[go: up one dir, main page]

CN114170143A - Method for aneurysm detection and rupture risk prediction in digital subtraction angiography - Google Patents

Method for aneurysm detection and rupture risk prediction in digital subtraction angiography Download PDF

Info

Publication number
CN114170143A
CN114170143A CN202111332597.6A CN202111332597A CN114170143A CN 114170143 A CN114170143 A CN 114170143A CN 202111332597 A CN202111332597 A CN 202111332597A CN 114170143 A CN114170143 A CN 114170143A
Authority
CN
China
Prior art keywords
aneurysm
image
detection
filter
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111332597.6A
Other languages
Chinese (zh)
Inventor
余锦华
胡涛
雷宇
顾宇翔
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fudan University
Original Assignee
Fudan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fudan University filed Critical Fudan University
Priority to CN202111332597.6A priority Critical patent/CN114170143A/en
Publication of CN114170143A publication Critical patent/CN114170143A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular
    • G06T2207/30104—Vascular flow; Blood flow; Perfusion

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

本发明涉及一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，包括步骤：1)从DSA影像上自动提取出一帧图像；2)根据黑塞矩阵特征值构建多尺度动脉瘤检测滤波器；3)采用参数优化后的滤波器对图像上的动脉瘤进行增强，根据滤波器响应强度的均值和动脉瘤的形状特点判别动脉瘤；4)提取动脉瘤的血流灌注特征、纹理特征和强度特征；5)采用迭代稀疏表示对未破裂和破裂动脉瘤的特征进行特征筛选，得到鉴别性更高的特征，再采用稀疏表示进行分类决策。与现有技术相比，本发明不仅能够准确检测动脉瘤，而且无需人工设置滤波器检测参数，也能对动脉瘤的破裂风险获得较高的预测精度，具有较强鲁棒性，对动脉瘤的精确诊断和破裂预测提供了参考。

Figure 202111332597

The invention relates to a method for aneurysm detection and rupture risk prediction in digital subtraction angiography, comprising the steps of: 1) automatically extracting a frame of image from a DSA image; 2) constructing a multi-scale aneurysm according to Hessian matrix eigenvalues detection filter; 3) the aneurysm on the image is enhanced by the filter after parameter optimization, and the aneurysm is identified according to the mean value of the filter response intensity and the shape characteristics of the aneurysm; 4) the blood perfusion characteristics of the aneurysm are extracted, Texture features and intensity features; 5) Iterative sparse representation is used to screen the features of unruptured and ruptured aneurysms to obtain more discriminative features, and then sparse representation is used to make classification decisions. Compared with the prior art, the present invention can not only accurately detect the aneurysm, but also can obtain higher prediction accuracy for the rupture risk of the aneurysm without manually setting filter detection parameters, and has strong robustness. provide a reference for accurate diagnosis and rupture prediction.

