Individualized variable parameter closed-loop ultrasonic stimulation device and method based on deep learning

Abstract

The invention discloses individualized variable parameter closed-loop ultrasonic stimulation equipment and method based on deep learning, and belongs to the technical field of transcranial ultrasonic stimulation. The individualized variable parameter closed-loop ultrasonic stimulation equipment comprises an electroencephalogram signal acquisition module, an electroencephalogram signal processing module, an electroencephalogram signal deep learning judgment module and an ultrasonic stimulation module; the electroencephalogram signal acquisition module acquires electroencephalogram signals of an experimental subject within a specified finite time and transmits the electroencephalogram signals to the electroencephalogram signal processing module; the electroencephalogram signal processing module preprocesses and displays the electroencephalogram signals in real time, splits the electroencephalogram signals according to specified time, and stores all electroencephalogram signal segments in a picture format; the electroencephalogram signal deep learning discrimination module discriminates an electroencephalogram abnormal level and appoints effective ultrasonic stimulation parameters; and the ultrasonic stimulation module performs variable parameter ultrasonic stimulation on the experimental subject. According to the method, deep learning training is carried out on large sample electroencephalogram signals, an electroencephalogram signal deep learning model is established, the electroencephalogram abnormal level of any other experimental subject is judged, and variable parameter closed-loop ultrasonic stimulation is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of transcranial ultrasonic stimulation, and in particular to an individualized variable parameter closed-loop ultrasonic stimulation device and method based on deep learning. Background Art

[0002] In recent years, brain stimulation technologies, such as deep brain stimulation, optogenetic stimulation, and transcranial magnetic stimulation, have developed rapidly (Krauset et al., 2016; Milosevic et al., 2018; Zhang et al., 2018). Among them, transcranial ultrasound stimulation (TUS) is a brain regulation technology that uses specific ultrasound to act on the cranial nervous system, produces biomechanical effects on neurons, affects brain metabolism and neural electrical activity, and thus causes a series of physiological and biochemical reactions. It has two results on cranial nerves: excitation and inhibition, with or without damage, high spatial resolution, and high penetration depth (Bystritsky et al., 2011; Fomenko et al., 2018). TUS has attracted great attention and has been used in a large number of animal and human experiments (Bystritsky and Korb, 2015; Niu et al., 2018; Yu et al., 2019). Compared with open-loop brain stimulation, closed-loop brain stimulation can regulate brain nerves in real time through brain states, achieving targeted stimulation on demand (Muller et al., 2012; Cagnan et al., 2017; Zhou et al., 2019). With the self-developed ultrasound stimulation module, the experimental subjects can be subjected to multi-parameter adjustable ultrasound stimulation. Deep learning aims to learn the inherent laws and representation levels of sample data. The information obtained during the learning and training process is of great help in the interpretation of data such as text, images and sounds. The ultimate goal of deep learning is to enable machines to have analytical learning capabilities like humans and to be able to recognize data such as text, images and sounds. When judging the abnormal level of EEG of experimental subjects, ordinary manual judgment methods require a lot of time and effort. The deep learning training model of convolutional neural network CNN (LecunY et al., 1989; Zhou SS et al., 2013) is used to analyze the EEG signals of experimental subjects collected in real time, which can quickly and accurately judge the abnormal level of EEG of experimental subjects and specify effective ultrasound stimulation parameters. Therefore, we need an individualized variable parameter closed-loop ultrasound stimulation device and method based on deep learning. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an individualized variable parameter closed-loop ultrasonic stimulation device and method based on deep learning. By performing deep learning training on a large sample of EEG signals, a deep learning model of EEG signals is established, the EEG abnormality level of any other experimental subject is judged, and variable parameter closed-loop ultrasonic stimulation is performed.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A deep learning-based individualized variable parameter closed-loop ultrasonic stimulation device, comprising an electroencephalogram signal acquisition module, an electroencephalogram signal processing module, an electroencephalogram signal deep learning discrimination module and an ultrasonic stimulation module; The EEG signal acquisition module is used to acquire EEG signals of the experimental subject within a specified limited time according to a specified sampling frequency and transmit them to the EEG signal processing module; The EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; The EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the EEG abnormality level of the experimental subject and specifies effective ultrasonic stimulation parameters; The ultrasonic stimulation module performs variable parameter closed-loop ultrasonic stimulation on the experimental object.

[0005] A further improvement of the technical solution of the present invention is that the EEG signal acquisition module includes an electrode module, a front-end amplifier, a microelectrode AC amplifier, a neural signal processor, Ethernet communication and an FPGA control module; the electrode module acquires the original EEG signal of the experimental subject; the front-end amplifier receives and amplifies the EEG signal sent by the electrode module, and sends the amplified EEG signal to the microelectrode AC amplifier; the microelectrode AC amplifier receives the EEG signal sent by the front-end amplifier and amplifies and filters it, and sends the processed EEG signal to the neural signal processor; the neural signal processor is used to convert the EEG signal into a corresponding digital signal, and send the converted digital signal to the FPGA control module; the FPGA control module is used to control the acquisition time and frequency, and transmit the digital signal converted by the neural signal processor to the EEG signal processing module via Ethernet.

[0006] A further improvement of the technical solution of the present invention is that the EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting the EEG signals according to specified time, and finally saving all EEG signal segments as picture format.

[0007] A further improvement of the technical solution of the present invention lies in that: the EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the highest EEG abnormality level corresponding to all EEG signal segments of the experimental subject, and sends the effective ultrasonic stimulation parameters corresponding to the highest EEG abnormality level to the ultrasonic stimulation module through serial port communication; the deep learning training model needs to iteratively learn a large sample of EEGs of each EEG abnormality level in advance and obtain a higher discrimination accuracy, and also needs to use a large number of new EEG segments to verify the discrimination accuracy of the model; the effective ultrasonic stimulation parameters are derived from animal experiments and clinical trial data.

[0008] The further improvement of the technical solution of the present invention is that: the ultrasonic stimulation module includes a signal generating module, a driving module, an E-class power amplifier module, an L-type impedance matching module, an ultrasonic transducer, a collimator and an acoustic coupling agent; the signal generating module generates a PWM waveform corresponding to the duty cycle and time length according to the specified ultrasonic stimulation parameters using the output comparison function of the timer peripheral of the single-chip microcomputer, and the ultrasonic stimulation parameters include the fundamental frequency FF, the pulse width PD, the pulse repetition frequency PRF, the stimulation duration SD, the stimulation interval ISI, the total stimulation time TT, and the sound intensity I; the specified parameter PWM waveform controls the fast switching of the field effect tube in the E-class power amplifier module through the driving module, so as to achieve high-efficiency power amplification; the L-type impedance matching module matches the higher impedance of the customized ultrasonic transducer to the smaller load design impedance of the E-class power amplifier module; the ultrasonic transducer receives the power-amplified electrical signal and generates an ultrasonic stimulation signal of the specified parameters; the ultrasonic stimulation signal is transmitted to the target stimulation position of the experimental object through the collimator and the acoustic coupling agent.

[0009] A method for individualized variable parameter closed-loop ultrasound stimulation based on deep learning, comprising the following steps: S1, the EEG signal acquisition module collects the EEG signals of the experimental subject within a specified limited time according to the specified sampling frequency and transmits them to the EEG signal processing module; S2, the EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; S3, the EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the highest EEG abnormality level corresponding to all EEG signal segments of the experimental subject, and sends the effective ultrasound stimulation parameters corresponding to the highest EEG abnormality level to the ultrasound stimulation module through serial port communication; S4. The ultrasonic stimulation module performs ultrasonic stimulation on the experimental object according to the specified effective ultrasonic stimulation parameters.

[0010] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is: 1. The present application provides an individualized variable parameter closed-loop ultrasonic stimulation device and method based on deep learning. The EEG signal acquisition module collects the EEG signals of the experimental subject within a specified limited time according to the specified sampling frequency and transmits them to the EEG signal processing module; the EEG signal processing module is used for preprocessing and real-time display of EEG signals, and splits the EEG signals according to the specified time, and finally saves all EEG signal fragments as image format; the EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the EEG abnormality level of the experimental subject and specify effective ultrasonic stimulation parameters; the ultrasonic stimulation module performs variable parameter closed-loop ultrasonic stimulation on the experimental subject.

[0011] 2. The device of the present invention can be applied to a variety of scenarios. It uses a deep learning training model to analyze the EEG signals of the experimental subjects collected in real time, and can quickly and accurately determine the level of EEG abnormality of the experimental subjects and perform effective ultrasonic stimulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a block diagram of the principle of an individualized variable parameter closed-loop ultrasound stimulation device based on deep learning; Figure 2A , Figure 2B These are the EEG signals of epileptic mice and normal mice after treatment; Figure 3A , Figure 3B , Figure 3C These are the EEG signals of epileptic mice in the early, middle and late stages of closed-loop ultrasound stimulation; Figure 4 It is the mathematical model of a single residual block of the ResNet network; Figure 5 This is the basic schematic diagram of the Class E power amplifier module in the ultrasonic stimulation module. DETAILED DESCRIPTION

[0013] The present invention is further described in detail below with reference to the accompanying drawings and embodiments: like Figure 1 The figure shows a principle block diagram of the individualized variable parameter closed-loop ultrasonic stimulation device based on deep learning of the present application. The individualized variable parameter closed-loop ultrasonic stimulation device based on deep learning includes an EEG signal acquisition module, an EEG signal processing module, an EEG signal deep learning discrimination module and an ultrasonic stimulation module; The EEG signal acquisition module is used to collect EEG signals of the experimental subject within a specified limited time according to a specified sampling frequency and transmit them to the EEG signal processing module; The EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; The EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the EEG abnormality level of the experimental subject and specify effective ultrasound stimulation parameters; The ultrasonic stimulation module performs variable parameter closed-loop ultrasonic stimulation on the experimental object.

[0014] In this embodiment, the EEG signal acquisition module includes an electrode module, a front-end amplifier, a microelectrode AC amplifier, a neural signal processor, Ethernet communication and an FPGA control module; the electrode module acquires the original EEG signal of the epileptic mouse; the front-end amplifier receives and amplifies the EEG signal sent by the electrode module, and sends the amplified EEG signal to the microelectrode AC amplifier; the microelectrode AC amplifier adopts the model of Microelectrode AC Amplifier\MODEL1800, which is used to receive the EEG signal sent by the front-end amplifier and amplify and filter it, and send the processed EEG signal to the neural signal processor; the neural signal processor adopts the Cerebus-128 neural signal processor of Blackrock Microsystems of the United States, which is used to convert the EEG signal into a corresponding digital signal, and send the converted digital signal to the FPGA control module; the FPGA control module has a built-in Xilinx series Zynq UltraScale+MPSoC heterogeneous multi-core processing SoC chip, which is used to control the acquisition time and acquisition frequency, and transmit the digital signal converted by the neural signal processor to the EEG signal processing module via Ethernet.

[0015] In this embodiment, the EEG signal processing module first pre-processes the received EEG signal, including intercepting the low-frequency EEG in the specified frequency band, removing the 5Hz baseline drift and the 50Hz power frequency interference, then displays the EEG signal in the host computer interface, and then splits the EEG signal according to the 5s time length, and finally saves all the EEG signal segments as image format. Figures 2 and 3 are partial EEG signal segments after processing.

[0016] In this embodiment, the EEG signal deep learning training model selects the ResNet residual network as the basic network, selects the cross-entropy loss function (Cross-Entropy Loss) as the loss function, and selects the stochastic gradient descent function (Stochastic Gradient Descent, SGD) as the optimization function. The basic structure of the Resnet residual network consists of multiple residual blocks (class Residual). The mathematical model of a single ResNet residual block is as follows Figure 4As shown in the figure, x is the input of the residual block, the Weight Layer is the weight layer, F(x) is the output before the introduction of the "shortcut branch", and W(x) is the final output of the residual block after the introduction of the "shortcut branch". The EEG signal deep learning model needs to iteratively train all EEG signal fragments of different EEG abnormality levels of the processed large sample experimental subjects in advance. The model will update the learning parameters according to the gap between the autonomous deduction results and the given standard results. Through iterative learning of large sample EEG fragments, the model's discrimination accuracy for the training picture set gradually increases. When the accuracy exceeds 95%, the model's discrimination accuracy is considered to meet the standard. It is also necessary to use a large number of new EEG fragments to verify the discrimination accuracy of the model. If the discrimination accuracy of the verification picture set is greater than the discrimination accuracy of the training picture set, the verification is considered to have passed. The experimental subjects are epileptic mice. Examples of EEG segments learned by the model are shown in Figures 2 and 3. The learned features are mainly the number of spikes in 1-4 Hz delta waves, 4-8 Hz theta waves, and 8-13 Hz alpha waves. The more spikes there are, the higher the level of epileptic EEG abnormality is. The EEG abnormality level is classified into normal and 1-5 levels of EEG abnormality. Variable parameter ultrasound stimulation is performed on a large sample of epileptic mice with different levels of EEG abnormality in advance, and the improvement effect of the epileptic mice after stimulation is observed. The effective ultrasound stimulation parameters are recorded in grades, and the corresponding relationship between the EEG abnormality level and the effective ultrasound stimulation parameters is established. The EEG signal deep learning training model discriminates the EEG abnormality level for all newly collected EEG segments, and sends the effective ultrasound stimulation parameters corresponding to the highest EEG abnormality level to the ultrasound stimulation module through serial communication.

[0017] In this embodiment, the ultrasonic stimulation module includes a signal generating module, a driving module, an E-class power amplifier module, an L-type impedance matching module, an ultrasonic transducer, a collimator and an acoustic coupling agent; the signal generating module has a built-in STM32F103C8T6 chip, and generates a PWM waveform corresponding to the duty cycle and time length according to the specified ultrasonic stimulation parameters by using the output comparison function of the chip's timer peripheral; the ultrasonic stimulation parameters include base frequency FF, pulse width PD, pulse repetition frequency PRF, stimulation duration SD, stimulation interval ISI, total stimulation time TT, and sound intensity I; the sound intensity I can be adjusted by the working voltage of the E-class power amplifier module; the driving module adopts a UCC27525DR field effect transistor gate driving chip, and the specified parameter PWM waveform controls the fast switching of the field effect transistor in the E-class power amplifier module through the driving module, thereby achieving high-efficiency power amplification; Figure 5 The basic schematic diagram of the class E power amplifier module includes a drive signal V in, field effect tube Q1, working power supply VDD, choke coil RFC, parallel capacitor C1, series resonant capacitor Cr, series resonant inductor Lr, phase shift inductor Lx, standard design load R load ; The field effect tube model in the class E power amplifier module is N-channel IRFP240PBF; the L-type impedance matching module matches the higher impedance of the customized ultrasonic transducer to the smaller load design impedance of the class E power amplifier module; the ultrasonic transducer uses a focused ultrasonic transducer of model SF25-0.5C40-FD50-F2.0-R9, with a center frequency of 0.5MHz, a chip diameter of 40mm, a focusing depth of 50mm, and an impedance of 500 ohms, which converts the received high-power electrical signal into an ultrasonic stimulation signal of specified parameters; the ultrasonic stimulation signal is transmitted to the target stimulation position of the epileptic mouse through a collimator and an acoustic coupling agent.

[0018] The device and method for using individualized variable parameter closed-loop ultrasound stimulation based on deep learning specifically include the following steps: S1 EEG signal acquisition module collects EEG signals of the experimental subject within a specified limited time according to the specified sampling frequency and transmits them to the EEG signal processing module; The S2 EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; The S3 EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the highest EEG abnormality level corresponding to all EEG signal segments of the experimental subject, and sends the effective ultrasound stimulation parameters corresponding to the highest EEG abnormality level to the ultrasound stimulation module through serial port communication; The S4 ultrasonic stimulation module performs ultrasonic stimulation on the experimental object according to the specified effective ultrasonic stimulation parameters.

[0019] After ultrasound stimulation, the EEG signals of epileptic mice changed. The changed EEG signals were identified again to determine the new abnormal EEG level and whether ultrasound stimulation was needed. If stimulation was still needed, a new set of ultrasound stimulation parameters was specified based on the effect of reducing the abnormal EEG level of epileptic mice, thus forming an individualized variable parameter closed-loop ultrasound stimulation based on deep learning. Figure 3A , Figure 3B , Figure 3C These are the EEG signals of epileptic mice in the early, middle and late stages of closed-loop ultrasound stimulation. The results show that after a period of closed-loop ultrasound stimulation, the severity of epilepsy in mice gradually decreased, and closed-loop ultrasound stimulation has a significant effect on improving the severity of epilepsy in mice.

[0020] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A personalized variable parameter closed-loop ultrasound stimulation device based on deep learning, characterized in that: It includes an EEG signal acquisition module, an EEG signal processing module, an EEG signal deep learning discrimination module and an ultrasonic stimulation module; The EEG signal acquisition module is used to acquire EEG signals of the experimental subject within a specified limited time according to a specified sampling frequency and transmit them to the EEG signal processing module; The EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; The EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the EEG abnormality level of the experimental subject and specifies effective ultrasonic stimulation parameters; The ultrasonic stimulation module performs variable parameter closed-loop ultrasonic stimulation on the experimental object.

2. The deep learning-based individualized variable parameter closed-loop ultrasound stimulation device according to claim 1, characterized in that: The EEG signal acquisition module includes an electrode module, a front-end amplifier, a microelectrode AC amplifier, a neural signal processor, an Ethernet communication and an FPGA control module; The electrode module collects the original EEG signals of the experimental subjects; the front-end amplifier receives and amplifies the EEG signals sent by the electrode module, and sends the amplified EEG signals to the microelectrode AC amplifier; the microelectrode AC amplifier receives the EEG signals sent by the front-end amplifier and amplifies and filters them, and sends the processed EEG signals to the neural signal processor; the neural signal processor is used to convert the EEG signals into corresponding digital signals, and send the converted digital signals to the FPGA control module; the FPGA control module is used to control the acquisition time and frequency, and transmit the digital signals converted by the neural signal processor to the EEG signal processing module via Ethernet.

3. The individualized variable parameter closed-loop ultrasound stimulation device based on deep learning according to claim 1, characterized in that: The EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as picture format.

4. The deep learning-based individualized variable parameter closed-loop ultrasound stimulation device according to claim 1, characterized in that: The EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the highest EEG abnormality level corresponding to all EEG signal segments of the experimental subject, and sends the effective ultrasonic stimulation parameters corresponding to the highest EEG abnormality level to the ultrasonic stimulation module through serial port communication; the deep learning training model needs to iteratively learn a large number of sample EEGs of each EEG abnormality level in advance and obtain a high discrimination accuracy, and also needs to use a large number of new EEG segments to verify the discrimination accuracy of the model; the effective ultrasonic stimulation parameters are derived from animal experiments and clinical trial data.

5. The deep learning-based individualized variable parameter closed-loop ultrasound stimulation device according to claim 1, characterized in that: The ultrasonic stimulation module includes a signal generation module, a driving module, an E-class power amplifier module, an L-type impedance matching module, an ultrasonic transducer, a collimator and an acoustic coupling agent; the signal generation module generates a PWM waveform corresponding to the duty cycle and time length according to the specified ultrasonic stimulation parameters by using the output comparison function of the timer peripheral of the single-chip microcomputer, and the ultrasonic stimulation parameters include the fundamental frequency FF, the pulse width PD, the pulse repetition frequency PRF, the stimulation duration SD, the stimulation interval ISI, the total stimulation time TT, and the sound intensity I; the specified parameter PWM waveform controls the fast switching of the field effect tube in the E-class power amplifier module through the driving module, so as to achieve high-efficiency power amplification; the L-type impedance matching module matches the higher impedance of the customized ultrasonic transducer to the smaller load design impedance of the E-class power amplifier module; the ultrasonic transducer receives the power-amplified electrical signal and generates an ultrasonic stimulation signal with specified parameters; the ultrasonic stimulation signal is transmitted to the target stimulation position of the experimental object through the collimator and the acoustic coupling agent.

6. A method for individualized variable parameter closed-loop ultrasound stimulation based on deep learning, characterized in that: Using the individualized variable parameter closed-loop ultrasound stimulation device based on deep learning as described in any one of claims 1 to 5 comprises the following steps: S1, the EEG signal acquisition module collects the EEG signals of the experimental subject within a specified limited time according to the specified sampling frequency and transmits them to the EEG signal processing module; S2, the EEG signal processing module is used for preprocessing and real-time display of EEG signals, splitting EEG signals according to specified time, and finally saving all EEG signal segments as image format; S3, the EEG signal deep learning discrimination module uses the EEG signal deep learning model to discriminate the highest EEG abnormality level corresponding to all EEG signal segments of the experimental subject, and sends the effective ultrasound stimulation parameters corresponding to the highest EEG abnormality level to the ultrasound stimulation module through serial port communication; S4. The ultrasonic stimulation module performs ultrasonic stimulation on the experimental object according to the specified effective ultrasonic stimulation parameters.

Images