module 22 adopts a structure similar to that of the center point positioning network, both of which use a V-Net structure composed of upsampling and downsampling modules with jump connections. Unlike the center point detection task, the segmentation network ultimately outputs a probability map of two channels, corresponding to the background and the foreground of the CMBs area respectively. Finally, based on the size of the corresponding probability values of the two channels, it is determined whether each position corresponds to the foreground or the background, and finally the CMBs segmentation result is obtained.

[0048] Alternatively, the model can better locate and segment larger CMBs than smaller ones. To better locate smaller CMBs, we assign different weights to CMBs of different volumes based on their volume. CMBs are hemorrhage points around blood vessels. In images, CMBs have a spherical geometry, so their volume is positively correlated with their size. By assigning different weights to different CMBs based on size, we can help the network focus more on tissue regions with smaller spherical structures.

[0049] Therefore, based on the weights set by the lesion area size, a lesion area segmentation result is obtained from the brain SWI scan image. Specifically, based on the size of the initially detected cerebral microbleeding lesion area, a corresponding weight is assigned to the corresponding cerebral microbleeding lesion area, and finally a lesion area segmentation result is obtained. The lesion area segmentation result includes: one or more cerebral microbleeding lesion areas.

[0050] Preferably, to improve the network's ability to locate and segment small CMBs, during model training, larger weights are assigned to smaller CMBs and smaller weights are assigned to larger CMBs. The weight calculation formula is W = 1 - 0.5 * VVmax - Vmin. This improves the network's performance in locating and segmenting small, difficult-to-segment CMBs. Therefore, the smaller the lesion area, the larger the corresponding weight.

[0051] Optional, such as Figure 3 As shown, the loss functions for these two parallel tasks constrain model training based on the geometric features of CMBs (coordinates, size, and shape). For the detection task, we directly regress the coordinates of the CMB centers and use a weighted Hausdorff distance to maximize the probability of the detection network detecting the CMB centers, thereby achieving the goal of CMB detection. For the segmentation task, we use dice loss and focal loss to predict the CMB regions while addressing the issue of the CMBs occupying a small overall volume. For the final merged output, we use the same loss function as the segmentation task to train the final segmentation task. For the entire network, we assign different weights to different CMBs based on their different volumes, enabling the network to achieve higher segmentation accuracy for relatively small CMBs.

[0052] Optionally, the fusion module adopts a CNN network.

[0053] In order to better illustrate the above-mentioned method for localizing and segmenting cerebral microbleeding lesions, the present invention provides the following specific embodiments.

[0054] Example 1: A method for locating and segmenting cerebral microbleeding lesions. The method comprises:

[0055] Obtain a brain SWI scan image of a patient with small vessel disease to be segmented;

[0056] Based on a multi-task, multi-view constrained CMBs localization and segmentation network, corresponding CMBs localization and segmentation results are obtained according to the brain SWI scan images of the patients with small vessel disease;

[0057] During model training, the overall network assigns greater weight to smaller CMBs and less weight to larger CMBs. The weight is calculated as W = 1-0.5*VVmax-Vmin. This improves the network's performance in localizing and segmenting small, challenging CMBs. The network comprises two parallel localization and segmentation networks: 1) a centerpoint localization network to obtain the coordinates of the CMB centerpoints; 2) a lesion segmentation network to obtain the CMB lesion regions. These two parallel tasks utilize geometric information from multiple viewpoints to constrain the CMB centerpoints and lesion shapes. Finally, a CNN fusion module combines the results of these two branches into the final CMB localization and segmentation results.

[0058] We validated the performance of our proposed algorithm using 105 clinically collected brain SWI scans from patients with small vessel disease. Each image contained at least one CMB, and the 105 data sets contained a total of 822 microbleeds. We implemented the proposed method in Python, building the model using a PyTorch deep learning network. We trained the model on two NVIDIA V100s GPUs. Our proposed method can predict CMBs in one data set in just 2.6 seconds.

[0059] We use accuracy, sensitivity, and average false positive rate to measure the performance of our algorithm. Experimental results show that our algorithm can significantly reduce the false positive rate while achieving similar accuracy to previous methods. It also achieves a Dice value of 0.7461.

[0060] In the prior art, only the segmentation network is used to directly segment CMBs. Due to the small size of CMBs, the segmentation network may miss many detections. Using only the detection network without shape constraints may result in a high false positive prediction rate. However, this embodiment simultaneously detects and segments CMBs, ultimately using a fusion model to fuse the results of the two models. The two tasks are highly correlated, and using a parallel structure allows the two models to mutually enhance each other, addressing both missed detections in the segmentation task and the high false positive detection rate of the detection model. Comparative experiments also show that our parallel structure can indeed improve segmentation performance.