Figure 202111332597

Description

Method for aneurysm detection and rupture risk prediction in digital subtraction angiography
Technical Field
The invention relates to the technical field of computer-aided diagnosis, in particular to an aneurysm detection and rupture risk prediction method in Digital Subtraction Angiography (DSA).
Background
Intracranial aneurysms are cerebrovascular diseases that are a serious threat to the life and health of a patient and usually occur around arteries in the bottom of the brain. In chinese Magnetic Resonance Angiography (MRA) screening, about 7% of adults between 35 and 75 years of age have aneurysms. Rupture of an intracranial aneurysm causes Subarachnoid Hemorrhage (SAH), with high morbidity and mortality. Although aneurysm rupture is a rare event, early detection is critical to avoid aneurysm rupture.
DSA is the "gold standard" for diagnosing aneurysms, which provides higher image resolution and higher sensitivity for detection of microaneurysms. The treatment of aneurysms is also a matter of controversy, and whether they rupture is an important factor in determining the performance of the operation, because rupture of an aneurysm during the operation also poses a life risk to the patient, and therefore, it is necessary to predict the risk of rupture of an intracranial aneurysm. Many studies use statistical methods to analyze risk factors of aneurysm rupture, including shape, size, location, etc. of the aneurysm, but the specific cause of its rupture remains unclear.
In current studies, aneurysm detection is rarely performed directly on DSA, but the sensitivity of detection on other modalities is also not high. Most aneurysm rupture prediction methods extract some features according to the shape, size and position of the aneurysm, and then establish a rupture risk prediction function, or establish a classification prediction model by using a Machine learning method to find out several important factors related to rupture risk. However, these rupture prediction methods are not complete in feature extraction of the aneurysm, and the accuracy of prediction is not high, and further, artificial extraction of the aneurysm is required.
Disclosure of Invention
The present invention is directed to overcoming the above-mentioned drawbacks of the prior art and providing a method for detecting an aneurysm and predicting the risk of rupture in digital subtraction angiography.
The purpose of the invention can be realized by the following technical scheme:
a method for aneurysm detection and risk of rupture prediction in digital subtraction angiography, comprising the steps of:
1) automatically extracting a frame of image from the DSA image, and performing noise removal and normalization pretreatment on the extracted image;
2) constructing a multi-scale aneurysm detection filter according to the response difference of the image blackplug matrix characteristic value to different structures, and automatically searching the detection parameters of the filter by adopting Bayesian optimization;
3) the method comprises the steps of enhancing the aneurysm on an image by using a filter after parameter optimization, judging the aneurysm according to the mean value of filter response intensity and the shape characteristic of the aneurysm, removing the detected aneurysm by using a region growing method, and performing cyclic detection on the image from which the aneurysm is removed;
4) extracting blood flow perfusion characteristics, texture characteristics and intensity characteristics of the detected aneurysm;
5) and (3) performing feature screening on the features of the unbroken and broken aneurysms by adopting iterative sparse representation to obtain features with higher discriminativity, and then performing classification decision by adopting sparse representation.
Further, the step 1) specifically comprises the following steps:
11) extracting images from the artery phase and the capillary phase of the DSA, removing image noise by adopting Gaussian filtering, enhancing blood vessels on the image by adopting a Frangi filter, obtaining a binary blood vessel image by adopting an Otsu threshold segmentation method, and corroding the binary image by utilizing a morphological corrosion method to remove small spots in the binary image;
12) marking a connected region of the image, thinning the main blood vessel by taking the maximum connected region of the binary image as the main blood vessel to obtain a blood vessel center line, eliminating blood vessel branches which are perpendicular to the trend of the blood vessel and have pixel points less than 10, and calculating the length of the blood vessel center line of all image frames;
13) in the arterial phase, the length of the center line of the blood vessel is gradually increased, when the arterial phase is nearly finished, the length of the center line reaches the first maximum value, the image at the moment is extracted to detect the aneurysm, and normalization processing is carried out on the extracted image.
Further, in step 2), the multi-scale aneurysm detection filter is composed of eigenvalues of an image blackplug matrix, and the blackplug matrix is defined as a convolution of a pixel gray value and a gaussian function derivative, and then:
Figure BDA0003349451870000031
wherein, H is a black plug matrix, and I (x) is a coordinate point x ═ x in the two-dimensional image1,x2]TG (x, s) is a Gaussian function, and
Figure BDA0003349451870000032
s represents a scale parameter and x represents a convolution.
Further, in the step 2), the aneurysm detection filter BpThe expression of (a) is:
Figure BDA0003349451870000033
Figure BDA0003349451870000034
Figure BDA0003349451870000035
wherein λ is2(x, s) denotes. lambda.2(x) The value at the s-scale, τ being a parameter determining the strength of the filter response, λ1、λ2Two eigenvalues of a black plug matrix of pixels in the image, and lambda1|≤|λ2|，B1And λρIs an intermediate parameter.
Further, in the step 2), a maximum response value of the filter is obtained by comparing a characteristic value of each point x in the image under the s scale, and the detection parameters of the filter specifically include:
Figure BDA0003349451870000036
wherein A isrAnd P is the area and perimeter, V, of the measured object, respectivelymeanFor the average value of the filter response intensity, the larger the value of the detection parameter F, the more likely it is an aneurysm.
Further, in the step 2), in the aneurysm detection process, a bayesian optimization method is adopted to automatically find two unknown detection parameters τ and s of the filter, the target is to search the maximum value of the detection parameter F, the corresponding parameter is the detection parameter of the filter, and in order to obtain the maximum value of the detection parameter F, the filter parameter is adjusted and the F value corresponding to each group of parameters is calculated for comparison.
The optimal parameter set can be automatically and quickly found by adopting Bayesian optimization without manually selecting or setting any parameter, and after the filter parameter is found, the filter parameter is substituted into the filter for aneurysm detection.
Further, in the step 4), the aneurysms of the five continuous frames of images, including ruptured aneurysms and unbroken aneurysms, are extracted, and corresponding blood perfusion characteristics, intensity characteristics and texture characteristics are respectively extracted. .
Further, the extraction of the blood perfusion characteristics of the aneurysm is specifically as follows:
selecting three different positions on the image, namely the interior of the aneurysm, the edge of the aneurysm and the exterior of the aneurysm, selecting an interested region with the size of 5 x 5 at each position, drawing a time-density curve of the interested region at each position, and extracting blood flow perfusion characteristics from the time-density curve.
Further, the extraction of the intensity features and texture features of the aneurysm is specifically as follows:
and extracting intensity and texture features according to the difference of the aneurysm gray level and texture on the image, wherein the intensity features describe the statistical distribution of voxel intensity in the image, and the texture features describe the texture difference of the image respectively based on a gray level co-occurrence matrix, a gray level run-length matrix, a gray level size area matrix and a neighborhood gray level difference matrix.
Further, in the step 5), a method of iterative sparse representation is adopted to select features closely related to sample labels, specifically, an OMP algorithm is adopted to solve an optimization problem, a sliding window strategy is adopted in the sparse representation method, and information of all samples in a window is utilized, so that an iterative process specifically includes the following steps:
51) calculating coefficients after the kth iteration
Figure BDA0003349451870000041
Then there are:
Figure BDA0003349451870000042
wherein, gkAs a standard value for the kth iteration, FkThe sample characteristic of the kth iteration is sigma, which is a small constant;
52) calculating the coefficient after the k iteration
Figure BDA0003349451870000043
Average value of (2)
Figure BDA0003349451870000044
When the condition of M is satisfied(k)-M(k-1)||2< ε or K ═ K0Then the iteration stops, where ε is a small positive integer, K0Is the maximum iteration number;
53) coefficient M obtained by iteration(k)The method is used for feature selection, each feature corresponds to a score through iterative operation, the higher the score is, the more important the feature is, the final scores are sorted, a set number of features are selected for classification, the selected features are classified by adopting a sparse representation method, and then a sparse representation classification model is as follows:
Figure BDA0003349451870000045
wherein F represents the characteristics of the aneurysm to be tested, and F ═ FR,FU]，FR＝[f1,f2,…,fn1]For the feature set of the ruptured aneurysm in the training sample, n is the number of ruptured aneurysms, FU＝[f1,f2,…,fn2]To train the feature set for the unbroken aneurysm in the sample, n2 is the number of unbroken aneurysms, p is a sparsely represented control parameter,
Figure BDA0003349451870000046
for sparse representation of coefficients, when obtaining optimal sparse representation coefficients
Figure BDA0003349451870000047
According to the residual rα(f) The class to which the feature belongs is judged,then the residual error rα(f) The expression of (a) is:
Figure BDA0003349451870000048
wherein, deltaα(.) represent coefficients corresponding to the selected feature classes.
Compared with the prior art, the invention has the following advantages:
according to the characteristic that the intracranial aneurysm on the DSA image is in a similar circular shape, a multi-scale aneurysm detection filter is constructed based on the response difference situation of the characteristic value of a blackplug matrix to different object structures in the image, and the detection parameters of the filter are automatically searched by adopting a Bayesian optimization method, so that manual setting is not needed, and the manual intervention quantity is reduced.
And secondly, for the rupture risk prediction of the aneurysm, respectively extracting texture characteristics, intensity characteristics and blood flow perfusion characteristics of the aneurysm according to the morphological texture difference and the blood flow difference, wherein the characteristics can effectively reflect the difference between ruptured aneurysm and non-ruptured aneurysm.
Thirdly, the invention adopts a feature screening and classifying method based on sparse representation, can screen partial more distinctive features from the extracted features, and improves the classifying precision while reducing the calculated amount.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings used in the description will be briefly introduced, and it is obvious that the drawings in the following description are one embodiment of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a flow chart of an aneurysm detection and risk of rupture prediction method in digital subtraction angiography of the present invention.
Fig. 2 illustrates an aneurysm filter parameter extraction and aneurysm detection process in an embodiment.
Fig. 3 is a time-density plot of the interior, margins and exterior of an aneurysm in an example.
Fig. 4 is a process for predicting risk of aneurysm rupture.
Detailed Description
The aneurysm detection and rupture risk prediction method in DSA will be described in detail below with reference to the accompanying drawings and embodiments,
examples
As shown in fig. 1, the present invention provides a method for detecting an aneurysm and predicting a rupture risk in digital subtraction angiography, which comprises the following steps:
step 1, for each frame of image of DSA, starting from a first frame, firstly adopting Gaussian filtering to remove image noise, then utilizing a Frangi filter to enhance blood vessels, adopting an Otsu threshold segmentation method to obtain a binary blood vessel image, utilizing a flat disc structural element with the radius of 2 to corrode the binary image, removing small spots in the binary image, marking a connected region of the image, finding out the largest connected region as a main blood vessel, thinning the main blood vessel to obtain a center line of the blood vessel, eliminating blood vessel branches perpendicular to the trend of the blood vessel and with pixel points smaller than 10, calculating the length of the center line of the blood vessel, calculating each frame of image by using the same method, finding out a frame at the position of a first maximum value after the lengths of the center lines of all frames in the DSA image are obtained, and extracting the image of the frame to detect aneurysms;
step 2, normalizing the image with the noise removed, constructing an aneurysm detection filter based on the image blackplug matrix eigenvalue, searching filter detection parameters by a Bayesian optimization method, setting a tau value change range between 0.8 and 1 and a s value change range between 0 and 20 for two unknown parameters reflecting the target size and intensity in the filter, setting the search times as 50, namely searching and iterating 50 times in an image space, setting a Bayesian optimization acquisition function as a gain expectation, aiming at balancing between development and exploration and finding an optimal parameter value, so that tau and s corresponding to F as the maximum value are the optimal detection parameters of the filter, obtaining a group of detection parameters after the search is finished, each aneurysm corresponds to a group of optimal detection parameters, and removing the aneurysm by a region growing method after the aneurysm is detected, then repeating the previous steps in the rest images and then carrying out detection;
step 3, intercepting the detected aneurysms, wherein the aneurysms which are not ruptured form one group, the ruptured aneurysms form another group, then respectively extracting features, wherein the intensity features and the texture features obtained from each aneurysm are respectively 31 and 39, so that each frame of image has 70 features, extracting five continuous frames of images from a key frame, extracting the aneurysms of each frame of image, and then respectively extracting the features such as intensity, texture and the like, so that 350(5 × 70) features can be obtained; selecting ROI (region of interest) at three different positions of the aneurysm from DSA (digital image acquisition) images, namely the inside of the aneurysm, the edge of the aneurysm and the outside of the aneurysm, selecting an image area with the size of 5 multiplied by 5 at each position, drawing a time-density curve, obtaining blood flow perfusion information from the curve, obtaining 11 blood flow perfusion characteristics at the inside, the edge and the outside of the aneurysm, extracting 33 blood flow perfusion characteristics from each aneurysm, obtaining two characteristics 383(350+33) in total,
and 4, performing feature screening on the two extracted features by using a sparse representation method, wherein parameters during the specific screening are set as follows: maximum number of iterations K0350, the small positive number epsilon is 0.0001, the number of samples selected in each iterative operation is 5, then the importance of the features is ranked by using iterative sparse representation, the first 11 important features are selected at first, the accuracy rate is calculated by using a sparse representation classification model, then a feature is newly added in each calculation until the 80 th important feature is added, so that 70 different accuracy values are obtained in total, the feature number corresponding to the highest accuracy rate is used for final aneurysm rupture risk prediction, and 32 important features are finally screened out,
step 5, after sparse representation screening, extracting 32 important features from each aneurysm, training and testing by using a sparse representation classifier, setting control parameters to be 0.5 and residual error to be 0.001 when the sparse representation classifier is trained, solving a sparse representation classification model by using an orthogonal matching pursuit algorithm, and finally judging the class of the features according to the sparse representation residual error, wherein in the embodiment of the invention, the performance of the classifier is evaluated by using AUC (area Under classifier), accuracy, sensitivity and specificity,
the following describes a specific implementation procedure of aneurysm detection and rupture risk prediction according to this embodiment.
In the data set used in the present invention, there were 263 aneurysms in total, of which 138 non-ruptured aneurysms and 125 ruptured aneurysms, 287 were used as training set and the remaining 76 were used as test set (36 ruptured aneurysms and 40 non-ruptured aneurysms) during training, and for aneurysm rupture risk prediction, four different sets of characteristics were selected for comparative experiments:
(1) intensity features and texture features (ITF) of a single frame image;
(2) intensity, texture, and blood perfusion characteristics (ITPF) of the single frame image;
(3) intensity features and texture features (ITSF) of the five frame image;
(4) intensity features, texture features, and blood flow perfusion features (ITPSF) of the five frame images.
The present invention adopts the features of group (4), and tables 1 and 2 are the aneurysm rupture risk prediction results before and after feature selection, respectively.
TABLE 1 prediction of aneurysm rupture before feature selection
Figure BDA0003349451870000071
TABLE 2 prediction of aneurysm rupture following feature selection
Figure BDA0003349451870000072
As can be seen from tables 1 and 2, both accuracy and AUC were significantly improved after feature selection. In table 2, the method proposed by the present invention achieves the highest accuracy of 96.1%, the sensitivity also reaches 94.4%, and the AUC is 0.982. And among four different groups of characteristics, the accuracy rate is the lowest when the intensity characteristic and the texture characteristic of a single frame image are used, and the accuracy rate is only 90.8%. After time sequence information and blood flow perfusion characteristics are added, the accuracy and other evaluation indexes are improved, the accuracy and AUC of the second group in the table are higher than those of the third group, and the result shows that the perfusion characteristics can identify ruptured aneurysms and unbroken aneurysms better than the time sequence information.
In summary, the present invention provides a method for aneurysm detection and rupture risk prediction. Firstly, constructing a multi-scale aneurysm detection filter based on a blackplug matrix theory, automatically searching parameters of the filter by Bayesian optimization, and detecting the aneurysm according to the shape of the aneurysm and the response condition of the filter; then, extracting the characteristics of related texture, strength, blood perfusion and the like according to the morphological characteristics and the blood flow condition of the aneurysm; finally, feature screening and classification are carried out by using a sparse representation method, and an aneurysm rupture risk prediction model is established. Experimental results show that the framework provided by the invention can accurately detect the aneurysm and predict the rupture risk of the aneurysm, and has potential application value in neurosurgery.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1.一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，包括以下步骤：1. aneurysm detection and rupture risk prediction method in digital subtraction angiography, is characterized in that, comprises the following steps: 1)从DSA影像上自动提取出一帧图像，并对提取出的图像进行去除噪声和归一化预处理；1) Automatically extract a frame of image from the DSA image, and perform noise removal and normalization preprocessing on the extracted image; 2)根据图像黑塞矩阵特征值对不同结构的响应差异构建多尺度动脉瘤检测滤波器，并采用贝叶斯优化自动搜寻滤波器的检测参数；2) Build a multi-scale aneurysm detection filter according to the response difference of the eigenvalues of the image Hessian matrix to different structures, and use Bayesian optimization to automatically search for the detection parameters of the filter; 3)采用参数优化后的滤波器对图像上的动脉瘤进行增强，根据滤波器响应强度的均值和动脉瘤的形状特点判别动脉瘤，对于检测到的动脉瘤采用区域生长法去除，并在去除动脉瘤后的图像上循环检测；3) Use the filter after parameter optimization to enhance the aneurysm on the image, identify the aneurysm according to the mean value of the filter response intensity and the shape characteristics of the aneurysm, and remove the detected aneurysm by the regional growth method. Circulation detection on images after aneurysm; 4)截取检测到的动脉瘤，提取其血流灌注特征、纹理特征和强度特征；4) Intercept the detected aneurysm, and extract its blood perfusion feature, texture feature and intensity feature; 5)采用迭代稀疏表示对未破裂和破裂动脉瘤的特征进行特征筛选，得到鉴别性更高的特征，再采用稀疏表示进行分类决策。5) Use iterative sparse representation to screen the features of unruptured and ruptured aneurysms to obtain more discriminative features, and then use sparse representation to make classification decisions. 2.根据权利要求1所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤1)具体包括以下步骤：2. The method for aneurysm detection and rupture risk prediction in a digital subtraction angiography according to claim 1, wherein the step 1) specifically comprises the following steps: 11)从DSA的动脉期和毛细血管期提取图像，采用高斯滤波去除图像噪声，采用Frangi滤波器对影像上血管进行增强，然后采用Otsu阈值分割方法得到二值化的血管图像，利用形态学腐蚀方法对二值图像进行腐蚀处理，去除二值图像中的小斑点；11) Extract images from the arterial phase and capillary phase of DSA, use Gaussian filter to remove image noise, use Frangi filter to enhance blood vessels on the image, and then use Otsu threshold segmentation method to obtain binarized blood vessel images, using morphological erosion The method performs corrosion processing on the binary image to remove small spots in the binary image; 12)标记图像的连通区域，以二值图像的最大连通区域作为主血管，对主血管细化得到血管中心线，并消除垂直于血管走向和像素点小于10的血管分支，计算全部图像帧的血管中心线的长度；12) Mark the connected area of the image, take the largest connected area of the binary image as the main blood vessel, refine the main blood vessel to obtain the blood vessel center line, and eliminate the blood vessel branches perpendicular to the direction of the blood vessel and the pixel points less than 10, and calculate the total image frame. the length of the blood vessel centerline; 13)在动脉期，血管中心线的长度逐渐增大，在动脉期快结束时，中心线长度达到第一个极大值，提取该时刻的图像以检测动脉瘤，并对提取到的图像进行归一化处理。13) During the arterial phase, the length of the centerline of the blood vessel gradually increases. When the arterial phase is about to end, the length of the centerline reaches the first maximum value. The image at this moment is extracted to detect the aneurysm, and the extracted image is processed Normalized processing. 3.根据权利要求1所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤2)中，多尺度动脉瘤检测滤波器由图像黑塞矩阵特征值组成，黑塞矩阵定义为像素灰度值与高斯函数导数的卷积，则有：3. The method for aneurysm detection and rupture risk prediction in a digital subtraction angiography according to claim 1, characterized in that, in the step 2), the multi-scale aneurysm detection filter is determined by the image black plug. The matrix is composed of eigenvalues, and the Hessian matrix is defined as the convolution of the pixel gray value and the derivative of the Gaussian function, as follows:
Figure FDA0003349451860000021
Figure FDA0003349451860000021
 其中，H为黑塞矩阵，I(x)为二维图像中坐标点x＝[x1,x2]T的像素灰度值，G(x,s)为高斯函数，且
Figure FDA0003349451860000022
s表示尺度参数，*表示卷积。Among them, H is the Hessian matrix, I(x) is the pixel gray value of the coordinate point x=[x 1 , x 2 ] T in the two-dimensional image, G(x, s) is the Gaussian function, and
Figure FDA0003349451860000022
s represents the scale parameter, * represents convolution. 4.根据权利要求3所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤2)中，动脉瘤检测滤波器Bp的表达式为：4. The method for aneurysm detection and rupture risk prediction in a digital subtraction angiography according to claim 3, wherein in the step 2), the expression of the aneurysm detection filter B p is: :
Figure FDA0003349451860000023
Figure FDA0003349451860000023
Figure FDA0003349451860000024
Figure FDA0003349451860000024
Figure FDA0003349451860000025
Figure FDA0003349451860000025
 其中，λ2(x,s)表示λ2(x)在s尺度下的值，τ为决定滤波器响应强度的参数，λ1、λ2为图像中像素点黑塞矩阵的两个特征值，且|λ1|≤|λ2|，B1和λρ为中间参数。Among them, λ 2 (x,s) represents the value of λ 2 (x) at the s scale, τ is the parameter that determines the response strength of the filter, λ 1 , λ 2 are the two eigenvalues of the pixel Hessian matrix in the image , and |λ 1 |≤|λ 2 |, B 1 and λ ρ are intermediate parameters. 5.根据权利要求3所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤2)中，通过比较图像中每个点x在s尺度下的特征值，得到滤波器的最大响应值，所述的滤波器的检测参数具体为：5. A method for aneurysm detection and rupture risk prediction in digital subtraction angiography according to claim 3, characterized in that, in the step 2), by comparing each point x in the image at the s scale The eigenvalues below are obtained to obtain the maximum response value of the filter. The detection parameters of the filter are specifically:
Figure FDA0003349451860000026
Figure FDA0003349451860000026
 其中，Ar和P分别为被测目标的面积和周长，Vmean为滤波响应强度的平均值，检测参数F的值越大，则表示越可能是动脉瘤。Among them, A r and P are the area and perimeter of the measured target, respectively, and V mean is the average value of the filter response intensity. The larger the value of the detection parameter F, the more likely it is an aneurysm. 6.根据权利要求4所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤2)中，在动脉瘤检测过程中，采用贝叶斯优化方法自动寻找滤波器的两个检测参数τ和尺度s，目标为搜索检测参数F的最大值，其相对应的参数为滤波器的检测参数，为获得检测参数F的最大值，通过调整滤波器参数并计算每组参数对应的F值进行比较。6 . The method for aneurysm detection and rupture risk prediction in digital subtraction angiography according to claim 4 , wherein in the step 2), in the aneurysm detection process, a Bayesian method is adopted. 7 . The optimization method automatically finds the two detection parameters τ and scale s of the filter. The goal is to search for the maximum value of the detection parameter F, and the corresponding parameter is the detection parameter of the filter. In order to obtain the maximum value of the detection parameter F, the filter is adjusted by adjusting the filter. and calculate the corresponding F value of each group of parameters for comparison. 7.根据权利要求1所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤4)中，提取连续五帧图像的动脉瘤，包括破裂的动脉瘤和未破裂的动脉瘤，分别提取对应的血流灌注特征、强度特征和纹理特征。7 . The method for aneurysm detection and rupture risk prediction in digital subtraction angiography according to claim 1 , wherein in step 4), the aneurysm of five consecutive frames of images is extracted, including rupture The corresponding blood perfusion features, intensity features and texture features were extracted from the aneurysms and unruptured aneurysms, respectively. 8.根据权利要求7所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，提取动脉瘤的血流灌注特征具体为：8. a kind of aneurysm detection and rupture risk prediction method in a kind of digital subtraction angiography according to claim 7, is characterized in that, the blood perfusion characteristic of extraction aneurysm is specifically: 在影像上选择三个不同的位置，分别为动脉瘤内部、动脉瘤边缘和动脉瘤外部，在每个位置选取5*5大小的感兴趣区域，画出每个位置感兴趣区域的时间-密度曲线，然后从时间-密度曲线上提取血流灌注特征。Select three different positions on the image, namely the inside of the aneurysm, the edge of the aneurysm and the outside of the aneurysm, select a 5*5 area of interest at each position, and draw the time-density of the area of interest at each position curve, and then extract blood perfusion characteristics from the time-density curve. 9.根据权利要求7所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，提取动脉瘤的强度特征和纹理特征具体为：9. a kind of aneurysm detection and rupture risk prediction method in a kind of digital subtraction angiography according to claim 7, is characterized in that, the intensity feature and texture feature of extracting aneurysm are specifically: 根据影像上动脉瘤灰度和纹理的差异提取强度和纹理特征，强度特征描述图像中体素强度的统计分布，纹理特征分别基于灰度共生矩阵、灰度游程矩阵、灰度大小区域矩阵和邻域灰度差矩阵，描述图像的纹理差异。The intensity and texture features are extracted according to the difference between the grayscale and texture of the aneurysm on the image. The intensity feature describes the statistical distribution of voxel intensities in the image. Domain grayscale difference matrix, describing the texture differences of the image. 10.根据权利要求1所述的一种数字减影血管造影中的动脉瘤检测和破裂风险预测方法，其特征在于，所述的步骤5)中，采用迭代稀疏表示的方法选择与样本标签密切相关的特征，具体采用OMP算法求解优化问题，稀疏表示方法采用滑动窗口策略，利用窗口中所有样本的信息，则迭代过程具体包括以下步骤：10. A method for aneurysm detection and rupture risk prediction in digital subtraction angiography according to claim 1, characterized in that, in the step 5), an iterative sparse representation method is used to select a method that is closely related to the sample label. For related features, the OMP algorithm is used to solve the optimization problem. The sparse representation method adopts the sliding window strategy and uses the information of all samples in the window. The iterative process specifically includes the following steps: 51)计算第k次迭代后的系数
Figure FDA0003349451860000031
则有：51) Calculate the coefficients after the kth iteration
Figure FDA0003349451860000031
Then there are:
Figure FDA0003349451860000032
Figure FDA0003349451860000032
 其中，gk为第k次迭代的标准值，Fk为第k次迭代的样本特征，σ为一小常数；Among them, g k is the standard value of the k-th iteration, F k is the sample feature of the k-th iteration, and σ is a small constant; 52)计算第k次迭代后系数
Figure FDA0003349451860000033
的平均值
Figure FDA0003349451860000034
当满足||M(k)-M(k-1)||2＜ε或k＝K0时，迭代停止，其中，ε为一小正整数，K0为最大迭代次数；52) Calculate the coefficients after the kth iteration
Figure FDA0003349451860000033
average of
Figure FDA0003349451860000034
The iteration stops when ||M (k) -M (k-1) || 2 <ε or k=K 0 , where ε is a small positive integer, and K 0 is the maximum number of iterations; 53)通过迭代得到的系数M(k)用于特征选择，经过迭代运算，每个特征对应一个得分，得分越高表明特征越重要，对最终得分进行排序，选择设定数量的特征进行分类，采用稀疏表示的方法对所选特征进行分类，则稀疏表示分类模型为：53) The coefficient M (k) obtained by iteration is used for feature selection. After iterative operation, each feature corresponds to a score. The higher the score, the more important the feature is. The final score is sorted, and a set number of features are selected for classification. Using sparse representation to classify the selected features, the sparse representation classification model is:
Figure FDA0003349451860000035
Figure FDA0003349451860000035
 其中，f代表待测动脉瘤的特征，F＝[FR,FU]，FR＝[f1,f2,…,fn1]为训练样本中破裂动脉瘤的特征集，n为破裂动脉瘤的数目，FU＝[f1,f2,…,fn2]为训练样本中未破裂动脉瘤的特征集，n2为未破裂动脉瘤的数目，ρ为稀疏表示控制参数，
Figure FDA0003349451860000036
为稀疏表示系数，，当获得最优稀疏表示系数
Figure FDA0003349451860000037
时，根据残差rα(f)判断特征所属类别，则残差rα(f)的表达式为：Among them, f represents the feature of the aneurysm to be tested, F=[F R ,F U ], F R =[f 1 ,f 2 ,...,f n1 ] is the feature set of the ruptured aneurysm in the training sample, and n is the ruptured aneurysm The number of aneurysms, F U =[f 1 ,f 2 ,...,f n2 ] is the feature set of unruptured aneurysms in the training sample, n2 is the number of unruptured aneurysms, ρ is the sparse representation control parameter,
Figure FDA0003349451860000036
is the sparse representation coefficient, when the optimal sparse representation coefficient is obtained
Figure FDA0003349451860000037
When , the category of the feature is judged according to the residual r α (f), then the expression of the residual r α (f) is:
Figure FDA0003349451860000041
Figure FDA0003349451860000041
 其中，δα(.)表示与所选特征类别相对应的系数。where δ α (.) represents the coefficient corresponding to the selected feature class.
CN202111332597.6A 2021-11-11 2021-11-11 Method for aneurysm detection and rupture risk prediction in digital subtraction angiography Pending CN114170143A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111332597.6A CN114170143A (en) 2021-11-11 2021-11-11 Method for aneurysm detection and rupture risk prediction in digital subtraction angiography

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111332597.6A CN114170143A (en) 2021-11-11 2021-11-11 Method for aneurysm detection and rupture risk prediction in digital subtraction angiography

Publications (1)

Publication Number Publication Date
CN114170143A true CN114170143A (en) 2022-03-11

Family

ID=80478849

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111332597.6A Pending CN114170143A (en) 2021-11-11 2021-11-11 Method for aneurysm detection and rupture risk prediction in digital subtraction angiography

Country Status (1)

Country Link
CN (1) CN114170143A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115359051A (en) * 2022-10-19 2022-11-18 江苏诺阳家居科技有限公司 Aneurysm identification method based on pattern identification
US20230196568A1 (en) * 2021-12-20 2023-06-22 Compal Electronics, Inc. Angiography image determination method and angiography image determination device
CN119359556A (en) * 2024-12-23 2025-01-24 首都医科大学附属北京朝阳医院 A method for intelligent enhancement of fundus vascular images

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110490040A (en) * 2019-05-30 2019-11-22 浙江理工大学 A method of local vascular stenosis in identification DSA coronary artery images
CN111915538A (en) * 2020-08-19 2020-11-10 南京普爱医疗设备股份有限公司 Image enhancement method and system for digital blood vessel subtraction
CN112567378A (en) * 2018-05-27 2021-03-26 易鲁希德生物成像公司 Method and system for utilizing quantitative imaging
WO2021149875A1 (en) * 2020-01-20 2021-07-29 주식회사 아이도트 Cerebral aneurysm rupture prediction system

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112567378A (en) * 2018-05-27 2021-03-26 易鲁希德生物成像公司 Method and system for utilizing quantitative imaging
CN110490040A (en) * 2019-05-30 2019-11-22 浙江理工大学 A method of local vascular stenosis in identification DSA coronary artery images
WO2021149875A1 (en) * 2020-01-20 2021-07-29 주식회사 아이도트 Cerebral aneurysm rupture prediction system
CN111915538A (en) * 2020-08-19 2020-11-10 南京普爱医疗设备股份有限公司 Image enhancement method and system for digital blood vessel subtraction

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
TAO HU 等: ""Automatic detection of intracranial aneurysms in 3D‑DSA based on a Bayesian optimized flter"", 《BIOMEDICAL ENGINEERINGONLINE》, 15 September 2020 (2020-09-15), pages 1 - 18 *
TAO HU等: ""A framework for intracranial aneurysm detection and rupture analysis on DSA"", 《JOURNAL OF CLINICAL NEUROSCIENCE》, 30 September 2023 (2023-09-30), pages 101 - 107 *
吴国庆等: ""基于稀疏表示体系的原发性脑部淋巴 瘤和胶质母细胞瘤图像鉴别"", 《生物医学工程学杂志》, no. 05, 25 October 2018 (2018-10-25), pages 754 - 759 *
吴国庆等: "基于稀疏表示体系的原发性脑部淋巴瘤和胶质母细胞瘤图像鉴别", 生物医学工程学杂志, no. 05, 25 October 2018 (2018-10-25), pages 754 - 759 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20230196568A1 (en) * 2021-12-20 2023-06-22 Compal Electronics, Inc. Angiography image determination method and angiography image determination device
CN115359051A (en) * 2022-10-19 2022-11-18 江苏诺阳家居科技有限公司 Aneurysm identification method based on pattern identification
CN115359051B (en) * 2022-10-19 2023-12-15 江苏诺阳家居科技有限公司 Aneurysm identification method based on pattern identification
CN119359556A (en) * 2024-12-23 2025-01-24 首都医科大学附属北京朝阳医院 A method for intelligent enhancement of fundus vascular images
CN119359556B (en) * 2024-12-23 2025-06-10 首都医科大学附属北京朝阳医院 Fundus blood vessel image intelligent enhancement method

Similar Documents

Publication Publication Date Title
Dubost et al. Enlarged perivascular spaces in brain MRI: automated quantification in four regions
Wan et al. Accurate segmentation of overlapping cells in cervical cytology with deep convolutional neural networks
CN114170143A (en) Method for aneurysm detection and rupture risk prediction in digital subtraction angiography
CN109190690B (en) Detection and identification method of cerebral microbleeds in SWI images based on machine learning
Mahapatra Analyzing training information from random forests for improved image segmentation
CN107644420B (en) Vascular image segmentation method and MRI system based on centerline extraction
Dai et al. A new approach to segment both main and peripheral retinal vessels based on gray-voting and gaussian mixture model
Zhang et al. Structured learning for 3-D perivascular space segmentation using vascular features
Kuo et al. Automatic lung nodule detection system using image processing techniques in computed tomography
CN101667289B (en) Retinal Image Segmentation Method Based on NSCT Feature Extraction and Supervised Classification
US7949162B2 (en) System and method for solid component evaluation in mixed ground glass nodules
CN108122221B (en) Segmentation method and device for cerebral ischemia area in diffusion weighted imaging image
Yang et al. Discriminative dictionary learning for retinal vessel segmentation using fusion of multiple features
CN112037221B (en) Multi-domain co-adaptation training method for cervical cancer TCT slice positive cell detection model
Lawrence et al. Computer assisted detection of polycystic ovary morphology in ultrasound images
Gao et al. An effective retinal blood vessel segmentation by using automatic random walks based on centerline extraction
CN114359288A (en) Medical image cerebral aneurysm detection and positioning method based on artificial intelligence
Ananth et al. An Advanced Low-cost Blood Cancer Detection System.
Peng et al. PombeX: robust cell segmentation for fission yeast transillumination images
CN112819765A (en) Liver image processing method
Jeslin et al. AGWO-CNN Classification for Computer-Assisted Diagnosis of Brain Tumors.
Dash et al. Retinal blood vessels extraction from fundus images using an automated method
Pundikal et al. Microaneurysms Detection Using Grey Wolf Optimizer and Modified K-Nearest Neighbor for Early Diagnosis of Diabetic Retinopathy.
Pundikal et al. Detection of Microaneurysms Using Grey Wolf Optimization for Early Diagnosis of Diabetic Retinopathy.
CN113269756B (en) Method and device for optimizing retinal vessel segmentation based on multi-scale matched filtering and particle swarm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination