Carotid artery ROI localization method and device

Abstract

This disclosure relates to a method and apparatus for locating a carotid artery region of interest (ROI). The method includes: acquiring a three-dimensional head and carotid artery MRI image; slicing the three-dimensional head and carotid artery MRI image in a preset direction to obtain a sequence of sliced ​​images in the preset direction; for each preset direction, inputting the sliced ​​images in the sliced ​​image sequence into a pre-trained classification model for classification to obtain the probability that each sliced ​​image belongs to a carotid artery ROI region category; generating a one-dimensional probability vector of the sliced ​​image sequence based on the probability of each sliced ​​image; iteratively processing the one-dimensional probability vector using a sliding window to determine the range in which the mean probability satisfies a preset condition; and determining the carotid artery ROI location result of the three-dimensional head and carotid artery MRI image based on the range corresponding to each preset direction. According to the technical solution of this disclosure, the efficiency and accuracy of carotid artery ROI location can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and in particular to a method and apparatus for locating the ROI of the carotid artery. Background Technology

[0002] Ischemic stroke is characterized by high incidence, disability and mortality rates. Rupture of carotid atherosclerotic plaques is an important cause of ischemic stroke. Therefore, early detection of the stability of carotid atherosclerotic plaques can effectively prevent the occurrence of ischemic stroke.

[0003] Imaging is a crucial tool for assessing carotid artery plaques. Currently, high-resolution MRI of the carotid artery wall is a non-invasive technique for evaluating carotid artery plaques. By suppressing intraluminal blood flow signals and perivascular fat signals, it directly displays the carotid artery wall structure, clearly defining the luminal morphology and sub-millimeter plaque characteristics, showing good consistency with histopathology. However, due to the massive amount of data in high-resolution MRI images, manual interpretation is not only time-consuming and laborious but also easily influenced by subjective factors. Therefore, utilizing artificial intelligence (AI) technology has become an important means to improve diagnostic efficiency and accuracy. The first step in AI technology is to accurately locate and segment plaques, thus providing a reliable foundation for further analysis and evaluation. However, carotid artery plaques only occupy a small portion of the raw high-resolution MRI data. How to accurately crop the region of interest (ROI) of the carotid artery wall from the complete image is a significant challenge. Achieving plaque localization using the Region of Interest (ROI) has become a key challenge in clinical practice and research.

[0004] In related technologies, carotid artery ROI extraction methods are usually combined with deep learning models, such as using cascaded segmentation networks and object detection to extract ROI. However, due to the limited size of the dataset, uneven data distribution, and imbalanced sample labels, deep learning-based methods are prone to overfitting, which can lead to missed detections or incorrect judgments. Furthermore, existing methods usually require cascading multiple models or introducing complex object detection structures, which makes the processing cumbersome and difficult to efficiently meet the needs of clinical work. Summary of the Invention

[0005] To address the aforementioned technical problems, this disclosure provides a method and apparatus for locating the carotid artery ROI.

[0006] In a first aspect, embodiments of this disclosure provide a method for locating a carotid artery region of interest (ROI), including:

[0007] A three-dimensional head and carotid artery MRI image is acquired, and the three-dimensional head and carotid artery MRI image is sliced ​​in a preset direction to obtain a slice image sequence in the preset direction;

[0008] For each preset direction, the slice images in the slice image sequence are input into a pre-trained classification model for classification, and the probability of each slice image belonging to the carotid artery ROI region category is obtained;

[0009] Based on the probability of each slice image, a one-dimensional probability vector of the slice image sequence is generated;

[0010] By iteratively processing the one-dimensional probability vector through a sliding window, the range in which the mean probability satisfies the preset condition is determined.

[0011] Based on the range corresponding to each preset direction, the carotid artery ROI localization result of the three-dimensional head and carotid artery MRI image is determined.

[0012] Secondly, embodiments of this disclosure provide a carotid artery ROI localization device, comprising:

[0013] The acquisition module is used to acquire three-dimensional head and carotid artery MRI images and slice the three-dimensional head and carotid artery MRI images in a preset direction to obtain a slice image sequence in the preset direction.