[0061] Similar in principle to the above-mentioned embodiment, the present invention provides a system for localizing and segmenting cerebral microbleeding lesions.

[0062] The following provides specific embodiments in conjunction with the accompanying drawings:

[0063] like Figure 4 A schematic structural diagram of a cerebral microbleed lesion localization and segmentation system in an embodiment of the present invention is shown.

[0064] The system comprises:

[0065] An image acquisition module 41 is used to acquire a SWI scan image of the brain to be located;

[0066] The localization and segmentation module 42 is connected to the image acquisition module 41 and is configured to obtain corresponding localization and segmentation results of cerebral microbleeding lesions based on the brain SWI scan image, based on a multi-task, multi-view constraint-based cerebral microbleeding lesion localization and segmentation model. Since the implementation principles of this cerebral microbleeding lesion localization and segmentation system have been described in the previous embodiment, they will not be repeated here.

[0067] Optionally, the training method of the brain microbleeding lesion localization and segmentation model includes: based on the brain microbleeding lesion volume constraint, training according to multiple lesion localization and segmentation training samples to obtain the brain microbleeding lesion localization and segmentation model; wherein each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleeding lesion localization and segmentation result.

[0068] Optionally, the brain microbleed lesion localization and segmentation model includes: a center point detection module, used to obtain the corresponding center point localization result based on the input brain SWI scan image; a lesion area segmentation module, used to obtain the corresponding lesion area segmentation result based on the input brain SWI scan image; a fusion module, connecting the center point localization module and the lesion area segmentation module, used to fuse the center point localization result and the lesion area segmentation result to obtain the brain microbleed lesion localization and segmentation result.

[0069] Optionally, obtaining the corresponding center point positioning result based on the input brain SWI scan image includes: obtaining a center point positioning probability map based on the brain SWI scan image based on the weighted Hausdorff distance; wherein, the center point positioning probability map includes: the position data of each point and its corresponding brain microbleed lesion center point probability value.

[0070] Optionally, obtaining the corresponding lesion area segmentation result based on the input brain SWI scan image includes: obtaining the lesion area segmentation result based on the brain SWI scan image based on a weight set by the lesion area size; wherein, the lesion area segmentation result includes: one or more brain microbleeding lesion areas.

[0071] Optionally, the smaller the size of the lesion area is, the greater the corresponding weight is.

[0072] like Figure 5 A schematic structural diagram of a cerebral microbleed lesion localization and segmentation terminal 10 according to an embodiment of the present invention is shown.

[0073] The cerebral microbleed lesion positioning and segmentation terminal 50 includes: a memory 51 and a processor 52. The memory 51 is used to store computer programs; the processor 52 runs the computer program to implement the following Figure 2 The method for localizing and segmenting cerebral microbleeding lesions.

[0074] Optionally, the number of the memories 51 can be one or more, the number of the processors 52 can be one or more, and Figure 5 Take one as an example.

[0075] Optionally, the processor 52 in the cerebral microbleed lesion localization and segmentation terminal 50 may be configured as follows: Figure 1 In the above steps, one or more instructions corresponding to the process of the application are loaded into the memory 51, and the processor 52 runs the application stored in the first memory 51, thereby achieving the following Figure 2 Various functions in the cerebral microbleed lesion localization and segmentation method.

[0076] Optionally, the memory 51 may include, but is not limited to, high-speed random access memory and non-volatile memory, such as one or more disk storage devices, flash memory devices, or other non-volatile solid-state storage devices; the processor 52 may include, but is not limited to, a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0077] Optionally, the processor 52 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0078] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed, implements the following Figure 1 The computer-readable storage medium may include, but is not limited to, a floppy disk, an optical disk, a CD-ROM (Compact Disc Read-Only Memory), a magneto-optical disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), an EPROM (Erasable Programmable Read-Only Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a magnetic or optical card, a flash memory, or other types of media / machine-readable media suitable for storing machine-executable instructions. The computer-readable storage medium may be a product not connected to a computer device, or a component connected to a computer device for use.

