Artificial intelligence-based method for measuring aneurysms in cerebral angiography images

Abstract

This invention discloses an artificial intelligence-based method for measuring aneurysms in cranial arterial angiography images, comprising the following steps: S1, acquiring keyframes of DSA images; S2, segmenting the keyframes using a deep neural network to distinguish between the background and blood vessels; S3, determining the aneurysm location region based on the segmentation results, obtaining the contour of the region, and simultaneously obtaining the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w; S4, based on the obtained aneurysm contour shape, calculating the aneurysm centroid O by spatial moments, drawing a perpendicular line through the centroid O and intersecting the contour boundary, traversing all angles, and taking the largest intersection line segment as the maximum diameter l of the aneurysm, drawing a perpendicular line from the aneurysm diameter width w and intersecting the contour boundary, the maximum height of which is the aneurysm height h. This invention achieves fully automated measurement of the diameter of cerebral vascular aneurysms through a deep neural network trained on a large amount of data, solving the problems of slow speed, poor accuracy, and large influence of subjective factors in manual measurement.

Description

Technical Field

[0001] This invention relates to the field of medical imaging, and more particularly to a method for measuring aneurysms in cranial angiography images based on artificial intelligence. Background Technology

[0002] Intracranial aneurysms are localized abnormal enlargements of cerebral arteries, causing aneurysmal protrusions into the arterial wall. Intracranial aneurysms often result from congenital defects in the local arterial wall and increased intraluminal pressure, leading to cystic bulging and are the leading cause of subarachnoid hemorrhage. Statistics show that the incidence of intracranial aneurysms in my country is approximately 7%. When an intracranial aneurysm ruptures and bleeds, it causes brain damage, potentially leading to a series of sequelae such as hemiplegia, slowed reflexes, and aphasia. In a small number of patients with acute onset and critical conditions, even with timely treatment, death may occur during the acute phase, or the patient may survive but still suffer severe disability.

[0003] In the treatment of intracranial aneurysms, indicators such as aneurysm diameter, aneurysm width, maximum aneurysm diameter, and aneurysm height are of significant clinical importance, serving as crucial criteria for diagnosing giant aneurysms and formulating surgical strategies. High-quality three-dimensional reconstruction of six vessels in cerebral angiography (DSA) remains the gold standard for diagnosis. DSA not only determines the location and anatomical structure of the aneurysm but also provides valuable information such as adjacent branches, collateral circulation, extracranial circulation (for surgical planning), and distal cerebral perfusion.

[0004] In the prior art, CN 109389637B discloses a method and system for measuring morphological parameters of intracranial aneurysm images. This method solves the problems of automated measurement and inconsistent measurement of morphological parameters in intracranial aneurysm images. The method includes: acquiring the centerline of the intracranial aneurysm-bearing vessel, segmented intracranial aneurysm images, and images of the intracranial aneurysm-bearing vessel; generating the intracranial aneurysm surface using the segmented intracranial aneurysm images and calculating the aneurysm neck center; and measuring the morphological parameters of the intracranial aneurysm image. However, this method requires a large amount of prior information, such as the centerline and radius of the intracranial vessel. Obtaining this information still relies on experienced personnel manually measuring using relevant software, resulting in slow measurement speed and unsatisfactory accuracy. Furthermore, the subjective factors of the measuring personnel have a significant impact on the measurement results. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes an artificial intelligence-based method for measuring aneurysms in cranial arterial angiography images. This method enables fully automated measurement of the diameter of cerebral aneurysms, solving the problems of slow speed, poor accuracy, and significant subjective influence in manual measurement.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a method for measuring aneurysms in cranial angiography images based on artificial intelligence, comprising the following steps:

[0008] S1. Acquire keyframes of DSA image;

[0009] S2. Use a deep network to segment keyframes and distinguish between background and blood vessels. The deep network uses a convolutional neural network (CNN). The entire neural network is composed of several shortcut connection units. The input feature of each shortcut connection unit is x, and the output is represented as F(x). Finally, F(x) and x are added together to obtain H(x). Between two adjacent weight layers, there are convolutional layers, batch normalization layers, and activation function layers.

[0010] S3. Determine the location of the aneurysm based on the segmentation results, obtain the outline of the region, and at the same time obtain the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w.

