Ultrasonic image nakagami model training method and ultrasonic image temperature estimation method

Abstract

The application discloses an ultrasonic image Nakagami model training method and an ultrasonic image temperature estimation method, which are methods for estimating temperature by using ultrasonic images of biological tissues; firstly, in the method, ultrasonic images and corresponding temperature information at different temperatures are collected through microwave ablation experiments of biological tissues; then, the ultrasonic images are subjected to filtering processing to remove noise interference, and an envelope image is obtained by using Hilbert transform; subsequently, the characteristic parameters of the Nakagami model are calculated on the envelope image by using a sliding window method; then, the calculated characteristic parameters and the collected ultrasonic image temperatures are used as inputs and outputs of a neural network, and a back propagation neural network model is trained; for a to-be-measured ultrasonic image sample, the characteristic parameters of the Nakagami model are calculated, and then the characteristic parameters are input into the trained back propagation neural network model, so that the temperature estimation result of the to-be-measured sample can be obtained.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, and specifically relates to a method for processing the temperature of medical ultrasound images. More specifically, it is a method for calculating the temperature of ultrasound images using the Nakagami model and neural networks. Background Technology

[0002] In cancer treatment, microwave hyperthermia has been proven to effectively compensate for the shortcomings of surgery, chemotherapy, and radiotherapy alone, and can effectively improve the remission rate of malignant tumors. However, the dosage of hyperthermia remains an issue. Temperature monitoring is crucial during hyperthermia; if the temperature is too low, it will not kill the tumor, while if the temperature is too high, it will damage surrounding normal tissues while killing tumor cells. Currently, the main clinical technique for measuring temperature in tumor hyperthermia is invasive, involving inserting temperature sensors such as thermocouples, thermistors, and optical fibers into the area to be measured. This method is damaging to tissues, causing patient discomfort, and is difficult to obtain the temperature field distribution of the tissue, and is susceptible to interference from high-frequency electromagnetic waves. Ultrasonic thermometry, on the other hand, has advantages such as minimal harm to the human body, providing deep imaging of the body, good real-time performance, low cost, and compatibility with hyperthermia equipment. It has been proven to be a non-invasive method for temperature monitoring during tumor hyperthermia.

[0003] The principle of ultrasonic thermometry is to infer the body temperature based on the changes in the acoustic properties of biological tissue before and after heating (acoustic properties include sound velocity, nonlinear parameters, and scattering amplitude). Numerous studies have shown that the degree of change in the probability distribution of ultrasonic backscattered signals is related to temperature. Statistical models used to analyze ultrasonic backscattered signals mainly include the Rayleigh distribution, Nakagami distribution, and zero-difference K-distribution. The Nakagami distribution model, which includes pre-Rayleigh, post-Rayleigh, and post-Rayleigh distributions, can more comprehensively describe the statistical situation of various backscattered signals that may occur within biological tissues, and its calculation is relatively simple. The high accuracy of ultrasonic thermometry is not only related to the instrument's manufacturing process but also to the relationship between acoustic property parameters and temperature. Therefore, a high degree of fit between acoustic property parameters and temperature is particularly important. However, traditional methods mainly use the least squares method for fitting, which is based on a linear model. Applying this method to the fitting of ultrasonic data will produce significant errors, limiting its clinical use. Summary of the Invention

[0004] The purpose of this invention is to provide a method for training a Nakagami model of ultrasound images and a method for estimating the temperature of ultrasound images, for non-invasive temperature monitoring in microwave hyperthermia, and to solve the problems of low fitting degree and large estimation error caused by the use of least squares method for fitting in the prior art.

[0005] The technical solution provided by this invention is as follows:

[0006] According to a first aspect of this disclosure, the present invention provides a method for training a Nakagami model of ultrasound images, comprising the following steps:

[0007] Step 1: Acquire ultrasound images and corresponding temperature information at different temperatures through microwave ablation experiments on ex vivo biological tissues;

[0008] Step 2: The ultrasound images acquired in Step 1 are processed using median filtering and thresholding to remove noise and interference; then, a Hilbert transform is performed on the ROIs in the ultrasound images to obtain the envelope image.

[0009] Step 3: Calculate the Nakagami model feature parameters m and ARCN of the envelope image obtained in Step 2 using the sliding window method. ARCN is the absolute value of the ratio transformation of m.