[0014] The classification module is used to input the slice images in the slice image sequence into a pre-trained classification model for each preset direction to classify them and obtain the probability that each slice image belongs to the carotid artery ROI region category.

[0015] The generation module is used to generate a one-dimensional probability vector of the slice image sequence based on the probability of each slice image;

[0016] The determination module is used to perform iterative processing on the one-dimensional probability vector through a sliding window to determine the range in which the mean probability satisfies a preset condition.

[0017] The positioning module is used to determine the carotid artery ROI positioning result of the three-dimensional head and carotid artery MRI image based on the range corresponding to each preset direction.

[0018] Compared with the prior art, the technical solution provided in this disclosure has the following advantages: It determines the probability of each direction of the slice image belonging to the carotid artery ROI region category through a classification model. Based on the one-dimensional probability vector of the slice image sequence, a sliding window is used to determine the range of the probability mean satisfying preset conditions on the one-dimensional probability vector. Then, based on the range of each direction, the carotid artery ROI localization result of the three-dimensional head and carotid artery MRI image is obtained. Therefore, for cases where data imbalance leads to model overfitting and missed detection on one side, this method uses the probability mean of a sliding window to locate the ROI based on the one-dimensional probability vector of the slice image sequence, which can accurately locate the left and right carotid artery ROI regions, improving the accuracy of carotid artery ROI localization and reducing missed detections caused by data imbalance. Compared with target detection methods, it improves processing efficiency and the efficiency of carotid artery ROI localization. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0020] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a method for locating a carotid artery ROI according to an embodiment of this disclosure.

[0022] Figure 2 A schematic diagram of a one-dimensional probability vector provided in an embodiment of this disclosure;

[0023] Figure 3 This is a schematic diagram of a positioning process provided in an embodiment of the present disclosure;

[0024] Figure 4 This is a schematic diagram of a carotid artery ROI positioning device provided in an embodiment of this disclosure. Detailed Implementation

[0025] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0026] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0027] Figure 1 This is a flowchart illustrating a carotid artery ROI localization method provided in an embodiment of this disclosure. The method provided in this embodiment can be executed by a carotid artery ROI localization device, which can be implemented in software and / or hardware and can be integrated into any electronic device with computing capabilities.

[0028] like Figure 1 As shown, the carotid artery ROI localization method provided in this embodiment may include:

[0029] Step 101: Acquire three-dimensional head and carotid artery MRI images, and slice the three-dimensional head and carotid artery MRI images in a preset direction to obtain a slice image sequence in the preset direction.

[0030] In this embodiment, the three-dimensional head and carotid artery MRI (Magnetic Resonance Imaging) image consists of multiple image sequences. Optionally, the three-dimensional head and carotid artery MRI image includes four image sequences. The three-dimensional head and carotid artery MRI image is preprocessed, and then sliced ​​in any preset direction. The resulting slice image sequence includes multiple slice images, each of which is a multimodal overlapping image, such as a four-modal overlapping image.

[0031] As an example, the three-dimensional head and carotid artery MRI images include T1WI images (T1-weighted images), T2WI images (T2-weighted images), T1WI enhanced images, and time-of-flight (TOF) images. Before slicing the three-dimensional head and carotid artery MRI images in a preset orientation, the preprocessing includes isotropic interpolation of the three-dimensional head and carotid artery MRI images to ensure that the above four sequences of images achieve consistency in spatial resolution.

[0032] The preset directions can be set as needed. For example, the preset directions include axial, sagittal, and coronal. The images are sliced ​​according to the three directions to generate two-dimensional image data, providing multi-view input for deep learning models.

[0033] Optionally, before slicing the three-dimensional head and carotid artery MRI images in a preset direction, quantile normalization is performed on the three-dimensional head and carotid artery MRI images. The upper limit of quantile normalization is 99 and the lower limit is 1. By applying the normalization method to the three-dimensional head and carotid artery MRI images, the consistency of numerical distribution between different modalities and slices is ensured, thereby improving the model convergence speed and robustness.