[0079] In summary, the present invention's method, system, and terminal for cerebral microbleed lesion localization and segmentation utilize a multi-task, multi-view constraint-based cerebral microbleed lesion localization and segmentation model to obtain corresponding cerebral microbleed lesion localization and segmentation results based on brain SWI scan images. This invention utilizes a novel multi-task, multi-view constraint learning framework to achieve rapid and accurate CMBs localization and segmentation. Therefore, this invention effectively overcomes the shortcomings of existing technologies and possesses high industrial value.

[0080] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, any equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A method for locating and segmenting cerebral microbleeding lesions, characterized in that: The method comprises: Obtaining the brain SWI scan image to be segmented; A cerebral microbleeding lesion localization and segmentation model based on multi-task and multi-view constraints obtains corresponding cerebral microbleeding lesion localization and segmentation results according to the brain SWI scan image; the cerebral microbleeding lesion localization and segmentation model includes: A center point detection module is configured to obtain a corresponding center point positioning result based on an input brain SWI scan image; the obtaining of the corresponding center point positioning result based on the input brain SWI scan image comprises: obtaining a center point positioning probability map based on the brain SWI scan image based on a weighted Hausdorff distance; wherein the center point positioning probability map comprises: position data of each point and its corresponding probability value of the center point of the cerebral microbleeding lesion; A lesion region segmentation module is configured to obtain a corresponding lesion region segmentation result based on an input brain SWI scan image; the obtaining of the corresponding lesion region segmentation result based on the input brain SWI scan image comprises: obtaining a lesion region segmentation result based on the brain SWI scan image based on a weight set according to the lesion region size; wherein the lesion region segmentation result includes: one or more cerebral microbleed lesion regions; the smaller the lesion region size, the greater the corresponding weight set; A fusion module is connected to the center point positioning module and the lesion area segmentation module, and is used to fuse the center point positioning result and the lesion area segmentation result to obtain the cerebral microbleeding lesion positioning and segmentation result.

2. The method for locating and segmenting cerebral microbleeding lesions according to claim 1, characterized in that: The cerebral microbleed lesion localization and segmentation model training method includes: Based on the volume constraint of cerebral microbleeding lesions, a cerebral microbleeding lesion localization and segmentation model was obtained by training multiple lesion localization and segmentation training samples; Each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleed lesion localization and segmentation result.

3. The method for locating and segmenting cerebral microbleeding lesions according to claim 1, characterized in that: The smaller the size of the lesion area, the greater the corresponding weight.

4. A cerebral microbleed lesion localization and segmentation system, characterized in that: The system comprises: An image acquisition module, used for acquiring a SWI scan image of the brain to be located; A positioning and segmentation module, connected to the image acquisition module, is used to obtain a corresponding cerebral microbleeding lesion positioning and segmentation result based on the brain SWI scan image based on a multi-task and multi-view constraint cerebral microbleeding lesion positioning and segmentation model; The cerebral microbleed lesion localization and segmentation model includes: A center point detection module is configured to obtain a corresponding center point positioning result based on an input brain SWI scan image; the obtaining of the corresponding center point positioning result based on the input brain SWI scan image comprises: obtaining a center point positioning probability map based on the brain SWI scan image based on a weighted Hausdorff distance; wherein the center point positioning probability map comprises: position data of each point and its corresponding probability value of the center point of the cerebral microbleeding lesion; A lesion region segmentation module is configured to obtain a corresponding lesion region segmentation result based on an input brain SWI scan image; the obtaining of the corresponding lesion region segmentation result based on the input brain SWI scan image comprises: obtaining a lesion region segmentation result based on the brain SWI scan image based on a weight set according to the lesion region size; wherein the lesion region segmentation result includes: one or more cerebral microbleed lesion regions; the smaller the lesion region size, the greater the corresponding weight set; A fusion module is connected to the center point positioning module and the lesion area segmentation module, and is used to fuse the center point positioning result and the lesion area segmentation result to obtain the cerebral microbleeding lesion positioning and segmentation result.

5. The cerebral microbleed lesion localization and segmentation system according to claim 4, characterized in that: The cerebral microbleed lesion localization and segmentation model training method includes: Based on the volume constraint of cerebral microbleeding lesions, a cerebral microbleeding lesion localization and segmentation model was obtained by training multiple lesion localization and segmentation training samples; Each lesion localization and segmentation training sample includes: a brain SWI scan image sample and its corresponding brain microbleed lesion localization and segmentation result.

6. A terminal for locating and segmenting cerebral microbleeding lesions, characterized in that: include: one or more memories and one or more processors; The one or more memories are used to store computer programs; The one or more processors are connected to the memory and are configured to run the computer program to perform the method according to any one of claims 1 to 3.

Images