[0010] Step 4: Build a backpropagation neural network model, which includes an input layer, hidden layers, and an output layer; use the Nakagami model feature parameters ARCN calculated in Step 3 as the input of the neural network, and the corresponding temperature data as the output of the neural network, and train the model to obtain the backpropagation neural network model.

[0011] In one exemplary embodiment of this disclosure, the microwave ablation experiment of biological tissue in step 1 is as follows: Fresh, isolated pork tenderloin is selected as the biological tissue, and the initial temperature of the pork is 25°C; the isolated pork is heated using a radiofrequency ablation device, temperature data during the heating process is acquired using a temperature measuring device, and ultrasound image data is acquired using an ultrasonic measuring device; the time of the temperature measuring device and the data acquisition time of the ultrasonic measuring device are calibrated to be consistent; and ultrasound images and corresponding temperature data are acquired throughout the heating process through the experiment.

[0012] In one exemplary embodiment of this disclosure, in step 2, the threshold range is 0-200.

[0013] In an exemplary embodiment of this disclosure, in step 3, the window size is set to three times the pulse length of the ultrasonic transducer, and the sliding window moves across the entire envelope image with a step size of 50% window overlap pixels, and the feature parameter m within the ROI region is calculated; the ARCN parameter is calculated using the initial temperature of the pork tissue as a reference temperature to reflect the relationship between the change in the envelope statistical distribution and the temperature.

[0014] In one exemplary embodiment of this disclosure, in step 4, the number of nodes in the input layer and output layer of the neural network is 1, the number of neurons in the intermediate hidden layer is set to 15, and the learning rate LR is set to 0.01.

[0015] According to a second aspect of this disclosure, the present invention further provides a method for estimating the temperature of an ultrasound image using the above-mentioned trained Nakagami model of an ultrasound image. The method includes the following steps: for an ultrasound image sample to be tested, firstly, its characteristic parameters are calculated according to step 3, and then the characteristic parameters are input into the backpropagation neural network model trained in step 4 to obtain the Nakagami model temperature estimation result of the sample to be tested.

[0016] The beneficial effects of this invention are as follows: This invention discloses a method for training a Nakagami model for ultrasound images and a method for estimating the temperature of ultrasound images, which is a method for estimating the temperature using ultrasound images of biological tissues. First, in this method, ultrasound images and corresponding temperature information at different temperatures are acquired through microwave ablation experiments on biological tissues. Then, the ultrasound images are filtered to remove noise interference, and the envelope image is obtained using Hilbert transform. Next, the sliding window method is used to calculate the feature parameters of the Nakagami model on the envelope image. Then, the calculated feature parameters and the acquired ultrasound image temperature are used as the input and output of the neural network to train a backpropagation neural network model. For the ultrasound image sample to be tested, its Nakagami model feature parameters are calculated, and then the feature parameters are input into the trained backpropagation neural network model to obtain the temperature estimation result of the sample to be tested. Attached Figure Description

[0017] Figure 1 This is a flowchart of the Nakagami model training method for ultrasound images according to the present invention.

[0018] Figure 2 These are images before and after the ultrasound image denoising process in steps 1 and 2 of this invention.

[0019] Figure 3 This is a schematic diagram of the ROI selected in step 2 of the present invention.

[0020] Figure 4 This is a diagram showing the neural network fitting results of this invention.

[0021] Figure 5 This is a graph showing the least squares fitting result in the existing technology.

[0022] Figure 6 This is a flowchart of the method for estimating temperature in ultrasound images using the Nakagami model of ultrasound images, as per the present invention.

[0023] Figure 7 This is a comparison chart of the temperature estimated by the neural network of this invention and the actual temperature measured by the thermocouple.

[0024] Figure 8 This is a diagram of the experimental measurement system of the present invention. Detailed Implementation

[0025] Example 1

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and mechanisms are omitted in the following description.

[0027] The present invention provides a method for estimating temperature using a Nakagami model based on a neural network for ultrasound images. This method estimates temperature using ultrasound images of biological tissues. First, the method acquires ultrasound images and corresponding temperature information at different temperatures through microwave ablation experiments on biological tissues. Next, the ultrasound images are filtered to remove noise interference, and the envelope image is obtained using Hilbert transform. Then, the sliding window method is used to calculate the feature parameters of the Nakagami model on the envelope image. Finally, the calculated feature parameters and the acquired ultrasound image temperatures are used as the input and output of the neural network to train a backpropagation neural network model.

[0028] like Figure 1 The diagram shows a flowchart of the training method of the present invention; the method of the present invention specifically includes the following steps:

[0029] Step 1: Ultrasonic images and corresponding temperature information at different temperatures were acquired through microwave ablation experiments on ex vivo biological tissues.

[0030] Specifically, the microwave ablation experiment of biological tissue in step 1 is as follows:

[0031] Because pork tissue is similar in composition to human tissue and is easy to obtain, the biological tissue used in this invention is fresh, excised pork tenderloin. The initial temperature of the pork is 25°C, and the size of the pork is 10cm*5cm*5cm.

[0032] Setting up an experimental measurement system, such as Figure 8 As shown; the equipment used in the microwave ablation experiment mainly includes radiofrequency ablation equipment (with radiofrequency ablation needle), temperature measurement equipment (thermocouple and matching temperature probe), and ultrasonic measurement equipment; this experiment consists of two steps:

[0033] (1) Fixation of biological tissue and placement of instruments: First, select a smooth plane and place the excised pork horizontally. Then, select the center position of the pork tissue and insert the radiofrequency ablation needle horizontally to a depth of about 2 cm. At a horizontal distance of about 1 cm from the radiofrequency ablation needle, insert the temperature probe of the thermocouple horizontally to a depth of about 2 cm. The needle tip of the ablation needle and the temperature probe of the thermocouple are on the same horizontal plane. Finally, apply coupling agent to the surface of the pork tissue and place the ultrasound probe vertically and close to the surface of the pork tissue. During the acquisition of experimental data, the positions of the ablation needle, temperature probe, ultrasound probe and pork tissue remain unchanged.

[0034] (2) Experimental data acquisition: First, the time of the temperature measurement system and the data acquisition time of the ultrasound image (ultrasound measurement equipment) were aligned; then, the temperature measurement system, the ultrasound measurement system and the microwave ablation equipment were turned on simultaneously, and the microwave ablation power was set to 5W and the ablation time to 6 minutes. During this period, the ultrasound image data was continuously acquired until the ablation was completed. The ultrasound images and their corresponding temperature data during the entire heating process were obtained through this experiment.

[0035] Because the ultrasound images acquired in step 1 contained many white streaks (which, upon examination, were identified as pork fascia tissue), as well as the tips of microwave ablation needles and thermocouple temperature probes, such as Figure 2 As shown in (a), this greatly interferes with the calculation of the Nakagami feature parameters. Therefore, the present invention performs the following processing.

[0036] Step 2: The ultrasound images acquired in Step 1 are processed using median filtering and thresholding to remove noise and interference. In a preferred embodiment, the threshold range is 0-200. The processed ultrasound images are shown below. Figure 2 As shown in (b);

[0037] Then, a Hilbert transform was performed on the ultrasound image to obtain the envelope image by selecting a Region of Interest (ROI). The ROIs selected in this experiment are as follows: Figure 3 As shown.

[0038] The Hilbert transform is further explained below:

[0039] The Hilbert transform first requires the construction of an analytic signal. The purpose is to transform a real signal into a complex signal. Let's assume the signal is... Substituting the signal into the analytical signal yields:

[0040]

[0041] In the formula It is a complex carrier signal. For a complex envelope, the absolute value of the analytic signal is the envelope signal:

[0042]

[0043] Step 3: Calculate the Nakagami model feature parameters m and ARCN of the envelope image obtained in Step 2 using the sliding window method. ARCN is the absolute value of the ratio transformation of m. In a preferred embodiment: set the window size to three times the pulse length of the ultrasonic transducer, and move the sliding window across the entire envelope image with a step size of 50% window overlap pixels to calculate the feature parameters m within the ROI region. Using the initial temperature of pork tissue as the reference temperature (25℃), calculate the ARCN parameter to reflect the relationship between the change in the envelope statistical distribution and the temperature.

[0044] In step 3, the probability density function of the Nakagami distribution model is:

[0045]

[0046] In the formula and These are the gamma function and the unit step function, respectively. Represents statistical expectation. Scale parameters associated with the Nakagami distribution. The Nakagami parameter m can be represented as:

[0047]

[0048]

[0049] The ARCN expression used to reflect envelope changes is:

[0050]

[0051] In the formula and These are the Nakagami shape parameters m at the reference temperature and the current temperature, respectively. This represents the multiplication factor that changes as a function of the initial value of m. and Proportional.

[0052] Step 4: Construct a backpropagation neural network model. This neural network has three layers, including an input layer, a hidden layer, and an output layer. Use the Nakagami model feature parameters ARCN calculated in Step 3 as the input of the neural network and the corresponding temperature data as the output of the neural network. After training, a backpropagation neural network model with good fit is obtained. In a preferred embodiment, the number of nodes in the input layer and the output layer of this model is 1, the number of neurons in the intermediate hidden layer is set to 15, and the learning rate LR is set to 0.01.

[0053] like Figure 4 The figure shown is a graph illustrating the fitting degree of the neural network model of this invention; as shown... Figure 5 The figure shown is a diagram of the fitting results of existing polynomial models based on the least squares method; by comparison, it can be clearly seen that the neural network model of the present invention has a higher fitting degree.

[0054] This invention further provides a method for estimating the temperature of ultrasound images using the Nakagami model of the trained ultrasound images described above, such as... Figure 6 The flowchart shown is as follows. The method includes the following steps: For the ultrasound image sample to be tested, firstly calculate its feature parameters according to step 3, and then input the feature parameters into the backpropagation neural network model trained in step 4 to obtain the Nakagami model temperature estimation result of the sample to be tested.

[0055] The evaluation metric for the temperature estimation performance of the Nakagami model is relative error. The relative error of the neural network model estimation is compared with that of traditional polynomial fitting estimation, as shown in Table 1 below.

[0056] Table 1. Relative errors of polynomial fitting and neural network temperature estimation

[0057]

[0058] Figure 7 The graph shows a comparison between the temperature estimated by the neural network of this invention and the actual temperature measured by the thermocouple. It can be seen that the average error of the temperature estimation is within 1℃. Overall, the method of this invention has significant advantages and can improve the accuracy of ultrasonic temperature measurement.

[0059] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

Claims

1. An ultrasound image Nakagami model training method, characterized by, The method comprises the following steps: Step 1: Collecting ultrasonic images and corresponding temperature information at different temperatures through an in-vitro microwave ablation experiment of biological tissue; the microwave ablation experiment of biological tissue is as follows: fresh in-vitro pork tenderloin is selected as the biological tissue, and the initial temperature of the pork is 25 DEG C; a radio frequency ablation device is used to heat the in-vitro pork, a temperature measuring device is used to obtain temperature data in the heating process of the in-vitro pork, and an ultrasonic measuring device is used to obtain ultrasonic image data; the temperature measuring device time is corrected to be consistent with the ultrasonic measuring device data acquisition time; through the experiment, ultrasonic images and corresponding temperature data in the whole heating process are obtained; Step 2: The ultrasonic images collected in step 1 are processed by using the median filtering and threshold setting method to remove noise and interference; then, a ROI is selected on the ultrasonic image to obtain an envelope image through Hilbert transform; Step 3: The Nakagami model feature parameter m and ARCN of the envelope image obtained in step 2 are calculated by using the sliding window method, and the ARCN is the absolute value of the m value ratio transformation; Step 4: A back propagation neural network model is built, and the neural network comprises an input layer, a hidden layer and an output layer; The Nakagami model feature parameter ARCN calculated in step 3 is taken as the input of the neural network, and the corresponding temperature data is taken as the output of the neural network, and the back propagation neural network model is obtained after training. 2.The method of training an ultrasound image Nakagami model according to claim 1, characterized in that, In step 2, the threshold range is 0-200. 3.The method of training an ultrasound image Nakagami model according to claim 1, characterized in that, In step 3, the size of the window is set to be three times the length of the ultrasonic transducer pulse, the sliding window moves on the whole envelope image with a step of 50% window overlap rate pixels, and the feature parameter m in the ROI region is calculated; The initial temperature of the pork tissue is taken as the reference temperature to calculate the ARCN parameter, which is used to reflect the relationship between the change of envelope statistical distribution and temperature. 4.The method of claim 1, wherein, In step 4, the number of nodes of the input layer and the output layer of the neural network is 1, the number of neurons in the middle hidden layer is set to 15, and the learning rate LR is set to 0.

01.

5. An ultrasound image temperature estimation method, characterized by, The ultrasonic image Nakagami model obtained by the training method of any one of claims 1-4 comprises the following steps: for the ultrasonic image sample to be measured, first, the feature parameter thereof is calculated according to step 3, Then, the feature parameter is input into the back propagation neural network model trained in step 4, and the Nakagami model temperature estimation result of the sample to be measured can be obtained.

Images