[0034] Step 102: For each preset direction, input the slice images in the slice image sequence into the pre-trained classification model for classification, and obtain the probability that each slice image belongs to the carotid artery ROI region category.

[0035] Step 103: Generate a one-dimensional probability vector for the slice image sequence based on the probability of each slice image.

[0036] In this embodiment, a corresponding classification model is trained for each preset direction. The input of the classification model is a slice image, and the output is the classification result of the slice image. The classification result indicates whether the slice image belongs to the carotid artery ROI region category. Specifically, it can be the probability that the slice image contains the carotid artery ROI region. Thus, N probabilities are obtained for N slice images in any preset direction. The N probabilities are used to generate a one-dimensional probability vector according to the slice order.

[0037] The training process of the classification model includes: acquiring training data, which consists of positive and negative samples. Positive samples are slices containing carotid artery walls or atherosclerotic plaques, while those not containing these are considered negative samples. The model is then trained on a pre-defined classification network based on the training data to obtain the classification model. Optionally, a pre-trained ResNet can be used as the base classification network to learn classification on the generated two-dimensional slice images. Positive samples are any slice images containing carotid artery walls or atherosclerotic plaques. To address the imbalance between positive and negative samples, the model training loss function employs a focal loss function to enhance the model's ability to learn the minority class and reduce the false negative rate.

[0038] It should be noted that the above implementation of the classification network is not limited to ResNet; other deep learning models such as Transformer can also be used for classification, and no restrictions are imposed here.

[0039] As an example, for the three directions of axial, sagittal and coronal, classification models corresponding to the axial direction, the sagittal direction and the coronal direction are trained respectively. Then, for each direction, the slice image in that direction is classified by the corresponding classification model to obtain the probability of each slice image and generate a one-dimensional probability vector for that direction.

[0040] Step 104: Iterate over the one-dimensional probability vector using a sliding window to determine the range in which the mean probability satisfies the preset conditions.

[0041] In this embodiment, the mean probability within the window is calculated by sliding a sliding window of a specified size on a one-dimensional probability vector. The positioning range is determined on the one-dimensional probability vector based on the mean probability within the window and preset conditions. The probability on the one-dimensional probability vector represents the probability that the corresponding slice image belongs to the carotid artery ROI region category. Optionally, the mean probability satisfies preset conditions, including the maximum mean probability.

[0042] The positioning process is explained below.

[0043] In one embodiment of this disclosure, iterative processing is performed on a one-dimensional probability vector using a sliding window to determine the range in which the probability mean satisfies a preset condition. This includes: determining a first range with the largest probability mean on the one-dimensional probability vector using a first sliding window of a first size; determining a second range with the largest probability mean within the first range of the one-dimensional probability vector using a second sliding window of a second size; roughly locating the carotid artery region within the second range of the one-dimensional probability vector using a third sliding window of a third size to obtain a coarse location range for each side of the carotid artery; and determining the range with the largest probability mean within the coarse location range of the one-dimensional probability vector using a fourth sliding window of a fourth size. In the carotid artery ROI localization scenario, carotid artery sample annotation involves marking plaque portions. However, the location and length of plaques vary among different patients. For example, when plaques are distributed along the z-axis on the same side, one patient's plaque length is 7, while another patient's plaque length is 80. The uneven distribution of the markings themselves often leads to the model overfitting to one side of the carotid artery. In this case, directly using traditional methods will result in missed detections. This method is based on the anatomical symmetry of the left and right carotid arteries. It first uses a longer, non-overlapping sliding window to constrain the range of the left and right carotid arteries, and then uses a smaller sliding window to accurately locate the carotid arteries within the coarse localization range.

