A method for identifying cardiac model parameters based on longitudinal displacement of atrioventricular plane

Abstract

A method for identifying parameters of a heart model based on longitudinal displacement of the atrioventricular plane includes: using clinically acquired cardiac nuclear magnetic resonance images, calibrating key points of the atrioventricular plane on the image, using a tracking algorithm to track the key points to obtain atrioventricular plane displacement, and then inputting the atrioventricular plane displacement parameters into the heart model, and using a Kalman filter to identify important physiological parameters of the heart. The method of the present invention can overcome the problem of many abnormal values ​​when using segment software to track the position of key points of the atrioventricular plane, and by optimizing the detection method, the atrioventricular plane displacement curve is more consistent with its physiological performance; a new heart model is constructed, which can more accurately depict the dynamic characteristics of the heart, and then the important physiological parameters of the heart are identified through a Kalman filter. The method is simple and has high identification accuracy.

Description

Technical Field

[0001] The present invention relates to cardiac medicine, image processing technology and cardiac dynamics model, and in particular to a cardiac model parameter identification method based on longitudinal displacement of the atrioventricular plane. Background Art

[0002] Atrioventricular plane displacement (AVPD) is an indicator of cardiac systolic and diastolic function, accounting for 60% of the left ventricle and 80% of the right ventricle. AVPD is usually measured clinically on echocardiography as mitral and tricuspid annular plane systolic excursion (MAPSE and TAPSE), but it is not widely used in cardiovascular magnetic resonance (CMR). So far, there is no robust automatic algorithm that allows clinical measurement of AVPD in CMR and input within a single time frame. In 2017, Felicia Seemann proposed a method for calculating the atrioventricular plane displacement of the heart. Felicia Seemann used segment software to calculate the atrioventricular plane of the heart. Due to the limitations of the tracking algorithm and data processing method, the calculated atrioventricular plane displacement of the heart had many abnormal values, resulting in the atrioventricular plane displacement curve not being physiologically consistent.

[0003] The heart is a complex system with fluid, mechanical and electrophysiological coupling. The mathematical model of the heart can not only help understand the regulatory mechanism of the circulatory system, but also be used to quantitatively analyze cardiovascular diseases and select appropriate treatment plans. Heart models include distributed parameter models and lumped parameter models. The former usually uses the finite element method to simulate the excitation propagation timing and stress-strain distribution of the three-dimensional heart, so as to accurately describe the local movement of the ventricular wall. The latter focuses more on the macroscopic hemodynamic characteristics of the heart cavity, and has the advantages of small calculation amount and simple structure. Therefore, it is widely used in research fields such as heart-vascular coupling and left ventricular assist device control. Sagawa et al. found that the ESPVR (End-systolic pressure-volume relationship) of the left ventricle is insensitive to heart rate and afterload. Based on this principle, ventricular contraction can be represented as a time-varying elastic cavity. As the most commonly used lumped parameter model of the heart, the time-varying elastic model uses a complex nonlinear function to fit the pressure-volume relationship in the ventricle. Based on this, a variety of improved ventricular models have been proposed one after another. Liu Lingling et al. added a parameter to represent myocardial impedance; Wen Taiyang et al. proposed a time-varying resistance model of the mitral valve; Zhong et al. divided ventricular elasticity into active elasticity and passive elasticity, trying to further reveal the physiological mechanism of ventricular elasticity. However, the biggest disadvantage of such models is that they cannot reflect the mechanical movement process of the ventricular wall, so the description of dynamic processes such as heart-pump coupling is not accurate. Summary of the invention

[0004] The purpose of the present invention is to provide a simple and highly accurate method for identifying heart model parameters. Through clinically acquired cardiac magnetic resonance images, key points of the atrioventricular plane are calibrated on the image, and the key points are tracked using a tracking algorithm to obtain the atrioventricular plane displacement. The atrioventricular plane displacement parameters are then input into the heart model, and important physiological parameters of the heart can be identified through a Kalman filter.

[0005] The technical solution of the present invention is as follows:

[0006] A method for identifying parameters of a heart model based on longitudinal displacement of an atrioventricular plane comprises:

[0007] S1: construct the heart model;

[0008] S2: Obtain cardiac magnetic resonance images clinically, mark the key points of the atrioventricular plane on the images, and use a tracking algorithm to track the key points to obtain the atrioventricular plane displacement;

[0009] S3: Input the atrioventricular plane displacement parameters into the heart model and identify the physiological parameters of the heart through the Kalman filter.

[0010] Furthermore, the heart model is:

[0011]

[0012]

[0013] Among them, V AVP represents the displacement velocity of the atrioventricular plane, A represents the rate of change of atrioventricular plane displacement, LV is the equivalent piston area on the ventricular side, A LA is the equivalent piston area on the atrial side, F C is myocardial strength, Q LV is the left ventricular flow, P LV The pressure in the left ventricle, represents the rate of change of left ventricular pressure, P LA Indicates the pressure of the left atrium, C LV represents the left ventricular compliance, K RAD Represents the radial function coefficient.

