High-frequency electrosurgical unit control device and control system

Abstract

This disclosure presents embodiments of a high-frequency electrosurgical unit control device and control system. One specific embodiment of the device includes: a control unit configured to control a high-frequency power generator to output high-frequency electrical energy; a data acquisition unit configured to perform multi-dimensional data acquisition of human tissue; a signal acquisition unit configured to acquire sound waves and extract features from the sound wave signals; a feature extraction unit configured to extract features from tissue impedance information sequences and human tissue image sequences respectively; a feature fusion unit configured to fuse sound wave feature tensors, tissue impedance feature tensors, and human tissue feature tensors; a generation unit configured to generate high-frequency electrosurgical unit control commands; and an adjustment unit configured to adjust the output power and modulate the waveform of the high-frequency electrical energy. This embodiment can reduce excessive thermal damage during surgery or improve cutting and coagulation efficiency.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of automatic control technology, specifically to a high-frequency electrosurgical unit control device and control system. Background Technology

[0002] A high-frequency electrosurgical unit (HFEMU) is an electronic surgical device that uses high-frequency, high-voltage current to cut and coagulate biological tissues, and it is widely used in various surgeries. Its basic principle is to use the high-frequency, high-voltage current generated at the tip of the effective electrode to cause the contacted tissue cells to rupture, vaporize, or coagulate in a very short time, thereby achieving the purpose of cutting or hemostasis. Currently, the common method for controlling the operation of a HFEMU is as follows: the surgeon performs the corresponding surgical cutting or coagulation operation based on a preset power mode (such as pure cutting, mixed cutting, or electrocoagulation).

[0003] However, when using the above method, the following technical problems often arise:

[0004] Traditional fixed power output or simple manual adjustment modes cannot adapt to tissue changes and are prone to causing excessive thermal damage or insufficient cutting and coagulation efficiency during surgery (for example, excessive power may lead to excessive carbonization, adhesion or deep thermal damage to tissue; while insufficient power may lead to low cutting efficiency or inadequate coagulation). Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure provide a high-frequency electrosurgical unit control device and control system to solve the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a high-frequency electrosurgical unit control device, comprising: a control unit configured to, in response to receiving a start command characterizing the start of operation of the high-frequency electrosurgical unit, control a high-frequency power generator to output high-frequency electrical energy to a surgical electrode based on preset initial operating parameters of the high-frequency electrosurgical unit; a data acquisition unit configured to perform multi-dimensional data acquisition on human tissue in the target surgical area using an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence and a human tissue image sequence; a signal acquisition unit configured to acquire sound waves generated by the surgical electrode acting on the human tissue using a sound wave signal acquisition device, to obtain a sound wave signal, and to extract features from the sound wave signal to obtain a sound wave feature tensor; and a feature extraction unit configured to extract features from the tissue impedance information. Feature extraction is performed on the information sequence and the aforementioned human tissue image sequence to obtain a tissue impedance feature tensor and a human tissue feature tensor, respectively. A feature fusion unit is configured to perform feature fusion processing on the aforementioned acoustic feature tensor, the aforementioned tissue impedance feature tensor, and the aforementioned human tissue feature tensor to obtain a fused feature tensor. A generation unit is configured to generate high-frequency electrosurgical control commands based on the aforementioned fused feature tensor, wherein the aforementioned high-frequency electrosurgical control commands include: an output power sequence and a waveform modulation parameter sequence. An adjustment unit is configured to adjust the output power of the aforementioned high-frequency power generator based on the aforementioned output power sequence, and to control the waveform modulation module to perform waveform modulation on the high-frequency electrical energy output by the aforementioned high-frequency power generator based on the aforementioned waveform modulation parameter sequence, so as to realize closed-loop control of the high-frequency electrosurgical unit.

[0008] Secondly, some embodiments of this disclosure provide a high-frequency electrosurgical control system, which includes: a high-frequency power generator, a waveform modulation module, an impedance sensor, an optical imaging sensor, an acoustic signal acquisition device, a main controller, and surgical electrodes, wherein: the high-frequency power generator is electrically connected to the surgical electrodes; the waveform modulation module is communicatively connected to the high-frequency power generator; the high-frequency power generator, the waveform modulation module, the impedance sensor, the optical imaging sensor, and the acoustic signal acquisition device are all communicatively connected to the main controller; the acoustic signal acquisition device is used to acquire acoustic signals generated by tissue vaporization and rupture when high-frequency electrical energy is applied to human tissue; the optical imaging sensor is used to acquire micro-signals under the surface of human tissue. The system observes structural images to provide early warning of subcutaneous blood vessels. The waveform modulation module receives waveform modulation parameters from the main controller and modulates the high-frequency power output from the high-frequency power generator. The impedance sensor monitors the tissue impedance value and its changes between the surgical electrode and the tissue circuit in real time. The high-frequency power generator receives the output power from the main controller and outputs high-frequency power of corresponding amplitude. The surgical electrode receives the modulated high-frequency power output from the high-frequency power generator and applies the modulated high-frequency power to the human tissue. The main controller receives data collected by each sensor, runs a control algorithm based on the data collected by each sensor, and sends control commands to the high-frequency power generator and the waveform modulation module.

[0009] The various embodiments of this disclosure have the following beneficial effects: the high-frequency electrosurgical control device of some embodiments of this disclosure can reduce excessive thermal damage during surgery or improve cutting and coagulation efficiency. Specifically, the reason for excessive thermal damage or insufficient cutting and coagulation efficiency during surgery is that traditional fixed power output or simple manual adjustment mode cannot adapt to tissue changes and is prone to causing excessive thermal damage or insufficient cutting and coagulation efficiency during surgery (for example, excessive power may lead to excessive carbonization, adhesion, or deep thermal damage to tissue; excessive power may lead to low cutting efficiency or insufficient coagulation). Based on this, the high-frequency electrosurgical control device of some embodiments of this disclosure firstly, has a control unit configured to, in response to receiving a start command characterizing the start of high-frequency electrosurgical operation, control a high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on preset initial operating parameters of the high-frequency electrosurgical. Thus, cutting and coagulation operations on human tissue can be performed based on a conventional fixed high-frequency electrosurgical working mode and fixed power. Then, a data acquisition unit is configured to perform multi-dimensional data acquisition on the human tissue of the target surgical area through an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence and a human tissue image sequence. Therefore, a tissue impedance information sequence characterizing the dynamic changes in the electrical properties of human tissue under the action of electric current can be obtained, as well as a human tissue image sequence characterizing the microstructure beneath the tissue surface. Next, the signal acquisition unit is configured to acquire the sound waves generated by the surgical electrodes acting on the human tissue using a sound wave signal acquisition device, obtaining sound wave signals, and extracting features from these sound wave signals to obtain sound wave feature tensors. This allows for the acquisition of sound wave information and the conversion from raw sound waves to physical acoustic features. Next, the feature extraction unit is configured to extract features from the tissue impedance information sequence and the human tissue image sequence respectively, obtaining tissue impedance feature tensors and human tissue feature tensors. This allows for the conversion of impedance information and tissue information from raw data to feature information. Then, the feature fusion unit is configured to perform feature fusion processing on the sound wave feature tensors, the tissue impedance feature tensors, and the human tissue feature tensors to obtain fused feature tensors. This yields a fused feature tensor, which aligns and fuses features from different physical dimensions (electric, optical, and acoustic) to form a unified, more comprehensive fused feature tensor. The generation unit is then configured to generate high-frequency electrosurgical control commands based on this fused feature tensor. These commands include an output power sequence and a waveform modulation parameter sequence. This allows for the generation of control commands corresponding to the actual working scenario and requirements.Finally, the adjustment unit is configured to adjust the output power of the high-frequency power generator according to the aforementioned output power sequence, and to control the waveform modulation module to modulate the high-frequency electrical energy output by the high-frequency power generator according to the aforementioned waveform modulation parameter sequence, thereby achieving closed-loop control of the high-frequency electrosurgical unit. Thus, the amplification gain of the high-frequency power generator can be adjusted in real time according to the output power, changing the intensity of the output energy, and the operating mode of the waveform modulation module can be controlled in real time according to the waveform modulation parameters, changing the form of the output energy. The two work together to enable the high-frequency electrical energy output to the surgical electrode to dynamically and adaptively match the current actual tissue state and surgical requirements, thereby reducing excessive thermal damage during surgery or improving cutting and coagulation efficiency (automatically enhancing coagulation effect upon encountering blood vessels). Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0011] Figure 1 This is a schematic diagram of an application scenario of a high-frequency electrosurgical unit control device according to some embodiments of this disclosure;

[0012] Figure 2 This is a flowchart of some embodiments of the high-frequency electrosurgical control device according to the present disclosure;

[0013] Figure 3 This is a module structure diagram of a preset depth feature extraction module according to some embodiments of the high-frequency electrosurgical control device of this disclosure;

[0014] Figure 4 This is a flowchart illustrating the feature fusion of acoustic wave feature tensor, tissue impedance feature tensor, and human tissue feature tensor according to some embodiments of the high-frequency electrosurgical control device disclosed herein;

[0015] Figure 5 This is a system component structure diagram of the high-frequency electrosurgical control system according to this disclosure;

[0016] Figure 6 This is a schematic diagram of the structure of some embodiments of the high-frequency electrosurgical control device according to the present disclosure. Detailed Implementation

[0017] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0018] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0020] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0021] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0022] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Figure 1 This is a schematic diagram of an application scenario of a high-frequency electrosurgical unit control device according to some embodiments of this disclosure.

[0024] exist Figure 1In the application scenario, firstly, the computing device 101, in response to receiving a start command representing the start of operation of the high-frequency electrosurgical unit, controls the high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on the preset initial operating parameters 102 of the high-frequency electrosurgical unit. Then, the computing device 101 can perform multi-dimensional data acquisition on the human tissue of the target surgical area using an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence 103 and a human tissue image sequence 104. Next, the computing device 101 can use an acoustic signal acquisition device to acquire the acoustic waves generated by the surgical electrode acting on the human tissue, obtaining an acoustic signal 105, and perform feature extraction on the acoustic signal 105 to obtain an acoustic feature tensor 108. Secondly, the computing device 101 can perform feature extraction on the tissue impedance information sequence 103 and the human tissue image sequence 104, respectively, to obtain a tissue impedance feature tensor 106 and a human tissue feature tensor 107. Then, the computing device 101 can perform feature fusion processing on the aforementioned acoustic wave feature tensor 108, the aforementioned tissue impedance feature tensor 106, and the aforementioned human tissue feature tensor 107 to obtain a fused feature tensor 109. Afterwards, the computing device 101 can generate a high-frequency electrosurgical control command 110 based on the fused feature tensor 109, wherein the high-frequency electrosurgical control command 110 includes an output power sequence and a waveform modulation parameter sequence. Finally, the computing device 101 can adjust the output power of the high-frequency power generator based on the output power sequence, and control the waveform modulation module to modulate the high-frequency electrical energy output by the high-frequency power generator based on the waveform modulation parameter sequence, thereby achieving closed-loop control of the high-frequency electrosurgical unit.

[0025] It should be noted that the aforementioned computing device 101 can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here. It should be understood that... Figure 1 The number of computing devices in the system can be arbitrary, depending on the implementation requirements.

[0026] refer to Figure 2 The diagram illustrates a flow 200 of some embodiments of a high-frequency electrosurgical control device according to the present disclosure. This high-frequency electrosurgical control device includes the following steps:

[0027] Step 201: In response to receiving a start command indicating that the high-frequency electrosurgical unit has started working, the high-frequency power generator is controlled to output high-frequency electrical energy to the surgical electrode based on the preset initial working parameters of the high-frequency electrosurgical unit.

[0028] In some embodiments, the actuator of the high-frequency electrosurgical control device (e.g. Figure 1 The computing device 101 can, in response to receiving a start command characterizing the start of operation of the high-frequency electrosurgical unit, control the high-frequency power generator to output high-frequency electrical energy to the surgical electrodes based on preset initial operating parameters of the high-frequency electrosurgical unit. The initial operating parameters of the high-frequency electrosurgical unit include: initial output power and initial waveform modulation parameters. The initial output power can be the power of the high-frequency power generator at its initial operation. The initial waveform modulation parameters can be parameters used by the waveform modulation module to initially modulate the high-frequency electrical energy generated by the high-frequency power generator based on the initial output power. The initial waveform modulation parameters can include, but are not limited to: waveform mode, carrier frequency, and duty cycle. The waveform mode can include: pure cutting waveform, mixed cutting waveform, and coagulation waveform. For example, the initial operating parameters of the high-frequency electrosurgical unit can be: 80W (initial output power), pure cutting waveform (waveform mode), 400kHz (carrier frequency), and 100% (duty cycle). The high-frequency power generator can be a power generator (radio frequency generator, also understood as the main unit of the high-frequency electrosurgical unit) capable of generating high-frequency electrical energy according to the operating parameters of the high-frequency electrosurgical unit. The waveform modulation module described above can be a module capable of waveform modulation of the high-frequency electrical energy generated by the high-frequency power generator based on the output power, according to waveform modulation parameters. For example, the waveform modulation module can be a chip used for waveform modulation.

[0029] In some optional implementations of certain embodiments, based on preset initial operating parameters of the high-frequency electrosurgical unit, the execution entity can control the high-frequency power generator to output high-frequency electrical energy to the surgical electrode through the following steps:

[0030] The first step is to control the high-frequency power generator to output high-frequency electrical energy based on the initial output power. In practice, the executing entity can output the initial output power to the high-frequency power generator to control the high-frequency power generator to output high-frequency electrical energy.

[0031] The second step involves controlling the waveform modulation module to modulate the high-frequency electrical energy output from the high-frequency power generator based on the initial waveform modulation parameters, and then outputting the modulated high-frequency electrical energy to the surgical electrode. In practice, the executing entity can output the initial waveform modulation parameters to the waveform modulation module to control the module to modulate the high-frequency electrical energy output from the high-frequency power generator and output the modulated high-frequency electrical energy to the surgical electrode.

[0032] As an example, the aforementioned actuator can, in response to receiving a start command (which can be a digital signal) sent by the surgeon via a foot switch, send the initial output power and initial waveform modulation parameters to the high-frequency power generator and waveform modulation module respectively via a digital communication interface (such as SPI, I2C, or CAN bus). The high-frequency power generator generates an initial high-frequency (e.g., 470kHz) sine wave based on the initial output power. The waveform modulation module modulates the initial high-frequency (e.g., 470kHz) sine wave according to the initial waveform modulation parameters, forming a cutting wave, a coagulation wave, or a mixed wave. Then, the waveform-modulated high-frequency electrical energy can be output to the surgical electrodes via a high-voltage cable.

[0033] Optionally, before controlling the high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on the preset initial operating parameters of the high-frequency electrosurgical unit, the aforementioned execution entity may also perform the following steps:

[0034] The first step involves acquiring image data of the target surgical area using an image acquisition device to obtain an image of the area to be operated on. This image acquisition device can be any device capable of acquiring images of the target surgical area. For example, it could be a medical camera. The target surgical area can be a region of human tissue requiring surgery. The location of the image acquisition device is not specifically limited. For example, it can be integrated with the surgical light in the operating room. Specifically, for laparoscopic surgeries, it can be integrated with an endoscope system. The image of the area to be operated on can be an image of the human tissue region awaiting surgery, acquired by the image acquisition device.

[0035] The second step involves inputting the image of the surgical area into a pre-trained human tissue recognition model to obtain the human tissue recognition result. This result can represent a specific human tissue location. For example, the result could be the anus or abdomen. The human tissue recognition model can be a network model capable of identifying different human tissue locations; that is, it can take the image of the surgical area as input and output the human tissue recognition result. This model may include one input layer (for preprocessing the input data, such as normalizing the input image size), five convolutional layers, three fully connected layers, and one output layer (fully connected layer + Softmax function). For example, this model could be an AlexNet model trained on a human tissue dataset. The human tissue data in this dataset includes: human tissue images and human tissue labels. For example, the human tissue label could be the abdomen.

[0036] The third step is to determine the initial operating parameters of the high-frequency electrosurgical unit based on the aforementioned human tissue identification results. In practice, the executing entity can determine the initial operating parameters of the high-frequency electrosurgical unit by querying a tissue location operating parameter mapping table. This tissue location operating parameter mapping table represents the mapping relationship between human tissue locations and the operating parameters of the high-frequency electrosurgical unit. For example, the tissue location operating parameter mapping table could be: [Abdomen: Power: 80W, Waveform: Pure cutting waveform, Frequency: 400kHz, Duty Cycle: 100%; Face: Power: 20-35W, Waveform: Pure cutting waveform, Frequency: 400kHz, Duty Cycle: 100%].

[0037] Step 202: Multidimensional data acquisition of human tissue in the target surgical area is performed using impedance sensors and optical imaging sensors to obtain tissue impedance information sequences and human tissue image sequences.

[0038] In some embodiments, the aforementioned execution entity can perform multidimensional data acquisition on the human tissue of the target surgical area using an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence and a human tissue image sequence. In practice, the aforementioned execution entity can perform multidimensional data acquisition on the human tissue of the target surgical area using an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence and a human tissue image sequence. The aforementioned optical imaging sensor can be a miniature optical coherence tomography (OCT) probe. Here, the aforementioned optical imaging sensor can be used to detect the distribution of blood vessels in human tissue. The aforementioned impedance sensor can be a sensor for impedance detection of the human tissue to be operated on. The aforementioned tissue impedance information sequence can be a sequence composed of various tissue impedance information characterizing the electrical properties of human tissue acquired by the aforementioned impedance sensor in chronological order. The aforementioned human tissue image sequence can be a sequence composed of various human tissue images characterizing the internal anatomical structures of the human body acquired by the aforementioned optical imaging sensor in chronological order.

[0039] Step 203: Acquire the sound wave generated by the surgical electrode acting on human tissue using a sound wave signal acquisition device to obtain the sound wave signal, and extract the features of the sound wave signal to obtain the sound wave feature tensor.

[0040] In some embodiments, the aforementioned executing entity can acquire sound waves generated by the surgical electrode acting on human tissue using a sound wave signal acquisition device to obtain sound wave signals, and extract features from the sound wave signals to obtain sound wave feature tensors. The aforementioned sound wave signal acquisition device can be any device capable of acquiring sound waves generated by the surgical electrode acting on human tissue. For example, the aforementioned sound wave signal acquisition device can be an acoustic emission sensor.

[0041] In practice, the aforementioned executing entity can extract features from the acoustic signal through the following steps:

[0042] The first step is to pre-emphasize the aforementioned acoustic signal to obtain a pre-emphasized acoustic signal. In practice, the aforementioned execution entity can use a first-order FIR high-pass digital filter to pre-emphasize the aforementioned acoustic signal to obtain a pre-emphasized acoustic signal.

[0043] The second step involves framing the pre-emphasized acoustic signal to obtain a framed acoustic signal sequence. In practice, the executing entity can first perform framing on the pre-emphasized acoustic signal using a preset frame length threshold and a preset frame shift threshold, resulting in a set of framed acoustic signals. The preset frame shift threshold is less than the preset frame length threshold, and the frame shift is the displacement of the subsequent frame relative to the previous frame. The preset frame shift threshold can be a pre-set frame shift threshold, for example, 100. The preset frame length threshold can also be a pre-set frame length threshold, for example, 200. Then, the executing entity can sort the individual framed acoustic signals included in the framed acoustic signal set to obtain a framed acoustic signal sequence. In practice, the executing entity can sort the individual framed acoustic signals included in the framed acoustic signal set according to their generation time to obtain a framed acoustic signal sequence.

[0044] The third step is to window the aforementioned framed acoustic signal sequence to obtain a windowed acoustic signal sequence. In practice, the execution entity can use a window function to window the aforementioned framed acoustic signal sequence to obtain a windowed acoustic signal sequence. The window function can be any of the following: rectangular window, Hanning window, Hamming window, and Kaiser window.

[0045] The fourth step involves performing frequency domain transformation on the windowed acoustic signal sequence to obtain the acoustic spectrum. In practice, the execution entity can use short-time Fourier transform to perform frequency domain transformation on the windowed acoustic signal sequence to obtain the acoustic spectrum.

[0046] The fifth step is to generate a sound energy spectrum based on the aforementioned sound wave spectrum diagram. In practice, firstly, for each sound wave spectrum in the aforementioned sound wave spectrum diagram, the executing entity can determine the square of the modulus of the aforementioned sound wave spectrum as the sound energy spectrum. Then, the determined sound energy spectra can be used as the sound energy spectrum diagram.

[0047] Step 6: Based on the aforementioned acoustic energy spectrum, generate a logarithmic acoustic energy spectrum. In practice, the aforementioned execution entity can perform logarithmic transformation on each acoustic energy spectrum included in the acoustic energy spectrum to obtain the logarithmically transformed acoustic energy spectrum as the acoustic logarithmic energy spectrum.

[0048] Step 7: Perform a time-domain transformation on the aforementioned logarithmic energy spectrum of the sound wave to obtain a sequence of sound wave characteristic parameters. In practice, the aforementioned execution entity can perform a discrete cosine transform on the logarithmic energy spectrum of the sound wave to obtain a logarithmic energy spectrum of the sound wave after the discrete cosine transform, which can then be used as a sequence of sound wave characteristic parameters.

[0049] Step 8: Based on the above sequence of acoustic wave characteristic parameters, determine the sequence of differential acoustic wave characteristic parameters. In practice, the sequence of differential acoustic wave characteristic parameters can be expressed by the following formula:

[0050] .

[0051] Among them, the above This can represent a preset time difference. The k above can be represented as a value from 1 to... Temporary variables for loops. (The above) This can represent a preset order. Here, the preset order can be 12. It can represent the first element in the sequence of acoustic wave characteristic parameters. Each acoustic wave characteristic parameter. The above It can represent the first element in the sequence of acoustic wave characteristic parameters. Each acoustic wave characteristic parameter. The above It can represent the first element in the sequence of acoustic wave characteristic parameters. Each acoustic wave characteristic parameter. The above It can represent the first Each acoustic wave differential characteristic parameter.

[0052] Step nine: Based on the aforementioned acoustic wave characteristic parameter sequence and the aforementioned acoustic wave differential characteristic parameter sequence, generate an acoustic wave update differential parameter sequence. In practice, for each acoustic wave characteristic parameter in the aforementioned acoustic wave characteristic parameter sequence, the executing entity can determine the acoustic wave update differential parameter by summing the aforementioned acoustic wave characteristic parameter with the corresponding acoustic wave differential characteristic parameter in the aforementioned acoustic wave differential characteristic parameter sequence. Then, the determined acoustic wave update differential parameters can be used to form an acoustic wave update differential parameter sequence.

[0053] Step 10: Based on the aforementioned acoustic wave update difference parameter sequence, generate the acoustic wave feature tensor. In practice, the aforementioned execution entity can input the acoustic wave update difference parameter sequence into a preset feature extraction model to obtain the acoustic wave feature tensor. The aforementioned preset feature extraction model can be a pre-defined feature extraction model. It can be used to compress acoustic wave signals of unequal length into feature tensors that can be classified. For example, the aforementioned preset feature extraction model can be GMM-UBM (Gaussian Mixture Model-Universal Background Model).

[0054] In some optional implementations of certain embodiments, the aforementioned execution entity may perform feature extraction on the aforementioned acoustic signal through the following steps:

[0055] The first step is to perform acoustic encoding processing on the aforementioned acoustic signal to obtain the acoustic feature tensor. In practice, the aforementioned execution entity can perform acoustic encoding processing on the aforementioned acoustic signal using an image-bind multimodal large model to obtain the acoustic feature tensor.

[0056] The second step involves convolving the aforementioned acoustic feature tensor to obtain the acoustic convolutional feature tensor. In practice, the execution entity can use a convolution module to convolve the acoustic feature tensor to obtain the acoustic convolutional feature tensor. Here, the size of the convolution kernel corresponding to the convolution module can be 7*7, and the convolution stride can be 2.

[0057] The third step is to perform batch normalization on the above acoustic wave convolution feature tensor to obtain the normalized acoustic wave feature tensor.

[0058] The fourth step involves generating an acoustic activation feature tensor based on the normalized acoustic feature tensor and the preset activation function. In practice, the executing entity can input the normalized acoustic feature tensor into the preset activation function to obtain the acoustic activation feature tensor. The preset activation function can be a pre-defined activation function. Here, the preset activation function can be a ReLU activation function.

[0059] The fifth step is to perform global max pooling on the above acoustic wave activation feature tensor to obtain the max pooled acoustic wave feature tensor.

[0060] The sixth step is to perform fine feature extraction on the above max-pooling acoustic wave feature tensor to obtain the acoustic wave feature tensor.

[0061] In the process of further feature extraction, multiple network layers are typically used, with the output of the previous layer serving as the input to the next. However, as the depth increases, feature information is often lost. Therefore, the following solution is proposed:

[0062] In practice, the aforementioned executing entity can perform fine feature extraction on the max-pooling acoustic feature tensor through the following steps:

[0063] The first step involves performing depth feature extraction on the max-pooling acoustic feature tensor to obtain a first acoustic depth feature tensor, wherein the number of channels in the max-pooling acoustic feature tensor is the same as the number of channels in the first acoustic depth feature tensor. In practice, the execution entity can use a first preset depth feature extraction module to perform depth feature extraction on the max-pooling acoustic feature tensor to obtain the first acoustic depth feature tensor. This first preset depth feature extraction module may include: a batch normalization layer, a ReLU activation function layer, and a convolutional layer (1x1 kernel size, stride 1). Here, the batch normalization layer, ReLU activation function layer, and convolutional layer (1x1 kernel size, stride 1) in the first branch are used for feature dimensionality reduction and information fusion, compressing the high-dimensional, mixed input into a low-dimensional, unified feature space through 1x1 convolution. The second branch's batch normalization layer, ReLU activation function layer, and convolutional layer (with a kernel size of 1*1 and a stride of 1) are used to perform another nonlinear transformation (BN->ReLU->Conv) on the fused compact features to generate the new features that this layer will actually contribute.

[0064] The second step is to perform channel splicing on the above-mentioned maximum pooling acoustic wave feature tensor and the above-mentioned first acoustic wave depth feature tensor to obtain the first spliced ​​acoustic wave feature tensor.

[0065] The third step involves performing depth feature extraction on the first spliced ​​acoustic wave feature tensor to obtain a second acoustic wave depth feature tensor, wherein the number of channels in the second acoustic wave depth feature tensor is the same as the number of channels in the first acoustic wave depth feature tensor. In practice, the execution entity can use a second preset depth feature extraction module to perform depth feature extraction on the first spliced ​​acoustic wave feature tensor to obtain the second acoustic wave depth feature tensor. Here, the second preset depth feature extraction module has the same module structure as the first preset depth feature extraction module.

[0066] The fourth step is to perform channel splicing processing on the first spliced ​​acoustic wave feature tensor and the second acoustic wave depth feature tensor to obtain the second spliced ​​acoustic wave feature tensor.

[0067] The fifth step involves performing depth feature extraction on the second spliced ​​acoustic wave feature tensor to obtain a third acoustic wave depth feature tensor, wherein the number of channels in the third acoustic wave depth feature tensor is the same as the number of channels in the second acoustic wave depth feature tensor. In practice, the execution entity can utilize a third preset depth feature extraction module to perform depth feature extraction on the second spliced ​​acoustic wave feature tensor to obtain the third acoustic wave depth feature tensor. Here, the module structure of the third preset depth feature extraction module is the same as the module structure of the first preset depth feature extraction module.

[0068] The sixth step is to perform channel splicing processing on the second spliced ​​acoustic wave feature tensor and the third acoustic wave depth feature tensor to obtain the third spliced ​​acoustic wave feature tensor.

[0069] It should be noted that the number of operations for performing depth feature extraction on the acoustic feature tensor using the preset depth feature extraction module is not specifically limited. For example, depth feature extraction can be performed based on three preset depth feature extraction modules, or it can be performed based on six preset depth feature extraction modules as needed. Furthermore, the input to the depth feature extraction module is obtained by concatenating the output data from the previous depth feature extraction modules, and the output of the current depth feature extraction module has the same number of channels as the output data from the previous depth feature extraction modules. Specifically, the module structure of the preset depth feature extraction module can be found in [reference needed]. Figure 3 .

[0070] Step 7: Perform feature compression processing on the aforementioned third-joined acoustic wave feature tensor to obtain the feature-compressed third-joined acoustic wave feature tensor as the acoustic wave feature tensor. In practice, the aforementioned execution entity can perform feature compression processing on the aforementioned third-joined acoustic wave feature tensor using a preset feature compression module to obtain the feature-compressed third-joined acoustic wave feature tensor as the acoustic wave feature tensor. The preset feature compression module can be a feature compression module capable of performing channel compression and spatial compression on the features of the third-joined acoustic wave feature tensor. Here, the module structure of the preset feature compression module may include: a batch normalization layer, a ReLU activation function layer, a convolutional layer, and a max-pooling layer.

[0071] Steps one through seven of the above scheme and their related contents constitute an inventive point of this disclosure, solving the aforementioned technical problem: "loss of feature information." The loss of feature information is often caused by the fact that, during further feature extraction, multiple network layers are typically used, with the output of the previous layer serving as the input to the next. As the depth increases, feature information is often lost. Solving this problem reduces the loss of feature information. To achieve this, firstly, depth feature extraction is performed on the max-pooling acoustic feature tensor to obtain a first acoustic depth feature tensor, where the number of channels corresponding to the max-pooling acoustic feature tensor is the same as the number of channels corresponding to the first acoustic depth feature tensor. This completes the first depth transformation of the features, generating a new feature tensor, while controlling complexity through a bottleneck structure (two branch structures). Then, channel concatenation is performed on the max-pooling acoustic feature tensor and the first acoustic depth feature tensor to obtain a first concatenated acoustic feature tensor. This enables feature reuse. Subsequent layers can not only learn from the output features of their current layer but also directly access the original feature information, preventing early features from being lost in deeper networks and mitigating gradient vanishing. Next, depth feature extraction is performed on the first concatenated acoustic wave feature tensor to obtain a second acoustic wave depth feature tensor, where the number of channels in the second acoustic wave depth feature tensor is the same as that in the first acoustic wave depth feature tensor. This completes the second depth transformation of the features, further generating new feature tensors, while the complexity is further controlled through a bottleneck structure (two branch structures). Next, channel concatenation is performed on the first and second concatenated acoustic wave feature tensors to obtain another second concatenated acoustic wave feature tensor. This allows for feature reuse. Then, depth feature extraction is performed on the second concatenated acoustic wave feature tensor to obtain a third acoustic wave depth feature tensor, where the number of channels in the third acoustic wave depth feature tensor is the same as that in the second acoustic wave depth feature tensor. This completes the third depth transformation of the features, further generating new feature tensors, while the complexity is further controlled through a bottleneck structure (two branch structures). Next, the second and third spliced ​​acoustic feature tensors are processed by channel splicing to obtain a third spliced ​​acoustic feature tensor. This allows for multiple reuse of features. Finally, the third spliced ​​acoustic feature tensor is compressed to obtain a compressed third spliced ​​acoustic feature tensor as the acoustic feature tensor. Thus, after sufficient feature reuse and growth, feature compression can be performed, thereby controlling the model's complexity and obtaining a highly condensed acoustic feature tensor.Because of the multiple iterations of feature extraction, feature concatenation, and further extraction and concatenation of input features, feature reuse can be achieved, improving information utilization. Furthermore, the learning of each new feature layer is built upon the complete context of all previous layers of features (from the original input to the latest abstraction), enriching the feature representation (the final output features can fuse information from multiple semantic levels, from low to high). Also, because the feature extraction module adopts a dual-branch structure, it can force dimensionality reduction and feature fusion based on the first branch, controlling model complexity, and generate new features based on the second branch, thereby reducing information loss.

[0072] Step 204: Extract features from the tissue impedance information sequence and the human tissue image sequence to obtain the tissue impedance feature tensor and the human tissue feature tensor.

[0073] In some embodiments, the aforementioned execution entity can extract features from the tissue impedance information sequence and the human tissue image sequence respectively to obtain a tissue impedance feature tensor and a human tissue feature tensor. In practice, the aforementioned execution entity can extract features from the tissue impedance information sequence using a BiLSTM (Bi-directional Long Short-Term Memory) network (with the Softmax layer removed) to obtain a tissue impedance feature tensor, and extract features from the human tissue image sequence using a ResNet network (Residual Network) (e.g., ResNet50 with the final fully connected classification layer (containing Softmax) removed) to obtain a human tissue feature tensor.

[0074] In some optional implementations of certain embodiments, the execution entity may perform feature extraction on the tissue impedance information sequence and the human tissue image sequence respectively through the following steps:

[0075] The first step is to generate a sequence of tissue impedance change rate and a sequence of tissue impedance acceleration corresponding to the aforementioned tissue impedance information sequence. Specifically, the tissue impedance information in the aforementioned tissue impedance information sequence corresponds one-to-one with the tissue impedance change rate in the aforementioned tissue impedance change rate sequence. Similarly, the tissue impedance information in the aforementioned tissue impedance information sequence corresponds one-to-one with the tissue impedance acceleration in the aforementioned tissue impedance acceleration sequence. In practice, since the impedance sensor collects tissue impedance information from the surgical electrode acting on human tissue in real time, for each tissue impedance information at the current moment, a windowed tissue impedance information sequence can be constructed based on the tissue impedance information at the current moment and before the current moment, using a preset sliding window size. The tissue impedance change rate corresponding to the tissue impedance information at the current moment is then generated based on the following formula, resulting in the tissue impedance change rate sequence corresponding to the aforementioned tissue impedance information sequence:

[0076] .

[0077] Among them, the above This can represent the window organization impedance information in the window organization impedance information sequence. The above... This can represent the sequence number corresponding to the window organization impedance information in the window organization impedance information sequence. The above... This can represent the number of window organization impedance information entries in the window organization impedance information sequence. The above... It can represent the rate of change of tissue impedance corresponding to the tissue impedance information at the current moment.

[0078] As an example, the aforementioned execution entity can generate a tissue impedance acceleration sequence based on the tissue impedance information sequence through the following steps:

[0079] The first step is to generate an instantaneous impedance difference information sequence based on the aforementioned tissue impedance information sequence. In practice, firstly, for each pair of adjacent tissue impedance information in the aforementioned tissue impedance information sequence, the executing entity can determine the instantaneous impedance difference information as the difference between the tissue impedance information at the second position and the tissue impedance information at the first position in the two adjacent tissue impedance information sequences. Then, the obtained instantaneous impedance difference information can be determined as an instantaneous impedance difference information sequence. Here, the instantaneous impedance difference information corresponding to the first tissue impedance information in the aforementioned tissue impedance information sequence can be set to 0, or it can be determined based on the difference between the first tissue impedance information and the tissue impedance information collected at the previous moment corresponding to the real-time collected first tissue impedance information.

[0080] The second step involves generating a tissue impedance acceleration sequence based on the aforementioned instantaneous impedance difference information sequence. In practice, since the impedance sensor acquires tissue impedance information from the surgical electrode applied to human tissue in real time, the corresponding instantaneous impedance difference information can also be generated in real time. For each instantaneous impedance difference at the current moment, a window instantaneous impedance difference information sequence can be constructed based on the instantaneous impedance difference information at the current moment and before the current moment, using a preset sliding window size. Then, tissue impedance acceleration is generated using the same steps as generating the tissue impedance change rate (i.e., using the same formula), resulting in the tissue impedance acceleration sequence.

[0081] The second step involves combining the aforementioned tissue impedance information sequence, tissue impedance change rate sequence, and tissue impedance acceleration sequence to obtain a combined tissue impedance information sequence. In practice, the executing entity can perform the following combined processing steps for each tissue impedance information in the aforementioned tissue impedance information sequence:

[0082] The first step is to determine the tissue impedance change rate corresponding to the tissue impedance information in the above tissue impedance change rate sequence as the target tissue impedance change rate.

[0083] The second step is to determine the tissue impedance acceleration corresponding to the tissue impedance information in the above tissue impedance acceleration sequence as the target tissue impedance acceleration.

[0084] The third step involves combining the aforementioned tissue impedance information, the target tissue impedance change rate, and the target tissue impedance acceleration to obtain combined tissue impedance information. Here, the combination process can be achieved through splicing.

[0085] The fourth step is to determine the obtained tissue impedance combination information into a tissue impedance combination information sequence.

[0086] The third step involves performing a fully connected operation on the aforementioned tissue impedance combination information sequence to obtain a fully connected tissue impedance combination information sequence as a fully connected tissue impedance feature tensor. Here, the fully connected operation on the aforementioned tissue impedance combination information sequence is used to map the discrete or continuous original attributes to a higher-dimensional, denser continuous tensor space.

[0087] The fourth step involves performing a fully connected process on the aforementioned fully connected tissue impedance feature tensor to obtain a fully connected tissue impedance feature tensor as the tissue impedance feature tensor. Here, performing a fully connected process on the aforementioned fully connected tissue impedance feature tensor is used to deepen the basic features and construct a more complex feature representation.

[0088] The fifth step involves image encoding of the aforementioned human tissue image sequence to obtain the tissue image feature tensor. In practice, the aforementioned execution entity can use the image-bind multimodal large model to perform image encoding on the aforementioned human tissue image sequence to obtain the tissue image feature tensor.

[0089] The sixth step is to extract human tissue features from the above tissue image feature tensor to obtain the human tissue feature tensor.

[0090] In practice, the aforementioned executing entity can extract human tissue features from the aforementioned tissue image feature tensor through the following steps to obtain the human tissue feature tensor:

[0091] The first step is to perform convolution processing on the aforementioned tissue image feature tensor to obtain the first tissue image convolutional feature tensor. In practice, the aforementioned execution entity can use a convolution module to perform convolution processing on the aforementioned tissue image feature tensor to obtain the first tissue image convolutional feature tensor. Here, the size of the convolution kernel corresponding to the convolution module can be 1*1, and the convolution stride can be 1.

[0092] The second step involves generating a first tissue image activation feature tensor based on the aforementioned first tissue image convolutional feature tensor and a preset activation function. In practice, the executing entity can input the first tissue image convolutional feature tensor into the preset activation function to obtain the first tissue image activation feature tensor. Here, the preset activation function can be a ReLU activation function.

[0093] The third step is to perform global average pooling on the activation feature tensor of the first tissue image to obtain the global average pooled tissue image feature tensor.

[0094] The fourth step involves convolving the aforementioned global average pooling tissue image feature tensor to obtain the second tissue image convolutional feature tensor. In practice, the execution entity can use a convolution module to convolve the aforementioned global average pooling tissue image feature tensor to obtain the second tissue image convolutional feature tensor. Here, the size of the convolution kernel corresponding to the convolution module can be 7*7, and the convolution stride can be 2.

[0095] The fifth step is to perform batch normalization on the convolutional feature tensor of the second tissue image to obtain the normalized tissue image feature tensor.

[0096] Step 6: Based on the normalized tissue image feature tensor and the preset activation function, generate the second tissue image activation feature tensor. In practice, the executing entity can input the normalized tissue image feature tensor into the preset activation function to obtain the second tissue image activation feature tensor.

[0097] Step 7: Perform global max pooling on the activation feature tensor of the second tissue image to obtain the max pooled tissue image feature tensor.

[0098] Step 8: Perform fine feature extraction on the above maximum pooling tissue image feature tensor to obtain the human tissue feature tensor.

[0099] In practice, the aforementioned executing entity can perform fine feature extraction on the max-pooled tissue image feature tensor through the following steps:

[0100] The first step involves performing depth feature extraction on the max-pooling tissue image feature tensor to obtain a first tissue image depth feature tensor, wherein the number of channels in the max-pooling tissue image feature tensor is the same as the number of channels in the first tissue image depth feature tensor. In practice, the execution entity can use a fourth preset depth feature extraction module to perform depth feature extraction on the max-pooling tissue image feature tensor to obtain the first tissue image depth feature tensor. Here, the depth feature extraction module may include: a batch normalization layer, a ReLU activation function layer, and a convolutional layer (1*1 kernel size, stride 1).

[0101] The second step involves performing channel stitching on the max-pooled tissue image feature tensor and the first tissue image depth feature tensor to obtain the first stitched tissue image feature tensor.

[0102] The third step involves extracting depth features from the first stitched tissue image feature tensor to obtain a second tissue image depth feature tensor, wherein the number of channels in the second tissue image depth feature tensor is the same as the number of channels in the first tissue image depth feature tensor. In practice, the executing entity can use a fifth preset depth feature extraction module to extract depth features from the first stitched tissue image feature tensor to obtain the second tissue image depth feature tensor. Here, the fifth preset depth feature extraction module has the same module structure as the fourth preset depth feature extraction module.

[0103] The fourth step involves performing channel stitching on the first stitched tissue image feature tensor and the second tissue image depth feature tensor to obtain the second stitched tissue image feature tensor.

[0104] The fifth step involves extracting depth features from the second stitched tissue image feature tensor to obtain a third tissue image depth feature tensor, wherein the number of channels in the third tissue image depth feature tensor is the same as the number of channels in the second tissue image depth feature tensor. In practice, the executing entity can utilize a sixth preset depth feature extraction module to extract depth features from the second stitched tissue image feature tensor to obtain the third tissue image depth feature tensor. Here, the module structure of the sixth preset depth feature extraction module is the same as that of the fourth preset depth feature extraction module.

[0105] The sixth step involves performing channel stitching on the second stitched tissue image feature tensor and the third tissue image depth feature tensor to obtain the third stitched tissue image feature tensor.

[0106] Step 7: Perform feature compression processing on the aforementioned third-stitched tissue image feature tensor to obtain the feature-compressed third-stitched tissue image feature tensor as the human tissue feature tensor. In practice, the aforementioned execution entity can perform feature compression processing on the aforementioned third-stitched tissue image feature tensor using a preset feature compression module to obtain the feature-compressed third-stitched tissue image feature tensor as the human tissue feature tensor. The preset feature compression module can be a feature compression module capable of performing channel compression and spatial compression on the features of the third-stitched tissue image feature tensor. Here, the module structure of the preset feature compression module may include: a batch normalization layer, a ReLU activation function layer, a convolutional layer, and a max-pooling layer.

[0107] Step 205: Perform feature fusion processing on the acoustic wave feature tensor, tissue impedance feature tensor, and human tissue feature tensor to obtain the fused feature tensor.

[0108] In some embodiments, the aforementioned execution entity may perform feature fusion processing on the acoustic feature tensor, the tissue impedance feature tensor, and the human tissue feature tensor to obtain a fused feature tensor.

[0109] As an example, the aforementioned execution entity can perform feature fusion processing on the acoustic feature tensor, tissue impedance feature tensor, and human tissue feature tensor through the following steps:

[0110] The first step involves normalizing the aforementioned acoustic wave feature tensor, tissue impedance feature tensor, and human tissue feature tensor to obtain normalized acoustic wave feature tensor, normalized tissue impedance feature tensor, and normalized human tissue feature tensor, respectively. The normalization process can include, but is not limited to, layer normalization and batch normalization. Here, the normalization process can be layer normalization. In practice, the executing entity can perform layer normalization on the acoustic wave feature tensor, tissue impedance feature tensor, and human tissue feature tensor, respectively, to obtain normalized acoustic wave feature tensor, normalized tissue impedance feature tensor, and normalized human tissue feature tensor.

[0111] The second step involves inputting the normalized acoustic wave feature tensor, the normalized tissue impedance feature tensor, and the normalized human tissue feature tensor into a pre-trained weight network to obtain acoustic wave weights, impedance weights, and tissue weights. The weight network can be a neural network that generates weights corresponding to the input feature tensors. Specifically, the weight network may include three attention networks and a multi-classification function layer (softmax layer), where the three attention networks share parameters. The attention network may include a fully connected layer, a ReLU activation function layer, and a fully connected layer. The acoustic wave weights can be weights corresponding to the normalized acoustic wave feature tensor. The impedance weights can be weights corresponding to the normalized tissue impedance feature tensor. The tissue weights can be weights corresponding to the normalized human tissue feature tensor. Here, the multi-classification function layer is used to normalize the weights obtained from the three attention networks.

[0112] The third step involves generating updated acoustic feature tensors, updated tissue impedance feature tensors, and updated human tissue feature tensors based on the aforementioned normalized acoustic feature tensor, normalized tissue impedance feature tensor, normalized human tissue feature tensor, acoustic weights, impedance weights, and tissue weights. In practice, firstly, the executing entity can determine the updated acoustic feature tensor as the product of the normalized acoustic feature tensor and the acoustic weights. Then, it can determine the updated tissue impedance feature tensor as the product of the normalized tissue impedance feature tensor and the impedance weights. Finally, it can determine the updated human tissue feature tensor as the product of the normalized human tissue feature tensor and the tissue weights.

[0113] The fourth step involves performing feature splicing on the updated acoustic wave feature tensor, the updated tissue impedance feature tensor, and the updated human tissue feature tensor to obtain a fused feature tensor. Here, the splicing process can be channel splicing.

[0114] In practice, when aligning different features, feature fusion is often performed by direct concatenation. This is especially true for multimodal data, where features from different modalities naturally exhibit spatiotemporal asynchrony and semantic gaps (neurons at the same location in different feature tensors encode different information). Direct concatenation can easily lead to information loss or noise introduction during feature fusion. Therefore, the following solution is proposed:

[0115] In some optional implementations of certain embodiments, the execution entity may perform feature fusion processing on the acoustic wave feature tensor, the tissue impedance feature tensor, and the human tissue feature tensor through the following steps:

[0116] The first step is to concatenate the aforementioned acoustic wave characteristic tensor and the aforementioned tissue impedance characteristic tensor to obtain the acoustic wave impedance characteristic tensor. Here, the concatenation process can be channel concatenation.

[0117] The second step involves generating an acoustic impedance cross-fusion feature tensor based on the first preset cross-fusion function and the aforementioned acoustic impedance feature tensor. In practice, the executing entity can input the aforementioned acoustic impedance feature tensor into the first preset cross-fusion function to obtain the acoustic impedance cross-fusion feature tensor. The first preset cross-fusion function can be a pre-defined cross-fusion function. For example, the first preset cross-fusion function can be a multi-head self-attention function (MSA function).

[0118] The third step involves concatenating the aforementioned human tissue feature tensor and the aforementioned tissue impedance feature tensor to obtain the human tissue impedance feature tensor. Here, the concatenation process can be channel concatenation.

[0119] The fourth step involves generating a human tissue impedance cross-fusion feature tensor based on the second preset cross-fusion function and the aforementioned human tissue impedance feature tensor. In practice, the executing entity can input the aforementioned human tissue impedance feature tensor into the aforementioned second preset cross-fusion function to obtain the human tissue impedance cross-fusion feature tensor. The aforementioned second preset cross-fusion function can be a pre-defined cross-fusion function. Here, the aforementioned second preset cross-fusion function can be the same as the aforementioned first preset cross-fusion function.

[0120] The fifth step involves fusing and aligning the aforementioned acoustic impedance cross-fusion feature tensor and the aforementioned human tissue impedance cross-fusion feature tensor to obtain the cross-fusion feature tensor. In practice, the executing entity can use the following formula to fusing and align the aforementioned acoustic impedance cross-fusion feature tensor and the aforementioned human tissue impedance cross-fusion feature tensor to obtain the cross-fusion feature tensor:

[0121] .

[0122] .

[0123] Among them, the above This can represent the total number of samples corresponding to the aforementioned acoustic impedance cross-fusion feature tensor and the aforementioned human tissue impedance cross-fusion feature tensor. This can represent a single sample within the total number of samples. (The above...) This can represent the first weight vector. (The above...) This can represent the impedance cross-fusion characteristic tensor of human tissue. The above... This can represent the acoustic impedance cross-fusion characteristic tensor. The above... It can be used to represent the cross-fusion feature tensor. The above This can represent the first element in the human tissue impedance cross-fusion feature tensor. The feature vector corresponding to the nth sample, i.e., the nth feature vector in the human tissue impedance cross-fusion feature tensor (matrix). Row vectors. The above It can represent the first characteristic tensor of the acoustic impedance cross-fusion. The feature vector corresponding to the nth sample, i.e., the nth feature vector in the acoustic impedance cross-fusion feature tensor (matrix). Row vectors. It can represent The absolute value of. The above. It can represent The absolute value of. The above. This can be represented as the sum of the absolute values ​​of the activation function values ​​of each sample included in the human tissue impedance cross-fusion feature tensor along the channel dimension. The above... It can be represented as the summation of the absolute values ​​of the activation function values ​​of each sample included in the acoustic impedance cross-fusion feature tensor along the channel dimension. It can represent The diagonal matrix above. It can represent A diagonal matrix.

[0124] Step 6: Perform fusion and alignment processing on the aforementioned cross-fusion feature tensor and the aforementioned acoustic wave feature tensor to obtain the fused feature tensor. In practice, the aforementioned execution entity can perform fusion and alignment processing on the aforementioned cross-fusion feature tensor and the aforementioned acoustic wave feature tensor to obtain the fused feature tensor through the following steps:

[0125] The first step is to align the aforementioned cross-fusion feature tensor with the aforementioned acoustic feature tensor using the following formula to obtain the aligned feature tensor:

[0126]

[0127] .

[0128] Among them, the above This can represent the alignment feature tensor. (The above...) This can represent the characteristic tensor of sound waves. (The above...) This can represent the second weight vector. (The above...) It can represent The absolute value of. The above. It can represent The absolute value of. The above. It can represent The diagonal matrix above. It can represent The diagonal matrix above. It can be represented as the sum of the absolute values ​​of the activation function values ​​of each sample included in the cross-fusion feature tensor along the channel dimension. ,here, It can represent the first characteristic tensor of sound waves. The feature vector corresponding to the nth sample, i.e., the nth feature tensor (matrix) of the sound wave. Row vectors. The above It can represent the first in the cross-fusion feature tensor The feature vector corresponding to the nth sample, i.e., the nth feature tensor (matrix) in the cross-fusion feature tensor (matrix). Row vectors.

[0129] The second step involves performing cross-modal fusion processing on the alignment feature tensor to obtain a fused feature tensor. In practice, the execution entity can input the alignment feature tensor into the Transformer module to obtain the fused feature tensor.

[0130] As an example, the steps for feature fusion processing of the aforementioned acoustic wave feature tensor, tissue impedance feature tensor, and human tissue feature tensor can be referred to... Figure 4 .

[0131] The first to sixth steps and related content described above constitute an inventive point of this disclosure, solving the aforementioned technical problem: "information loss or noise introduction during feature fusion." The reason for information loss or noise introduction during feature fusion is often that, when aligning different features, a direct concatenation method is frequently used for feature fusion. This is especially problematic for multimodal data, where features from different modalities naturally exhibit spatiotemporal asynchrony and semantic gaps (neurons at the same location in different feature tensors encode different information). Direct concatenation easily leads to information loss or noise introduction during feature fusion. Solving these factors can reduce information loss or noise introduction during feature fusion. To achieve this, firstly, the aforementioned acoustic wave feature tensor and the aforementioned tissue impedance feature tensor are concatenated to obtain an acoustic wave impedance feature tensor. Thus, acoustic and electrical information with original feature independence can be concatenated to obtain an acoustic wave impedance feature tensor representing shallow fusion. Then, based on the first preset cross-fusion function and the aforementioned acoustic wave impedance feature tensor, an acoustic wave impedance cross-fusion feature tensor is generated. Therefore, intramodal data feature interaction can be performed to capture the nonlinear relationship between acoustic and impedance features, while simultaneously enhancing the weights of important feature channels and suppressing noise through a self-attention mechanism. Next, the aforementioned human tissue feature tensor and the aforementioned tissue impedance feature tensor are concatenated to obtain the human tissue impedance feature tensor. Thus, visual and electrical information with independent original features can be concatenated to obtain a human tissue impedance feature tensor representing shallow fusion. Next, based on the second preset cross-fusion function and the aforementioned human tissue impedance feature tensor, a human tissue impedance cross-fusion feature tensor is generated. This allows for intramodal data feature interaction, capturing the nonlinear relationship between visual and impedance features, while simultaneously enhancing the weights of important feature channels and suppressing noise through a self-attention mechanism. Then, the aforementioned acoustic impedance cross-fusion feature tensor and the aforementioned human tissue impedance cross-fusion feature tensor are fused and aligned to obtain a cross-fusion feature tensor. Thus, the same channels in different branches may encode information of different intensities; by dynamically setting different weights for different channels, semantic alignment between different modal data can be achieved. Finally, the aforementioned cross-fusion feature tensor and the aforementioned acoustic feature tensor are fused and aligned to obtain the fused feature tensor. This allows for deep feature fusion with the acoustic feature tensor, which represents temporal characteristics, further enhancing the temporal properties and expressive power of the features, resulting in more robust and discriminative multimodal features. Furthermore, by initially fusing the acoustic feature tensor and the tissue impedance feature tensor, and by initially fusing the human tissue feature tensor and the tissue impedance feature tensor, a preliminary spatiotemporal correlation between multimodal data and features can be achieved.Furthermore, by incorporating a temporally relevant acoustic feature tensor, the initially fused features are further fused, achieving refined feature calibration and context-aware deep integration, thus eliminating ambiguity and noise from the initial fusion. This reduces information loss and noise introduction during feature fusion.

[0132] Step 206: Generate high-frequency electrosurgical control commands based on the fused feature tensor.

[0133] In some embodiments, the aforementioned execution entity can generate high-frequency electrosurgical control commands based on the fused feature tensor. These high-frequency electrosurgical control commands may include an output power sequence and a waveform modulation parameter sequence.

[0134] In practice, the aforementioned executing entity can generate high-frequency electrosurgical control commands based on the fused feature tensor through the following steps:

[0135] The first step is to perform a fully connected operation on the aforementioned fused feature tensor to obtain a first fully connected fused feature tensor. Here, the fully connected operation on the fused feature tensor is used to perform feature dimensionality reduction and higher-order combination of features.

[0136] The second step involves inputting the first fully connected fusion feature tensor into the preset activation function to obtain the activated fusion feature tensor. Here, the preset activation function can be a ReLU activation function. Inputting the first fully connected fusion feature tensor into the preset activation function introduces a nonlinear transformation, enhancing the model's expressive power and enabling data mapping between features (activating key neurons and filtering out irrelevant or contradictory neurons).

[0137] The third step involves performing a fully connected processing on the aforementioned activation fusion feature tensor to obtain the high-frequency electrosurgical control commands. Here, the fully connected processing is used to map the high-dimensional features to 2D high-frequency electrosurgical control commands.

[0138] Step 207: Adjust the output power of the high-frequency power generator according to the output power sequence, and control the waveform modulation module to modulate the high-frequency electrical energy output by the high-frequency power generator according to the waveform modulation parameter sequence, so as to realize the closed-loop control of the high-frequency electric knife.

[0139] In some embodiments, the aforementioned execution entity can adjust the output power of the high-frequency power generator according to the output power sequence, and control the waveform modulation module to modulate the high-frequency electrical energy output by the high-frequency power generator according to the waveform modulation parameter sequence, thereby realizing closed-loop control of the high-frequency electrosurgical unit. In practice, firstly, for each output power in the output power sequence, the aforementioned execution entity can send the output power to the high-frequency power generator, which generates a sine wave corresponding to the output power. Simultaneously, the aforementioned execution entity can send the waveform modulation parameters corresponding to the output power in the waveform modulation parameter sequence to the waveform modulation module. The waveform modulation module modulates the sine wave output by the high-frequency power generator according to the waveform modulation parameters, and sends the waveform-modulated high-frequency electrical energy to the surgical electrode, thereby realizing closed-loop control of the high-frequency electrosurgical unit.

[0140] refer to Figure 5 , Figure 5 A high-frequency electrosurgical unit control system is shown, comprising: a high-frequency power generator, a waveform modulation module, an impedance sensor, an optical imaging sensor, an acoustic signal acquisition device, a main controller, and surgical electrodes, wherein:

[0141] The aforementioned high-frequency power generator is electrically connected to the aforementioned surgical electrode;

[0142] The waveform modulation module is communicatively connected to the high-frequency power generator.

[0143] The aforementioned high-frequency power generator, waveform modulation module, impedance sensor, optical imaging sensor, and acoustic signal acquisition device are all communicatively connected to the aforementioned main controller.

[0144] The aforementioned acoustic signal acquisition equipment is used to acquire acoustic signals generated by tissue vaporization and rupture when high-frequency electrical energy is applied to human tissue.

[0145] The aforementioned optical imaging sensor is used to acquire images of the microstructure beneath the surface of human tissues, providing early warning of subcutaneous blood vessels;

[0146] The waveform modulation module is used to receive the waveform modulation parameters sent by the main controller and to perform waveform modulation on the high-frequency electrical energy output by the high-frequency power generator.

[0147] The aforementioned impedance sensor is used to monitor the tissue impedance value and its changes between the surgical electrode and the tissue circuit in real time;

[0148] The aforementioned high-frequency power generator is used to receive the output power from the aforementioned main controller and output high-frequency electrical energy of corresponding amplitude;

[0149] The surgical electrodes described above are used to receive modulated high-frequency electrical energy output from the high-frequency power generator, and to apply modulated high-frequency electrical energy to human tissue.

[0150] The main controller receives data from various sensors, runs control algorithms based on this data, and sends control commands to the high-frequency power generator and waveform modulation module. A system component structure diagram of the high-frequency electrosurgical control system can be found by referring to... Figure 5 . Figure 5 The black control line indicates a communication connection. The red control line indicates an electrical connection.

[0151] like Figure 6 As shown, a high-frequency electrosurgical unit control device 600 in some embodiments includes: a control unit 601, a data acquisition unit 602, a signal acquisition unit 603, a feature extraction unit 604, a feature fusion unit 605, a generation unit 606, and an adjustment unit 607. The control unit 601 is configured to, in response to receiving a start command indicating the start of operation of the high-frequency electrosurgical unit, control a high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on preset initial operating parameters of the high-frequency electrosurgical unit; the data acquisition unit 602 is configured to perform multi-dimensional data acquisition on the human tissue of the target surgical area using an impedance sensor and an optical imaging sensor, respectively, to obtain a tissue impedance information sequence and a human tissue image sequence; the signal acquisition unit 603 is configured to acquire sound waves generated by the surgical electrode acting on the human tissue using a sound wave signal acquisition device, to obtain a sound wave signal, and to extract features from the sound wave signal to obtain a sound wave feature tensor; the feature extraction unit 604 is configured to extract features from the tissue impedance information sequence and the human tissue image sequence... The columns are subjected to feature extraction to obtain tissue impedance feature tensors and human tissue feature tensors respectively; the feature fusion unit 605 is configured to perform feature fusion processing on the above-mentioned acoustic feature tensors, the above-mentioned tissue impedance feature tensors and the above-mentioned human tissue feature tensors to obtain fused feature tensors; the generation unit 606 is configured to generate high-frequency electrosurgical control commands based on the above-mentioned fused feature tensors, wherein the above-mentioned high-frequency electrosurgical control commands include: an output power sequence and a waveform modulation parameter sequence; the adjustment unit 607 is configured to adjust the output power of the above-mentioned high-frequency power generator according to the above-mentioned output power sequence, and to control the waveform modulation module to perform waveform modulation on the high-frequency electrical energy output by the above-mentioned high-frequency power generator according to the above-mentioned waveform modulation parameter sequence, so as to realize closed-loop control of the high-frequency electrosurgical.

[0152] It is understandable that the units described in the high-frequency electrosurgical control device 600 are similar to those in the reference device. Figure 2 The steps in the described embodiments correspond to each other.

Claims

1. A high-frequency electrosurgical unit control device, characterized in that, include: The control unit is configured to, in response to receiving a start command characterizing the start of operation of the high-frequency electrosurgical unit, control the high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on preset initial operating parameters of the high-frequency electrosurgical unit; The data acquisition unit is configured to acquire multidimensional data of human tissue in the target surgical area through impedance sensors and optical imaging sensors, respectively, to obtain tissue impedance information sequence and human tissue image sequence. The signal acquisition unit is configured to acquire sound waves generated by the surgical electrode acting on human tissue via a sound wave signal acquisition device, obtain sound wave signals, and extract features from the sound wave signals to obtain sound wave feature tensors. The step of extracting features from the sound wave signals to obtain the sound wave feature tensors includes: The acoustic signal is subjected to acoustic coding processing to obtain an initial acoustic feature tensor; The initial acoustic wave feature tensor is convolved to obtain the acoustic wave convolution feature tensor; Batch normalization is performed on the acoustic wave convolution feature tensor to obtain a normalized acoustic wave feature tensor. Based on the normalized acoustic wave feature tensor and the preset activation function, an acoustic wave activation feature tensor is generated. The acoustic wave activation feature tensor is subjected to global max pooling to obtain the max pooled acoustic wave feature tensor. Fine feature extraction is performed on the max-pooling acoustic feature tensor to obtain the acoustic feature tensor. The feature extraction unit is configured to extract features from the tissue impedance information sequence and the human tissue image sequence respectively, to obtain a tissue impedance feature tensor and a human tissue feature tensor, wherein the step of extracting features from the tissue impedance information sequence and the human tissue image sequence respectively to obtain the tissue impedance feature tensor and the human tissue feature tensor includes: Generate a tissue impedance change rate sequence and a tissue impedance acceleration sequence corresponding to the tissue impedance information sequence; The tissue impedance information sequence, the tissue impedance change rate sequence, and the tissue impedance acceleration sequence are combined to obtain a combined tissue impedance information sequence. The tissue impedance combination information sequence is fully connected to obtain the fully connected tissue impedance combination information sequence as the fully connected tissue impedance feature tensor. The fully connected tissue impedance feature tensor is fully connected to obtain the fully connected tissue impedance feature tensor as the tissue impedance feature tensor. The human tissue image sequence is subjected to image encoding processing to obtain the tissue image feature tensor; Human tissue feature extraction is performed on the tissue image feature tensor to obtain a human tissue feature tensor, wherein the process of extracting human tissue features from the tissue image feature tensor to obtain the human tissue feature tensor includes: The tissue image feature tensor is convolved to obtain the first tissue image convolutional feature tensor; Based on the first tissue image convolutional feature tensor and the preset activation function, generate the first tissue image activation feature tensor; The activation feature tensor of the first tissue image is subjected to global average pooling to obtain the global average pooling tissue image feature tensor. The global average pooling tissue image feature tensor is convolved to obtain the second tissue image convolutional feature tensor. Batch normalization is performed on the second tissue image convolutional feature tensor to obtain the normalized tissue image feature tensor; A second tissue image activation feature tensor is generated based on the normalized tissue image feature tensor and the preset activation function. Global max pooling is performed on the second tissue image activation feature tensor to obtain the max pooled tissue image feature tensor. Fine feature extraction is performed on the max-pooled tissue image feature tensor to obtain the human tissue feature tensor; The feature fusion unit is configured to perform feature fusion processing on the acoustic feature tensor, the tissue impedance feature tensor, and the human tissue feature tensor to obtain a fused feature tensor. The generation unit is configured to generate high-frequency electrosurgical control instructions based on the fused feature tensor, wherein the high-frequency electrosurgical control instructions include: an output power sequence and a waveform modulation parameter sequence; The adjustment unit is configured to adjust the output power of the high-frequency power generator according to the output power sequence, and to control the waveform modulation module to modulate the high-frequency electrical energy output by the high-frequency power generator according to the waveform modulation parameter sequence, so as to realize closed-loop control of the high-frequency electrosurgical unit.

2. The apparatus according to claim 1, characterized in that, Before controlling the high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on the preset initial operating parameters of the high-frequency electrosurgical unit, the method further includes: Image data of the target surgical area is acquired using an image acquisition device to obtain an image of the area to be operated on. The image of the area to be operated on is input into a pre-trained human tissue recognition model to obtain the human tissue recognition result; Based on the human tissue identification results, the initial operating parameters of the high-frequency electrosurgical unit are determined.

3. The apparatus according to claim 1, characterized in that, The initial operating parameters of the high-frequency electrosurgical unit include: initial output power and initial waveform modulation parameters; And the control of the high-frequency power generator to output high-frequency electrical energy to the surgical electrode based on the preset initial operating parameters of the high-frequency electrosurgical unit, including: Based on the initial output power, control the high-frequency power generator to output high-frequency electrical energy; Based on the initial waveform modulation parameters, the waveform modulation module is controlled to modulate the high-frequency electrical energy output by the high-frequency power generator and output the waveform-modulated high-frequency electrical energy to the surgical electrode.

4. A high-frequency electrosurgical control system, applied to the high-frequency electrosurgical control device according to any one of claims 1 to 3, characterized in that, The high-frequency electrosurgical control system includes: a high-frequency power generator, a waveform modulation module, an impedance sensor, an optical imaging sensor, an acoustic signal acquisition device, a main controller, and surgical electrodes, wherein: The high-frequency power generator is electrically connected to the surgical electrode; The waveform modulation module is communicatively connected to the high-frequency power generator; The high-frequency power generator, the waveform modulation module, the impedance sensor, the optical imaging sensor, and the acoustic signal acquisition device are all communicatively connected to the main controller. The acoustic signal acquisition device is used to acquire acoustic signals generated by tissue vaporization and rupture when high-frequency electrical energy is applied to human tissue. The optical imaging sensor is used to acquire images of the microstructure beneath the surface of human tissue and to provide early warning of subcutaneous blood vessels. The waveform modulation module is used to receive waveform modulation parameters sent by the main controller and to perform waveform modulation on the high-frequency electrical energy output by the high-frequency power generator. The impedance sensor is used to monitor the tissue impedance value and its changes between the surgical electrode and the tissue circuit in real time. The high-frequency power generator is used to receive the output power from the main controller and output high-frequency electrical energy of corresponding amplitude. The surgical electrode is used to receive modulated high-frequency electrical energy output from the high-frequency power generator, and to apply modulated high-frequency electrical energy to human tissue. The main controller is used to receive data collected by each sensor, run control algorithms based on the data collected by each sensor, and send control commands to the high-frequency power generator and the waveform modulation module.

Images