[0044] In this model, the first dimension is larger than the second dimension, the second dimension is larger than the third dimension, and the third dimension is larger than the fourth dimension. Optionally, the first and second dimensions are determined by statistically analyzing the spatial distribution characteristics of ROIs in a large number of segmentation probability maps to ensure that sufficient potential ROI regions are covered. Then, on the one-dimensional probability vector, a non-overlapping sliding window of the third dimension is used to coarsely locate the left and right carotid artery regions. This process is mainly used to exclude low-confidence regions and quickly narrow down the search space. The coarse localization range can be one or two, and the number of coarse localization ranges is determined according to the direction. Iterative processing: Based on the coarse localization results, a sliding window of the fourth dimension is further executed within the coarse localization range of each carotid artery. This step can be executed iteratively to gradually optimize the final ROI localization result based on the local peak information of the context probability, effectively avoiding the omission of important structural regions, improving the localization accuracy and completeness, achieving more precise localization of the ROI regions of the left and right carotid arteries, and reducing redundant candidate regions, ensuring the accuracy and reliability of ROI localization.

[0045] The following example illustrates this concept using a diagram of a one-dimensional probability vector. A diagram of a one-dimensional probability vector in a predetermined direction is shown below. Figure 2 It should be noted that, Figure 2 The length of the one-dimensional probability vector and the size of each sliding window are only examples and are not limited here. Figure 2 In the diagram, 22 represents the second sliding window of the second size, the first sliding window is not shown, 23 represents the third sliding window of the third size, 24 represents the fourth sliding window of the fourth size, and 25 represents the range determined on the one-dimensional probability vector.

[0046] As an example, taking the X-direction as an example, the length of the one-dimensional probability vector is 512, the first dimension is 320, the second dimension is 240, where the carotid artery statistics of the sample are obtained, 320 is determined by the range of the statistics of all people in the sample, and 240 is determined by the length of the statistics of a single person in the sample, the third dimension is 100, and the fourth dimension is 64. A first range of length 320 is determined on the one-dimensional probability vector through a first sliding window, a second range of length 240 is determined within the first range through a second sliding window, and then a coarse positioning range of length 100 is determined within the second range through a third sliding window. Further, two fourth sliding windows are used in each coarse positioning range to determine the ROI positioning range in the X-direction. The constraint condition in this step also includes the intersection-union ratio between the two fourth sliding windows, for example, the intersection-union ratio between the two fourth sliding windows is less than or equal to 0.5.

[0047] Optionally, the carotid artery region is coarsely located within the second range of the one-dimensional probability vector to obtain a coarse localization range for each carotid artery. This includes: when there are two coarse localization ranges corresponding to a preset direction, combining the constraint of the intersection-union ratio (IUU) between two third sliding windows, the two coarse localization ranges with the highest probability mean are determined within the second range using the two third sliding windows. The IUU between the two third sliding windows is greater than a threshold, and the threshold is negative. This threshold can be determined experimentally by setting the IUU between the two third sliding windows to be greater than a certain negative value, such as -0.25, to constrain a certain distance between the two coarse localization ranges, avoiding including the area between the left and right carotid arteries that is easily misdetected in the coarse localization range, and further improving the accuracy of carotid artery localization.

[0048] It should be noted that the above method for determining the range on a one-dimensional probability vector is an improved implementation of the concept of NMS (Non-Maximum Suppression). ROI localization is achieved by iteratively performing NMS processing on a one-dimensional probability vector using sliding windows of different sizes. Optionally, other strategies can also be introduced, but no specific restrictions are imposed here.

[0049] Step 105: Determine the carotid artery ROI localization result of the three-dimensional head and carotid artery MRI image according to the range corresponding to each preset direction.

[0050] In this embodiment, after determining the ROI positioning range of each preset direction, the ROI positioning range of each preset direction can be combined with the preset direction to convert the carotid artery ROI positioning result of the three-dimensional head and carotid artery MRI image.

[0051] As an example, taking axial, sagittal, and coronal views, the carotid artery region of interest (ROI) location in a 3D head and carotid MRI image is determined based on the range corresponding to each preset direction. This includes determining the carotid artery ROI location based on the ranges corresponding to the axial, sagittal, and coronal views. In this example, the Cartesian product is calculated using the ranges corresponding to the axial, sagittal, and coronal views to determine the carotid artery ROI location in the 3D head and carotid MRI image. The actual processing flow is as follows: Figure 3 As shown, Figure 3 Cartesian product, Hadamard product, and interactive NMS.

