Closed-loop ultrasonic stimulation system and method based on waking animal single neuron calcium activity

Abstract

The invention discloses a closed-loop ultrasonic stimulation system and method based on the calcium activity of a single neuron of a conscious animal, and belongs to the field of biomedical engineering, and the system comprises an ultrasonic stimulation module, a neuron calcium signal collection module, a neuron calcium signal processing module, and a closed-loop feedback control module. According to the system, neuronal calcium signals of a motor cortex are collected, motion correction, denoising, deconvolution and de-mixing are carried out on the signals, effective neuronal activity data are obtained through recognition, the data are transmitted to a feedback control module and compared with a given threshold value, corresponding ultrasonic parameters are selected according to a comparison result, and ultrasonic stimulation is carried out on a conscious animal. And the neuron signal is in a normal value range. According to the closed-loop ultrasonic stimulation method based on the waking animal single neuron calcium activity, an ultrasonic stimulation module can adjust stimulation parameters according to whether collected neuron calcium activity signals reach a given value or not, and closed-loop adjustment of the neuron calcium activity is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical engineering, and specifically relates to a closed-loop ultrasonic stimulation system, and in particular to a closed-loop ultrasonic stimulation system based on single neuron calcium activity of awake animals. Background Art

[0002] Calcium imaging is a widely used method in neuroscience research. It indirectly reflects the activity of neurons by recording changes in calcium ion concentrations in neurons. This technology has been applied to a variety of experimental environments, including in vitro and in vivo animal models. In particular, calcium imaging studies in awake animals can observe the activity of neurons in natural behavior states in real time, making this technology of great value in studying the relationship between neural circuits, neuronal activity patterns and behavior. Ultrasound technology has received increasing attention in the field of neuromodulation in recent years. Focused ultrasound (FUS) can be used to non-invasively modulate neuronal activity in specific brain regions. Compared with electrical stimulation or optogenetics, ultrasound has deeper tissue penetration and higher spatial selectivity. Therefore, ultrasound neuromodulation is considered to be a highly promising neuromodulatory method that can achieve precise regulation of specific brain regions without damaging tissues. A closed-loop control system refers to a system that dynamically adjusts stimulation parameters through real-time feedback regulation. In neuroscience research, a closed-loop system can adjust the stimulation strategy according to the real-time neural activity of the animal, thereby achieving more precise regulation. Combining calcium imaging with a closed-loop control system, neuronal activity can be monitored in real time and external stimuli can be adjusted through feedback, making neural regulation more precise and personalized.

[0003] Two-photon fluorescence imaging technology is used to detect neuronal calcium signaling activity. The working principle of two-photon laser scanning microscopy (TPLSM) is based on two-photon absorption in nonlinear optical phenomena. Specifically, when the fluorescent molecules in the sample are exposed to high-intensity laser radiation, these molecules can absorb two long-wavelength photons at the same time. The energy of these two photons is added up, causing the electrons in the molecules to transition to an excited state. After an extremely short excited state lifetime, these electrons will emit a photon with a shorter wavelength, thereby generating fluorescence. Two-photon fluorescence microscopy platform for neuronal Ca 2+ Imaging.

[0004] Currently, the existing ultrasound stimulation systems include open-loop ultrasound stimulation systems and closed-loop ultrasound stimulation systems, which are used to perform ultrasound stimulation on the brain areas of awake animals to achieve a regulatory effect.

[0005] Patent CN118436916A discloses a positioning system and method for transcranial ultrasonic neuromodulation. The invention first performs magnetic resonance positioning on the experimenter, plans the position of the ultrasonic transducer, and then uses optical navigation to calibrate and record the position of each component for ultrasonic stimulation to facilitate the next stimulation. Patent CN118356596A discloses an ultrasonic focusing navigation method and related equipment based on type A ultrasonic signals. The invention uses type A ultrasonic signals to assist in focusing ultrasound to locate the target focal area, achieving more accurate and effective focused stimulation. Patent CN118161767A discloses a use of low-intensity pulsed ultrasound to increase blood pressure in hemorrhagic shock. The invention monitors cerebral blood flow by analyzing the changes in blood pressure and heart rate of normal and hemorrhagic hypotensive rats under stimulation with different LIPU parameters, and analyzes the changes in metabolites after neck stimulation through peripheral blood metabolomics, confirming that low-intensity pulsed ultrasound stimulation of the carotid triangle can increase the blood pressure of hemorrhagic hypotensive rats. Patent CN117205457A discloses that by transmitting ultrasound for imaging and receiving echo signals to convert them into corresponding image information, combined with the detection of at least one biological signal of the treatment object, dual positioning functions of imaging and functional science are realized, and the accuracy of positioning and ablation is improved. Patent CN221384935U discloses a wearable ultrasound stimulation device for treating diabetic neuropathy. This invention realizes the effective relief or treatment of diabetic peripheral neuropathy by ultrasound, especially for neuropathy in multiple and irregularly shaped areas. It has good versatility, can be carried with you, and can be used anytime and anywhere according to the needs of the patient, without side effects, and is convenient, safe and effective.

[0006] All of the above inventions belong to open-loop ultrasonic stimulation systems. During the ultrasonic stimulation process, human participation is required to adjust the stimulation parameters. Moreover, these inventions have not realized a closed-loop ultrasonic stimulation system, and it is difficult to achieve the purpose of real-time ultrasonic stimulation regulation of awake animals.

[0007] Patent CN118416414A discloses a portable adaptive non-invasive ultrasonic motor cortex stimulation system. After the invention sends ultrasonic stimulation to the target stimulation site, it collects the acquired motion-evoked electric signals, determines whether the target site is accurately stimulated according to the motion-evoked electric signals, and changes the three-dimensional coordinates of the ultrasonic signal to calibrate the stimulation site when the target site is not accurately stimulated to achieve accurate stimulation of the target site. Patent CN117771544A discloses a cranial nerve stimulation detection device and method based on an adaptive sliding mode control algorithm. The invention performs ultrasonic and electrical stimulation on awake animals, collects neuronal fluorescence signals, obtains electrical signals through a data processing module, and adjusts the intensity of ultrasonic and electrical stimulation signals according to the electrical signals to achieve the purpose of regulating neuronal activity. Patent CN116849680A discloses a closed-loop ultrasonic-light combined visual stimulation system and method. The invention uses light waves to stimulate the visual nerves in combination with ultrasound waves to achieve dual stimulation of the visual nerves, collects physiological electrical signals and uses them as a basis for adjusting stimulation parameters, and efficiently stimulates the visual nerves. Patent CN116637020A discloses a real-time sleep rhythm closed-loop control system based on ultrasonic stimulation of the Shenmen acupoint. The system extracts data features by collecting and processing signals, and feeds back to the ultrasonic stimulation module to adjust the parameters, thereby realizing closed-loop control of EEG in the NREM sleep stage and improving sleep problems. Patent CN116649983A discloses a mouse EEG abnormality detection and ultrasonic stimulation device based on CNN technology, which performs targeted transcranial ultrasonic stimulation on mice with abnormal EEG signals, collects and processes the EEG signals of mice and transmits them to a computer host, which performs filtering, training and application of a convolutional neural network model (CNN), and controls ultrasonic parameters to achieve a closed-loop ultrasonic stimulation system.

[0008] The above patents all belong to closed-loop ultrasonic stimulation systems, which collect physiological signals such as motion-evoked electrical signals, neuronal fluorescence signals or EEG signals, feed them back to the ultrasonic stimulation system, and use them as a basis to adjust the stimulation parameters in real time to achieve appropriate ultrasonic stimulation for awake animals. However, although these patents have closed-loop characteristics, they still have some shortcomings. Most signal detection only stays at the electrical signal level and does not go deep into the neuronal activity level; and a small number of patents that can reach the neuronal level cannot achieve simultaneous stimulation and detection due to device limitations, making it impossible to monitor and evaluate the stimulation effect in real time. Summary of the invention

[0009] The present invention proposes a closed-loop ultrasonic stimulation system based on the calcium activity of single neurons in awake animals, which aims to provide feedback of the activity signals of neurons in awake animals, and form a closed-loop system in conjunction with ultrasonic stimulation. When the calcium signals of neurons in awake animals are abnormal, the system can monitor and collect the calcium activity of neurons in real time, and use a hollow focused ultrasonic transducer to perform ultrasonic stimulation on awake animals, thereby effectively regulating abnormal neuronal conditions.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0011] A closed-loop ultrasonic stimulation system based on single neuron calcium activity of awake animals comprises an ultrasonic stimulation module, a neuron calcium signal acquisition module, a neuron calcium signal processing and analysis module and a closed-loop feedback control module; the output of the ultrasonic stimulation module is connected to the input of the neuron calcium signal acquisition module, the output of the neuron calcium signal acquisition module is connected to the input of the neuron calcium signal processing and analysis module, the output of the neuron calcium signal processing and analysis module is connected to the input of the closed-loop feedback control module, and the output of the closed-loop feedback control module is connected to the input of the ultrasonic stimulation module.

[0012] A further improvement of the technical solution of the present invention is that: the ultrasonic stimulation module is used to generate ultrasonic stimulation to the brain area of ​​conscious animals, and at the same time receives ultrasonic parameters output by a closed-loop feedback control module; the neuron calcium signal acquisition module is used to acquire cerebral cortical neuron calcium signals through a two-photon device; the neuron calcium signal processing and analysis module is used to process the neuron calcium signal data acquired by the neuron calcium signal acquisition module, remove noise and drift through preprocessing, obtain the fluorescence change rate, and analyze the neuron calcium activity; the closed-loop feedback control module is used to analyze and identify the activity of cerebral cortical neurons through the ResNet deep learning algorithm, and then select appropriate ultrasonic parameters to feed back to the ultrasonic stimulation module, so as to adjust the neuron activity of conscious animals to be within a normal range.

[0013] A further improvement of the technical solution of the present invention is that the ultrasonic stimulation module includes two signal generators, a power amplifier and a hollow focused ultrasonic transducer. The signal generator, the power amplifier and the hollow focused ultrasonic transducer are all connected to a computer. The hollow focused ultrasonic transducer is used to perform precise ultrasonic stimulation on the cerebral cortex of awake animals. The two signal generators set the ultrasonic parameters from the closed-loop feedback control module to generate ultrasonic signals, thereby realizing two-level stimulation parameter setting.

[0014] A further improvement of the technical solution of the present invention is that: the signal generator 1 sets the first-level ultrasonic parameters, including: pulse repetition frequency PRF, amplitude 1Amp1, number of cycles 1cyc1, burst period Burst, and square wave waveform; the output signal of the signal generator 1 is transmitted to the signal generator 2; the signal generator 2 sets the second-level ultrasonic parameters, including: fundamental frequency FF, amplitude 2Amp2, number of cycles 2cyc2, and sine wave waveform; the output signal of the signal generator 2 is transmitted to the power amplifier, and the output of the signal generator 2 is connected to the power amplifier; after the power amplifier amplifies the new ultrasonic stimulation generated in real time, it is transmitted to the hollow focused ultrasonic transducer; the hollow focused ultrasonic transducer is placed between the two-photon lens of the two-photon device and the skull window of the mouse, so as to accurately stimulate the target brain area without affecting the two-photon observation of neuronal calcium signal activity.

[0015] A further improvement of the technical solution of the present invention is that: the neuronal calcium signal acquisition module uses ZEN 2 software to control the LSM880 system to take time series pictures. During the neuronal calcium ion imaging process, the mouse remains awake at all times, and the whiskers can move freely for information acquisition. The fluorescent signal is captured by the detector of the microscope and converted into an electrical signal; after the electrical signal is amplified and processed, the neuronal calcium signal changes can be displayed in real time on a computer, and the collected signal data can be saved and transmitted to the neuronal calcium signal processing and analysis module.

[0016] The further improvement of the technical solution of the present invention is that the hardware parameters and software parameters are set to: 20× two-photon lens, 1.0 numerical aperture, 2× zoom magnification, time series mode, resolution 512×512 pixels, laser wavelength of 920nm, and each frame shooting time is 0.465 seconds.

[0017] The technical solution of the present invention is further improved in that: the neuron calcium signal processing and analysis module is used to perform a series of fine preprocessing and feature extraction operations on the collected neuron calcium signals, and use the CaImAn algorithm on the Python platform to perform quantitative analysis and image processing on neuron activity; preprocess the calcium fluorescence video, identify the neuron cell body, reconstruct the activity curve of the target neuron, △F / F0 curve, and determine the Ca by calculating the peak intensity of △F / F0. 2+ This process effectively filters out unnecessary signal components, and the signal parts that are closely related to the calcium activity of specific neurons and have important analytical value are accurately retained and highlighted.

[0018] A closed-loop ultrasound stimulation method based on calcium activity of single neurons in awake animals, comprising the following steps:

[0019] Step S1, collecting raw calcium signals of cerebral cortical neuronal activity of awake animals;

[0020] Step S2, the neuron calcium signal processing and analysis module performs image format conversion, motion correction, noise removal, source extraction, fluorescence change rate calculation, and drift removal on the cerebral cortex neuron activity calcium signal to obtain the processed calcium signal of the neuron activity of the awake animal;

[0021] Step S3, the closed-loop feedback control module compares the calcium signal value after the processing of the neuronal activity of the awake animal with a given value;

[0022] Step S4: If the processed calcium signal value is outside the given range, the deviation between the signal value and the given value is identified and analyzed by the deep learning ResNet model, and appropriate ultrasound parameters are selected and fed back to the ultrasound signal generator;

[0023] Step S5: The ultrasonic device uses the feedback ultrasonic parameters to give ultrasonic stimulation to the awake animal, regulate the neuronal calcium activity signal of the awake animal, and achieve closed-loop ultrasonic stimulation of the calcium activity of a single neuron in the awake animal.

[0024] The further improvement of the technical solution of the present invention is that: the closed-loop feedback control module adopts a deep learning ResNet model to control the calcium activity of neurons in awake animals so that it reaches a given threshold range within a certain period of time; the specific steps are:

[0025] Step S3.1, data preparation and preprocessing: collect data, collect data of different excitement levels from awake animals, increase data diversity by removing noise, blurring, rotating, flipping, scaling, cropping, and color transformation methods, and improve the generalization ability of the model; scale the pixel values ​​of the image data to a uniform range, and normalize the data so that the network can learn better;

[0026] Step S3.2, load and preprocess data: use appropriate data loaders and framework built-in functions to load the preprocessed image data, and divide the data set into training set, validation set and test set;

[0027] Step S3.3, build ResNet-18 model: Import the ResNet-18 architecture into the PyTorch deep learning framework, and adjust the size of the input layer to match the size of the data as needed. Because the output of the control system is a classification label, the fully connected layer of ResNet-18 needs to be retained as the output layer;

[0028] Step S3.4, create loss function and optimizer: select a loss function suitable for the task, use cross entropy loss, and set the optimizer Adam to adjust the network weights to minimize the loss;

[0029] Step S3.5, training model: Train the ResNet-18 model using the training data, monitor the performance on the validation set to avoid overfitting, and adjust the learning rate, batch size, or other hyperparameters as appropriate;

[0030] Step S3.6, test and evaluate the model:

[0031] Evaluate the performance of the model on the test set, calculate its accuracy and other evaluation indicators, analyze the model's prediction results, and determine whether it meets the requirements of the control task;

[0032] Step S3.7, deploy the model: deploy the trained ResNet-18 model to the control system, adjust the control strategy according to the real-time image data, and implement image-based control decisions.

[0033] A further improvement of the technical solution of the present invention is that the operation steps in the neuronal calcium signal processing and analysis module include:

[0034] Step S2.1 Data import and check: The raw calcium signal data collected by the Zeiss two-photon microscope was converted into 16-bit format using ImageJ and then imported into the dedicated data algorithm CaImAn. A preliminary check was performed on the imported data, including confirmation of data integrity, format correctness, and signal intensity range.

[0035] Step S2.2: removing noise: removing noise by filtering technology, and determining the filtering method according to the specific data characteristics;

[0036] Step S2.3 motion correction: align each frame in the acquired video sequence with the reference frame, i.e., the first frame, to eliminate image offset caused by animal motion or imaging device jitter, and then transform each frame image into the same coordinate system as the reference frame through rigid body transformation of translation and rotation;

[0037] Step S2.4 Source extraction: First, image segmentation is performed, using threshold segmentation and edge detection algorithms to separate neurons from the background. Next, region definition is performed to define the neuron region of interest (ROI). Then, feature extraction is performed to extract features related to neuron activity from the ROI, including fluorescence intensity and shape, to accurately identify and locate the position and morphology of neurons, and to extract feature information directly related to neuron activity, providing a basis for subsequent analysis.

[0038] Step S2.5 calculates the fluorescence change rate: determine the basic fluorescence intensity F0, select the average fluorescence intensity in a period of time before stimulation as the basic fluorescence intensity F0; calculate the fluorescence change value ΔF, for each frame image, calculate the difference ΔF between its fluorescence intensity F and F0; calculate the fluorescence change rate ΔF / F0, divide ΔF by F0 to obtain the fluorescence change rate ΔF / F0, which reflects the relative change degree of neuronal activity and quantifies the response degree of neuronal calcium activity to stimulation.

[0039] Due to the adoption of the above technical solution, the technical effects achieved by the present invention are as follows:

[0040] Compared with other open-loop ultrasonic stimulation systems, the closed-loop ultrasonic stimulation system based on single neuron calcium activity of awake animals in the present application can adjust ultrasonic stimulation parameters in real time according to the feedback collected neuron calcium activity signal, so as to achieve closed-loop ultrasonic stimulation of single neuron calcium activity of awake animals; compared with other closed-loop ultrasonic stimulation systems, the feedback signal of the present invention is a neuron calcium activity signal, and a closed-loop ultrasonic stimulation system is formed based on this feedback signal, so that the ultrasonic stimulation effect can be observed more directly from the neuron level.

[0041] The present invention uses the ResNet deep learning algorithm. A large number of neuronal calcium activity signal waveforms are first used to train the deep learning model. The system uses the trained model to identify and analyze the neuronal calcium activity signals fed back by the closed-loop system, and then calculates and adjusts the ultrasonic stimulation parameters based on the results of the identification and analysis, thereby adjusting the ultrasonic stimulation in real time.

[0042] The ultrasonic stimulation module of the present invention adopts a hollow focused ultrasonic transducer, which can accurately stimulate the target brain area and can realize ultrasonic stimulation of awake animals while collecting information data through the hollow focused ultrasonic transducer with a two-photon lens. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a structural block diagram of the closed-loop ultrasonic stimulation system based on calcium activity of single neurons in awake animals according to the present invention;

[0044] Figure 2 It is a flow chart of the closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to the present invention;

[0045] Figure 3 is a top view of an awake animal in the closed-loop ultrasonic stimulation system based on calcium activity of single neurons in awake animals according to the present invention;

[0046] Figure 4 This is a schematic diagram of the physical connection of the closed-loop ultrasonic stimulation system based on calcium activity of single neurons in awake animals according to the present invention;

[0047] Figure 5This is an image of the neuron calcium signal processing and analysis process of the closed-loop ultrasonic stimulation system based on single neuron calcium activity in awake animals described in the present invention;

[0048] Figure 6 It is a waveform diagram of the effect of regulating neuronal calcium activity of animals in resting state and excited state by the closed-loop ultrasonic stimulation system based on single neuron calcium activity of awake animals described in the present invention;

[0049] Among them, 1. Signal generator 1, 2. Signal generator 2, 3. Computer, 4. Power amplifier, 5. Conscious animal, 6. Two-photon lens, 7. Hollow focused ultrasound transducer, 8. Skull window. DETAILED DESCRIPTION

[0050] The present invention is described in detail below with reference to the accompanying drawings

[0051] like Figure 1 As shown, a closed-loop ultrasonic stimulation system based on single neuron calcium activity of awake animals includes an ultrasonic stimulation module, a neuron calcium signal acquisition module, a neuron calcium signal processing and analysis module, and a closed-loop feedback control module; the output of the ultrasonic stimulation module is connected to the input of the neuron calcium signal acquisition module, the output of the neuron calcium signal acquisition module is connected to the input of the neuron calcium signal processing and analysis module, the output of the neuron calcium signal processing and analysis module is connected to the input of the closed-loop feedback control module, and the output of the closed-loop feedback control module is connected to the input of the ultrasonic stimulation module.

[0052] The ultrasonic stimulation module is used to generate ultrasonic stimulation to the brain area of ​​awake animals, and at the same time receives ultrasonic parameters output by the closed-loop feedback control module; the neuron calcium signal acquisition module is used to acquire neuron calcium signals of the cerebral cortex through a two-photon device; the neuron calcium signal processing and analysis module is used to process the neuron calcium signal data acquired by the neuron calcium signal acquisition module, remove noise and drift through preprocessing, obtain the fluorescence change rate, and analyze the neuron calcium activity; the closed-loop feedback control module is used to analyze and identify the activity of cerebral cortex neurons through the ResNet deep learning algorithm, and then select appropriate ultrasonic parameters to feed back to the ultrasonic stimulation module, so as to adjust the neuron activity of awake animals to be within a normal range.

[0053] A closed-loop ultrasound stimulation method based on calcium activity of single neurons in awake animals, specifically comprising the following steps:

[0054] Step S1, collecting raw calcium signals of neuronal activity in the cerebral cortex of awake animals;

[0055] Step S2, converting the image format of the cerebral cortical neuron activity calcium signal, performing motion correction, removing noise, extracting sources, calculating the fluorescence change rate, and removing drift to obtain the processed calcium signal of the neuron activity of awake animals;

[0056] Step S3, comparing the calcium signal value after the neuronal activity processing of the awake animal with a given value;

[0057] Step S4: If the processed calcium signal value is outside the given range, the deviation between the signal value and the given value is identified and analyzed by the deep learning ResNet model, and appropriate ultrasound parameters are selected and fed back to the ultrasound signal generator;

[0058] Step S5, the ultrasonic device uses the feedback ultrasonic parameters to give ultrasonic stimulation to the awake animal, so as to adjust the neuronal calcium activity signal of the awake animal and realize closed-loop ultrasonic stimulation of single neuron calcium activity of the awake animal.

[0059] Fix the awake GCaMP6f transgenic mouse on the stage of the microscope, make sure its head is stable and motionless, and observe the activity of pyramidal neurons.

[0060] The ultrasonic stimulation module is composed of two signal generators, a power amplifier and a hollow focused ultrasonic transducer. The hollow focused ultrasonic transducer is used to perform precise ultrasonic stimulation on the cerebral cortex of awake animals. Two signal generators set the ultrasonic parameters from the feedback control module to generate ultrasonic signals to achieve two-level stimulation parameter setting. Signal generator 1 sets the first-level ultrasonic parameters, including: pulse repetition frequency (PRF), amplitude 1 (Amp1), number of cycles 1 (cyc1), burst period (Burst), and waveform (square wave). The output signal of signal generator 1 is transmitted to signal generator 2. Signal generator 2 sets the second-level ultrasonic parameters, including: fundamental frequency (FF), amplitude 2 (Amp2), number of cycles 2 (cyc2), and waveform (sine wave). The output signal of signal generator 2 is transmitted to power amplifier 4, and the output of signal generator 2 is connected to power amplifier 4. After power amplifier 4 amplifies and generates new ultrasonic stimulation in real time, it is transmitted to hollow focused ultrasonic transducer 7. Hollow focused ultrasonic transducer 7 is placed between the two-photon lens of the two-photon device and the mouse skull window 8, so as to accurately stimulate the target brain area without affecting the two-photon observation of neuronal calcium signal activity. The power amplifier 4 performs amplification to generate new ultrasonic stimulation in real time.

[0061] The neuronal calcium signal acquisition module uses ZEN 2 (black edition) software to control the LSM880 system to capture time series images. The hardware and software parameters are set as follows: 20× two-photon lens, 1.0 numerical aperture, 2× zoom magnification, time series mode, resolution 512×512 pixels, laser wavelength 920nm, and each frame capture time of 0.465 seconds. During the neuronal calcium ion imaging process, the mouse remains awake at all times, and its whiskers can move freely for information collection. The fluorescent signal is captured by the microscope detector and converted into an electrical signal. After the electrical signal is amplified and processed, the changes in the neuronal calcium signal can be displayed in real time on the computer. Figure 5 -(1). Save the acquired signal data and transmit it to the calcium signal processing and analysis module.

[0062] The calcium signal processing and analysis module is used to perform a series of sophisticated preprocessing and feature extraction operations on the collected neuronal calcium signals. The CaImAn (Calcium Imaging Analysis) algorithm is used on the Python platform. This algorithm is a tool for analyzing neuronal calcium imaging data and can be used for quantitative analysis and image processing of neuronal activity. This toolkit can preprocess calcium fluorescence videos, identify neuronal cell bodies, and reconstruct the activity curve (ΔF / F0 curve) of the target neuron. △ The peak intensity of F / F0 is used to determine the maximum value of Ca2+ activity. Through this process, unnecessary signal components are effectively filtered out, while the signal parts that are closely related to the calcium activity of specific neurons and have important analytical value are accurately retained and highlighted. The data quality is significantly improved, which provides an accurate basis for the subsequent in-depth exploration of the calcium activity characteristics of neurons and the study of the regulatory effect of ultrasound stimulation on neuronal calcium activity.

[0063] The operation steps in the calcium signal processing and analysis module include: ① Data import and inspection: The raw calcium signal data collected by the Zeiss two-photon microscope is converted into 16-bit format using ImageJ and then imported into the special data algorithm CaImAn. Perform a preliminary check on the imported data, including confirming data integrity, format correctness, and signal intensity range. ② Noise removal: Calcium signal data often contains various noises, such as global trend changes caused by light source fluctuations and detector response changes. These noises are removed by filtering technology to improve signal quality. The filtering method is determined according to the specific data characteristics to avoid signal distortion caused by excessive smoothing. ③ Motion correction: Align each frame in the acquired video sequence with the reference frame, that is, the first frame, to eliminate image offset caused by animal movement or imaging device jitter. Then, each frame of the image is transformed into the same coordinate system as the reference frame through rigid body transformations such as translation and rotation. Improve the spatial consistency and temporal continuity of the data, providing an accurate basis for subsequent analysis. ④ Source extraction: First, image segmentation is performed, that is, using threshold segmentation, edge detection and other algorithms to separate neurons from the background. Secondly, the region definition automatically defines the region of interest (ROI), which usually corresponds to the cell body or dendrites of the neuron. Then feature extraction extracts features related to neuronal activity from the ROI, such as fluorescence intensity, shape, etc. Accurately identify and locate the position and morphology of neurons, extract feature information directly related to neuronal activity, and provide a basis for subsequent analysis. ⑤ Calculate the fluorescence change rate: Determine the basic fluorescence intensity (F0). Usually the average fluorescence intensity in a period of time before stimulation is selected as the basic fluorescence intensity F0; calculate the fluorescence change value (ΔF). For each frame of the image, calculate the difference ΔF between its fluorescence intensity F and F0; calculate the fluorescence change rate (ΔF / F0): Divide ΔF by F0 to obtain the fluorescence change rate ΔF / F0, which reflects the relative degree of change in neuronal activity. Quantify the response of neuronal calcium activity to stimulation. The final processing and analysis results are as follows: Figure 5 -(2) Neuronal calcium signal results.

[0064] The closed-loop feedback control module is located after the neuronal calcium signal processing and analysis module. It uses the trained deep learning model to match the processed and analyzed calcium signal data features with the pre-set normal range group, low excitability group, and high excitability group. Through the precise matching algorithm, the module can quickly and accurately determine the group to which the current calcium signal data belongs, retrieve the most optimized ultrasound stimulation parameter combination from the preset ultrasound stimulation parameter library, and feed it back to the ultrasound stimulation module to realize a closed-loop ultrasound system.

[0065] The deep learning model used in the closed-loop feedback control module is ResNet-18 (residual network, with a depth of 18 layers), which is a popular convolutional neural network (CNN) architecture that is widely used in image recognition and other visual tasks. The introduction of residual connections solves the gradient vanishing or gradient exploding problems that may occur in the deep network during training, allowing the construction of deeper networks to improve the performance of the model. Usage steps:

[0066] ① Data preparation and preprocessing: Collect data from awake animals with different levels of excitement. Increase the diversity of data by removing noise, blurring, rotating, flipping, scaling, cropping, color conversion, etc. to improve the generalization ability of the model.

[0067] Scale the pixel values ​​of the image data to a uniform range and normalize the data so that the network can learn better; ② Load and preprocess data: Use appropriate data loaders and framework built-in functions to load preprocessed image data, and divide the data set into training set, validation set, and test set; ③ Build a ResNet-18 model: Import the ResNet-18 architecture in the PyTorch deep learning framework, and adjust the size of the input layer as needed to match the size of the data. Because the output of the control system is a classification label, it is necessary to retain the fully connected layer of ResNet-18 as the output layer; ④ Create a loss function and optimizer: Select a loss function suitable for the task, use cross-entropy loss, and set the optimizer Adam to adjust the network weights to minimize the loss; ⑤ Train the model: Train the ResNet-18 model using training data, monitor the performance on the validation set to avoid overfitting, and adjust the learning rate, batch size, or other hyperparameters as appropriate. ⑥ Test and evaluate the model:

[0068] Evaluate the performance of the model on the test set and calculate its accuracy and other evaluation indicators. Analyze the prediction results of the model to determine whether it meets the requirements of the control task. ⑦ Deploy the model: Deploy the trained ResNet-18 model to the control system. Adjust the control strategy according to the real-time image data to achieve image-based control decisions.

[0069] Detection of the regulatory effect of a closed-loop ultrasound stimulation system on calcium activity of single neurons in awake animals

[0070] Resting state data:

[0071] Based on the closed-loop ultrasound stimulation system for single neuron calcium activity in awake animals, the neuron calcium signal activity of resting animals was regulated. After signal acquisition, data processing and analysis, the ΔF / F data was obtained. 16 segments of 10 seconds of data were intercepted and averaged to draw a waveform diagram as shown in the figure below. Figure 6-(1)

[0072] Excited state data:

[0073] Based on the closed-loop ultrasound stimulation system for single neuron calcium activity in awake animals, neuron calcium signal activity in excited animals is regulated. The data of 10 seconds before and after the stimulation are intercepted, and the number of spikes is calculated to draw a bar graph as shown in the figure. Figure 6 -(2), pre and post are the fluorescence change rate of neuronal calcium signal before and after regulation, ΔF / F, and the average number of spikes within 10 s.

[0074] Adjustment effect:

[0075] By the above Figure 6 -(1) The waveform shows that when ultrasound is used to stimulate neurons in the resting state, the neuronal calcium signal ΔF / F0 is low before stimulation, and the ΔF / F0 value after ultrasound stimulation is significantly increased compared with that before stimulation, that is, the neuronal Ca 2+ Activity is significantly enhanced and neuronal excitability is effectively regulated.

[0076] By the above Figure 6 -(2) The bar graph shows that when the neuron is stimulated with ultrasound in the excited state, the number of peaks of neuronal calcium signal ΔF / F0 per unit time after stimulation is lower than that before stimulation, that is, neuronal Ca 2+ Activity is inhibited and neuronal excitability is effectively regulated.

[0077] This indicates that the closed-loop ultrasound stimulation system has the function of regulating neuronal calcium activity, and the specific function is that ultrasound stimulation can enhance the excitability of resting neurons, and ultrasound stimulation can inhibit the excitability of excited neurons.

Claims

1. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals, characterized in that: It includes an ultrasonic stimulation module, a neuron calcium signal acquisition module, a neuron calcium signal processing and analysis module and a closed-loop feedback control module; the output of the ultrasonic stimulation module is connected to the input of the neuron calcium signal acquisition module, the output of the neuron calcium signal acquisition module is connected to the input of the neuron calcium signal processing and analysis module, the output of the neuron calcium signal processing and analysis module is connected to the input of the closed-loop feedback control module, and the output of the closed-loop feedback control module is connected to the input of the ultrasonic stimulation module.

2. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 1, characterized in that: The ultrasonic stimulation module is used to generate ultrasonic stimulation to the brain area of ​​awake animals, and at the same time receives ultrasonic parameters output by the closed-loop feedback control module; the neuron calcium signal acquisition module is used to acquire neuron calcium signals of the cerebral cortex through a two-photon device; the neuron calcium signal processing and analysis module is used to process the neuron calcium signal data acquired by the neuron calcium signal acquisition module, remove noise and drift through preprocessing, obtain the fluorescence change rate, and analyze the neuron calcium activity; the closed-loop feedback control module is used to analyze and identify the activity of cerebral cortex neurons through the ResNet deep learning algorithm, and then select appropriate ultrasonic parameters to feed back to the ultrasonic stimulation module, so as to adjust the neuron activity of awake animals to be within a normal range.

3. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 1, characterized in that: The ultrasonic stimulation module comprises two signal generators, a power amplifier (4) and a hollow focused ultrasonic transducer (7); the signal generator, the power amplifier (4) and the hollow focused ultrasonic transducer (7) are all connected to a computer (3); the hollow focused ultrasonic transducer (7) is used to perform precise ultrasonic stimulation on the cerebral cortex of a conscious animal (5); the two signal generators set the ultrasonic parameters from the closed-loop feedback control module to generate ultrasonic signals, thereby realizing two-level stimulation parameter setting.

4. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 3, characterized in that: The signal generator 1 (1) sets the first-level ultrasonic parameters, including: pulse repetition frequency PRF, amplitude 1Amp1, number of cycles 1cyc1, burst period Burst, and square wave waveform. The output signal of the signal generator 1 (1) is transmitted to the signal generator 2 (2). The signal generator 2 (2) sets the second-level ultrasonic parameters, including: base frequency FF, amplitude 2Amp2, number of cycles 2cyc2, and sine wave waveform. The output signal of the signal generator 2 (2) is transmitted to the power amplifier (4), and the output of the signal generator 2 (2) is connected to the power amplifier (4). After the power amplifier (4) amplifies the new ultrasonic stimulation generated in real time, it is transmitted to the hollow focused ultrasonic transducer (7). The hollow focused ultrasonic transducer (7) is placed between the two-photon lens (6) of the two-photon device and the skull window (8) of the mouse, so as to accurately stimulate the target brain area without affecting the two-photon observation of neuronal calcium signal activity.

5. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 4, characterized in that: The neuronal calcium signal acquisition module uses ZEN 2 software to control the LSM880 system to take time series pictures. During the neuronal calcium ion imaging process, the mouse remains awake at all times, and the whiskers can move freely for information acquisition. The fluorescent signal is captured by the detector of the microscope and converted into an electrical signal; after the electrical signal is amplified and processed, the changes in the neuronal calcium signal can be displayed in real time on the computer, and the collected signal data can be saved and transmitted to the neuronal calcium signal processing and analysis module.

6. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 5, characterized in that: The hardware and software parameters were set as follows: 20× two-photon lens, 1.0 numerical aperture, 2× zoom magnification, timeseries mode, resolution 512×512 pixels, laser wavelength 920nm, and each frame capture time of 0.465 seconds.

7. A closed-loop ultrasound stimulation system based on calcium activity of single neurons in awake animals according to claim 2, characterized in that: The neuron calcium signal processing and analysis module is used to perform a series of fine preprocessing and feature extraction operations on the collected neuron calcium signals, and use the CaImAn algorithm on the Python platform to perform quantitative analysis and image processing on neuron activity; preprocess the calcium fluorescence video, identify the neuron cell body, reconstruct the activity curve of the target neuron, the △F / F0 curve, and calculate △ The peak intensity of F / F0 was used to determine the Ca 2+ This process effectively filters out unnecessary signal components, and the signal parts that are closely related to the calcium activity of specific neurons and have important analytical value are accurately retained and highlighted.

8. A closed-loop ultrasound stimulation method based on calcium activity of single neurons in awake animals, characterized in that: The following steps are involved: Step S1, collecting raw calcium signals of cerebral cortical neuronal activity of awake animals; Step S2, the neuron calcium signal processing and analysis module performs image format conversion, motion correction, noise removal, source extraction, fluorescence change rate calculation, and drift removal on the cerebral cortex neuron activity calcium signal to obtain the processed calcium signal of the neuron activity of the awake animal; Step S3, the closed-loop feedback control module compares the calcium signal value after the processing of the neuronal activity of the awake animal with a given value; Step S4: If the processed calcium signal value is outside the given range, the deviation between the signal value and the given value is identified and analyzed by the deep learning ResNet model, and appropriate ultrasound parameters are selected and fed back to the ultrasound signal generator; Step S5: The ultrasonic device uses the feedback ultrasonic parameters to give ultrasonic stimulation to the awake animal, regulate the neuronal calcium activity signal of the awake animal, and achieve closed-loop ultrasonic stimulation of the calcium activity of a single neuron in the awake animal.

9. The closed-loop ultrasound stimulation method based on calcium activity of single neurons in awake animals according to claim 3, characterized in that: The closed-loop feedback control module adopts the deep learning ResNet model to control the calcium activity of neurons in awake animals so that it reaches a given threshold range within a certain period of time; the specific steps are: Step S3.1, data preparation and preprocessing: collect data, collect data of different excitement levels from awake animals, increase data diversity by removing noise, blurring, rotating, flipping, scaling, cropping, and color transformation methods, and improve the generalization ability of the model; scale the pixel values ​​of the image data to a uniform range, and normalize the data so that the network can learn better; Step S3.2, load and preprocess data: use appropriate data loaders and framework built-in functions to load the preprocessed image data, and divide the data set into training set, validation set and test set; Step S3.3, build ResNet-18 model: Import the ResNet-18 architecture into the PyTorch deep learning framework, and adjust the size of the input layer to match the size of the data as needed. Because the output of the control system is a classification label, the fully connected layer of ResNet-18 needs to be retained as the output layer; Step S3.4, create loss function and optimizer: select a loss function suitable for the task, use cross entropy loss, and set the optimizer Adam to adjust the network weights to minimize the loss; Step S3.5, training model: Train the ResNet-18 model using the training data, monitor the performance on the validation set to avoid overfitting, and adjust the learning rate, batch size, or other hyperparameters as appropriate; Step S3.6, test and evaluate the model: Evaluate the performance of the model on the test set, calculate its accuracy and other evaluation indicators, analyze the model's prediction results, and determine whether it meets the requirements of the control task; Step S3.7, deploy the model: deploy the trained ResNet-18 model to the control system, adjust the control strategy according to the real-time image data, and implement image-based control decisions.

10. The closed-loop ultrasound stimulation method based on calcium activity of single neurons in awake animals according to claim 3, characterized in that: The operation steps in the neuronal calcium signal processing and analysis module include: Step S2.1 Data import and check: The raw calcium signal data collected by the Zeiss two-photon microscope was converted into 16-bit format using ImageJ and then imported into the dedicated data algorithm CaImAn. The imported data was preliminarily checked, including confirming the data integrity, format correctness, and signal intensity range; Step S2.2: removing noise: removing noise by filtering technology, and determining the filtering method according to the specific data characteristics; Step S2.3 motion correction: align each frame in the acquired video sequence with the reference frame, i.e., the first frame, to eliminate image offset caused by animal motion or imaging device jitter, and then transform each frame image into the same coordinate system as the reference frame through rigid body transformation of translation and rotation; Step S2.4 Source extraction: First, image segmentation is performed, using threshold segmentation and edge detection algorithms to separate neurons from the background. Next, region definition is performed to define the neuron region of interest (ROI). Then, feature extraction is performed to extract features related to neuron activity from the ROI, including fluorescence intensity and shape, to accurately identify and locate the position and morphology of neurons, and to extract feature information directly related to neuron activity, providing a basis for subsequent analysis. Step S2.5 calculates the fluorescence change rate: determine the basic fluorescence intensity F0, select the average fluorescence intensity in a period of time before stimulation as the basic fluorescence intensity F0; calculate the fluorescence change value ΔF, for each frame image, calculate the difference ΔF between its fluorescence intensity F and F0; calculate the fluorescence change rate ΔF / F0, divide ΔF by F0 to obtain the fluorescence change rate ΔF / F0, which reflects the relative change degree of neuronal activity and quantifies the response degree of neuronal calcium activity to stimulation.

Images