Semantic segmentation-based aortic dissection rupture risk intelligent identification method and system

Abstract

The invention relates to the technical field of image data processing, in particular to an intelligent aortic dissection rupture risk identification method and system based on semantic segmentation, and the method comprises the steps: collecting CT images of a same patient at different time points; acquiring a plurality of connected domains of an aorta region in the CT image; obtaining the cavity wall morphological irregularity of the cavity wall connected domain; determining a noise cavity region based on cavity wall morphological irregularity; obtaining a non-fracture prediction coefficient by using the true and false connected domains of the noise cavity region; determining a risk trend index according to the non-fracture prediction coefficient; according to the risk trend index, determining a noise cavity area with a fracture risk; based on the risk trend index of the noise cavity region with the fracture risk and the cavity wall morphological irregularity, obtaining a risk prediction coefficient of the noise cavity region with the fracture risk; and obtaining a risk level evaluation result according to the risk prediction coefficient. The risk identification accuracy can be improved.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, specifically to a method and system for intelligent identification of aortic dissection rupture risk based on semantic segmentation. Background Technology

[0002] Aortic dissection occurs when the intima of the aorta tears, allowing blood to flow through the rupture into the vessel wall's media, forming a false lumen parallel to the original true lumen. This false lumen is only covered by a thin adventitia, making it structurally fragile. It not only severely impairs normal blood flow but can also cause ischemia in important branch vessels. If the adventitia of the false lumen cannot withstand the blood flow pressure and ruptures, blood will rush into the pleural cavity, abdominal cavity, or pericardial cavity instantly, leading to sudden death. This is the most critical complication of aortic dissection. Therefore, accurate identification of the risks before rupture occurs is crucial for gaining time for clinical intervention and improving patient prognosis.

[0003] Currently, the automatic identification of aortic dissection rupture risk mainly relies on semantic segmentation technology of CT images. This method performs pixel-level classification of CT images to extract structures such as the true lumen and false lumen, and then performs morphological analysis based on the segmentation results. The maximum diameter of the aortic dissection is usually used as the main static indicator for assessing rupture risk. However, this type of method has significant limitations: first, it relies on static measurements at a single time point, lacking individualized baseline controls and failing to reflect differences between patients; second, it does not incorporate temporal information, making it difficult to capture the dynamic evolution trend of the dissection morphology; and third, it uses only the overall diameter as a criterion, ignoring the detailed features of the local lumen wall morphology, thus easily leading to false positives or false negatives in clinical applications, resulting in low accuracy in risk identification. Summary of the Invention

[0004] This invention provides a method and system for intelligent identification of aortic dissection rupture risk based on semantic segmentation, in order to solve existing problems.

[0005] The present invention employs the following technical solution for the intelligent identification method and system for aortic dissection rupture risk based on semantic segmentation:

[0006] One embodiment of the present invention provides an intelligent identification method for the risk of aortic dissection rupture based on semantic segmentation, the method comprising the following steps:

[0007] CT images of the same patient were acquired at different time points; one time point corresponds to one CT image.

[0008] Semantic segmentation is performed on CT images to obtain multiple connected domains in the aortic region of the CT images; wherein the multiple connected domains include at least a true lumen connected domain, a false lumen connected domain, and a lumen wall connected domain;

[0009] The cavity wall morphological irregularity of the cavity wall connected domain is obtained by using the Hough circle detection results of the cavity wall connected domain, as well as the edge pixels and geometric center of the cavity wall connected domain.

[0010] The noise cavity region was determined based on the morphological irregularity of the cavity wall;

[0011] By utilizing the true and false connected components of the noise cavity region, the non-broken prediction coefficients of the noise cavity region are obtained;

[0012] The risk trend index of the noise cavity region is determined based on the non-rupture prediction coefficient of the noise cavity region.

[0013] Based on the risk trend index of the noise cavity region, the noise cavity regions with a risk of rupture are identified;

[0014] Based on the risk trend index of the noise cavity region with rupture risk and the morphological irregularity of the cavity wall, the risk prediction coefficient of the noise cavity region with rupture risk is obtained.

[0015] Based on the risk prediction coefficient of the noise cavity region at risk of rupture, the risk level assessment result is obtained.

[0016] Furthermore, the specific steps for obtaining the cavity wall morphological irregularity of the cavity wall connected region using the Hough circle detection results of the cavity wall connected region, as well as the edge pixels and geometric center of the cavity wall connected region, are as follows:

[0017] For each edge pixel in the cavity wall connected domain, calculate the absolute value of the difference between the x-coordinate of the edge pixel and the x-coordinate of the geometric center of the cavity wall connected domain, as well as the absolute value of the difference between the y-coordinate of the edge pixel and the y-coordinate of the geometric center of the cavity wall connected domain. Then sum the two absolute values ​​to obtain the difference between each edge pixel and the geometric center of the cavity wall connected domain.

[0018] The difference between all edge pixels of the cavity wall connected region and the geometric center is summed to obtain the difference between the cavity wall connected region and the geometric center.

[0019] The ratio of the side length of the cavity wall connected domain to the Hough circle detection result of the cavity wall connected domain is calculated, and the ratio is multiplied by the difference between the cavity wall connected domain and the geometric center to obtain the cavity wall morphological irregularity of the cavity wall connected domain.

[0020] Furthermore, the specific steps for determining the noise cavity region based on the morphological irregularity of the cavity wall are as follows:

[0021] When the morphological irregularity of the cavity wall after normalization of the cavity wall connected domain is greater than or equal to the preset irregularity threshold, the cavity wall connected domain is determined to be a noisy cavity region.

[0022] Furthermore, the specific steps for obtaining the non-broken prediction coefficients of the noise cavity region using the true and false connected components of the noise cavity region are as follows:

[0023] For each noise cavity region, calculate the ratio of the area of ​​the true cavity connected region to the area of ​​the false cavity connected region, as well as the mean of the Hough circle detection results of the true cavity connected region and the false cavity connected region. Then, determine the non-breaking prediction coefficient of each noise cavity region by multiplying the ratio and the mean.

[0024] Furthermore, the specific steps for determining the risk trend index of the noise cavity region based on the non-fracture prediction coefficient of the noise cavity region are as follows:

[0025] Obtain the unbroken prediction coefficients of each noise cavity region at different time points, and construct the unbroken prediction coefficient sequence of each noise cavity region in the order of time points;

[0026] For the non-rupture prediction coefficient sequence of each noise cavity region, calculate the ratio of two adjacent elements, sum the ratios of all two adjacent elements, divide the sum by the first quantity, and obtain the risk trend index of each noise cavity region; where the first quantity is the number of elements minus 1.

[0027] Furthermore, the specific steps for determining the noise cavity region at risk of rupture based on the risk trend index of the noise cavity region are as follows:

[0028] When the risk trend index of each noise cavity region is greater than the preset index threshold, it is determined that there is a risk of rupture in the noise cavity region.

[0029] Furthermore, it also includes:

[0030] When the risk trend index of each noise cavity region is less than or equal to the preset index threshold, it is determined that there is no risk of rupture in the noise cavity region.

[0031] Furthermore, the specific steps for obtaining the risk prediction coefficient for a noise cavity region at risk of rupture, based on the risk trend index of the region and the morphological irregularity of the cavity wall, are as follows:

[0032] For a noise cavity region at risk of rupture, the mean value of the cavity wall morphological irregularity at different time points is calculated, and the product of the mean value and the risk trend index of the noise cavity region is determined as the risk prediction coefficient of the noise cavity region.

[0033] Furthermore, the specific steps for obtaining the risk level assessment result based on the risk prediction coefficient of the noise cavity region with a risk of rupture are as follows:

[0034] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within the preset first threshold range, the risk level is judged to be low and regular monitoring is required.

[0035] When the normalized risk prediction coefficient of the noise cavity region at risk of rupture is within the preset second threshold range, the risk level is judged to be medium and further evaluation is required.

[0036] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within the preset third threshold range, the risk level is judged to be high and preventive measures need to be taken.

[0037] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture falls within the preset fourth threshold range, the risk level is determined to be extremely high, requiring emergency handling and intervention.

[0038] Furthermore, the description also includes:

[0039] When the morphological irregularity of the cavity wall after normalization is less than the preset irregularity threshold, the cavity wall connected region is determined to be a non-noise cavity region.

[0040] The beneficial effects of the technical solution of this invention are as follows: This invention proposes an intelligent identification method and system for aortic dissection rupture risk based on semantic segmentation. By acquiring CT images of the same patient at multiple time points and performing semantic segmentation processing, fine-grained labeling results of various aortic structures are obtained. Then, based on the morphological irregularities of the lumen wall, high-risk areas (i.e., noisy lumen areas) are extracted. By analyzing the area ratio of the true lumen to the false lumen within this area, and combining this with multi-time-point image data for dynamic trend calculation, the risk progression determination and its quantified risk coefficient are finally obtained, thereby achieving intelligent, dynamic, and individualized identification of aortic dissection rupture risk. This invention can improve the accuracy of risk identification. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the steps of the intelligent identification method for aortic dissection rupture risk based on semantic segmentation of the present invention.

[0043] Figure 2 This is a schematic diagram of the normal cavity wall and the interlayer cavity wall in this invention. Detailed Implementation

[0044] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent identification method and system for aortic dissection rupture risk based on semantic segmentation proposed in this invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent identification method and system for aortic dissection rupture risk based on semantic segmentation provided by this invention.

[0047] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent identification of aortic dissection rupture risk based on semantic segmentation, according to an embodiment of the present invention. The method includes the following steps:

[0048] Step S001: Acquire CT images of the same patient at different time points; where each time point corresponds to one CT image.

[0049] It should be noted that this embodiment acquires three-dimensional volumetric image data obtained from CT scans of the same patient at multiple different time points. Each time point corresponds to a complete CT scan examination, and its output is usually regarded as a whole three-dimensional data unit (e.g., a DICOM series or a NIFTI format three-dimensional image file) in medical image processing. This data unit logically corresponds to "one" three-dimensional image, but internally it contains multiple consecutively arranged cross-sectional slice images in the z-axis direction, which together represent the complete three-dimensional anatomical structure of the aorta at that moment.

[0050] For example, during regular follow-up of a patient who has undergone aortic dissection, all slice data generated from the first CT scan performed one month post-surgery are reconstructed and integrated to constitute a time point. The corresponding 3D CT image; the data from the second examination three months post-surgery constitutes the time point. The corresponding 3D CT image.

[0051] Step S002: Perform semantic segmentation on the CT image to obtain multiple connected domains of the aortic region in the CT image; wherein, the multiple connected domains include at least the true lumen connected domain, the false lumen connected domain, and the lumen wall connected domain.

[0052] It should be noted that semantic segmentation is a key technology in the field of computer vision. Its goal is to classify each pixel in an image, thereby achieving a refined understanding of the image content. This technology learns features and patterns in images through deep neural networks, enabling it to automatically identify and segment different target objects.

[0053] During training, the network receives a large amount of pixel-level labeled image data. In this labeled data, each pixel is assigned a label to indicate its category. By continuously adjusting the network parameters to minimize the difference between the predicted segmentation results and the true labels, the model gradually learns the ability to accurately classify pixels.

[0054] Semantic segmentation technology plays a crucial role in medical image analysis, particularly in the intelligent identification of aortic dissection. By performing pixel-level segmentation on CT images, key structures of the aorta, such as the true lumen, false lumen, lumen wall (i.e., vessel wall), intimal tear, and surrounding hematoma, can be automatically and accurately delineated, forming corresponding region masks. These precise segmentation results provide a foundation for subsequent analysis.

[0055] Commonly used semantic segmentation models include U-Net, U-Net++, DeepLabV3+, and Swin-Unet, which incorporates the Transformer architecture. These models, trained on large datasets of labeled aortic images, are able to achieve robust and accurate pixel-level classification of structures in this region.

[0056] Specifically, multiple temporal CT images were acquired and semantic segmentation was performed to obtain the labeling results of different regions of the aorta: First, the images of the same patient at different time points were input. Multiple CT images acquired (total) The images were preprocessed, including resampling all images to the same resolution of 1×1×1 mm³, intensity normalization (adjusting window width and window level and normalizing HU values ​​to the [0, 1] interval), and using affine or elastic registration algorithms (such as Elastix) to match the images at each time point with the reference time point (…). The images were spatially aligned. To improve the model's temporal stability, a temporal consistency loss function was constructed during training to encourage stable segmentation results at adjacent time points when image content changes are small. The AdamW optimizer was used during model training, with a cosine annealing hot restart strategy for learning rate scheduling (e.g., an initial period of 10, multiplied by 2), and a batch size of 2 (to accommodate the memory limitations of 3D data), for a total of 300 training cycles. To prevent overfitting, Dropout with a dropout rate of 0.3 and spatial Dropout3D with a dropout rate of 0.2 were applied for regularization, and a temporal consistency data augmentation strategy was employed (i.e., applying the same random transformation to images from all time points of the same patient). The post-segmentation processing included: topological correction of the segmentation results to ensure the single connectivity of the true lumen and the possible multi-connectivity of the false lumen; applying Gaussian filtering (e.g., σ=1.5) along the temporal dimension to smooth temporal fluctuations; and extracting cross-sectional features along the aortic centerline to ensure spatial comparability of features at different time points. Finally, the output file (e.g., NIFTI format) contains fine structural markers of the aorta (such as the true lumen, false lumen, and vessel wall) at each time point. The network outputs the class probability of each pixel and performs thresholding to obtain the final pixel-level classification label, thus providing accurate anatomical structural basis for subsequent risk prediction and diagnostic analysis.

[0057] Thus, after obtaining the labeling results of different regions of the aorta at each time point, the true and false lumen regions and other lesion regions of the aorta are further analyzed based on the classification labels of different regions.

[0058] By combining the above steps with multi-temporal CT image data and semantic segmentation technology, different regions of the aorta can be accurately marked, providing a basis for further risk prediction and diagnosis.

[0059] Step S003: Using the Hough circle detection results of the cavity wall connected domain, as well as the edge pixels and geometric center of the cavity wall connected domain, obtain the cavity wall morphological irregularity of the cavity wall connected domain.

[0060] It should be noted that: by obtaining the aortic region marking results, extracting the noisy cavity region based on the morphological irregularity of the cavity wall, analyzing the ratio of true to false area of ​​the noisy cavity, and combining the results of multi-time series CT images, the dynamic change value is determined, thereby obtaining the false judgment result and its corresponding risk prediction coefficient, which achieves the purpose of intelligent identification of aortic dissection rupture risk.

[0061] First, the aortic structural markers, obtained through semantic segmentation and arranged in temporal order, are acquired. The markers accurately delineate the relevant anatomical structures of the aorta, mainly including key regions such as the true lumen, false lumen, and vessel wall (luminal wall).

[0062] Each labeled region in an image is represented as one or more connected components, which are sets of pixels with the same category label that are spatially adjacent. Each connected component contains information such as its specific spatial location, the number of pixels it covers (directly related to the area of ​​the region), and the CT value (pixel value) corresponding to the pixel.

[0063] Existing methods for identifying aortic dissection risk typically calculate the maximum diameter of the entire aorta or dissection region based on the segmentation results described above, and compare this static measurement with a standardized clinical threshold to infer the risk of rupture. The core assumption is that a larger diameter results in greater stress on the vessel wall, thus increasing the risk of rupture. However, this method leads to static false positives.

[0064] Furthermore, the occurrence and development of aortic dissection does not uniformly affect the entire aorta. Therefore, in order to achieve accurate risk identification, this embodiment needs to combine the specific features of the aorta itself and its aortic wall on CT images when aortic dissection or rupture precursors occur, and extract the target lumen wall (i.e. the lumen wall to be analyzed).

[0065] First, the arterial lumen wall is analyzed. When aortic dissection or precursors to rupture occur, the aorta exhibits irregular, concave characteristics. Specifically, dissection causes concavity or irregular structures on the inner wall of the aorta. These morphological changes can be detected in CT images, especially in high-resolution images. Characteristically, the inner wall of the aortic lumen may appear uneven or abnormally tortuous, indicating the occurrence of dissection. Therefore, regions exhibiting these characteristics are extracted and designated as the noisy lumen region, or noise lumen region. The noise lumen region indicates the possible presence of aortic dissection or precursors to rupture in the aortic region.

[0066] Specifically, this includes: for each edge pixel in the cavity wall connected domain, calculating the absolute value of the difference between the x-coordinate of the edge pixel and the x-coordinate of the geometric center of the cavity wall connected domain, and the absolute value of the difference between the y-coordinate of the edge pixel and the y-coordinate of the geometric center of the cavity wall connected domain, and summing the two absolute values ​​to obtain the difference value between each edge pixel in the cavity wall connected domain and the geometric center.

[0067] It should be noted that: Figure 2 This is a schematic diagram of the normal cavity wall and the interlayer cavity wall in this invention, as shown below. Figure 2 As shown, in addition to the depression, the cavity wall also exhibits slight bends, meaning that the edges of the cavity wall's connected domains are distributed rather randomly, and the side lengths increase due to this randomness (the diameter increases in existing methods).

[0068] The random arrangement of edges can be quantified as follows: for a cavity wall connected region, the coordinates of the edge pixels are... ( Indicates the first term in the connected component (number of connected edge pixels), geometric center coordinates are: The analysis focuses on the difference between the coordinates of edge pixels and their geometric center. The significance of this difference lies in the fact that if the edges are regular, the difference tends to be 0 (consistent circle radii). Conversely, the difference accumulates to a higher value. Specifically, the coordinates of the edge pixels and their geometric center within the cavity wall connected domain are obtained using a binary mask derived from semantic segmentation: all pixel coordinates constituting the contour of this connected domain are extracted using an edge detection algorithm; its geometric center (centroid) is obtained by calculating the arithmetic mean of all pixel coordinates within the mask.

[0069] At this point, the difference in the horizontal coordinates is: The difference in the ordinates is: ; It is denoted as the difference between each edge pixel of the cavity wall connected domain and the geometric center.

[0070] The difference between the edge pixels of the cavity wall connected region and the geometric center is summed to obtain the difference between the cavity wall connected region and the geometric center.

[0071] It should be noted that: The difference between the connected domain of the cavity wall and the geometric center is denoted as . Indicates the first The absolute value of the difference between the x-coordinate of each edge pixel and the x-coordinate of the geometric center of the connected region of the cavity wall. Indicates the first The absolute value of the difference between the ordinate of each edge pixel and the ordinate of the geometric center of the connected region of the cavity wall. This represents the number of edge pixels in the connected domain of the cavity wall.

[0072] The ratio of the side length of the cavity wall connected domain to the Hough circle detection result of the cavity wall connected domain is calculated, and the ratio is multiplied by the difference between the cavity wall connected domain and the geometric center to obtain the cavity wall morphological irregularity of the cavity wall connected domain.

[0073] It should be noted that for each CT image, the image contains the acquisition time, multiple connected components and their geometry, and the area and side length of each connected component.

[0074] Obtain the geometry of each connected component, perform Hough circle detection on it, and obtain the Hough circle detection result. The closer the detection result is to 1, the closer the shape of the connected component is to a circle.

[0075] Hough circle detection identifies circular regions using a parameter space voting mechanism: for each edge point in a connected component, the algorithm traverses all possible circles (defined by their center coordinates and radius) that might pass through that point, accumulating votes in the corresponding parameter space. The center and radius of the ideal circle receive the most edge point votes, thus forming a significant peak in the accumulator. Finally, the algorithm uses the circle corresponding to the peak as the detection result and generates a roundness confidence score between 0 and 1 by comparing the voting score with the score of the theoretically perfect circle. The higher the score, the closer the geometry of the connected component is to a standard circle. Hough circle detection is a well-known technique and will not be elaborated upon here.

[0076] Therefore, the higher the Hough circle test result, the more regular the morphological features of the aortic lumen wall, that is, the lower the indentation or irregularity characteristics of the aforementioned lumen wall features.

[0077] The expression for the cavity wall morphological irregularity of the cavity wall connected domain is:

[0078]

[0079] in, Indicates the first Cavity wall morphological irregularities of connected domains within the cavity wall Indicates the first The side length of the connected domain of the cavity wall Indicates the first Hough circle detection results for each cavity wall connected domain.

[0080] This indicates the irregularity of the cavity wall morphology from a macroscopic visual perspective. The larger the value, the higher the Hough circle test score corresponding to the connected domain of the lumen wall, and the closer the shape is to a circle, meaning the lower the morphological irregularity of the lumen wall. Conversely, the smaller the value, the higher the irregularity. The longer the side length of the connected domain, the greater the stretching of the arterial lumen wall, the larger the diameter, and the higher the irregularity.

[0081] The summation indicates some noise points that may exist at the edges of the connected regions in a microscopic visual sense, i.e., irregular points on the arterial wall. The larger the summation, the less regular the distribution of the corresponding edge pixels relative to the geometric center, and the more random the distribution.

[0082] Step S004: Determine the noise cavity region based on the morphological irregularity of the cavity wall.

[0083] Specifically, this includes: when the morphological irregularity of the cavity wall after normalization of the cavity wall connected domain is greater than or equal to a preset irregularity threshold, the cavity wall connected domain is determined to be a noisy cavity region.

[0084] It should be noted that each cavity wall connected domain has a morphological irregularity, which is normalized.

[0085] The preset irregularity threshold is set according to the specific situation, and is preferably 0.1 here.

[0086] For a normal arterial wall, its morphological regularity is relatively high, so its normalized morphological irregularity is generally less than 0.1. Therefore, when the normalized morphological irregularity of the arterial wall is ≥0.1, the connected region of the arterial wall is determined to be a noisy cavity region.

[0087] When the morphological irregularity of the cavity wall after normalization is less than the preset irregularity threshold, the cavity wall connected region is determined to be a non-noise cavity region.

[0088] It should be noted that when the normalized cavity wall morphological irregularity is <0.1, the cavity wall connected region is determined to be a non-noise cavity region.

[0089] Step S005: Utilize the true and false connected components of the noise cavity region to obtain the non-fragmented prediction coefficients of the noise cavity region.

[0090] It is important to note that after identifying the noisy lumen region, further analysis of the area ratio between the corresponding true lumen and false lumen is necessary. The principle is as follows: after aortic dissection, blood flows through the intimal tear into the vessel wall's media, forming a false lumen coexisting with the original true lumen. On CT images, the true lumen appears high-density due to rapid blood flow and adequate contrast filling; the false lumen, on the other hand, appears low-density due to blood stasis, poor contrast filling, or the presence of thrombi. The continued existence and expansion of the false lumen exerts pressure on its adventitia. When the area of ​​the false lumen significantly increases relative to the true lumen (i.e., the true lumen / false lumen area ratio decreases), it indicates that the false lumen dominates, the abnormal stress on the vessel wall is more concentrated, and the risk of rupture increases accordingly. Therefore, for each identified noisy lumen region, it is necessary to spatially locate and extract the directly adjacent true lumen and false lumen connected regions, and then calculate their area ratio as one of the core predictive factors for quantifying the risk of this local area.

[0091] The connected components of the true cavity and false cavity corresponding to each noisy cavity region are automatically determined by analyzing the spatial adjacency relationship between the noisy cavity region and the connected components of each category in the segmentation result. Specifically, based on the pixel position of the noisy cavity region on the two-dimensional cross-sectional image, a neighborhood search is performed towards the inside of the blood vessel lumen. Connected components belonging to non-blood vessel wall pixels that are directly in contact with one side edge are identified as true cavities, while connected components belonging to non-blood vessel wall pixels that are directly in contact with the other side edge are identified as false cavities.

[0092] Specifically, this includes: for each noise cavity region corresponding to the true cavity connected domain and the false cavity connected domain, calculating the ratio of the area of ​​the true cavity connected domain to the area of ​​the false cavity connected domain, as well as the mean of the Hough circle detection results of the true cavity connected domain and the false cavity connected domain, and determining the non-breaking prediction coefficient of each noise cavity region by multiplying the ratio and the mean.

[0093] The expression for the unbroken prediction coefficient of each noise cavity region is:

[0094]

[0095] in, Indicates the first The non-fracture prediction coefficients for each noise cavity region , They represent the first The areas of the true cavity connected regions and the areas of the false cavity connected regions corresponding to each noise cavity region. , They represent the first The Hough circle detection results for the true cavity connected domain and the false cavity connected domain corresponding to each noise cavity region.

[0096] The smaller the ratio of the true lumen area to the false lumen area, the higher the proportion of the false lumen, and the higher the risk of aortic dissection rupture.

[0097] The shape of the true and false cavities is related to the risk of rupture. Certain shapes, such as elliptical cavities, are more prone to interstitial rupture, and their corresponding rupture risks are higher. Therefore, the lower the mean Hough circle test score of the true and false cavities, the more the shape of the true and false cavities deviates from a regular circle (e.g., elliptical, irregular, etc.), and the higher the corresponding rupture risk.

[0098] In summary, the lower the non-rupture prediction coefficient mentioned above, the higher the probability of rupture.

[0099] Step S006: Determine the risk trend index of the noise cavity region based on the non-rupture prediction coefficient of the noise cavity region.

[0100] Specifically, this includes: obtaining the unbroken prediction coefficients of each noise cavity region at different time points, and constructing a sequence of unbroken prediction coefficients for each noise cavity region in the order of time points.

[0101] It should be noted that: at different times The next The non-fracture prediction coefficients differ for each noise cavity region. Arranged chronologically, the following results were obtained: The sequence of unbroken prediction coefficients for each noise cavity region.

[0102] For the non-rupture prediction coefficient sequence of each noise cavity region, calculate the ratio of two adjacent elements, sum the ratios of all two adjacent elements, divide the sum by the first quantity, and obtain the risk trend index of each noise cavity region; where the first quantity is the number of elements minus 1.

[0103] It should be noted that: by combining multi-time-series CT image results, false positive results and their corresponding risk prediction coefficient values ​​are determined. Each element represents a non-rupture prediction coefficient in the non-rupture prediction coefficient sequence, and the number of elements represents the number of non-rupture prediction coefficients in the sequence.

[0104] Get the first result corresponding to the current detection result The unbroken prediction coefficients of each noise cavity region are dynamically iterated over multiple frames to track their changes:

[0105]

[0106] in, Indicates the first Risk trend index for each noise cavity region; This indicates the total number of tests minus 1, which is the first quantity. Indicates the first In the unbroken prediction coefficient sequence of each noise cavity region, time points The corresponding non-rupture prediction coefficient, i.e. the h-th non-rupture prediction coefficient in the non-rupture prediction coefficient sequence; Indicates the first In the unbroken prediction coefficient sequence of each noise cavity region, time points The corresponding non-rupture prediction coefficient is the (h+1)th non-rupture prediction coefficient in the non-rupture prediction coefficient sequence. The total number of detections represents the number of non-rupture prediction coefficients in the non-rupture prediction coefficient sequence.

[0107] Step S007: Based on the risk trend index of the noise cavity region, identify the noise cavity regions with a risk of rupture.

[0108] Specifically, this includes: when the risk trend index of each noise cavity region is greater than a preset index threshold, it is determined that the noise cavity region is at risk of rupture.

[0109] It should be noted that the preset index threshold is set according to the specific situation, and here it is preferably 1.

[0110] when Compare A smaller ratio, i.e., a ratio greater than 1, indicates that the non-rupture predictive coefficient of aortic dissection is continuously decreasing, which in turn indicates that the risk of rupture is continuously increasing. When U > 1, it indicates that the risk is progressing, meaning that there is a risk of rupture in the aortic cavity region.

[0111] When the risk trend index of each noise cavity region is less than or equal to the preset index threshold, it is determined that there is no risk of rupture in the noise cavity region.

[0112] It should be noted that when U≤1, it indicates that the risk is stabilizing or mitigating, and there is no risk of rupture in the noise cavity region for the time being.

[0113] Step S008: Based on the risk trend index of the noise cavity region with rupture risk and the morphological irregularity of the cavity wall, obtain the risk prediction coefficient of the noise cavity region with rupture risk.

[0114] Specifically, this includes: for noise cavity regions with a risk of rupture, calculating the mean value of the cavity wall morphological irregularity at different time points, and multiplying the mean value by the risk trend index of the noise cavity region to determine the risk prediction coefficient of the noise cavity region.

[0115] It should be noted that: for the noise cavity region where U > 1, the risk prediction coefficient is calculated, and the expression for the risk prediction coefficient of the noise cavity region is:

[0116]

[0117] in, This represents the risk prediction coefficient for the noise cavity region s, which is at risk of rupture. The risk trend index indicates the risk of rupture in the noise cavity region s. The mean value of the cavity wall morphological irregularity of the cavity region s with rupture risk at different time points indicates that the higher the mean value, the greater the rupture risk.

[0118] Step S009: Obtain the risk level assessment result based on the risk prediction coefficient of the noise cavity region with the risk of rupture.

[0119] It should be noted that: As a risk prediction coefficient, each noise cavity region with a risk of rupture has one. For each After normalization, the final risk prediction coefficient is obtained. The higher this value, the higher the risk index for predicting aortic dissection rupture obtained after processing CT images.

[0120] Specifically, this includes: when the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within a preset first threshold range, the risk level is judged to be low, and regular monitoring is required.

[0121] It should be noted that the preset first threshold range is set according to specific circumstances, and is preferably [0, 0.4) here. A value in the range [0, 0.4) indicates a low risk.

[0122] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within the preset second threshold range, the risk level is judged to be moderate, and further evaluation is required.

[0123] It should be noted that the preset second threshold range is set according to specific circumstances, and is preferably [0.4, 0.6] here.

[0124] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture falls within the preset third threshold range, the risk level is considered high, and preventive measures need to be taken.

[0125] It should be noted that the preset third threshold range is set according to specific circumstances, and is preferably [0.6, 0.8] here.

[0126] When the normalized risk prediction coefficient of a noise cavity region at risk of rupture falls within the preset fourth threshold range, the risk level is determined to be extremely high, requiring emergency handling and intervention.

[0127] It should be noted that the preset fourth threshold range is set according to the specific situation, and is preferably [0.8,1] here.

[0128] This classification only applies to risk prediction coefficients, is not a diagnostic result, and has no medical effect.

[0129] One embodiment of the present invention provides an intelligent identification system for the risk of aortic dissection rupture based on semantic segmentation. The system includes the following modules:

[0130] The acquisition module is used to acquire CT images of the same patient at different time points; one time point corresponds to one CT image.

[0131] The analysis module is used to perform semantic segmentation on CT images and obtain multiple connected components in the aortic region of the CT images; wherein, the multiple connected components include at least the true lumen connected component, the false lumen connected component, and the lumen wall connected component;

[0132] The cavity wall morphological irregularity of the cavity wall connected domain is obtained by using the Hough circle detection results of the cavity wall connected domain, as well as the edge pixels and geometric center of the cavity wall connected domain.

[0133] The noise cavity region was determined based on the morphological irregularity of the cavity wall;

[0134] By utilizing the true and false connected components of the noise cavity region, the non-broken prediction coefficients of the noise cavity region are obtained;

[0135] The risk trend index of the noise cavity region is determined based on the non-rupture prediction coefficient of the noise cavity region.

[0136] Based on the risk trend index of the noise cavity region, the noise cavity regions with a risk of rupture are identified;

[0137] Based on the risk trend index of the noise cavity region with rupture risk and the morphological irregularity of the cavity wall, the risk prediction coefficient of the noise cavity region with rupture risk is obtained.

[0138] The assessment module is used to obtain the risk level assessment result based on the risk prediction coefficient of the noise cavity region with the risk of rupture.

[0139] In summary, in this embodiment of the invention, by integrating semantic segmentation technology with dynamic feature analysis of aortic dissection, specific morphological manifestations related to the risk of rupture in CT images can be accurately identified, thereby overcoming the false positive or false negative misjudgments caused by traditional methods being detached from clinical practice, and achieving intelligent and accurate identification of the risk of aortic dissection rupture.

[0140] Traditional identification methods typically rely on static geometric parameters (such as diameter) from a single time phase and use a uniform threshold for judgment. This static, one-size-fits-all assessment approach fails to reflect individual anatomical and pathological differences. In contrast, this invention introduces multi-temporal image analysis. By comprehensively analyzing the morphological changes of different regions of the aorta at different time points, it extracts their dynamic evolution characteristics and determines the risk based on the temporal visual performance of the vessel wall, thereby obtaining an individualized rupture risk prediction coefficient.

[0141] In the risk assessment process, this invention focuses on the typical imaging features of the vessel wall before rupture, including the irregularity of the local vessel wall morphology and the related ratio of the true lumen to the false lumen area. By quantifying these features and their trends over time, the reliability and practicality of the analysis results can be effectively improved.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent identification of aortic dissection rupture risk based on semantic segmentation, characterized in that, The method includes the following steps: CT images of the same patient were acquired at different time points; one time point corresponds to one CT image. Semantic segmentation is performed on CT images to obtain multiple connected domains in the aortic region of the CT images; wherein the multiple connected domains include at least a true lumen connected domain, a false lumen connected domain, and a lumen wall connected domain; The cavity wall morphological irregularity of the cavity wall connected domain is obtained by using the Hough circle detection results of the cavity wall connected domain, as well as the edge pixels and geometric center of the cavity wall connected domain. The noise cavity region was determined based on the morphological irregularity of the cavity wall; By utilizing the true and false connected components of the noise cavity region, the non-broken prediction coefficients of the noise cavity region are obtained; The risk trend index of the noise cavity region is determined based on the non-rupture prediction coefficient of the noise cavity region. Based on the risk trend index of the noise cavity region, the noise cavity regions with a risk of rupture are identified; Based on the risk trend index of the noise cavity region with rupture risk and the morphological irregularity of the cavity wall, the risk prediction coefficient of the noise cavity region with rupture risk is obtained. Based on the risk prediction coefficient of the noise cavity region at risk of rupture, the risk level assessment result is obtained.

2. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for obtaining the morphological irregularity of the cavity wall connected region by utilizing the Hough circle detection results of the cavity wall connected region, as well as the edge pixels and geometric center of the cavity wall connected region, are as follows: For each edge pixel in the cavity wall connected domain, calculate the absolute value of the difference between the x-coordinate of the edge pixel and the x-coordinate of the geometric center of the cavity wall connected domain, as well as the absolute value of the difference between the y-coordinate of the edge pixel and the y-coordinate of the geometric center of the cavity wall connected domain. Then sum the two absolute values ​​to obtain the difference between each edge pixel and the geometric center of the cavity wall connected domain. The difference between all edge pixels of the cavity wall connected region and the geometric center is summed to obtain the difference between the cavity wall connected region and the geometric center. The ratio of the side length of the cavity wall connected domain to the Hough circle detection result of the cavity wall connected domain is calculated, and the ratio is multiplied by the difference between the cavity wall connected domain and the geometric center to obtain the cavity wall morphological irregularity of the cavity wall connected domain.

3. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for determining the noise cavity region based on the morphological irregularity of the cavity wall are as follows: When the morphological irregularity of the cavity wall after normalization of the cavity wall connected domain is greater than or equal to the preset irregularity threshold, the cavity wall connected domain is determined to be a noisy cavity region.

4. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for obtaining the non-broken prediction coefficients of the noise cavity region by utilizing the true and false connected components of the noise cavity region are as follows: For each noise cavity region, calculate the ratio of the area of ​​the true cavity connected region to the area of ​​the false cavity connected region, as well as the mean of the Hough circle detection results of the true cavity connected region and the false cavity connected region. Then, determine the non-breaking prediction coefficient of each noise cavity region by multiplying the ratio and the mean.

5. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for determining the risk trend index of the noise cavity region based on the non-fracture prediction coefficient of the noise cavity region are as follows: Obtain the unbroken prediction coefficients of each noise cavity region at different time points, and construct the unbroken prediction coefficient sequence of each noise cavity region in the order of time points; For the unbroken prediction coefficient sequence of each noise cavity region, calculate the ratio of two adjacent elements, sum the ratios of all two adjacent elements, divide the sum by the first quantity, and obtain the risk trend index of each noise cavity region. The first quantity is the number of elements minus 1.

6. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for determining the noise cavity region at risk of rupture based on the risk trend index of the noise cavity region are as follows: When the risk trend index of each noise cavity region is greater than the preset index threshold, it is determined that there is a risk of rupture in the noise cavity region.

7. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 6, characterized in that, Also includes: When the risk trend index of each noise cavity region is less than or equal to the preset index threshold, it is determined that there is no risk of rupture in the noise cavity region.

8. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for obtaining the risk prediction coefficient for a noise cavity region at risk of rupture, based on the risk trend index of the region and the morphological irregularity of the cavity wall, are as follows: For a noise cavity region at risk of rupture, the mean value of the cavity wall morphological irregularity at different time points is calculated, and the product of the mean value and the risk trend index of the noise cavity region is determined as the risk prediction coefficient of the noise cavity region.

9. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 1, characterized in that, The specific steps for obtaining the risk level assessment result based on the risk prediction coefficient of the noise cavity region with a risk of rupture are as follows: When the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within the preset first threshold range, the risk level is judged to be low and regular monitoring is required. When the normalized risk prediction coefficient of the noise cavity region at risk of rupture is within the preset second threshold range, the risk level is judged to be medium and further evaluation is required. When the normalized risk prediction coefficient of a noise cavity region at risk of rupture is within the preset third threshold range, the risk level is judged to be high and preventive measures need to be taken. When the normalized risk prediction coefficient of a noise cavity region at risk of rupture falls within the preset fourth threshold range, the risk level is determined to be extremely high, requiring emergency handling and intervention.

10. The intelligent identification method for aortic dissection rupture risk based on semantic segmentation according to claim 3, characterized in that, Also includes: When the morphological irregularity of the cavity wall after normalization is less than the preset irregularity threshold, the cavity wall connected region is determined to be a non-noise cavity region.

Images