[0052] In one embodiment of this disclosure, after determining the carotid artery ROI localization result of the three-dimensional head and carotid artery MRI image, the method further includes: automatically cropping the three-dimensional head and carotid artery MRI image according to the carotid artery ROI localization result to obtain the carotid artery ROI image cropping result; and visualizing the carotid artery ROI image cropping result, the visualization display form including multimodal overlay map and slice comparison map, which facilitates intuitive verification of the localization accuracy and medical relevance of the ROI region. In this embodiment, by outputting high-quality cropped ROIs, these ROI regions, after precise localization and processing, can be used for subsequent medical research or clinical diagnosis, providing doctors with high-precision image data support and assisting in early disease diagnosis and treatment decisions.

[0053] According to the technical solution of this disclosure, a classification model is used to determine the probability that slice images in each direction belong to the carotid artery ROI region category. Based on the one-dimensional probability vector of the slice image sequence, a sliding window is used to determine the range of the probability mean satisfying preset conditions on the one-dimensional probability vector. Then, the carotid artery ROI localization result of the three-dimensional head and carotid artery MRI image is obtained according to the localization range in each direction. Therefore, for cases where data imbalance leads to model overfitting and missed detection on one side, this method performs ROI localization based on the probability mean of the one-dimensional probability vector of the slice image sequence through a sliding window, which can accurately locate the left and right carotid artery ROI regions. This method improves the accuracy of carotid artery ROI localization and reduces missed detections due to data imbalance. Compared to target detection methods, it eliminates the need for complex target detection model structures, improving processing efficiency and carotid artery ROI localization efficiency. This allows for the stable generation of ROI regions covering key carotid artery structures (including the vessel wall and lesion areas) with zero missed detections, effectively reducing background interference from a large amount of irrelevant tissue. Compared to traditional methods, this workflow not only improves localization accuracy but also significantly reduces interference from invalid information in subsequent image analysis, contributing to improved model performance in downstream tasks (such as segmentation and quantitative analysis). Furthermore, this method is applicable to multi-sequence data from high-resolution MRI of the head and carotid arteries and can be extended to other related clinical localization tasks, demonstrating good versatility.

[0054] Figure 4 This is a schematic diagram of the structure of a carotid artery ROI localization device provided in an embodiment of this disclosure, as shown below. Figure 4 As shown, the carotid artery ROI localization device includes: an acquisition module 41, a classification module 42, a generation module 43, a determination module 44, and a localization module 45.

[0055] The acquisition module 41 is used to acquire three-dimensional head and carotid artery MRI images and slice the three-dimensional head and carotid artery MRI images in a preset direction to obtain a slice image sequence in the preset direction.

[0056] The classification module 42 is used to input the slice images in the slice image sequence into the pre-trained classification model for each preset direction to classify them and obtain the probability that each slice image belongs to the carotid artery ROI region category.

[0057] The generation module 43 is used to generate a one-dimensional probability vector of the slice image sequence based on the probability of each slice image;

[0058] The determination module 44 is used to perform iterative processing on a one-dimensional probability vector through a sliding window to determine the range in which the mean probability satisfies a preset condition.

[0059] The positioning module 45 is used to determine the carotid artery ROI positioning result of the three-dimensional head and carotid artery MRI image according to the range corresponding to each preset direction.

[0060] In one embodiment of this disclosure, the preset directions include axial, sagittal, and coronal positions, and the positioning module 45 is specifically used for:

[0061] Based on the ranges corresponding to the axial, sagittal, and coronal views, the carotid artery ROI localization results of the three-dimensional head and carotid artery MRI images were determined.

[0062] In one embodiment of this disclosure, the device further includes:

[0063] The training module is used to acquire training data; the training data includes positive class samples and negative class samples, and the positive class samples are slices containing carotid artery walls or atherosclerotic plaques;

[0064] The pre-defined classification network is trained based on the training data to obtain the classification model.