[0014] Furthermore, the step S2 comprises:

[0015] S2-1: Acquire cardiac magnetic resonance images and select tracking areas;

[0016] S2-2: Perform linear enhancement and mean filtering on the selected area, apply the FAST algorithm to replace the input points, and use K-means clustering to select several groups of input points;

[0017] S2-3: Track the input point, obtain the corresponding atrioventricular plane displacement, and calculate the average value of the atrioventricular plane displacement as a parameter for evaluating the atrioventricular plane displacement of the heart.

[0018] Further, the step S2-3 includes:

[0019] The displacement of each tracking point on the long axis is fitted to obtain the atrioventricular plane displayed on each long axis. The multi-point tracking method is adopted to fully describe the atrioventricular plane. The automatic selection of input points is realized in multi-point tracking and the atrioventricular plane displacement is calculated. The average value of the atrioventricular plane displacement is calculated as the parameter for evaluating the atrioventricular plane displacement of the heart.

[0020] Furthermore, the tracking implementation method includes:

[0021] The model and parameters are initialized according to the training image and the tracked target. First, the color histograms of the foreground and background of the sample are extracted to construct a binary mask. Secondly, CN features, HOG features and grayscale features are extracted, and the correlation filters with spatial constraints are obtained by combining the binary mask training. The feature map of each feature and the filter with spatial constraints are correlated, and the maximum response value of each channel is normalized as the channel reliability weight. In the target positioning stage, the trained filters and channel weights are used to obtain the filter responses of each channel of the detection image, and the final response results are obtained by summing these filter responses through an adaptive weighted summation strategy to complete the target tracking.

[0022] Compared with the prior art, the beneficial technical effects of the present invention are:

[0023] It overcomes the problem of many outliers when using segment software to track the key points of the atrioventricular plane;

[0024] The detection method has been optimized to make the atrioventricular plane displacement curve more consistent with its physiological performance;

[0025] A new heart model was constructed, which can more accurately characterize the dynamic characteristics of the heart. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the overall solution of the present invention;

[0027] Figure 2 It is the input point for tracking in 2CH, 3CH, and 4CH;

[0028] Figure 3 It is the definition of the chamber plane;

[0029] Figure 4 It is image preprocessing;

[0030] Figure 5 It is to automatically select the input point;

[0031] Figure 6 It is the axial and longitudinal contraction of the ventricle;

[0032] Figure 7 It is a heart model of the atrioventricular plane. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] Example 1

[0035] like Figure 1 As shown in the figure, a method for identifying heart model parameters based on longitudinal displacement of atrioventricular plane is firstly used to track the characteristic area of ​​atrioventricular plane by using the designed discriminant filter. Secondly, the 10 most representative groups of feature points in the area are selected as atrioventricular plane input points by using the FAST algorithm and the K-means clustering algorithm. Then, the average displacement of these 10 groups of atrioventricular plane input points is calculated as the parameter for evaluating the atrioventricular plane displacement of the heart. Finally, the unscented filter is used to identify the parameters of the heart model.

[0036] The method of the present invention specifically comprises:

[0037] Step 1: Track the input point

[0038] Use a discriminative correlation filter with channel and spatial reliability to track the input point. This discriminative correlation filter mainly includes two parts: learning update and target positioning. The model and parameters are initialized according to the training image and the tracked target. First, the color histogram of the foreground and background of the sample is extracted to construct a binary mask; secondly, CN features, HOG features and grayscale features are extracted, and combined with the binary mask training to obtain a correlation filter with spatial constraints. The parameters of this filter are concentrated in the mask area, which effectively suppresses background interference and highlights the target information; secondly, the feature map of each feature and the filter with spatial constraints are correlated, and the maximum response value of each channel is normalized as the channel reliability weight. In the target positioning stage, the trained filter and channel weights are used to obtain the filter responses of each channel of the detection image, and these filter responses are used through an adaptive weighted summation strategy to obtain the final response result to complete the target tracking. Figure 2As shown in the figure, the input points are manually placed on the three long-axis images. The cardiac MRI images are acquired by Siemens 1.5T MRI in Germany, and the characteristic areas of the atrioventricular plane are selected. The characteristic areas of the atrioventricular plane are selected on the 2CH, 3CH, and 4CH images respectively, and these areas are input into the tracker.

[0039] Step 2: Optimize the calculation method of atrioventricular plane displacement

[0040] Fitting the displacement of the tracking point on each long axis to obtain the atrioventricular plane displayed on each long axis. The atrioventricular plane is defined as follows: Figure 3 As shown in the figure, a multi-point tracking method is used to fully describe the atrioventricular plane. The automatic selection of input points and calculation of atrioventricular plane displacement are realized in multi-point tracking, including:

[0041] ① Manually select the approximate tracking area.

[0042] ② Perform linear enhancement and mean filtering on the selected area: Through image processing technology, the tracking area features are made more obvious. The present invention uses linear image enhancement technology to enhance image contrast and uses mean filtering to filter noise, such as Figure 4 shown.

[0043] ③ Use the FAST algorithm on the preprocessed image to obtain all the corner points of the image, and then use the K-means clustering algorithm to select the 10 most representative groups of corner points as input points, such as Figure 5 shown.

[0044] ④Track 10 groups of input points to obtain 10 groups of atrioventricular plane displacements, and finally calculate the average value of the 10 groups of atrioventricular plane displacements as the parameter for evaluating the atrioventricular plane displacement of the heart.

[0045] Step 3: Based on the left ventricular model, the atrioventricular plane displacement is used as input to obtain the hemodynamic variables using an unscented filter.

[0046] The comparison between axial contraction and longitudinal contraction of the ventricle Figure 6 As shown in the figure. The existing myocardial fiber model generally approximates the ventricular wall to a sphere. The radial contraction of the sphere produces a change in volume, so the volume of the heart will change significantly during the pumping process. However, clinical data show that the shape of the human ventricle is similar to an ellipsoid, with a long axis diameter of about 8.5 cm and a short axis diameter of about 5 cm. Due to the constraints of the pericardium, the atrioventricular plane will move significantly along the long axis of the ventricle during pumping, while the volume change of the entire heart is very limited.

[0047] Based on this principle, the left heart can be equivalent to a piston pump ( Figure 7The atrioventricular plane (AVP) is composed of soft tissues that wrap around the valves (aortic valve and mitral valve) and is located between the atria and ventricles. The equivalent piston area A on the ventricular side is LV Approximately equivalent piston area A on the atrial side LA During ventricular contraction, the myocardial force F C Vertically downward, the atrioventricular plane moves toward the apex, the blood in the ventricle is ejected through the aortic valve, and the blood in the pulmonary vein fills the atrium; during ventricular diastole, F C =0, the atrioventricular plane moves in the opposite direction under the action of the pressure difference on both sides of the piston, and the blood in the atrium flows into the ventricle through the mitral valve; during atrial contraction, F C Vertically upward, the ventricles continue to relax until the next ventricular systole.

[0048] The heart model is:

[0049]

[0050]

[0051] Among them, V AVP represents the displacement velocity of the atrioventricular plane, A represents the rate of change of atrioventricular plane displacement, LV is the equivalent piston area on the ventricular side, A LA is the equivalent piston area on the atrial side, F C is myocardial strength, Q LV is the left ventricular flow, P LV The pressure in the left ventricle, represents the rate of change of left ventricular pressure, P LA Indicates the pressure of the left atrium, C LV represents the left ventricular compliance, K RAD Represents the radial function coefficient.

[0052] An unscented filter is constructed based on the left heart model. The atrioventricular plane displacement and blood flow are used as the input of the unscented filter to identify the hemodynamic variables.

Claims

1. A method for identifying cardiac model parameters based on longitudinal displacement of the atrioventricular plane, characterized in that: include: S1: construct a heart model; the heart model is: Among them, V AVP represents the displacement velocity of the atrioventricular plane, A represents the rate of change of atrioventricular plane displacement, LV is the equivalent piston area on the ventricular side, A LA is the equivalent piston area on the atrial side, F C is myocardial strength, Q LV is the left ventricular flow, P LV The pressure in the left ventricle, represents the rate of change of left ventricular pressure, P LA Indicates the pressure of the left atrium, C LV represents the left ventricular compliance, K RAD represents the radial function coefficient; S2: Obtain cardiac magnetic resonance images clinically, mark the key points of the atrioventricular plane on the images, and use a tracking algorithm to track the key points to obtain the atrioventricular plane displacement; S2-1: Acquire cardiac magnetic resonance images and select tracking areas; S2-2: Perform linear enhancement and mean filtering on the selected area, apply the FAST algorithm to replace the input points, and use K-means clustering to select several groups of input points; S2-3: Track the input point, obtain the corresponding atrioventricular plane displacement, and calculate the average value of the atrioventricular plane displacement as a parameter for evaluating the atrioventricular plane displacement of the heart; S3: Input the atrioventricular plane displacement parameters into the heart model and identify the physiological parameters of the heart through the Kalman filter.

2. The method according to claim 1, characterized in that The step S2-3 comprises: The displacement of each tracking point on the long axis is fitted to obtain the atrioventricular plane displayed on each long axis. The multi-point tracking method is adopted to fully describe the atrioventricular plane. The automatic selection of input points is realized in multi-point tracking and the atrioventricular plane displacement is calculated. The average value of the atrioventricular plane displacement is calculated as the parameter for evaluating the atrioventricular plane displacement of the heart.

3. The method according to claim 1 or 2, characterized in that: The tracking implementation method includes: The model and parameters are initialized according to the training image and the tracked target. First, the color histograms of the foreground and background of the sample are extracted to construct a binary mask. Secondly, CN features, HOG features and grayscale features are extracted, and the correlation filters with spatial constraints are obtained by combining the binary mask training. The feature map of each feature and the filter with spatial constraints are correlated, and the maximum response value of each channel is normalized as the channel reliability weight. In the target positioning stage, the trained filters and channel weights are used to obtain the filter responses of each channel of the detection image, and the final response results are obtained by summing these filter responses through an adaptive weighted summation strategy to complete the target tracking.

Images