CN116934768A - Method and system for improving blood vessel segmentation accuracy in CTA image mode - Google Patents
Method and system for improving blood vessel segmentation accuracy in CTA image mode Download PDFInfo
- Publication number
- CN116934768A CN116934768A CN202311035425.1A CN202311035425A CN116934768A CN 116934768 A CN116934768 A CN 116934768A CN 202311035425 A CN202311035425 A CN 202311035425A CN 116934768 A CN116934768 A CN 116934768A
- Authority
- CN
- China
- Prior art keywords
- real
- blood vessel
- label
- vessel
- cta image
- 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.)
- Granted
Links
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 163
- 230000011218 segmentation Effects 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 56
- 230000006698 induction Effects 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 50
- 238000003384 imaging method Methods 0.000 claims description 14
- 230000008859 change Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010968 computed tomography angiography Methods 0.000 description 102
- 230000008569 process Effects 0.000 description 7
- 238000012549 training Methods 0.000 description 7
- 238000002059 diagnostic imaging Methods 0.000 description 6
- 230000002792 vascular Effects 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000008447 perception Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 210000005166 vasculature Anatomy 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000005452 bending Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 238000002583 angiography Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- 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
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The application provides a method and a system for improving blood vessel segmentation accuracy in a CTA image mode, wherein the method comprises the following steps: acquiring an initial CTA image and an initial vessel segmentation prediction label; performing data enhancement processing on the initial CTA image to obtain a target CTA image; generating a weight value of a real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value; combining the target CTA image and the corrected real blood vessel label to construct a real-order total variation operator; constructing a real-order total variation induction loss function according to the real-order total variation operator; and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induced loss function so as to correct the initial vessel segmentation prediction label. The application has the effect of improving the blood vessel segmentation accuracy when blood vessel segmentation is carried out in three-dimensional CTA scanning.
Description
Technical Field
The application belongs to the technical field of medical scanning imaging segmentation, and particularly relates to a method and a system for improving blood vessel segmentation accuracy in a CTA image mode.
Background
CT angiography (CTA) is a medical imaging technique combining CT enhancement technology with thin-layer, large-scale, fast scanning technology, and through reasonable post-processing, details of blood vessels at various parts of the whole body are clearly displayed. CTA is also called as a non-invasive blood vessel imaging technology in medicine, angiography is an interventional detection method, a developer is injected into blood vessels, the characteristic that X-rays penetrate through the developer is utilized, the displayed image of the developer under the X-rays is utilized to form blood vessel imaging, and CTA has the characteristics of no wound and simple operation.
After the medical imaging is acquired, the image may be manually segmented by the radiologist to obtain the vasculature image, but this process is time consuming and requires sufficient expertise. Image segmentation can be performed by a segmentation method based on deep learning, which is more accurate than the conventional manual method. However, with the continuous development of technology, medical imaging is gradually changed from traditional 2D imaging to high-resolution three-dimensional CTA medical imaging, and it is difficult to well extend to three-dimensional CTA medical imaging using a segmentation method based on deep learning to segment blood vessels having a complex topology.
In order to segment and mark the vasculature to the finest level in three-dimensional CTA medical imaging, especially the most clinically valuable of the fine vascular components. In the prior art, a patch-based model is typically used, i.e., a CTA scan is cut into overlapping small 3D patches, and then a segmentation task is performed on each patch. However, such patch-based methods may lose global context information and may lead to false positive segments on unwanted tissues, as these tissues are often difficult to distinguish locally from the intended tissue, ultimately resulting in poor vessel segmentation accuracy in the CTA scan image.
Disclosure of Invention
The application provides a method and a system for improving blood vessel segmentation accuracy in a CTA image mode, which are used for solving the problem that the blood vessel segmentation accuracy is not high due to the loss of global context information when blood vessel segmentation is carried out in CTA scanning.
In a first aspect, the present application provides a method for improving vessel segmentation accuracy in a CTA imaging modality, the method comprising the steps of:
acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image;
performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image;
generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value;
constructing a real-order total variation operator by combining the target CTA image and the corrected real blood vessel label, wherein the real-order total variation operator is used for quantifying the total complexity of all the target blood vessels;
constructing a real-order total variation induction loss function according to the real-order total variation operator;
and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
Optionally, the data enhancement processing is performed on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method, so as to obtain a target CTA image, which includes the following steps:
combining the initial CTA image and a preset real blood vessel label to calculate and obtain a change interval of a blood vessel pixel value in the initial CTA image;
and carrying out data enhancement processing on the initial CTA image based on the change interval and according to the random sampling weight to obtain a target CTA image.
Optionally, the calculation formula of the variation interval is:
wherein: τ represents the variation interval, Q represents the initial CTA image, V represents the real vessel label;
the expression of the target CTA image is:
Q VIIS ：＝Q+α·τV
wherein: q (Q) VIIS Representing the target CTA image, α represents random sampling weights.
Optionally, the generating the weight value of the real blood vessel label based on the blood vessel surface curvatures of all the target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value includes the following steps:
generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, wherein the expression of the weight value is as follows:
wherein: s represents the set of all points on the vessel wall of the target vessel, dist (·) represents the distance function, z represents any point on the vessel wall, y represents the point closest to S, and κ (y) represents the curvature of the y point;
correcting the real blood vessel label through the weight value, wherein the formula for correcting the real blood vessel label is as follows:
wherein: c (C) 2 (x) Represents any of the real vessel labels x corrected labels, |·| represents L 1 Norms.
Optionally, the step of constructing a real-order total variation operator by combining the target CTA image and the corrected real blood vessel label includes the following steps:
generating a fractional divergence operator based on pixels in the target CTA image and from Li Manliu-dimension fractional derivatives;
and combining the fractional divergence operator and the corrected real blood vessel label to construct a real-order total variation operator.
Optionally, the derivative expression of the Li Manliu wifery fractional derivative is:
wherein: d, d r Representing Li Manliu Utility fractional derivatives, r ε (0, 1), u:representing the Lebelger measurable set, +.>Represents a real number domain, Γ represents a gamma function, I represents a unit interval, x 0 E I represents the pixel.
Optionally, generating a fractional divergence operator div from the derivative expression r The expression of the real-order total variation operator is as follows:
TV r (V)：＝|div r V|
wherein: TV set r Representing the real-order total variation operator.
Optionally, the expression of the real-order total variation induction loss function is:
wherein:representing the real-order total variation induction loss function, and P represents the initial vessel segmentation prediction label.
Optionally, the method further comprises the steps of:
constructing a penalty loss function based on the real-order total variation induction loss function, wherein the expression of the penalty loss function is as follows:
wherein:represents the penalty loss function, lambda d And lambda (lambda) r All represent weighting coefficients, +.>Representing the dice loss.
In a second aspect, the present application also provides a system for improving the accuracy of vessel segmentation in a CTA imaging modality, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method as described in the first aspect above when executing the computer program.
The beneficial effects of the application are as follows:
the method for improving the blood vessel segmentation accuracy in the CTA image mode mainly comprises the following steps: acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image; performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image; generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value; combining the target CTA image and the corrected real blood vessel label to construct a real-order total variation operator; constructing a real-order total variation induction loss function according to the real-order total variation operator; and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
Through the weight correction in the steps, the method is beneficial to solving the imbalance problem among blood vessels with different complexity degrees in the CTA image, and global context perception can be provided in the blood vessel segmentation process by constructing a real-order total variation induction loss function, so that long-term geometric information of the blood vessels is better captured, the problem of discontinuity is relieved, and the segmentation precision in the blood vessel segmentation process in CTA scanning is further improved.
Drawings
FIG. 1 is a flow chart of a method for improving the accuracy of vessel segmentation in a CTA imaging modality according to the present application.