[0011] S4. Based on the obtained contour shape of the aneurysm, the centroid O of the aneurysm is obtained by calculating the spatial moments. A perpendicular line is drawn through the centroid O and intersects the boundary of the contour. After traversing all angles, the maximum intersection line segment is taken as the maximum diameter l of the aneurysm. A perpendicular line is drawn to the width w of the aneurysm and intersects the boundary of the contour. The maximum height is the height h of the aneurysm.

[0012] Furthermore, the quick connection process of the quick connection unit is represented as follows:

[0013] y=F(x,{w i})+x

[0014] Where y represents the output feature, x represents the input feature, and F(x,{w i}) represents the residual mapping function that needs to be trained, w i This indicates the weight of the layer.

[0015] Furthermore, the operation process of the batch normalization layer is as follows:

[0016] Subtracting the mean from the input features and dividing by its variance, can be expressed as:

[0017]

[0018] in Let E[x] represent the normalized features, x represent the input features, and E[x] represent the normalized features. (k) [] represents the mean of the input features. Represents the variance of the input features;

[0019] The normalized features are translated and scaled, as follows:

[0020]

[0021] Where γ (k) β is a learnable scaling parameter. (k) These are learnable translation parameters.

[0022] Furthermore, the activation function layer process can be represented as:

[0023] Y = G(x)

[0024] Where y is the output feature, x is the input feature, and G is the activation function.

[0025] On the other hand, the present invention also provides an artificial intelligence-based measurement system for aneurysms in cranial angiography images, comprising the following modules to implement the method described in any of the above:

[0026] The data processing module is used to acquire DSA image data, convert it into keyframes, and send them to the deep network segmentation module while storing them in memory;

[0027] The deep network segmentation module is used to acquire keyframes of DSA images and segment them to distinguish between the background and blood vessels;

[0028] The aneurysm region contour determination module is used to obtain the aneurysm region contour and simultaneously obtain the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w.

[0029] The measurement index calculation module is used to obtain the centroid O of the aneurysm region contour, the maximum diameter l of the aneurysm, and the height h of the aneurysm.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] This invention proposes an artificial intelligence-based method for measuring aneurysms in intracranial angiography images. This fully automated method utilizes a deep neural network trained on a large dataset to simultaneously distinguish between vascular and aneurysm regions, greatly simplifying subsequent measurements. The method calculates the diameter of the corresponding intracranial aneurysm from the angiography image. The results are more objective, requiring no human intervention and avoiding measurement errors caused by subjectivity. It also avoids discrepancies between different doctors and hospitals due to varying interpretation methods. The method is simple and easy to use, allowing physicians to perform coronary angiography procedures. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0033] Figure 1 A flowchart illustrating the method for measuring aneurysms in cranial arteriography images based on artificial intelligence, as provided in an embodiment of the present invention.

[0034] Figure 2 This is a comparison chart of the original data and segmentation results provided in an embodiment of the present invention. The left side of the chart shows the original data, and the right side shows the segmentation results. The black area represents the background area that is not of interest, the white area represents the normal intracranial blood vessel area, and the gray area represents the diseased intracranial hemangioma area.

[0035] Figure 3 This is a schematic diagram of the aneurysm outline provided in an embodiment of the present invention.

[0036] Figure 4 This is a schematic diagram of a quick connection unit provided in an embodiment of the present invention. Detailed Implementation

[0037] To better understand this technical solution, the technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described examples are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of the present invention.

[0038] This invention proposes an artificial intelligence-based method for measuring aneurysms in cranial angiography images, such as... Figure 1 As shown, it includes the following steps:

[0039] 1. Obtain appropriate DSA-affected keyframes;

[0040] 2. Deep network segmentation of keyframes to distinguish between background and blood vessels;

[0041] 3. Locate the aneurysm location area based on the segmentation results;

[0042] 4. Determine the centroid of the aneurysm and determine the various measurement indicators.

[0043] The following is a detailed explanation of each step.

[0044] The first step is to acquire DSA image data through the data processing module, convert it into a data stream that can be processed by subsequent modules, store it in memory, and send it to the deep network segmentation module.

[0045] The second step involves the deep network segmenting the keyframes of the acquired DSA images, distinguishing between the blood vessels and the background. A comparison of the original data and the segmentation results is shown in the image below. Figure 2 As shown. The black areas represent background areas that are not of interest, the white areas represent normal intracranial blood vessel areas, and the gray areas represent lesions such as intracranial hemangiomas.

[0046] The third step involves determining the aneurysm region based on the segmentation results, obtaining the outline of that region as shown in the image. Figure 3 As shown, while obtaining the outline, the length of the connection between the gray tumor area and the white blood vessel area can also be obtained, that is, the tumor diameter width w.

[0047] The fourth step is to obtain the centroid O of the aneurysm by calculating the spatial moments based on the obtained contour shape. Draw a perpendicular line through the centroid to intersect the contour boundary. After traversing all angles, take the largest intersection line segment as the maximum diameter l of the aneurysm. Draw a perpendicular line to the width of the aneurysm to intersect the contour. The maximum height is the height h of the aneurysm.

[0048] The above steps ultimately yield various measurement indicators for the aneurysm.

[0049] The following is a detailed introduction to deep learning networks.

[0050] 1. Training dataset

[0051] To achieve effective segmentation of intracranial blood vessels, we annotated over 100 cases of intracranial arterial angiography DSA data. A team of professional physicians marked the locations of blood vessels and aneurysms on the images, and further used different color markers to perform pixel-level fine annotation of the data. We used these data to train a network model, and then used the trained model to segment and identify blood vessels.

[0052] 2. Convolutional Networks

[0053] like Figure 4 The diagram shows a basic unit of a CNN called a shortcut connection, where X represents the input features, the output is F(X), and F(X) and X are summed. Between two adjacent weight layers are convolutional layers, batch normalization layers, and activation function layers. The entire neural network is composed of several such shortcut connection units connected together.

[0054] After receiving the DSA angiography image, the CNN module begins segmentation and recognition of blood vessels. First, the image passes through convolutional layers, where repeated convolution operations are performed on the input image to extract high-dimensional features. These features contain all the information used in the segmentation and recognition process.

[0055] 3. Residual Module

[0056] The deeper the neural network, the higher the dimensionality of the features it can learn; therefore, the number of layers has a significant impact on the neural network. However, as the number of layers in a neural network becomes increasingly deep, deeper models struggle to represent low-dimensional features, leading to problems such as gradient explosion and gradient vanishing. Shortcut connection units are a method to solve this problem. Let H(x) be the denoting expression.

[0057] = F(x) + x. In the extreme case, F(x) learns nothing, i.e., F(x) = 0, and H(x) = x. This ensures that shallow features are propagated backward, and the features learned by the entire network will not be too bad, thus solving the problems of gradient explosion and gradient vanishing.

[0058] The entire quick connection process can be represented as:

[0059] y=F(x,{w i})+x

[0060] Where y represents the output feature, x represents the input feature, and F(x,{w i}) represents the residual mapping function we need to train, w i This indicates the weight of the layer.

[0061] In the process of feature extraction from coronary angiography images, the problems mentioned above are encountered. As the number of layers in our designed neural network increases, the network struggles to represent low-dimensional features in the final output, leading to gradient explosion and vanishing gradient problems. Therefore, we use shortcut connection units in the task of extracting features from coronary angiography images. This allows the model to decide the dimensionality of the features it wants to extract, thus preserving as many useful low-dimensional intracranial vascular features as possible.

[0062] 4. Batch standardization layer

[0063] Batch normalization layers are a method to accelerate model training and significantly reduce time costs. Batch normalization normalizes features to a suitable distribution, accelerating network convergence. The specific operation is as follows:

[0064] The first step is to normalize the input features by subtracting their mean and then dividing by their variance. This process can be represented as follows:

[0065]

[0066] in Let E[x] represent the normalized features, x represent the input features, and E[x] represent the normalized features. (k) [] represents the mean of the input features. This represents the variance of the input features.

[0067] The second step is to translate and scale the normalized features, with the aim of allowing the network to learn the appropriate output. This process can be represented as follows:

[0068]

[0069] Where γ (k) β is a learnable scaling parameter. (k) These are learnable translation parameters.

[0070] 5. Activation function layer

[0071] Simply connecting these linear convolutional networks will produce the same result as a single convolutional unit. Therefore, in practical applications, activation function layers need to be introduced, such as... Figure 4 This is the graph of the activation function, and the specific process can be represented as follows:

[0072] Y = G(x)

[0073] Where y is the output feature, x is the input feature, and G is the activation function.

[0074] Corresponding to the above methods, the present invention also provides an artificial intelligence-based measurement system for aneurysms in cranial angiography images, comprising the following modules to implement the methods described in any of the above embodiments:

[0075] The data processing module is used to acquire DSA image data, convert it into keyframes, and send them to the deep network segmentation module while storing them in memory;

[0076] The deep network segmentation module is used to acquire keyframes of DSA images and segment them to distinguish between the background and blood vessels;

[0077] The aneurysm region contour determination module is used to obtain the aneurysm region contour and simultaneously obtain the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w.

[0078] The measurement index calculation module is used to obtain the centroid O of the aneurysm region contour, the maximum diameter l of the aneurysm, and the height h of the aneurysm.

[0079] By acquiring DSA images during coronary angiography and inputting them into the system, the system calculates and provides the measured diameter of the intracranial aneurysm corresponding to the angiographic image. This method yields more objective results and avoids discrepancies between different doctors and hospitals due to different interpretation methods. The measurement process requires no manual intervention, and the operation is simple and easy to implement, making it suitable for doctors to use during coronary angiography.

[0080] Corresponding to the system provided in the above embodiments of the present invention, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0081] Memory, used to store computer programs;

[0082] The processor, when executing a program stored in memory, implements the system flow provided in the above embodiments of the present invention.

[0083] The communication bus mentioned in the control equipment above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0084] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0085] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0086] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments, electronic device embodiments, computer-readable storage medium embodiments, and computer program product embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method of measuring an aneurysm in a cerebral arteriography image based on artificial intelligence, characterized by, Includes the following steps: S1. Acquire keyframes of DSA image; S2. Keyframes are segmented using a deep network to distinguish between the background and blood vessels; the black area represents the background region that is not of interest, the white area represents the normal intracranial blood vessel region, and the gray area represents the diseased intracranial hemangioma region; the deep network uses a convolutional neural network (CNN), and the entire neural network is composed of several shortcut connection units connected together. The input feature of each shortcut connection unit is x, and the output is represented as F(x). Finally, F(x) and x are added together to obtain H(x); between two adjacent weight layers, there are convolutional layers, batch normalization layers, and activation function layers; The quick connection process of the quick connection unit is represented as follows: ， where y represents the output feature, x represents the input feature, F(x, {w i}) represents the residual mapping function to be trained, and w i represents the weight of the layer; The batch normalization layer operation process is as follows: Subtracting the mean from the input features and dividing by its variance, can be expressed as: ， wherein denotes the normalized feature, x denotes the input feature, denotes the mean of the input feature, denotes the variance of the input feature; The normalized features are translated and scaled, as follows: ， wherein is a learnable scaling parameter, is a learnable translation parameter; The activation function layer process can be represented as: Y = G(x), Where y is the output feature, x is the input feature, and G is the activation function; S3. Determine the location of the aneurysm based on the segmentation results, obtain the outline of the region, and at the same time obtain the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w. S4. Based on the obtained contour shape of the aneurysm, the centroid O of the aneurysm is obtained by calculating the spatial moments. A perpendicular line is drawn through the centroid O and intersects the boundary of the contour. After traversing all angles, the maximum intersection line segment is taken as the maximum diameter l of the aneurysm. A perpendicular line is drawn to the width w of the aneurysm and intersects the boundary of the contour. The maximum height is the height h of the aneurysm.

2. An artificial intelligence-based aneurysm measuring system in a cerebral arteriography image, characterized by, The following modules are included to implement the method of claim 1: The data processing module is used to acquire DSA image data, convert it into keyframes, and send them to the deep network segmentation module while storing them in memory; The deep network segmentation module is used to acquire keyframes of DSA images and segment them to distinguish between the background and blood vessels; The aneurysm region contour determination module is used to obtain the aneurysm region contour and simultaneously obtain the length of the connection between the gray aneurysm region and the white blood vessel region, i.e., the aneurysm diameter width w. The measurement index calculation module is used to obtain the centroid O of the aneurysm region contour, the maximum diameter l of the aneurysm, and the height h of the aneurysm.

Images