[0065] In one embodiment of this disclosure, the loss function for model training is a focus loss function.

[0066] In one embodiment of this disclosure, the determining module 44 is specifically used for:

[0067] A first range with the largest probability mean is determined on a one-dimensional probability vector through a first sliding window of the first size;

[0068] Using a second sliding window of a second size, a second range with the largest probability mean is determined within a first range of the one-dimensional probability vector; the second size is smaller than the first size.

[0069] The carotid artery region is roughly located within the second range of the one-dimensional probability vector using a third sliding window of the third size, thus obtaining the coarse localization range of the carotid artery on each side; the third size is smaller than the second size.

[0070] The range with the maximum probability mean is determined by using a fourth sliding window of the fourth size within the coarse positioning range of the one-dimensional probability vector; the fourth size is smaller than the third size.

[0071] In one embodiment of this disclosure, the determining module 44 is specifically used for:

[0072] When there are two coarse positioning ranges, the crossover ratio between the two third sliding windows is greater than the threshold and the threshold is negative.

[0073] In one embodiment of this disclosure, the three-dimensional head and carotid artery MRI images include T1WI images, T2WI images, T1WI enhanced images, and time-of-flight images. The device also includes:

[0074] The preprocessing module is used to perform isotropic interpolation on three-dimensional head and carotid artery MRI images to ensure that the spatial resolution meets the consistency requirements.

[0075] In one embodiment of this disclosure, the device further includes:

[0076] The normalization module is used to perform quantile normalization on three-dimensional head and carotid artery MRI images; the upper limit of quantile normalization is 99, and the lower limit is 1.

[0077] In one embodiment of this disclosure, the device further includes:

[0078] The display module is used to automatically crop the three-dimensional head and carotid artery MRI image based on the carotid artery ROI localization result, and obtain the carotid artery ROI image cropping result;

[0079] The results of carotid artery ROI image cropping are visualized; the visualization formats include multimodal overlay images and slice comparison images.

[0080] The carotid artery ROI localization device provided in this disclosure can execute any carotid artery ROI localization method provided in this disclosure, and has the corresponding functional modules and beneficial effects for executing the method. Content not described in detail in the device embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.

[0081] This disclosure also provides an electronic device including one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the methods of the embodiments of this disclosure above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0082] In one example, the electronic device may also include input and output devices, which are interconnected via a bus system and / or other forms of connection. Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc. In addition, depending on the specific application, the electronic device may include any other suitable components such as a bus, input / output interfaces, etc.

[0083] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0084] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0085] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0086] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0088] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A carotid ROI positioning method, characterized by, The method comprises: acquiring a three-dimensional head and neck artery MRI image, and slicing the three-dimensional head and neck artery MRI image in a preset direction to obtain a slice image sequence in the preset direction; for each preset direction, inputting a slice image in the slice image sequence into a pre-trained classification model for classification to obtain a probability that each slice image belongs to a carotid artery ROI region category; generating a one-dimensional probability vector of the slice image sequence according to the probability of each slice image; determining a range in which a probability mean value meets a preset condition through iterative processing of the one-dimensional probability vector by a sliding window; determining a carotid artery ROI positioning result of the three-dimensional head and neck artery MRI image according to a ROI positioning range corresponding to each preset direction; the iterative processing of the one-dimensional probability vector by the sliding window to determine the range in which the probability mean value meets the preset condition comprises: determining a first range in which a probability mean value is maximum in the one-dimensional probability vector by a first sliding window of a first size; determining a second range in which a probability mean value is maximum in the first range of the one-dimensional probability vector by a second sliding window of a second size; the second size is smaller than the first size; coarsely positioning a carotid artery region in the second range of the one-dimensional probability vector by a third sliding window of a third size to obtain a coarse positioning range of each side carotid artery; the third size is smaller than the second size; determining a range in which a probability mean value is maximum in the coarse positioning range of the one-dimensional probability vector as the ROI positioning range corresponding to the preset direction by a fourth sliding window of a fourth size; the fourth size is smaller than the third size; the coarse positioning of the carotid artery region in the second range of the one-dimensional probability vector to obtain the coarse positioning range of each side carotid artery comprises: in the case that the number of the coarse positioning ranges is two, determining two coarse positioning ranges in which probability mean values are maximum in the second range by two third sliding windows, and the intersection over union between the two third sliding windows is greater than a threshold value and the threshold value is negative.