Fig. 2 is a diagram illustrating the weight size discrimination of the real blood vessel label according to the present application.
Fig. 3 is a diagram showing the weight value size discrimination of the real blood vessel label in the present application.
Fig. 4 is a schematic diagram of the vascular structures of three simulated vessels according to the present application.
Detailed Description
The application discloses a method for improving blood vessel segmentation accuracy in a CTA image mode.
Referring to fig. 1, the method for improving the segmentation accuracy of blood vessels in a CTA image modality specifically includes the following steps:
s101, acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image.
Wherein a dataset comprising the initial CTA image and the corresponding vessel segmentation label is first required. The dataset may be an existing dataset obtained from a medical image database. And then CTA scanning is carried out on the target person by using CT scanning equipment so as to acquire an initial CTA image. Ensuring that the scan parameters and image quality are suitable for vessel segmentation tasks. The initial CTA image is annotated to generate predictive labels for vessel segmentation. The initial CTA image and vessel segmentation labels also need to be preprocessed after the prediction label generation to facilitate subsequent training and segmentation tasks. The preprocessing steps may include image resampling, gray scale normalization, noise removal, image enhancement, and the like.
The data set is divided into a training set, a validation set and a test set. Typically, about 70-80% of the data is used for training, 10-15% is used for validation, and 10-15% is used for testing. And training a blood vessel segmentation model by using the divided training set and verification set. Training may be performed using deep learning methods such as Convolutional Neural Networks (CNNs) or U-Net, among others. The super parameters of the model, such as learning rate, batch size, network architecture, etc., are adjusted to obtain better performance. The trained model is evaluated using the test set. And calculating the vascular segmentation performance indexes of the model on the test set, such as accuracy, recall rate, F1 score and the like. Based on the evaluation results, model improvement and optimization can be performed. And finally, outputting the initial CTA image to a trained vessel segmentation model to obtain an initial vessel segmentation prediction label.
S102, carrying out data enhancement processing on the initial CTA image according to a preset real blood vessel label and through a blood vessel image intensity offset method to obtain a target CTA image.
The blood vessel image intensity offset method comprises Gaussian blur, histogram equalization, gray scale stretching and the like. These methods can change the contrast and brightness of the image to enhance the visibility of the blood vessels. The real vessel label is applied to the initial CTA image. The real vessel label can be superimposed or fused with the initial CTA image using image processing software or a programming library. Ensuring alignment of the vessel label with the image and maintaining label accuracy. The initial CTA image is enhanced using the selected vessel image intensity offset method. Appropriate parameters may be selected to adjust the contrast, brightness and detail of the image as desired. And obtaining a target CTA image by applying a blood vessel image intensity offset method and a real blood vessel label. Ensuring that the enhanced image still maintains the shape and position of the blood vessel while improving the visibility of the blood vessel. In another embodiment, the original dataset may also be augmented with the generated target CTA image. The target CTA image can be paired with the original CTA image to form a new training sample. Thus, the diversity of the data samples can be increased, and the robustness and performance of the model can be improved.
S103, generating a weight value of a real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value.
Wherein, the curvature of the surface of the blood vessel is an index for measuring the bending degree of the curve of the blood vessel. Curvature values for each vessel point may be calculated using curvature calculation algorithms, such as gaussian curvature or average curvature. A weight value is generated for each target vessel point based on the vessel surface curvature. Points of higher curvature of the vessel surface may represent bends or bifurcation of the vessel and should therefore be given a higher weight. A weight function may be defined that assigns a corresponding weight value to each point based on the magnitude of the curvature of the vessel surface. For example, the weight values may be adjusted using an exponential function or a linear function. Using the generated weight values, the real vessel label is corrected. For each target vessel point, its real vessel label is multiplied by a corresponding weight value to correct the strength of the label. Thus, the brightness or gray value of the label can be adjusted according to the bending degree of the blood vessel curve. The corrected real vessel label is applied to the target CTA image. The corrected real vessel label can be superimposed or fused with the target CTA image using image processing software or a programming library. Ensure the alignment of the label and the image and maintain the accuracy and consistency of the label.
