Pulmonary vascular motion compensation method based on space-time sequence

Abstract

The invention relates to the technical field of medical image processing, in particular to a pulmonary vascular motion compensation method based on a space-time sequence, and the method comprises the steps: constructing a four-dimensional space-time data set based on a three-dimensional CT image sequence and monitoring data; based on the four-dimensional spatiotemporal data set, calculating a motion field between adjacent data frames to obtain a continuous motion field based on pixels; according to the pixel-based continuous motion field, vascular motion is decomposed to obtain separated cardiopulmonary motion; calculating a predictive field and a local predictive field of vascular motion based on the separated cardiopulmonary motion; according to the local prediction field, adjusting imaging parameters of the angiography system; estimating a non-rigid motion field between the images through a frame registration method based on the separated cardiopulmonary motion and the adjusted imaging parameters; according to the non-rigid motion field, an image after motion compensation is generated through a multi-frame image fusion method, and high-precision compensation of complex pulmonary vascular motion is achieved by constructing a four-dimensional spatiotemporal data set and combining an innovative vascular motion decomposition model.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and more specifically, to a pulmonary vascular motion compensation method based on spatiotemporal sequence. Background Art

[0002] With the continuous development of medical imaging technology, pulmonary vascular imaging plays an increasingly important role in the diagnosis and treatment of diseases such as pulmonary embolism and pulmonary hypertension. However, due to the influence of respiratory motion and heart beats, obtaining high-quality pulmonary vascular images has always been a challenging problem. Existing pulmonary vascular motion compensation methods mainly focus on two directions: respiratory gating technology and image registration technology.

[0003] Respiratory gating technology monitors the patient's respiratory signals and acquires images at specific respiratory phases, thereby reducing the impact of respiratory motion. Although this method is simple and direct, it has several obvious disadvantages. First, it significantly prolongs the scanning time, increases the patient's discomfort and radiation dose. Second, it cannot completely eliminate the impact of heart pulsation, especially for pulmonary blood vessels close to the heart, the image quality is still unsatisfactory. In addition, for patients with irregular breathing, the effectiveness of this method is greatly reduced.

[0004] Image registration technology attempts to compensate for images by calculating the motion field in the image sequence. This method can theoretically handle the effects of respiratory motion and heart beats at the same time. However, most existing registration algorithms use global deformation models, which are difficult to accurately capture the local deformation of pulmonary vessels. In particular, small blood vessels are easily overlooked or mishandled during the registration process due to their weak signals. On the other hand, although some local registration algorithms can handle local deformation, they have high computational complexity and are difficult to be widely used in clinical practice.

[0005] In addition to the technical difficulties of motion compensation, existing methods also have shortcomings in image acquisition strategies. The traditional fixed sampling rate strategy cannot adapt to complex lung motion, often resulting in a decrease in image quality when the motion is intense, and causing unnecessary radiation doses when the motion is small. In addition, the existing contrast agent usage scheme also faces the problem of a trade-off between image quality and patient safety. Summary of the invention

[0006] In view of the above problems existing in the prior art, the present invention proposes a pulmonary vascular motion compensation method based on spatiotemporal sequence. The method aims to achieve accurate modeling and compensation of pulmonary vascular motion, while optimizing image acquisition strategy and contrast agent usage scheme, thereby improving image quality while protecting patient safety to the greatest extent.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] The pulmonary vascular motion compensation method based on spatiotemporal sequence includes:

[0009] The acquisition steps include:

[0010] Acquire the three-dimensional CT image sequence acquired by the angiography system and the monitoring data acquired by the respiratory / ECG monitoring equipment;

[0011] Processing steps include:

[0012] constructing a four-dimensional spatiotemporal data set based on the three-dimensional CT image sequence and the monitoring data;

[0013] Based on the four-dimensional spatiotemporal data set, calculating the motion field between adjacent data frames to obtain a pixel-based continuous motion field;

[0014] Decomposing the vascular motion according to the pixel-based continuous motion field to obtain separated cardiopulmonary motion;

[0015] Based on the separated cardiopulmonary motion, calculating a predicted field and a local predicted field of vascular motion;

[0016] According to the local prediction field, adaptive control is performed to adjust imaging parameters of the angiography system;

[0017] Output steps include:

[0018] Based on the separated cardiopulmonary motion and the adjusted imaging parameters, estimating a non-rigid motion field between images by a frame registration method;

[0019] According to the non-rigid motion field, a motion-compensated image is generated by a multi-frame image fusion method.

[0020] Preferably, the obtaining step specifically includes:

[0021] Acquire three-dimensional angiography data through an angiography system, and record the three-dimensional angiography data as I(t,r), where t represents different time points and r represents different spatial positions;

[0022] Initializing the three-dimensional CT image sequence;

[0023] The three-dimensional CT image sequence is initialized using a 3DFCM clustering method so that each pixel point obtains three initialization labels;

[0024] aligning the three-dimensional angiography data with the three-dimensional CT image sequence by a rigid alignment method;

[0025] The aligned three-dimensional CT image sequences are merged into the four-dimensional spatiotemporal dataset.

[0026] Preferably, the step of calculating the motion field between adjacent data frames specifically includes:

[0027] The motion field between the angiographic data I(t,r) and I(t+1,r) at two adjacent time intervals is calculated using a variational model based on non-rigid registration;

[0028] The motion field parameterization is decomposed into two parts: global and local, denoted as u = u g +u l ;

[0029] The pixel-based continuous motion field is obtained by minimizing the energy function E(u)=D(I, J, u)+αS(u), where D represents the data term, S represents the regularization term, and α is the weight coefficient.

[0030] Preferably, the step of decomposing the vascular motion specifically includes:

[0031] Constructing a vascular motion decomposition model:

[0032] E(u1,u2)=∑[w(r)(I(t+1,r)-I(t,r+u1+u2)) 2 ]+λR(u1,u2);

[0033] Where u1 represents lung motion, u2 represents heart motion, w(r) is the weight function, R(u1,u2) is the regularization term, and λ is the regularization parameter;

[0034] By optimizing the vascular motion decomposition model, separated cardiopulmonary motion is obtained.

[0035] Preferably, the step of calculating the prediction field and the local prediction field of vascular motion specifically includes:

[0036] Calculate the global prediction field: u g =argminE g (u);

[0037] Calculate the local prediction field: u l =argminE l (u);

[0038] Among them, E g and E l are the energy functions of the global and local prediction fields, respectively.

[0039] Preferably, the step of performing adaptive control specifically includes:

[0040] Calculate the adaptive sampling rate: s = f(u l );

[0041] Among them, f is based on the local prediction field u l Determined function;

[0042] The imaging rate of the angiography system is dynamically adjusted according to the adaptive sampling rate s.

[0043] Preferably, the step of estimating the non-rigid motion field between images by using a frame registration method specifically includes:

[0044] Frame registration based on cardiac motion estimation: u = argminE r (u);

[0045] Among them, E r is the energy function of the non-rigid motion field;

[0046] Combined with the initial estimation of the non-rigid motion field for frame registration: u = argminE′ r (u);

[0047] Among them, E′ r The energy function for the initialization estimate is considered.

[0048] Preferably, the method further comprises:

[0049] Image segmentation methods are used to transform the vascular motion decomposition problem into a voxel-level segmentation problem;

[0050] The segmentation problem is solved using a maximum expectation algorithm to obtain estimates of lung and heart motion.

[0051] Preferably, the method further comprises:

[0052] A convolutional neural network is used to solve the vascular motion decomposition problem;

[0053] Extracting image features of angiography data by using the convolutional neural network;

[0054] Based on the image features, estimates of lung and heart motion are calculated.

[0055] Preferably, the method further comprises:

[0056] On the injection side, a non-ionic X-ray non-developing agent and an ionic contrast agent are mixed and then injected into the body of the patient to be imaged with blood vessels;

[0057] Wherein, the non-ionic X-ray non-developing agent is used to reduce the radiation amount caused by X-ray fluoroscopy of pulmonary blood vessels during scanning;

[0058] The ionic contrast agent is used to increase the absorption of X-rays in the heart, blood vessels, and lungs, helping to identify tiny hemorrhages.

[0059] The method of the present invention has the following significant technical effects:

[0060] 1. High-precision motion compensation: This invention can accurately separate cardiopulmonary motion and achieve high-precision compensation for complex pulmonary vascular motion by constructing a four-dimensional spatiotemporal data set and combining it with an innovative vascular motion decomposition model. This method can not only process the overall motion of large blood vessels, but also capture the local deformation of small blood vessels, significantly improving the clarity and diagnostic value of the image.

[0061] 2. Adaptive sampling strategy: The adaptive control method proposed in this paper can dynamically adjust the sampling rate according to the real-time predicted motion situation. This strategy effectively reduces the radiation dose received by patients while ensuring image quality, and is particularly suitable for patients with chronic lung diseases who require long-term follow-up.

[0062] 3. Optimized contrast agent use: By mixing non-ionic and ionic contrast agents, the present invention significantly reduces the risk of adverse reactions associated with contrast agents while providing good image contrast. This innovative contrast agent strategy reflects the present invention's deep concern for patient safety.

[0063] 4. Improved computational efficiency: The motion estimation and decomposition algorithms used in the present invention, such as the improved non-rigid registration method and the feature extraction technology based on deep learning, greatly improve computational efficiency. This enables the present method to achieve near real-time processing in clinical practice, providing the possibility for doctors to make instant diagnoses.

[0064] 5. Wide applicability: The method of the present invention is not only applicable to CT angiography, but can also be extended to other modalities such as MR angiography. This wide applicability provides a unified solution for pulmonary vascular imaging in different clinical scenarios.

[0065] 6. Improved diagnostic value: By providing high-quality, high-resolution images of lung blood vessels, the present invention significantly improves the ability to detect microlesions. This is of great significance for the diagnosis of early pulmonary embolism, pulmonary hypertension and other diseases, and is expected to improve the early diagnosis rate and treatment effect of related diseases.

[0066] 7. Improved patient experience: Due to the use of adaptive sampling and optimized contrast agent strategy, the present invention can significantly shorten the scanning time and reduce patient discomfort. This not only improves patient compliance, but also makes it possible to conduct rapid examinations in emergency situations.

[0067] In summary, the pulmonary vascular motion compensation method based on spatiotemporal sequence proposed in the present invention comprehensively solves the problems existing in the prior art through innovative technical means and humanized design. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1The figure is an overall flow chart of the method of the present invention.

[0069] Figure 2 It is a detailed flow chart of the acquisition steps of the present invention.

[0070] Figure 3 Flowchart constructed for the four-dimensional spatiotemporal dataset of the present invention.

[0071] Figure 4 Flowchart of motion field calculation of the present invention.

[0072] Figure 5 Flow chart of the vascular motion decomposition of the present invention.

[0073] Figure 6 It is a flow chart of the adaptive control of the present invention.

[0074] Figure 7 Flow chart of the output step of the present invention. DETAILED DESCRIPTION

[0075] like Figure 1-7 As shown, first, the present invention provides a pulmonary vascular motion compensation method based on a spatiotemporal sequence. The method achieves accurate compensation for pulmonary vascular motion by combining data obtained by an angiography system and a respiratory / ECG monitoring device, thereby improving the quality and accuracy of pulmonary vascular imaging. The specific embodiments of the present invention will be described in detail below.

[0076] First, the method of the present invention includes an acquisition step, a processing step and an output step. In the acquisition step, the method acquires a three-dimensional CT image sequence acquired by an angiography system and monitoring data acquired by a respiratory / ECG monitoring device. These data provide a basis for subsequent processing.

[0077] In the processing step, the method first constructs a four-dimensional spatiotemporal data set based on the acquired three-dimensional CT image sequence and monitoring data. This step is crucial because it organically combines spatial and temporal information, laying the foundation for subsequent motion analysis. Preferably, the present invention uses an interpolation algorithm to construct a four-dimensional spatiotemporal data set to ensure the continuity and integrity of the data.

[0078] Next, based on the constructed four-dimensional spatiotemporal data set, this method calculates the motion field between adjacent data frames to obtain a pixel-based continuous motion field. This step uses a variational model based on non-rigid registration, and its mathematical expression is as follows:

[0079] E(u)=D(I,J,u)+αS(u)

[0080] Wherein, E(u) is the energy function, D represents the data term, S represents the regularization term, and α is the weight coefficient. In one embodiment of the present invention, the value range of α is 0.1 to 1.0, and the preferred value is 0.5. The selection of this range is based on a large amount of experimental data, and a good balance can be achieved between maintaining the smoothness and accuracy of the motion field.

[0081] Then, the method decomposes the vascular motion according to the obtained pixel-based continuous motion field to obtain the separated cardiopulmonary motion. This step adopts an innovative vascular motion decomposition model, whose mathematical expression is as follows:

[0082]

[0083] Wherein, u1 represents lung motion, u2 represents heart motion, w(r) is a weight function, R(u1,u2) is a regularization term, and λ is a regularization parameter. In a preferred embodiment of the present invention, the value range of λ is 0.01 to 0.1, which can effectively balance the accuracy and stability of motion decomposition.

[0084] Based on the separated cardiopulmonary motion, the method calculates the prediction field and local prediction field of vascular motion. This step is crucial for subsequent adaptive control. The present invention adopts a method combining the global prediction field and the local prediction field to capture the characteristics of vascular motion more comprehensively.

[0085] Based on the calculated local prediction field, the method performs adaptive control to adjust the imaging parameters of the angiography system. The innovation of this step is that it can dynamically adjust the imaging parameters according to the real-time motion prediction results, thereby improving the imaging quality and efficiency. For example, when intense motion is predicted, the system may increase the sampling rate; when the motion is small, the sampling rate may be reduced to reduce the radiation dose.

[0086] In the output step, the method estimates the non-rigid motion field between images through frame registration based on the separated cardiopulmonary motion and the adjusted imaging parameters. This step uses an improved non-rigid registration algorithm, and its objective function can be expressed as:

[0087] u=argminE r (u)

[0088] Among them, E r is the energy function of the non-rigid motion field. In one embodiment of the present invention, E r Image similarity terms, smoothness constraint terms, and volume preservation terms may be included to ensure the accuracy and physical rationality of the registration results.

[0089] Finally, the method generates a motion-compensated image by multi-frame image fusion based on the estimated non-rigid motion field. This step not only compensates for the image distortion caused by motion, but also improves the signal-to-noise ratio and resolution of the image through multi-frame fusion.

[0090] An important feature of the method of the present invention is the specific implementation of its acquisition step. Specifically, this step includes obtaining three-dimensional angiography data through an angiography system, and recording the three-dimensional angiography data as I(t,r), where t represents different time points and r represents different spatial positions. This representation method allows us to accurately capture the changes of blood vessels in time and space dimensions.

[0091] Next, the method initializes the three-dimensional CT image sequence. Preferably, the present invention uses the 3DFCM (Fuzzy C-Means) clustering method for initialization, so that each pixel obtains three initialization labels. The advantage of this method is that it can handle fuzzy boundaries, which is particularly important in medical image processing. In one embodiment, the fuzziness parameter m of the FCM algorithm can be set to 2, which is a value widely used in practice and can achieve a good balance between clustering effect and computational efficiency.

[0092] Subsequently, the method aligns the three-dimensional angiography data with the three-dimensional CT image sequence by a rigid alignment method. This step ensures the spatial consistency between images of different modalities and lays the foundation for subsequent motion analysis. In practical applications, a rigid registration algorithm based on mutual information can be used, and its objective function can be expressed as:

[0093] T*=argmaxMI(I fixed ,T(I moving ))

[0094] Among them, MI represents mutual information, I fixed and I moving denote fixed image and moving image respectively, and T denotes rigid transformation.

[0095] Finally, the method combines the aligned three-dimensional CT image sequences into the four-dimensional spatiotemporal data set. This step realizes the organic combination of spatial and temporal information, providing a comprehensive data basis for subsequent motion analysis.

[0096] Another key step of the method of the present invention is to calculate the motion field between adjacent data frames. Specifically, the method uses a variational model based on non-rigid registration to calculate the motion field between two adjacent time interval angiography data (t, r) and I(t+1, r). This method can effectively capture complex non-rigid motion and is particularly suitable for motion analysis of soft tissues such as pulmonary blood vessels.

[0097] In this step, the method decomposes the motion field parameterization into two parts: global and local, denoted as u = u g +u l The advantage of this decomposition method is that it can capture both large-scale global motion and small-scale local deformations simultaneously, thus providing a more comprehensive description of vascular motion.

[0098] By minimizing the energy function E(u) = D(I, J, u) + αS(u), the method obtains the pixel-based continuous motion field. Wherein, D represents the data term, which is used to measure the similarity between images; S represents the regularization term, which is used to ensure the smoothness of the motion field; α is the weight coefficient, which is used to balance the contribution of the data term and the regularization term. In practical applications, the value of α is usually between 0.1 and 1, and the optimal value can be determined by methods such as cross-validation.

[0099] In general, the pulmonary vascular motion compensation method based on spatiotemporal sequence provided by the present invention has the following advantages: first, it makes full use of spatiotemporal information to more accurately capture and compensate for complex vascular motion; second, through the adaptive control strategy, the imaging parameters can be dynamically adjusted according to the real-time motion prediction results, thereby improving the imaging quality and efficiency; finally, the application of multi-frame image fusion technology further improves the quality of the final image. These advantages make this method have broad application prospects in the field of pulmonary vascular imaging, especially in the diagnosis and treatment of diseases such as pulmonary embolism and pulmonary hypertension.

[0100] Next, the method of the present invention decomposes the vascular motion to separate the cardiopulmonary motion. This step is one of the core innovations of the present invention, which is achieved by constructing a vascular motion decomposition model. Specifically, the method adopts the following mathematical model:

[0101]

[0102] In this model, u1 represents lung motion, u2 represents heart motion, w(r) is a weight function, and R(u1,u2) is a regularization term. Preferably, the present invention adopts a spatially adaptive weight function w(r) to better handle the motion characteristics of different regions. For example, in the edge region of the lung, the weight may be smaller to reduce the impact of the boundary effect.

[0103] The design of the regularization term R(u1,u2) is crucial to obtain a smooth and physically reasonable motion field. In a preferred embodiment of the present invention, R(u1,u2) can be expressed as:

[0104]

[0105] in, represents the gradient operator. This form of regularization term can effectively suppress noise and discontinuity in the motion field. The parameter λ controls the balance between the data term and the regularization term. The practice of the present invention shows that the value range of λ is usually between 0.01 and 0.1. The specific value can be adjusted according to the quality of the image and the complexity of the motion. For example, when the image noise is large, the λ value can be appropriately increased to enhance the smoothness.

[0106] Based on the above vascular motion decomposition model, the method obtains separated cardiopulmonary motion by solving through optimization algorithm. Preferably, the present invention adopts alternating direction multiplier method (ADMM) for optimization solution, which has good convergence and computational efficiency when dealing with large-scale optimization problems.

[0107] After obtaining the separated cardiopulmonary motion, the method of the present invention further calculates the prediction field and local prediction field of vascular motion. This step is crucial for achieving subsequent adaptive control. Specifically, the method first calculates the global prediction field:

[0108] u g =argminE g (u)

[0109] Among them, E g is the energy function of the global prediction field. In one embodiment of the present invention, E g Can be designed as:

[0110]

[0111] This form of energy function not only takes into account the similarity between images, but also includes the smoothness constraint of the motion field.

[0112] Next, the method computes the local prediction field:

[0113] u l =argminE l (u)

[0114] Among them, E l is the energy function of the local prediction field. Preferably, E l Can be designed as:

[0115]

[0116] Here, w(r) is a local weight function used to emphasize the importance of local regions. The parameter β controls the smoothness of the local prediction field, and its value is usually smaller than α in the global prediction field to allow more local deformation.

[0117] Based on the calculated local prediction field, the method of the present invention performs adaptive control to adjust the imaging parameters of the angiography system. The core idea of ​​this step is to dynamically adjust the sampling strategy according to the predicted motion to achieve a balance between image quality and radiation dose.

[0118] Specifically, this method calculates the adaptive sampling rate:

[0119] s=f(u l )

[0120] Among them, f is based on the local prediction field u l In a preferred embodiment of the present invention, f can be designed as:

[0121] f(u l )=s0(1+γ||u l ||)

[0122] Here, s0 is the base sampling rate, γ is a tuning parameter, ||u l || represents the norm of the local prediction field. This design increases the sampling rate when violent motion is predicted, and reduces the sampling rate when the motion is small. Preferably, the value range of γ can be set to 0.1 to 1, and the specific value can be adjusted according to clinical needs and equipment performance. For example, for applications requiring high temporal resolution, a larger γ value can be selected; and for patients who are particularly sensitive to radiation dose, a smaller γ value can be selected. Based on the calculated adaptive sampling rate s, the method dynamically adjusts the imaging rate of the angiography system. This adaptive control strategy can not only improve image quality, but also effectively reduce unnecessary radiation dose, which is more patient-friendly.

[0123] Finally, the method of the present invention estimates the non-rigid motion field between images through a frame registration method and performs multi-frame image fusion. In the frame registration step, the method performs frame registration based on cardiac motion estimation:

[0124] u=argminE r (u)

[0125] Among them, E r is the energy function of the non-rigid motion field. In one embodiment of the present invention, E r Can be designed as:

[0126]

[0127] This form of energy function not only takes into account the similarity between images and the smoothness of the motion field, but also includes the volume preservation constraint (through the divergence term Parameters λ1 and λ2 control the strength of the smoothness constraint and the volume preservation constraint respectively, and their values ​​can be determined by methods such as cross-validation.

[0128] In order to further improve the accuracy of registration, this method also combines the initialization estimation of the non-rigid motion field for frame registration:

[0129] u=argminE r ′(u)

[0130] Among them, E′ r is the energy function considered for initialization estimation. Preferably, E′ r Can be designed as:

[0131] E r ′(u)=E r (u)+μ∫||uu init || 2 dr

[0132] Here, y init is the initial motion field estimate, and μ is a weight parameter used to control the influence of the initial estimate.

[0133] Through the above frame registration method, the present invention can effectively estimate the non-rigid motion field between images, laying a foundation for subsequent image fusion. Finally, the present method uses multi-frame image fusion technology such as weighted averaging to generate high-quality images after motion compensation.

[0134] In general, the pulmonary vascular motion compensation method based on spatiotemporal sequence provided by the present invention realizes accurate compensation for pulmonary vascular motion through innovative motion decomposition model, adaptive control strategy and high-precision frame registration technology. This method not only improves the quality and accuracy of pulmonary vascular imaging, but also reduces the radiation dose received by patients through adaptive sampling strategy. Therefore, the present invention has important application value in clinical diagnosis and treatment, especially for lung diseases that require precise vascular imaging, such as pulmonary embolism, pulmonary hypertension, etc., this method can provide more reliable diagnostic basis and treatment guidance. In a preferred embodiment of the present invention, the method also includes using image segmentation method to convert the vascular motion decomposition problem into a voxel-level segmentation problem. The advantage of this method is that it can process complex vascular structures more finely, thereby improving the accuracy of motion decomposition. Specifically, the method first uses a threshold-based method for image segmentation:

[0135]

[0136] Wherein, S(r) represents the segmentation result of the rth spatial position, I(r) represents the pixel intensity at the position, and T is the threshold. Preferably, the threshold T can be adaptively determined by the Otsu method, which can automatically adjust between different images to improve the robustness of the segmentation.

[0137] After obtaining the preliminary segmentation results, this method uses the maximum expectation algorithm to solve and obtain the estimation of lung and heart motion. The objective function of the maximum expectation algorithm can be expressed as:

[0138] Q(θθ (t) )=E Z [logp(X,Z|θ)|X,θ (t) ]

[0139] Wherein, θ represents the model parameters, X represents the observed data, and Z represents the latent variable. In the application of the present invention, X can be understood as image data, and Z corresponds to the labels of different tissue types.

[0140] Specifically, this method updates the model parameters in an iterative manner:

[0141]

[0142] In each iteration, the expectation step (E-step) calculates the posterior probability:

[0143]

[0144] The maximization step (M-step) updates the model parameters:

[0145]

[0146] In this way, the present invention can estimate the motion of the lungs and heart more accurately, thereby improving the accuracy of vascular motion compensation.

[0147] In another embodiment of the present invention, the method further comprises solving the vascular motion decomposition problem using a convolutional neural network. This deep learning-based method can automatically learn complex feature representations and performs well in processing large-scale medical image data.

[0148] Specifically, this method first extracts the image features of angiography data through a convolutional neural network. A typical convolutional layer can be expressed as:

[0149] F l =σ(W l *F l-1 +b l )

[0150] Among them, F lrepresents the feature map of the ιth layer, W l and b l are the convolution kernel and bias of the layer respectively, * represents the convolution operation, and σ is the activation function. Preferably, the present invention adopts ReLU as the activation function:

[0151] σ(x)=max(0,x)

[0152] This activation function can effectively alleviate the gradient vanishing problem and accelerate the network training process.

[0153] After extracting image features, this method obtains the final classification result through the fully connected layer and the softmax layer:

[0154] p(y|x)=softmax(Wx+b)

[0155] Where x is the input feature vector, W and b are the weight matrix and bias vector respectively, and y is the predicted category label.

[0156] Based on the above network structure, this method trains the model by minimizing the cross entropy loss function:

[0157]

[0158] Among them, y is the true label, p(y i |x i ) is the probability predicted by the model.

[0159] Through this convolutional neural network-based method, the present invention can automatically learn the complex characteristics of vascular motion, thereby achieving more accurate motion decomposition and compensation.

[0160] Finally, the method of the present invention also includes an important pretreatment step, that is, mixing the non-ionic X-ray non-developing agent and the ionic contrast agent on the injection side and injecting them into the body of the patient to be imaged. The design of this step reflects the consideration of patient safety in the practical application of the present invention.

[0161] Specifically, the role of non-ionic X-ray non-contrast agents is to reduce the radiation caused by X-ray fluoroscopy of pulmonary blood vessels during scanning. Such contrast agents usually have lower osmotic pressure and viscosity, which can reduce adverse effects on the human body. Preferably, the present invention uses iohexol as a non-ionic contrast agent, which has a molecular weight of about 821 and a concentration that can be selected between 300-370 mgI / mL, and the specific concentration can be adjusted according to the patient's weight and the examination site.

[0162] And the effect of ionic contrast agent is to improve the X-ray absorption of heart, blood vessel and lung, helps to identify tiny hemorrhage.This contrast agent has higher osmotic pressure usually, can provide better image contrast.Preferably, the present invention adopts iodixanol as ionic contrast agent, and its molecular weight is about 1550, and concentration can be selected between 320-400mgI / mL.

[0163] The mixing ratio of the two contrast agents needs to be adjusted according to the specific clinical needs and patient conditions. In general, the ratio of non-ionic contrast agents can be slightly higher to reduce the overall risk of adverse reactions. For example, a mixing ratio of 7:3 or 8:2 can be used. This mixing strategy can provide good image quality while protecting patient safety to the greatest extent.

[0164] By adopting this innovative contrast agent mixing strategy and combining it with the aforementioned motion compensation method, the present invention can not only obtain high-quality pulmonary vascular images, but also significantly reduce the radiation dose and the risk of adverse reactions associated with contrast agents received by patients. This makes the present method have broad application prospects in clinical practice, especially for patients with chronic lung diseases who need multiple follow-up examinations, the advantages of the present method are more obvious.

[0165] In general, the spatiotemporal sequence-based pulmonary vascular motion compensation method provided by the present invention comprehensively improves the quality and safety of pulmonary vascular imaging through innovative image segmentation technology, deep learning methods and contrast agent strategies.

[0166] The above description is only a preferred specific implementation manner of the present invention; however, the protection scope of the present invention is not limited thereto; any person familiar with the art who, within the scope disclosed by the present invention, makes equivalent replacements or changes based on the scheme and improved concepts of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A pulmonary vascular motion compensation method based on spatiotemporal sequence, characterized in that: include: The acquisition steps include: Acquire the three-dimensional CT image sequence acquired by the angiography system and the monitoring data acquired by the respiratory / ECG monitoring equipment; Processing steps include: constructing a four-dimensional spatiotemporal data set based on the three-dimensional CT image sequence and the monitoring data; Based on the four-dimensional spatiotemporal data set, calculating the motion field between adjacent data frames to obtain a pixel-based continuous motion field; Decomposing the vascular motion according to the pixel-based continuous motion field to obtain separated cardiopulmonary motion; Based on the separated cardiopulmonary motion, calculating a predicted field and a local predicted field of vascular motion; According to the local prediction field, adaptive control is performed to adjust imaging parameters of the angiography system; Output steps include: Based on the separated cardiopulmonary motion and the adjusted imaging parameters, estimating a non-rigid motion field between images by a frame registration method; According to the non-rigid motion field, a motion-compensated image is generated by a multi-frame image fusion method.

2. The method according to claim 1, characterized in that The acquisition step specifically includes: Acquire three-dimensional angiography data through an angiography system, and record the three-dimensional angiography data as I(t,r), where t represents different time points and r represents different spatial positions; Initializing the three-dimensional CT image sequence; The three-dimensional CT image sequence is initialized using a 3DFCM clustering method so that each pixel point obtains three initialization labels; aligning the three-dimensional angiography data with the three-dimensional CT image sequence by a rigid alignment method; The aligned three-dimensional CT image sequences are merged into the four-dimensional spatiotemporal dataset.

3. The method according to claim 1, characterized in that The step of calculating the motion field between adjacent data frames specifically includes: The motion field between the angiographic data I(t,r) and I(t+1,r) at two adjacent time intervals is calculated using a variational model based on non-rigid registration; The motion field parameterization is decomposed into two parts: global and local, denoted as u = u g +u l ; The pixel-based continuous motion field is obtained by minimizing the energy function E(u)=D(I, J, u)+αS(u), where D represents the data term, S represents the regularization term, and α is the weight coefficient.

4. The method according to claim 1, characterized in that: The step of decomposing the vascular movement specifically includes: Constructing a vascular motion decomposition model: E(u1,u2)=∑[w(r)(I(t+1,r)-I(t,r+u1+u2)) 2 ]+λR(u1,u2); Where u1 represents lung motion, u2 represents heart motion, w(r) is the weight function, R(u1,u2) is the regularization term, and λ is the regularization parameter; By optimizing the vascular motion decomposition model, separated cardiopulmonary motion is obtained.

5. The method according to claim 1, characterized in that: The step of calculating the prediction field and the local prediction field of the vascular motion specifically includes: Calculate the global prediction field: u g =argminE g (u); Calculate the local prediction field: u l =argminE l (u); Among them, E g and E l are the energy functions of the global and local prediction fields, respectively.

6. The method according to claim 1, characterized in that The steps of performing adaptive control specifically include: Calculate the adaptive sampling rate: s = f(u l ); Among them, f is based on the local prediction field u l Determined function; The imaging rate of the angiography system is dynamically adjusted according to the adaptive sampling rate s.

7. The method according to claim 1, characterized in that The step of estimating the non-rigid motion field between images by using the frame registration method specifically includes: Frame registration based on cardiac motion estimation: u = argminE r (u); Among them, E r is the energy function of the non-rigid motion field; Combine the initialization estimate of the non-rigid motion field for frame registration: u = argminE′r(u); Among them, E′ r The energy function for the initialization estimate is considered.

8. The method according to claim 4, characterized in that The method further comprises: Image segmentation methods are used to transform the vascular motion decomposition problem into a voxel-level segmentation problem; The segmentation problem is solved using a maximum expectation algorithm to obtain estimates of lung and heart motion.

9. The method according to claim 4, characterized in that The method further comprises: A convolutional neural network is used to solve the vascular motion decomposition problem; Extracting image features of angiography data by using the convolutional neural network; Based on the image features, estimates of lung and heart motion are calculated.

10. The method according to claim 1, characterized in that The method further comprises: On the injection side, a non-ionic X-ray non-developing agent and an ionic contrast agent are mixed and then injected into the body of the patient to be imaged with blood vessels; Wherein, the non-ionic X-ray non-developing agent is used to reduce the radiation amount caused by X-ray fluoroscopy of pulmonary blood vessels during scanning; The ionic contrast agent is used to increase the absorption of X-rays in the heart, blood vessels, and lungs, helping to identify tiny hemorrhages.

Images