An ultrasonic blood flow velocity measurement method and system based on a deep complex neural network

Abstract

The application discloses an ultrasonic blood flow velocity measurement method and system based on a deep complex neural network. The method comprises the following steps: a physical blood vessel model is built according to set simulation parameters, and original data are collected through the physical blood vessel model; the original data are original blood flow Doppler echo signals; the original data are divided into a training set, a verification set and a test set, and the training set is labeled; a deep complex neural network model is constructed; the deep complex neural network model is trained through the labeled training set; and blood flow velocity measurement is performed through the trained deep complex neural network model. The deep complex neural network is used to fuse the filtering and blood flow velocity measurement in two steps, so that the accuracy of blood flow velocity measurement is improved, the algorithm running time is reduced, and the efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic blood flow velocity measurement technology, and in particular to an ultrasonic blood flow velocity measurement method and system based on a deep complex neural network. Background Technology

[0002] Currently, methods for calculating blood flow velocity are mainly divided into two categories: parametric and non-parametric. Parametric methods directly calculate blood flow velocity from the demodulated ultrasound echo signal. This involves first mathematically modeling the ultrasound echo signal, then estimating the model parameters based on actual sampled data. By analyzing and filtering the estimated parameters, the desired blood flow velocity is finally obtained. Early signal models used were zero-pole models, with the most commonly used being the autoregressive (AR) model. The AR model expresses the input signal as a linear combination of past sampled values ​​and white noise, and its parameters are solved using algorithms such as Burg. However, due to limitations in imaging frame rate, the number of signal samples provided to the AR model is limited, resulting in low parameter estimation accuracy. If there is a significant difference between the mathematical model and the echo data, the parameter estimation algorithm will still introduce large estimation errors. Furthermore, even if the established model can approximate the input signal, classic spectral estimation algorithms often cannot accurately extract the blood flow signal and estimate the blood flow velocity due to the limitation of short data sets.

[0003] Because parametric methods are limited by the established mathematical models in blood flow signal extraction, non-parametric blood flow signal extraction methods, consisting of clutter filters and autocorrelation techniques, are more widely used in practice. Figure 2 As shown, the received ultrasound echo first passes through a clutter filter. The filtered signal is then used to calculate the Doppler frequency of the blood flow signal using autocorrelation techniques. This frequency is ultimately converted into blood flow velocity and used for imaging. Currently, the most popular clutter filter is based on eigenvalue decomposition. This method pre-selects an eigenvalue threshold or a frequency threshold. Eigenvectors with eigenvalues ​​greater than the given threshold or eigenvectors with frequencies lower than the threshold are used to construct a clutter subspace. The projection of the input signal onto the clutter subspace is considered the clutter component. Finally, the estimated clutter component is removed from the input signal. However, this method has the following drawbacks: First, the empirically selected eigenvalue threshold lacks adaptability, making it difficult to determine the dimension of the clutter subspace. Second, improper filter parameter settings can lead to incorrect filtering of blood flow signals, resulting in an incomplete blood flow velocity profile and significantly reduced estimation accuracy. Third, the method of first wall-filtering the entire echo signal and then using autocorrelation techniques to calculate blood flow velocity greatly increases computational complexity and results in low real-time performance, which is inconsistent with the high real-time requirements of clinical diagnosis. Summary of the Invention

[0004] The purpose of this invention is to provide an ultrasonic blood flow velocity measurement method and system based on deep complex neural networks, so as to solve the problem of low velocity measurement accuracy caused by the incorrect filtering of low-velocity blood flow by traditional non-parametric methods.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] An ultrasonic blood flow velocity measurement method based on a deep complex neural network includes:

[0007] A physical blood vessel model is built according to the set simulation parameters, and raw data is collected through the physical blood vessel model; the raw data is the raw blood flow Doppler echo signal.

[0008] The original data is divided into a training set, a validation set, and a test set, and the training set is labeled.

[0009] Construct a deep complex neural network model;

[0010] The deep complex neural network model is trained using the labeled training set;

[0011] Blood flow velocity is measured using a trained deep complex neural network model.

[0012] Optionally, before dividing the original data into training, validation, and test sets, the method further includes:

[0013] Perform a Hilbert transform on the original data.

[0014] Optionally, the annotation of the training set specifically includes:

[0015] The maximum blood flow velocity is calculated using pre-defined simulation parameters;

[0016] Blood flow velocity at different locations is determined based on the maximum blood flow velocity.

[0017] The training set is labeled based on the blood flow velocity at the different locations.

[0018] Optionally, the formula for calculating the maximum blood flow velocity is as follows:

[0019]

[0020] Among them, V max Where c is the maximum blood flow velocity, f is the speed of sound, and f is the maximum blood flow velocity. prf f0 is the pulse repetition frequency, f0 is the center frequency, and θ is the beam angle.

[0021] Optionally, the formula for calculating the blood flow velocity at different locations is as follows:

[0022]

[0023] Where v(d) is the blood flow velocity at different locations, d is the distance from the center line of the lumen, R is the radius of the blood vessel, and V max This represents the maximum blood flow velocity.

[0024] Optionally, the deep complex neural network model is constructed using the PyTorch deep learning framework, consisting of a three-layer convolutional layer and a two-layer fully connected layer.

[0025] Optionally, it also includes:

[0026] The trained deep complex neural network model is validated and tested using the validation set and the test set.

[0027] The present invention also provides an ultrasonic blood flow velocity measurement system based on a deep complex neural network, comprising:

[0028] The raw data acquisition module is used to build a physical blood vessel model based on the set simulation parameters and to acquire raw data through the physical blood vessel model; the raw data is the raw blood flow Doppler echo signal.

[0029] The partitioning and labeling module is used to partition the original data into a training set, a validation set, and a test set, and to label the training set.

[0030] The model building module is used to build deep complex neural network models;

[0031] The training module is used to train the deep complex neural network model using the labeled training set;

[0032] The blood flow velocity measurement module is used to measure blood flow velocity using a trained deep complex neural network model.

[0033] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] This invention provides an ultrasonic blood flow velocity measurement method based on a deep complex neural network, comprising: constructing a physical vascular model according to predefined simulation parameters, and acquiring raw data through the physical vascular model; the raw data being raw blood flow Doppler echo signals; dividing the raw data into a training set, a validation set, and a test set, and labeling the training set; constructing a deep complex neural network model; training the deep complex neural network model using the labeled training set; and performing blood flow velocity measurement using the trained deep complex neural network model. This invention uses a deep complex neural network to fuse filtering and blood flow velocity measurement in two steps, improving the accuracy of blood flow velocity measurement, reducing algorithm runtime, and increasing efficiency. It solves the problems of low accuracy, long algorithm runtime, and poor adaptability inherent in traditional non-parametric methods (i.e., methods that first filter and then use autocorrelation to measure blood flow velocity). Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart of the ultrasonic blood flow velocity measurement method based on a deep complex neural network according to the present invention;

[0037] Figure 2 This is a schematic diagram of a non-parametric blood flow velocity measurement method.

[0038] Figure 3 This is a structural diagram of the deep complex neural network model of the present invention;

[0039] Figure 4 This is a schematic diagram of the complex convolution operation in the deep complex neural network model of the present invention;

[0040] Figure 5 This is an evaluation diagram of the performance of the deep complex neural network model provided in the embodiments of the present invention;

[0041] Figure 6 This is a schematic diagram of a blood flow model provided in an embodiment of the present invention;

[0042] Figure 7 This is a schematic diagram of blood flow velocity imaging results provided in an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] The purpose of this invention is to provide an ultrasonic blood flow velocity measurement method and system based on deep complex neural networks, so as to solve the problem of low velocity measurement accuracy caused by the incorrect filtering of low-velocity blood flow by traditional non-parametric methods.

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1 As shown, the ultrasonic blood flow velocity measurement method based on a deep complex neural network provided by this invention includes the following steps:

[0047] Step 101: Build a physical blood vessel model according to the set simulation parameters, and collect raw data through the physical blood vessel model; the raw data is the raw blood flow Doppler echo signal.

[0048] Using Feild II ultrasonic simulation software, the main simulation parameters are set as follows:

[0049] Parameter name Parameter value Transducer type Linear array Number of array elements 128 Array element spacing 0.3mm Speed ​​of sound 1540m / s Pulse repetition frequency 4kHz Center frequency 5MHz Sampling frequency 100MHz Plane wave angular interval 3° Number of plane wave angles 3 plane wave angular range -3°—3°

[0050] Based on the parameters set above, a physical blood vessel model is built and scattering points are randomly placed. Then, plane waves are emitted to collect raw data and store it locally.

[0051] Step 102: Divide the original data into a training set, a validation set, and a test set, and label the training set.

[0052] The raw data (i.e. the raw Doppler echo signal) obtained in step 101 is subjected to Hilbert transform to obtain the I / Q analytical signal. The demodulated echo data is divided into training set, validation set and test set in a ratio of 7:2:1.

[0053] Calculate the maximum blood flow velocity V based on the set parameters. max :

[0054]

[0055] Where c is the speed of sound, in m / s, and f prfLet f0 be the pulse repetition frequency, f0 be the center frequency, and θ be the beam angle. Since the blood flow in the human carotid artery follows a laminar flow distribution, the blood flow profile is parabolic. Based on the calculated maximum blood flow velocity V... max Calculate the blood flow velocity v(d) at different locations:

[0056]

[0057] Where d is the distance from the center line of the lumen, and R is the radius of the blood vessel. The training set is carefully labeled using v(d) and then normalized.

[0058] Step 103: Construct a deep complex neural network model.

[0059] A model architecture with three convolutional layers and two fully connected layers was built based on the PyTorch deep learning framework. The overall structure of the deep complex neural network model is as follows: Figure 4 As shown.

[0060] Step 104: Train the deep complex neural network model using the labeled training set.

[0061] After setting hyperparameters such as the number of hidden neuron units, activation function, loss function, optimizer, learning rate, and number of training iterations, the model is trained using the Jupyter Notebook environment.

[0062] Step 105: Blood flow velocity is measured using a trained deep complex neural network model.

[0063] The calculated blood flow velocity is color-coded to obtain a two-dimensional blood flow velocity image within the field of view.

[0064] The process includes the following after step 105:

[0065] Step 106: Load the trained model, input the validation set data into the model to evaluate its performance, using the mean absolute error (MAE) as the evaluation metric.

[0066]

[0067] Where x (i) For the predicted value, y (i) This is the theoretical value.

[0068] Step 107: Load the trained model and input the test data into the model to obtain the blood flow velocity value v.

[0069] This invention maps the convolution operator of a conventional convolutional neural network to the complex domain, automatically extracting features from blood flow signals and non-stationary clutter signals from tissues and vessel walls. Using supervised learning and optimization-based criteria, the blood flow signal and clutter signal are calibrated using theoretical blood flow velocity and zero values, respectively. By training an optimized model, the invention automatically infers and calculates blood flow velocity from the echo signal. The main advantages of this invention are that it preserves low-velocity blood flow information, improving the accuracy of blood flow velocity measurements, and integrating filtering and blood flow velocity measurement into a single step, significantly reducing algorithm runtime and improving efficiency.

[0070] The specific implementation method is as follows:

[0071] (1) Establish as follows Figure 6 The blood flow model shown Figure 6 In the diagram, 1 represents the blood flow velocity profile, 2 represents the maximum velocity at the center of the blood flow velocity profile, and 3 represents the tilt angle of the blood vessel.

[0072] The cylindrical blood vessel is located 60 mm subcutaneously, with a lumen radius of R = 4 mm, a vessel tilt angle of 72°, and a maximum central velocity of V. max =0.9967 m / s, then the blood flow velocity profile from the upper wall to the lower wall is:

[0073]

[0074] Where 1≤d≤R

[0075] (2) Transmit plane waves to acquire echo data, and demodulate all blood flow Doppler signals respectively.

[0076] B(n) = R(n) + j*I(n)

[0077] Where R(n) and I(n) are the in-phase and quadrature components of B(n) after quadrature demodulation, respectively, and j is the imaginary unit.

[0078] (3) Divide the demodulated echo data into training set, validation set and test set in a ratio of 7:2:1.

[0079] (4) After multiplying the v(d) calculated in step (1) by the cosine of the beam angle, the blood vessel part is labeled, and the tissue and the wall part is labeled with 0. After standardizing the labeled data, it is input into the deep complex neural network.

[0080] (5) Use the PyTorch framework to build a deep complex neural network with three convolutional layers and two fully connected layers. The first convolutional layer consists of a complex convolutional layer, a batch normalization (BN) layer, a max pooling layer, and a sigmoid activation function. The second convolutional layer has the same structure as the first layer. The difference between the third layer and the first two layers is that the activation function used is the complex activation function (CReLU).

[0081] CReLU = max(0,x) + j*max(0,x)

[0082] The first fully connected layer uses CReLU activation, and the last fully connected layer is the output layer, which uses a linear function activation.

[0083] The input data is a complex vector H = x + iy, and the complex convolution kernel is W = A + iB. The input data is convolved with the complex convolution kernel to obtain:

[0084] W*H=(A*xB*y)+i(B*x+A*y)

[0085] Complex convolution operations in deep complex neural network models are as follows: Figure 3 As shown, the signal features after convolution pass through a max pooling layer. The max pooling layer extracts the most important features for calculating blood flow velocity from the feature matrix. After the main features are activated by a nonlinear activation function, they are linearly calculated by a fully connected layer and finally output.

[0086] (7) Select mean squared error (MSE) as the loss function, stochastic gradient optimization (SGD) as the optimizer, and set the learning rate to 0.0001. After setting the above hyperparameters, train the model for 100 epochs and save the optimal model.

[0087] (8) Load the optimal model from step (7), and evaluate the model using the validation set data partitioned in step (3). The result is shown in the figure below. Figure 5 As shown.

[0088] (9) Load the optimal model in step (7), use the test set data divided in step (3) to perform the test, and input the test data into the model to obtain the blood flow velocity value v.

[0089] (10) The blood flow velocity information v calculated in step (7) of color coding is used to obtain a two-dimensional blood flow velocity image within the field of view. The result is as follows: Figure 7 As shown.

[0090] Because the deep complex neural network model fully considers the low-velocity blood flow around the vessel wall during the training and learning phase, it avoids the problem of incomplete blood flow velocity profile caused by incorrect elimination of low-velocity blood flow, thus improving the accuracy of blood flow velocity measurement results.

[0091] The present invention also provides an ultrasonic blood flow velocity measurement system based on a deep complex neural network, comprising:

[0092] The raw data acquisition module is used to build a physical blood vessel model based on the set simulation parameters and to acquire raw data through the physical blood vessel model; the raw data is the raw blood flow Doppler echo signal.

[0093] The partitioning and labeling module is used to partition the original data into a training set, a validation set, and a test set, and to label the training set.

[0094] The model building module is used to build deep complex neural network models;

[0095] The training module is used to train the deep complex neural network model using the labeled training set;

[0096] The blood flow velocity measurement module is used to measure blood flow velocity using a trained deep complex neural network model.

[0097] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0098] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for measuring ultrasonic blood flow velocity based on a deep complex neural network, characterized in that, include: A physical blood vessel model is built based on the set simulation parameters, and raw data is collected through the physical blood vessel model. The raw data is then subjected to Hilbert transform. The raw data is the raw blood flow Doppler echo signal; The original data is divided into a training set, a validation set, and a test set, and the training set is labeled; specifically, this includes: calculating the maximum blood flow velocity using pre-defined simulation parameters; calculating the blood flow velocity at different locations based on the maximum blood flow velocity; and labeling the training set based on the blood flow velocity at different locations. Construct a deep complex neural network model; The deep complex neural network model is trained using the labeled training set; Blood flow velocity is measured using a trained deep complex neural network model.

2. The ultrasonic blood flow velocity measurement method based on a deep complex neural network according to claim 1, characterized in that, The formula for calculating the maximum blood flow velocity is as follows: Among them, V max Where c is the maximum blood flow velocity, f is the speed of sound, and f is the maximum blood flow velocity. prf f0 is the pulse repetition frequency, f0 is the center frequency, and θ is the beam angle.

3. The ultrasonic blood flow velocity measurement method based on a deep complex neural network according to claim 1, characterized in that, The formulas for calculating blood flow velocity at different locations are as follows: Where v(d) is the blood flow velocity at different locations, d is the distance from the center line of the lumen, R is the radius of the blood vessel, and V max This represents the maximum blood flow velocity.

4. The ultrasonic blood flow velocity measurement method based on a deep complex neural network according to claim 1, characterized in that, The deep complex neural network model consists of a three-layer convolutional layer and a two-layer fully connected layer built using the PyTorch deep learning framework.

5. The ultrasonic blood flow velocity measurement method based on a deep complex neural network according to claim 1, characterized in that, Also includes: The trained deep complex neural network model is validated and tested using the validation set and the test set.

6. An ultrasonic blood flow velocity measurement system based on a deep complex neural network, characterized in that, include: The raw data acquisition module is used to build a physical blood vessel model based on the set simulation parameters, acquire raw data through the physical blood vessel model, and perform Hilbert transform on the raw data; the raw data is the raw blood flow Doppler echo signal. The partitioning and labeling module is used to partition the original data into a training set, a validation set, and a test set, and to label the training set; specifically, it includes: calculating the maximum blood flow velocity using pre-defined simulation parameters; calculating the blood flow velocity at different locations based on the maximum blood flow velocity; and labeling the training set based on the blood flow velocity at different locations. The model building module is used to build deep complex neural network models; The training module is used to train the deep complex neural network model using the labeled training set; The blood flow velocity measurement module is used to measure blood flow velocity using a trained deep complex neural network model.

Images