Tumor motion prediction method and system based on electromyographic signals

Abstract

The invention belongs to the technical field of medical signal processing, and particularly relates to a tumor motion prediction method and system based on electromyographic signals. The system comprises an image guiding module which is configured to collect an electromyographic signal and tumor movement of a patient, and extract a respiration-related signal from the electromyographic signal by adopting a gating method, template subtraction, envelope calculation and offline filtering in sequence; a cross-correlation function is adopted to calculate correlation coefficients and look-ahead time of the respiratory correlation signals and tumor movement; the treatment delivery module is configured to collect an electromyographic signal of a patient and extract a respiration related signal from the electromyographic signal by adopting a gating method, template subtraction, envelope calculation and online filtering in sequence; and obtaining a prediction result of tumor movement based on the look-ahead time calculated by the image guiding module. The invention further provides a method for predicting tumor movement by applying the system. The method can compensate the system delay when the radiotherapy equipment tracks the tumor movement, and has a good application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical signal processing, and in particular relates to a tumor motion prediction method and system based on electromyographic signals. Background Art

[0002] Tumor radiotherapy is a local treatment method that uses radiation to treat tumors. Radiation includes α, β, and γ rays produced by radioactive isotopes and x-rays, electron beams, proton beams, and other particle beams produced by various x-ray therapy machines or accelerators. Radiotherapy also has a lethal effect on normal tissues, so the accuracy of radiotherapy is a very important topic in this field. However, chest and abdominal tumors will move with the patient's breathing, making it difficult to locate the tumor during radiotherapy. In addition, there are similar problems for non-tumor lesions. The lesions move with breathing, which will adversely affect the accuracy of robotic surgery. Therefore, respiratory movement is a common problem that affects the accuracy of chest and abdominal tumor radiotherapy and robotic surgery.

[0003] In the prior art, tumor motion tracking technology is mainly based on traditional images such as anatomical imaging systems, optical imaging systems, and electromagnetic imaging systems (for example, Chinese patent application: CN110168657A-Tumor tracking using intelligent tumor size change notification). This method is limited to tracking the historical tumor motion position that has already occurred or the position of the tumor motion reference object. After determining the tumor position, the radiotherapy accelerator or surgical robot also needs to determine the tumor position and adjust the treatment plan in real time, which means that there may be a large deviation between the tumor position tracked by traditional images and the tumor position treated by the accelerator irradiation or surgical robot.

[0004] Therefore, how to predict tumor movement faster and eliminate this deviation is an important topic. EMG signals have a 320ms look-ahead time relative to respiratory movement; the accelerator response time reported in clinical cases is mostly around 300ms, which means that EMG signals have the potential to compensate for the accelerator response time and predict the future position of lesion movement in advance, thereby performing more accurate lesion movement tracking. However, there is still a lack of relevant research in the existing technology, and it has not been possible to convert EMG signals into predictive information on lesion location. Summary of the invention

[0005] In view of the problems in the prior art, the present invention provides a method and system for predicting tumor motion based on electromyographic signals.

[0006] A tumor motion prediction system based on electromyographic signals, comprising:

[0007] The image guidance module is configured to: collect the patient's electromyographic signal and tumor motion, extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation and an offline filtering in sequence; calculate the correlation coefficient and the prospective time between the respiration-related signal and the tumor motion by using a cross-correlation function;

[0008] The treatment delivery module is configured to: collect the patient's electromyographic signal, and extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation, and online filtering in sequence; when the correlation coefficient is higher than a preset value, based on the look-ahead time calculated by the image-guided module, obtain a real-time prediction result of the upcoming position and movement trajectory of the tumor by sliding the respiration-related signal on the time axis.

[0009] Preferably, the gating method and template subtraction are used to clean the electromyographic signal and remove electrocardiographic signal artifacts.

[0010] Preferably, the envelope calculation is used to convert the electromyographic signal from which electrocardiographic signal artifacts are removed, to obtain the envelope of the electromyographic signal to represent the amplitude variation in the electromyographic signal.

[0011] Preferably, the offline filtering and the online filtering respectively smooth the envelope of the electromyographic signal to obtain the breathing-related signal in the electromyographic signal.

[0012] Preferably, the method of off-line filtering is selected from moving average smoothing.

[0013] And / or, the online filtering method is selected from online smoothing based on recurrent neural network.

[0014] Preferably, the recurrent neural network is selected from LSTM or GRU.

[0015] Preferably, during the online smoothing process, the output of the recurrent neural network is restricted in combination with the constraint conditions of spline interpolation.

[0016] Preferably, the preset value of the correlation coefficient is 0.9.

[0017] The present invention also provides a method for predicting tumor motion using the above-mentioned tumor motion prediction system based on electromyographic signals, comprising the following steps:

[0018] Step 1, in the image-guided stage, the patient's electromyographic signal and tumor motion are collected, and a gating method, template subtraction, envelope calculation and offline filtering are used in sequence to extract a respiratory-related signal from the electromyographic signal; a cross-correlation function is used to calculate a correlation coefficient and a look-ahead time between the respiratory-related signal and the tumor motion;

[0019] Step 2, during the treatment delivery phase, the patient's electromyographic signals are collected, and the respiration-related signals are extracted from the electromyographic signals using a gating method, template subtraction, envelope calculation, and online filtering in sequence; when the correlation coefficient is higher than a preset value, based on the look-ahead time calculated by the image-guided module, a real-time prediction result of the upcoming position and motion trajectory of the tumor is obtained by sliding the respiration-related signals on the time axis.

[0020] The present invention also provides a computer-readable storage medium, on which is stored: a computer program for implementing the above-mentioned tumor motion prediction system based on electromyographic signals, or a computer program for implementing the above-mentioned method.

[0021] The present invention aims to predict tumor movement using electromyographic signals, designs a process for data processing of electromyographic signals, and finally realizes the prediction of tumor movement through the prospective time of the respiration-related signal and tumor movement in the electromyographic signals. The present invention realizes the operation of predicting tumor movement using electromyographic signals for the first time. Compared with the tumor movement tracking technology based on traditional images (anatomical imaging, electromagnetic imaging, optical imaging), the tumor movement tracking technology based on electromyographic signals can reflect the future position of tumor movement in advance, which is conducive to solving the problem of reduced tumor movement tracking accuracy caused by the delay of radiotherapy equipment system. Therefore, the present invention has a good application prospect.

[0022] Obviously, according to the above contents of the present invention, in accordance with common technical knowledge and customary means in the art, without departing from the above basic technical ideas of the present invention, other various forms of modification, replacement or change may be made.

[0023] The above contents of the present invention are further described in detail below through specific implementation methods in the form of embodiments. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following examples. All technologies realized based on the above contents of the present invention belong to the scope of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of the process of Example 1 of the present invention;

[0025] Figure 2 1 and 2 are example diagrams of the data processing results of each step in Example 1 of the present invention, wherein (a) is the original data of the diaphragm surface electromyography signal, (b) is the electromyography signal after removing the electrocardiogram interference, (c) is the electromyography envelope signal, and (d) is the comparison between the electromyography predicted breathing signal and the actual breathing signal. DETAILED DESCRIPTION

[0026] It should be noted that the algorithms of data collection, transmission, storage and processing steps not specifically described in the embodiments, as well as the hardware structure, circuit connection, etc. not specifically described can all be implemented through the contents disclosed in the prior art.

[0027] Example 1 Tumor motion prediction method and system based on electromyographic signals

[0028] The system of this embodiment includes:

[0029] The image guidance module is configured to: collect the patient's electromyographic signal and tumor motion, extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation and an offline filtering in sequence; calculate the correlation coefficient and the prospective time between the respiration-related signal and the tumor motion by using a cross-correlation function;

[0030] The treatment delivery module is configured to: collect the patient's electromyographic signal, and extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation, and online filtering in sequence; when the correlation coefficient is higher than a preset value, based on the look-ahead time calculated by the image-guided module, obtain a real-time prediction result of the upcoming position and movement trajectory of the tumor by sliding the respiration-related signal on the time axis.

[0031] The process of using the above system to predict tumor motion is as follows: Figure 1 As shown, the specific steps include:

[0032] Step 1, in the image-guided stage, the patient's electromyographic signal and tumor motion are collected, and a gating method, template subtraction, envelope calculation and offline filtering are used in sequence to extract a respiratory-related signal from the electromyographic signal; a cross-correlation function is used to calculate a correlation coefficient and a look-ahead time between the respiratory-related signal and the tumor motion;

[0033] Step 2, during the treatment delivery phase, collect the patient's electromyographic signal, and use the gating method, template subtraction, envelope calculation and online filtering in sequence to extract the respiration-related signal from the electromyographic signal; when the correlation coefficient is higher than the preset value, based on the look-ahead time calculated by the image-guided module, the respiration-related signal is slid on the time axis to obtain a real-time prediction result of the upcoming position and movement trajectory of the tumor. If the correlation coefficient is lower than the preset value, remeasure after the patient's breathing stabilizes. As a preferred method, the preset value of the correlation coefficient is 0.9.

[0034] The above sliding operation is performed according to the following formula:

[0035] S sEMG [t] = S resp [t+t precursor ]

[0036] That is, the electromyographic signal S at the current time t sEMG , and calculate t+t precursor Breathing exercises resp , t precursor is the look-ahead time calculated in step 1.

[0037] In the above-mentioned electromyographic signal processing flow, the gating method and template subtraction are used to clean the electromyographic signal and remove the electrocardiographic signal artifacts. The envelope calculation is used to convert the electromyographic signal with the electrocardiographic signal artifacts removed to obtain the envelope of the electromyographic signal to represent the amplitude change in the electromyographic signal. The offline filtering and online filtering respectively smooth the envelope of the electromyographic signal to obtain the respiration-related signal in the electromyographic signal.

[0038] in,

[0039] The gating method was used to remove the PR interval component from the EMG signal, while the template subtraction method was used to remove the QT interval component, T wave, and U wave ( Figure 1 c):

[0040] S sEMG [t R -Δt g :t R ]=S sEMG [t R -2Δt g :t R -Δt g ]#(1)

[0041] The gating method refers to using a short stable signal to replace the signal interfered by the ECG signal. sEMG represents the electromyographic signal, t R represents the time point corresponding to the R peak, Δt g is the gating window. In this embodiment, the PR interval Δt g The data of the window with the length of Δt replaces the data of the PR interval to remove the interference of the ECG signal. g was set to 50 ms, which exceeds the 40 ms time delay for R-wave peak detection summarized in the review literature.

[0042] S sEMG [t R :t R +Δt m ]=S sEMG [t R :t R +Δt m ]-ηS ECG [t R :t R +Δtm ]#(2)

[0043] Among them, S ECG represents the average ECG signal obtained from the 10-second breath-hold task data, Δt m refers to the template subtraction window, and η corresponds to the number of S in each ECG cycle. sEMG [t R ] and S ECG [t R In this embodiment, Δt m is set to 500 milliseconds.

[0044] In order to eliminate the polarity change in sEMG, this embodiment calculates the envelope of the sEMG signal after removing the ECG. i :

[0045]

[0046] Where L refers to the envelope calculation window. In this study, L is set to 25 milliseconds, which does not exceed the gating window Δt1, thus avoiding the introduction of additional time delay. i represents the current time point, and j represents the end time point of the envelope calculation window.

[0047] On the basis of the above data processing, in step 1, the offline filtering method is selected from moving average smoothing to obtain a stable look-ahead time. Offline filtering can provide more accurate noise removal, which is convenient for accurate calculation of correlation coefficient and look-ahead time. Step 2 is an online intervention in the treatment process, which requires the use of an online filtering algorithm to reflect the real-time characteristics. The online filtering method is selected from online smoothing based on a recurrent neural network (RNN) to meet the real-time prediction requirements. The recurrent neural network is selected from LSTM or GRU, and the present embodiment is preferably a GRU with a relatively fast calculation speed. In the online smoothing process, since the RNN predicted value is a discrete variable, the predicted value will fluctuate greatly. Therefore, the output of the recurrent neural network is restricted in combination with the constraints of the spline interpolation to achieve the purpose of outputting data with smooth characteristics.

[0048] The results of the above data processing process are shown in the following example: Figure 2 shown.

[0049] The technical solution of the present invention is further illustrated by experiments below.

[0050] Experimental Example 1 Comparison of the difference in the calculation results of the forward time of electromyographic signals and respiratory movements in different time periods I. Experimental Method

[0051] Myoelectric signals and respiratory movements were acquired from 15 healthy adults using paired Ag-AgCl electrodes (Noraxon, Scottsdale, AZ, USA) at a sampling rate of 2000 Hz and a pressure sensor integrated in an elastic waistband (Anzai Medical, Tokyo, Japan) at a sampling rate of 40 Hz, respectively. For further analysis, the respiratory cycle curves were interpolated to 2000 Hz. Pairs of measurement electrodes were placed in a bilateral long configuration along the subcostal region, and the reference electrode was placed on the sternum, as shown in Figure 2. Figure 1 (a) The subjects were instructed to perform a breath-holding task of approximately 10 seconds to capture the template of the ECG signal, followed by a 10-minute free breathing task in the supine position.

[0052] The EMG signals were processed using MATLAB code to extract the respiration-related signals. The original diaphragm surface EMG signals were preprocessed using a 1-500 Hz bandpass filter and a 50 Hz notch filter to obtain the preliminarily preprocessed EMG signals.

[0053] Other experimental steps and settings were carried out according to the method described in Example 1.

[0054] 2. Experimental Results

[0055] The results are shown in Table 1, where r is the correlation coefficient between the EMG signal and the respiratory movement after the full 10-minute preprocessing. Δt0 is the prospective time between the EMG signal and the respiratory movement after the full 10-minute preprocessing. Δt1 and Δt2 are the prospective times calculated from the first 1 minute (during radiotherapy, this time period corresponds to step 1 of Example 1) and the last 9 minutes of data (during radiotherapy, this time period corresponds to step 2 of Example 1), respectively. There is no significant statistical difference between Δt1 and Δt2, indicating that the prospective time calculated in the image-guided stage of radiotherapy in Example 1 can be applied in the treatment delivery stage of radiotherapy.

[0056] Table 1

[0057]

[0058] It can be seen from the above embodiments and experimental examples that the present invention constructs a method and system for predicting tumor movement using electromyographic signals, which can compensate for the system delay of radiotherapy equipment in tracking tumor movement in the prior art and has good application prospects.

Claims

1. A tumor motion prediction system based on electromyographic signals, characterized in that: include: The image guidance module is configured to: collect the patient's electromyographic signal and tumor motion, extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation and an offline filtering in sequence; calculate the correlation coefficient and the prospective time between the respiration-related signal and the tumor motion by using a cross-correlation function; The treatment delivery module is configured to: collect the patient's electromyographic signal, and extract the respiration-related signal from the electromyographic signal by using a gating method, a template subtraction method, an envelope calculation, and online filtering in sequence; when the correlation coefficient is higher than a preset value, based on the look-ahead time calculated by the image-guided module, obtain a real-time prediction result of the upcoming position and movement trajectory of the tumor by sliding the respiration-related signal on the time axis.

2. The tumor motion prediction system based on electromyographic signals according to claim 1, characterized in that: The gating method and template subtraction method are used to clean the electromyographic signal and remove electrocardiographic signal artifacts.

3. The tumor motion prediction system based on electromyographic signals according to claim 1, characterized in that: The envelope calculation is used to convert the electromyographic signal from which the electrocardiographic signal artifacts are removed, and obtain the envelope of the electromyographic signal to represent the amplitude change in the electromyographic signal.

4. The tumor motion prediction system based on electromyographic signals according to claim 1, characterized in that: The off-line filtering and the on-line filtering respectively smooth the envelope of the electromyographic signal to obtain the respiration-related signal in the electromyographic signal.

5. The tumor motion prediction system based on electromyographic signals according to claim 1 or 4, characterized in that: The method of offline filtering is selected from moving average smoothing; And / or, the online filtering method is selected from online smoothing based on recurrent neural network.

6. The tumor motion prediction system based on electromyographic signals according to claim 5, characterized in that: The recurrent neural network is selected from LSTM or GRU.

7. The tumor motion prediction system based on electromyographic signals according to claim 5, characterized in that: During the online smoothing process, the output of the recurrent neural network is restricted in combination with the constraint conditions of spline interpolation.

8. The tumor motion prediction system based on electromyographic signals according to claim 1, characterized in that: The preset value of the correlation coefficient is 0.

9.

9. A method for predicting tumor motion using the tumor motion prediction system based on electromyographic signals as described in any one of claims 1 to 8, characterized in that: The steps include: Step 1, in the image-guided stage, the patient's electromyographic signals and tumor motion are collected, and a gating method, template subtraction, envelope calculation and offline filtering are used in sequence to extract a respiration-related signal from the electromyographic signals; Using a cross-correlation function to calculate the correlation coefficient and the look-ahead time between the respiration-related signal and the tumor motion; Step 2, during the treatment delivery phase, the patient's electromyographic signals are collected, and the respiration-related signals are extracted from the electromyographic signals using a gating method, template subtraction, envelope calculation, and online filtering in sequence; when the correlation coefficient is higher than a preset value, based on the look-ahead time calculated by the image-guided module, a real-time prediction result of the upcoming position and motion trajectory of the tumor is obtained by sliding the respiration-related signals on the time axis.

10. A computer-readable storage medium, characterized in that: Stored thereon are: a computer program for implementing the tumor motion prediction system based on electromyographic signals as described in any one of claims 1-8, or a computer program for implementing the method as described in claim 9.

Images