2. The method of claim 1, wherein, The preset direction comprises an axial position, a sagittal position and a coronal position, and the determination of the carotid artery ROI positioning result of the three-dimensional head and neck artery MRI image according to the range corresponding to each preset direction comprises: determining the carotid artery ROI positioning result of the three-dimensional head and neck artery MRI image according to the range corresponding to the axial position, the range corresponding to the sagittal position and the range corresponding to the coronal position.

3. The method of claim 1, wherein, The method further comprises: acquiring training data; the training data comprises positive class samples and negative class samples, and the positive class samples are slices containing carotid artery walls or atherosclerotic plaques; model training of a preset classification network based on the training data to obtain the classification model.

4. The method of claim 3, wherein, The loss function of the model training adopts a focal loss function.

5. The method of claim 1, wherein, The three-dimensional head and neck artery MRI image comprises a T1WI image, a T2WI image, a T1WI enhanced image and a time flight image, and the method further comprises, before the slicing of the three-dimensional head and neck artery MRI image in the preset direction: The three-dimensional head and neck artery MRI image is isotropically interpolated to make the spatial resolution meet a consistency condition.

6. The method of claim 5, wherein, Before the three-dimensional head and neck artery MRI image is sliced in the preset direction, the method further comprises: The three-dimensional head and neck artery MRI image is subjected to quantile normalization processing; wherein the upper limit of the quantile normalization processing is 99 and the lower limit is 1.

7. The method of claim 1, wherein, After the carotid artery ROI positioning result of the three-dimensional head and neck artery MRI image is determined, the method further comprises: According to the carotid artery ROI positioning result, the three-dimensional head and neck artery MRI image is automatically cropped to obtain a carotid artery ROI image cropping result; The carotid artery ROI image cropping result is visualized and displayed; the visualized display form includes a multi-modal overlap graph and a slice comparison graph.

8. A carotid ROI positioning device, characterized by, Comprise: The acquisition module is configured to acquire a three-dimensional head and neck artery MRI image, and slice the three-dimensional head and neck artery MRI image in a preset direction to obtain a slice image sequence in the preset direction; The classification module is configured to input a slice image in the slice image sequence into a pre-trained classification model for classification to obtain a probability that each slice image belongs to a carotid artery ROI region category for each preset direction; The generation module is configured to generate a one-dimensional probability vector of the slice image sequence according to the probability of each slice image; The determination module is configured to determine a range in which a probability mean value meets a preset condition by iterative processing on the one-dimensional probability vector through a sliding window; The positioning module is configured to determine a carotid artery ROI positioning result of the three-dimensional head and neck artery MRI image according to an ROI positioning range corresponding to each preset direction; The determination module is specifically configured to: Determine a first range with a maximum probability mean value on the one-dimensional probability vector through a first sliding window of a first size; Determine a second range with a maximum probability mean value within the first range of the one-dimensional probability vector through a second sliding window of a second size; the second size is smaller than the first size; Coarsely position a carotid artery region within the second range of the one-dimensional probability vector through a third sliding window of a third size to obtain a coarse positioning range of each carotid artery; the third size is smaller than the second size; Determine a range with a maximum probability mean value within the coarse positioning range of the one-dimensional probability vector as the ROI positioning range corresponding to the preset direction through a fourth sliding window of a fourth size; the fourth size is smaller than the third size; The determination module is specifically configured to: In a case where the number of coarse positioning ranges is two, determine two coarse positioning ranges with maximum probability mean values within the second range through two third sliding windows; the intersection over union between the two third sliding windows is greater than a threshold value and the threshold value is negative.

Images