S104, combining the target CTA image and the corrected real blood vessel label to construct a real-order total variation operator.
Wherein the real-order total variation operator is used to quantify the total complexity of all target vessels. The vasculature of the target vessel may be represented essentially by a tree having nodes, leaves, and branches, and generally having more complex geometries if the target vessel is vasculature with multiple branches, and thus the target vessel will have higher real-order total variation metric values.
S105, constructing a real-order total variation induction loss function according to the real-order total variation operator.
The global information is rapidly utilized by the real blood vessel label when the real total variation algorithm of each point in the blood vessel surface of the target blood vessel is evaluated, so that the constructed real total variation induction loss function can provide global context perception.
S106, performing energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
The implementation principle of the embodiment is as follows:
the method for improving the blood vessel segmentation accuracy in the CTA image mode mainly comprises the following steps: acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image; performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image; generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value; combining the target CTA image and the corrected real blood vessel label to construct a real-order total variation operator; constructing a real-order total variation induction loss function according to the real-order total variation operator; and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
Through the weight correction in the steps, the method is beneficial to solving the imbalance problem among blood vessels with different complexity degrees in the CTA image, and global context perception can be provided in the blood vessel segmentation process by constructing a real-order total variation induction loss function, so that long-term geometric information of the blood vessels is better captured, the problem of discontinuity is relieved, and the segmentation precision in the blood vessel segmentation process in CTA scanning is further improved.
In one embodiment, step S102 is to perform data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method, and the step of obtaining the target CTA image specifically includes the following steps:
combining the initial CTA image and a preset real blood vessel label to calculate and obtain a change interval of a blood vessel pixel value in the initial CTA image;
and carrying out data enhancement processing on the initial CTA image based on the change interval and according to the random sampling weight to obtain a target CTA image.
In the present embodiment, the calculation formula of the change interval is:
wherein: τ represents the change interval, Q represents the initial CTA image, V represents the real vessel label;
the expression of the target CTA image is:
Q VIIS ：＝Q+α·τV
wherein: q (Q) VIIS Representing the target CTA image, α represents the random sampling weight.
In one embodiment, in step S103, the weight value of the real blood vessel label is generated based on the blood vessel surface curvatures of all the target blood vessels in the target CTA image, and the real blood vessel label is corrected by the weight value. The expression of the weight value of the real blood vessel label is as follows:
wherein: s represents the set of all points on the vessel wall of the target vessel, dist (·) represents the distance function, z represents any point on the vessel wall, y represents the point closest to S, and κ (y) represents the curvature of the y point;
the real blood vessel label is corrected through the weight value, and the formula for correcting the real blood vessel label is as follows:
C 2 (x)＝ω(x)·|V|
wherein: c (C) 2 (x) Represents any real vessel label x corrected label, |·| represents L 1 Norms.
In this embodiment, although the fine vascular component in the CTA image is only a small part of the entire blood vessel, it is less than 5% in the coronary blood vessel and less than 10% in the head and neck blood vessel. However, these slim vessel sections are more prone to erroneous segmentation by neural networks trained using voxel matching metrics (such as die loss) because these metrics tend to favor thick vessel parts that occupy more voxels. Furthermore, elongated vascular components, as well as components with sharp turns and vascular branching points, often contain complex blood flows, which can significantly alter the image intensity and signal-to-noise ratio, further increasing the complexity of the segmentation task.
The concept of weight functions and weight spaces introduced in this embodiment therefore assigns higher weights to voxels of interest within the space. In particular, a weighting function based on the value of the vessel surface curvature may be introduced. Referring to fig. 2 and 3, curvature is a measure of the instantaneous rate of change of direction of a moving point on a surface: the greater the curvature, the greater this rate of change. Thus, a vessel part with a small radius, sharp turn or other complex geometry exhibits a higher curvature value than a vessel part with a simple geometry, thereby enabling the vessel label corresponding to the complex vessel part to have a higher weight value, and in fig. 2 and 3, a color closer to red indicates a higher weight value corresponding to a vessel and a color closer to blue indicates a lower weight value corresponding to a vessel.
In one embodiment, the step S104 of constructing the real-order total variation operator by combining the target CTA image and the corrected real blood vessel label specifically includes the following steps:
generating a fractional divergence operator based on pixels in the target CTA image and according to Li Manliu-dimension fractional order derivatives;
and combining the fractional divergence operator and the corrected real blood vessel label to construct a real-order total variation operator.
In this embodiment, the derivative expression of the Li Manliu wirl derivative is:
wherein: d, d r Representing Li Manliu Utility fractional derivatives, r ε (0, 1), u:representing the Lebelger measurable set, +.>Represents a real number domain, Γ represents a gamma function, I represents a unit interval, x 0 E I represents a pixel.
The specific expression of the gamma function is as follows:
generating fractional divergence operator div from derivative expression r The expression of the real-order total variation operator is:
TV r (V)：＝|div r V|
wherein: TV set r Representing the real-order total variation operator.
The expression of the real-order total variation induction loss function is:
wherein:representing the real-order total variation induced loss function, and P represents the initial vessel segmentation prediction label.
For example, referring to FIG. 4, assume that there are three vessels, vessel # 0, vessel # 1 and vessel # 2, in the CTA image, where vessel # 0 consists of a single vessel assembly that is isomorphic and equidistant, vessel # 0 has a length L, and vessel # 0 has a cylinder radius R < L. The vessel 1# is also composed of single vessel components, the vessel components are isomorphic and equidistant, the length of the vessel 1# is L-epsilon, epsilon represents the lost length of the vessel, epsilon is less than L, the radius of a cylinder where the vessel 1# is positioned is R less than L-epsilon, and the vessel 1# simulates the tail loss condition of the vessel. The vessel 2# consists of two vessel components, wherein the two vessel components are a pair of isomorphic and equidistant coaxial cylinders with the length of (L-epsilon)/2 and the radius of R < L, and the vessel 2# simulates the fracture condition of the vessel.
The spatial cylindrical coordinates of the three vessels were set as:
C 1 ：＝{(ρ，θ，z)：ρ≤R，z∈[0，L]}
fractional derivatives of the three blood vessels in the longitudinal direction are calculated by the derivative expression of Li Manliu-dimensional fractional derivatives respectively, and the calculation formula is as follows:
according to the fractional derivative, further calculating to obtain three real-order total variable operators of the blood vessels, wherein the real-order total variable operators of the blood vessel 0# are expressed as follows:
the real-order total variation operator for vessel 1# is expressed as:
the real-order total variation operator for vessel 2# is expressed as:
combining the above formulas with the expression of the actual total variation induced loss function, a conclusion can be drawn:
T r (Vessel1#)＜T r (Vessel2#)
according to the above conclusion, it is stated that: the real-order total variation induced loss function has a heavier penalty for vessels with breaks in the middle than for vessels with loss of tails.
In one embodiment, the method for improving the accuracy of vessel segmentation in a CTA imaging modality further comprises the steps of:
constructing a penalty loss function based on the real-order total variation induced loss function, wherein the expression of the penalty loss function is as follows:
wherein:represents a penalty loss function, lambda d And lambda (lambda) r All represent weighting coefficients, +.>Representing the dice loss.
In the present embodiment, lambda d Can be assigned a value of 1, lambda r The penalty loss function can assign 0.01, and because the blood vessel label used in the penalty loss function is a real blood vessel label corrected by the weight value, the penalty loss function can generate higher penalty for the special blood vessel with smaller radius, sharp turning characteristic and complex geometric shape in the CTA image, thereby being beneficial to improving the segmentation result of the special blood vessel.
The application also discloses a system for improving the blood vessel segmentation precision in the CTA image mode, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method for improving the blood vessel segmentation precision in the CTA image mode according to any one of the embodiments when executing the computer program.
The implementation principle of the embodiment is as follows:
by calling the program, the following steps are executed: acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image; performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image; generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value; combining the target CTA image and the corrected real blood vessel label to construct a real-order total variation operator; constructing a real-order total variation induction loss function according to the real-order total variation operator; and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
Through the weight correction in the steps, the method is beneficial to solving the imbalance problem among blood vessels with different complexity degrees in the CTA image, and global context perception can be provided in the blood vessel segmentation process by constructing a real-order total variation induction loss function, so that long-term geometric information of the blood vessels is better captured, the problem of discontinuity is relieved, and the segmentation precision in the blood vessel segmentation process in CTA scanning is further improved.
Those of ordinary skill in the art will appreciate that: the discussion of any of the embodiments above is merely exemplary and is not intended to suggest that the scope of protection of the application is limited to these examples; the technical features of the above embodiments or in the different embodiments may also be combined within the idea of the application, the steps may be implemented in any order and there are many other variations of the different aspects of one or more embodiments of the application as above, which are not provided in detail for the sake of brevity.
One or more embodiments of the present application are intended to embrace all such alternatives, modifications and variations as fall within the broad scope of the present application. Accordingly, any omissions, modifications, equivalents, improvements and others which are within the spirit and principles of the one or more embodiments of the application are intended to be included within the scope of the application.
Claims (10)
1. A method for improving vessel segmentation accuracy in a CTA image modality, comprising the steps of:
acquiring an initial CTA image required to be subjected to blood vessel segmentation and an initial blood vessel segmentation prediction label of the initial CTA image;
performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image;
generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, and correcting the real blood vessel label through the weight value;
constructing a real-order total variation operator by combining the target CTA image and the corrected real blood vessel label, wherein the real-order total variation operator is used for quantifying the total complexity of all the target blood vessels;
constructing a real-order total variation induction loss function according to the real-order total variation operator;
and carrying out energy matching on the initial vessel segmentation prediction label and the corrected real vessel label through the real-order total variation induction loss function so as to correct the initial vessel segmentation prediction label.
2. The method for improving the segmentation accuracy of blood vessels in a CTA image modality according to claim 1, wherein the step of performing data enhancement processing on the initial CTA image according to a preset real blood vessel label and by a blood vessel image intensity offset method to obtain a target CTA image comprises the following steps:
combining the initial CTA image and a preset real blood vessel label to calculate and obtain a change interval of a blood vessel pixel value in the initial CTA image;
and carrying out data enhancement processing on the initial CTA image based on the change interval and according to the random sampling weight to obtain a target CTA image.
3. The method for improving the segmentation accuracy of blood vessels in a CTA image modality according to claim 2, wherein the calculation formula of the variation interval is:
wherein: τ represents the variation interval, Q represents the initial CTA image, V represents the real vessel label;
the expression of the target CTA image is:
Q VIIS ：＝Q+α·τV
wherein: q (Q) VIIS Representing the target CTA image, α represents random sampling weights.
4. The method for improving the segmentation accuracy of blood vessels in a CTA image modality according to claim 1, wherein the generating the weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image and correcting the real blood vessel label by the weight value comprises the steps of:
generating a weight value of the real blood vessel label based on the blood vessel surface curvatures of all target blood vessels in the target CTA image, wherein the expression of the weight value is as follows:
wherein: s represents the set of all points on the vessel wall of the target vessel, dist (·) represents the distance function, z represents any point on the vessel wall, y represents the point closest to S, and κ (y) represents the curvature of the y point;
correcting the real blood vessel label through the weight value, wherein the formula for correcting the real blood vessel label is as follows:
wherein: c (C) 2 (x) Represents any of the real vessel labels x corrected labels, |·| represents L 1 Norms.
5. The method for improving vessel segmentation accuracy in a CTA imaging modality according to claim 1, wherein said combining the target CTA image and the corrected real vessel label to construct a real-order total variation operator comprises the steps of:
generating a fractional divergence operator based on pixels in the target CTA image and from Li Manliu-dimension fractional derivatives;
and combining the fractional divergence operator and the corrected real blood vessel label to construct a real-order total variation operator.
6. The method for improving vessel segmentation accuracy in a CTA imaging modality according to claim 5, wherein the derivative expression of the Li Manliu wifery fractional derivative is:
wherein: d, d r Represents the Li Manliu-dimensional fractional derivative,representing the Lebelger measurable set, +.>Represents a real number domain, Γ represents a gamma function, I represents a unit interval, x 0 E I represents the pixel.
7. The method for improving vessel segmentation accuracy in a CTA imaging modality of claim 6, wherein a fractional divergence operator div is generated from the derivative expression r The expression of the real-order total variation operator is as follows:
TV r (V)：＝|div r V|
wherein: TV set r Representing the real-order total variation operator.
8. The method for improving vessel segmentation accuracy in a CTA imaging modality according to claim 7, wherein the expression of the real-order total variation-induced loss function is:
wherein:representing the real-order total variation induction loss function, and P represents the initial vessel segmentation prediction label.
9. The method for improving vessel segmentation accuracy in a CTA imaging modality according to claim 8, further comprising the step of:
constructing a penalty loss function based on the real-order total variation induction loss function, wherein the expression of the penalty loss function is as follows:
wherein:represents the penalty loss function, lambda d And lambda (lambda) r All represent weighting coefficients, +.>Representing the dice loss.
10. A system for improving vessel segmentation accuracy in a CTA imaging modality, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 9 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311035425.1A CN116934768B (en) | 2023-08-16 | 2023-08-16 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311035425.1A CN116934768B (en) | 2023-08-16 | 2023-08-16 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116934768A true CN116934768A (en) | 2023-10-24 |
| CN116934768B CN116934768B (en) | 2024-05-10 |
Family
ID=88377260
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311035425.1A Active CN116934768B (en) | 2023-08-16 | 2023-08-16 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116934768B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160196666A1 (en) * | 2013-02-11 | 2016-07-07 | Angiometrix Corporation | Systems for detecting and tracking of objects and co-registration |
| WO2018208846A1 (en) * | 2017-05-08 | 2018-11-15 | Microvascular Health Solutions, LLC | Compositions, systems, and methods for assessing and improving vascular health and treatments involving the same |
| CN110782474A (en) * | 2019-11-04 | 2020-02-11 | 中国人民解放军总医院 | Deep learning-based method for predicting morphological change of liver tumor after ablation |
| CN113298800A (en) * | 2021-06-11 | 2021-08-24 | 沈阳东软智能医疗科技研究院有限公司 | Processing method, device and equipment of CT angiography CTA source image |
| US20220012859A1 (en) * | 2020-07-09 | 2022-01-13 | Henan University Of Technology | Method and device for parallel processing of retinal images |
| CN114202504A (en) * | 2021-09-24 | 2022-03-18 | 无锡祥生医疗科技股份有限公司 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
| WO2022161192A1 (en) * | 2021-02-01 | 2022-08-04 | 之江实验室 | Method for automatically segmenting left ventricle of spect three-dimensional reconstruction image |
| CN116309571A (en) * | 2023-05-18 | 2023-06-23 | 中国科学院自动化研究所 | Three-dimensional cerebrovascular segmentation method and device based on semi-supervised learning |
-
2023
- 2023-08-16 CN CN202311035425.1A patent/CN116934768B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160196666A1 (en) * | 2013-02-11 | 2016-07-07 | Angiometrix Corporation | Systems for detecting and tracking of objects and co-registration |
| WO2018208846A1 (en) * | 2017-05-08 | 2018-11-15 | Microvascular Health Solutions, LLC | Compositions, systems, and methods for assessing and improving vascular health and treatments involving the same |
| CN110782474A (en) * | 2019-11-04 | 2020-02-11 | 中国人民解放军总医院 | Deep learning-based method for predicting morphological change of liver tumor after ablation |
| US20220012859A1 (en) * | 2020-07-09 | 2022-01-13 | Henan University Of Technology | Method and device for parallel processing of retinal images |
| WO2022161192A1 (en) * | 2021-02-01 | 2022-08-04 | 之江实验室 | Method for automatically segmenting left ventricle of spect three-dimensional reconstruction image |
| CN113298800A (en) * | 2021-06-11 | 2021-08-24 | 沈阳东软智能医疗科技研究院有限公司 | Processing method, device and equipment of CT angiography CTA source image |
| CN114202504A (en) * | 2021-09-24 | 2022-03-18 | 无锡祥生医疗科技股份有限公司 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
| CN116309571A (en) * | 2023-05-18 | 2023-06-23 | 中国科学院自动化研究所 | Three-dimensional cerebrovascular segmentation method and device based on semi-supervised learning |
Non-Patent Citations (1)
| Title |
|---|
| 李越 等: "心血管造影图像分割方法综述", 计算机系统应用, vol. 29, no. 9, 30 September 2020 (2020-09-30), pages 1 - 4 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116934768B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Myatt et al. | Neuromantic–from semi-manual to semi-automatic reconstruction of neuron morphology | |
| CN107644420B (en) | Blood vessel image segmentation method based on centerline extraction and nuclear magnetic resonance imaging system | |
| CN111462047B (en) | Vascular parameter measurement method, vascular parameter measurement device, vascular parameter measurement computer device and vascular parameter measurement storage medium | |
| CN111429502B (en) | Method and system for generating a centerline of an object and computer readable medium | |
| Wang et al. | Novel 4-D open-curve active contour and curve completion approach for automated tree structure extraction | |
| CN112258530A (en) | Neural network-based computer-aided lung nodule automatic segmentation method | |
| CN110050281A (en) | Learn the annotation of the object in image | |
| US20120134552A1 (en) | Method for checking the segmentation of a structure in image data | |
| Wang et al. | Uncertainty-guided efficient interactive refinement of fetal brain segmentation from stacks of MRI slices | |
| CN101082983A (en) | Self-adapting medicine sequence image values inserting method based on interested region | |
| CN109949288A (en) | Tumor type determines system, method and storage medium | |
| CN113034507A (en) | CCTA image-based coronary artery three-dimensional segmentation method | |
| CN111383759A (en) | Automatic pneumonia diagnosis system | |
| CN115512110A (en) | Medical image tumor segmentation method related to cross-modal attention mechanism | |
| CN112529915A (en) | Brain tumor image segmentation method and system | |
| CN112802025A (en) | Liver tumor segmentation method and device under CT image | |
| CN114332132A (en) | Image segmentation method and device and computer equipment | |
| WO2015150320A1 (en) | Segmentation of tubular organ structures | |
| CN115830016A (en) | Medical image registration model training method and equipment | |
| Tan et al. | Automatic prostate segmentation based on fusion between deep network and variational methods | |
| Purnama et al. | Follicle detection on the usg images to support determination of polycystic ovary syndrome | |
| CN116934768B (en) | Method and system for improving blood vessel segmentation accuracy in CTA image mode | |
| CN116664594A (en) | Three-dimensional medical image two-stage segmentation method and device based on sharing CNN | |
| CN109872353A (en) | Based on the white light data and CT Registration of Measuring Data method for improving iteration closest approach algorithm | |
| El-Shafai et al. | Deep learning-based hair removal for improved diagnostics of skin diseases |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |