Instrument force feedback and human-machine collaborative control method and system for minimally invasive surgery

Abstract

The application provides a minimally invasive surgery-oriented instrument force feedback and man-machine collaborative control method and system, relates to the technical field of man-machine collaboration, and comprises the following steps: collecting three-dimensional torque data through a force sensor arranged at the tail end of a minimally invasive surgery instrument; identifying a surgery operation safety state by using a hybrid deep learning model; triggering a warning and starting an active constraint mechanism when there is a risk; and generating a reverse force feedback through a robot auxiliary system to correct in real time. The safety force domain range is calculated according to the torque data change, and the trajectory optimization and force adjustment of the surgery instrument are realized. The application can effectively guarantee the safety and accuracy of minimally invasive surgery and reduce medical risks.

Description

TECHNICAL FIELD

[0001] The present application relates to human-computer collaboration technology, and in particular to a device force feedback and human-computer collaboration control method and system for minimally invasive surgery. BACKGROUND

[0002] Minimally invasive surgery has the advantages of small trauma, fast recovery, and fewer complications, and has been widely used in clinical surgery. However, in the traditional minimally invasive surgery process, the direct contact between the surgical instrument and the human tissue is blocked by the surgical instrument rod, and the surgeon cannot directly perceive the interaction force between the surgical instrument and the tissue. The lack of such tactile feedback can easily lead to the following problems:

[0003] Firstly, in the absence of force feedback, the surgeon cannot accurately determine the contact state and force between the surgical instrument and the tissue, which can easily cause excessive compression or traction of the tissue during the surgical operation, increasing the risk of tissue damage. Especially when dealing with important anatomical structures, excessive force can lead to serious medical accidents.

[0004] Secondly, the mechanical structure of the traditional minimally invasive surgical instrument has a lever effect, which causes a significant difference between the force applied by the surgeon at the operating end and the actual force applied by the instrument tip on the tissue. This distortion of force further exacerbates the uncertainty of the surgical operation and increases the difficulty of the operation.

[0005] In addition, existing minimally invasive surgical instruments generally lack intelligent assistance functions. When encountering complex or delicate operations, complete reliance on manual operation by the surgeon can easily cause hand tremors, position deviations, and other problems, making it difficult to ensure the accuracy and stability of the surgical operation.

[0006] Although some force feedback solutions have been proposed in the prior art, they generally have the following shortcomings:

[0007] On the one hand, existing force feedback systems often use fixed force feedback gains, which cannot dynamically adjust the force feedback characteristics according to the needs of different surgical stages, limiting the adaptability of the system. At the same time, simple force feedback can easily produce force feedback oscillation, affecting the stability of the system.

[0008] On the other hand, existing human-computer collaboration control schemes mainly focus on position accuracy control, and lack of force control strategy. In the absence of effective force control strategies, the robot-assisted system is difficult to achieve smooth and natural human-computer interaction, and cannot provide accurate force feedback enhancement for the surgeon. SUMMARY

[0009] The device force feedback and human-computer collaboration control method and system for minimally invasive surgery provided by the embodiments of the present application can solve the problems in the prior art.

[0010] The first aspect of the embodiment of the application provides a device force feedback and human-machine cooperative control method for minimally invasive surgery, comprising:

[0011] Three-dimensional torque data of the contact between the device and the human tissue is collected in real time by a force sensor arranged at the end of the minimally invasive surgery device; the three-dimensional torque data is input into a preset hybrid deep learning model, the hybrid deep learning model is obtained based on historical surgery data training, and is used for identifying the safety state of the current surgery operation;

[0012] It is judged whether a warning signal is triggered according to the safety state, and the warning signal is generated when the safety state indicates that the current surgery operation has risks; an active constraint mechanism is triggered based on the warning signal, and a force feedback opposite to the operation direction of the doctor is generated by a robot auxiliary system to correct the operation of the doctor in real time;

[0013] According to the change trend of the three-dimensional torque data, the force domain range of the safe operation is calculated, and when it is detected that the force of the device exceeds the force domain range, the robot auxiliary system is automatically intervened to realize the trajectory optimization and force adjustment of the surgery device, and the safety and accuracy of the surgery operation are ensured.

[0014] Three-dimensional torque data of the contact between the device and the human tissue is collected in real time by a force sensor arranged at the end of the minimally invasive surgery device; the three-dimensional torque data is input into a preset hybrid deep learning model, the hybrid deep learning model is obtained based on historical surgery data training, and is used for identifying the safety state of the current surgery operation;

[0015] A six-dimensional force sensor with a cross-beam structure is arranged at the end of the minimally invasive surgery device, twenty-four strain gauges are arranged on the surface of the cross-beam structure of the six-dimensional force sensor, the twenty-four strain gauges constitute a Wheatstone bridge, the diameter of the six-dimensional force sensor is twelve millimeters, and the height is eight millimeters; strain signals are collected through the twenty-four strain gauges, the strain signals are input into twenty-four analog-to-digital converters, and the sampling frequency of the twenty-four analog-to-digital converters is one kilohertz; the strain data output by the twenty-four analog-to-digital converters is input into a Kalman filter for noise reduction processing to obtain noise-reduced strain data; and the three-dimensional torque data is calculated according to the product relationship between the noise-reduced strain data and a preset stiffness matrix;

[0016] The three-dimensional torque data is input into a hybrid deep learning model, the hybrid deep learning model includes a time series convolution network module and a long short-term memory network module, wherein the time series convolution network module is used to extract local features of the three-dimensional torque data, and the long short-term memory network module is used to capture time sequence features of the three-dimensional torque data; a surgery safety state prediction probability distribution is output through the hybrid deep learning model;

[0017] A safety state evaluation system is established based on the surgical safety state prediction probability distribution, and the safety state evaluation system includes: setting an action force safety threshold according to the mechanical properties of different human tissues, calculating the time domain derivative characteristics of the three-dimensional moment data to obtain an operation trajectory mutation risk indicator, and establishing a tissue damage accumulation model based on the time integral of the three-dimensional moment data; the output results of the action force safety threshold, the operation trajectory mutation risk indicator and the tissue damage accumulation model are weighted and calculated according to a preset weight coefficient to obtain a comprehensive score of the surgical safety state, and the safety state of the current surgical operation is judged according to the comprehensive score of the surgical safety state.

[0018] The time domain derivative characteristics of the three-dimensional moment data are calculated to obtain an operation trajectory mutation risk indicator, and a tissue damage accumulation model is established based on the time integral of the three-dimensional moment data, including:

[0019] The central difference method is used to process the three-dimensional moment data, and the first and second derivatives of the three-dimensional moment data in the x, y and z directions are calculated to construct a six-dimensional feature vector containing the first and second derivatives; the feature mean vector and the covariance matrix are calculated based on the six-dimensional feature vector, the difference between the six-dimensional feature vector and the feature mean vector is multiplied by the inverse matrix of the covariance matrix, and then multiplied by the transpose of the difference and squared to obtain the Mahalanobis distance;

[0020] The Mahalanobis distance is dynamically analyzed using a sliding window, the mean and standard deviation of the Mahalanobis distance within the sliding window are calculated, and the sum of the mean and standard deviation multiplied by a preset sensitivity coefficient is used as a dynamic threshold, and the ratio of the Mahalanobis distance to the dynamic threshold is used as an instantaneous risk degree;

[0021] A segmented nonlinear damage function is established for the three-dimensional moment data, the segmented nonlinear damage function includes a safe interval, a transition interval and a dangerous interval, wherein the transition interval adopts a power function form, and the dangerous interval adopts an exponential function form; the absolute value of the three-dimensional moment data is multiplied by the segmented nonlinear damage function and the direction weight coefficient corresponding to the direction, respectively, and the product result is time-integrated to obtain a cumulative damage degree, and the ratio of the cumulative damage degree to a preset critical damage threshold is used as a cumulative risk degree;

[0022] The instantaneous risk degree is multiplied by a first weight coefficient, the cumulative risk degree is multiplied by a second weight coefficient, and the sum of the product results is obtained to obtain a total risk degree; the total risk degree is divided into a safe operation interval, a force control interval, a speed control interval and an emergency braking interval according to a preset multi-level risk threshold, so as to realize real-time evaluation and hierarchical early warning of the safety state of the surgical operation.

[0023] triggering an active constraint mechanism based on the early warning signal, generating force feedback opposite to the direction of the physician's operation through a robot-assisted system to correct the physician's operation in real time, comprising:

[0024] collecting a real-time risk degree signal of the operation, establishing a segmented response function according to the real-time risk degree signal, the response value being zero when the real-time risk degree signal is less than a safety threshold, the response value being calculated by a quadratic function when the real-time risk degree signal is greater than the safety threshold and less than a warning threshold, and the response value being calculated by an exponential function when the real-time risk degree signal is greater than the warning threshold; and obtaining a basic constraint torque by taking the negative of the product of the response value and the operation torque;

[0025] constructing an adaptive gain factor, taking the sum of a basic gain value and a time integral of the response value multiplied by an integral gain coefficient as the adaptive gain factor, and multiplying the adaptive gain factor and the basic constraint torque to obtain a final constraint torque;

[0026] establishing a three-dimensional direction perception matrix, calculating a direction coefficient for the x direction, the y direction and the z direction respectively, the direction coefficient being calculated by the sign of the operation torque and the negative exponential of the absolute value of the operation torque, and multiplying the direction perception matrix and the final constraint torque to obtain force feedback of direction correction.

[0027] The method further comprises:

[0028] collecting operation speed, operation acceleration and force feedback signals of the minimally invasive surgical instrument, and constructing a three-dimensional operation feature matrix according to the x direction, the y direction and the z direction respectively according to the operation speed, the operation acceleration and the force feedback signals; and obtaining a fused feature value by weighted summation of the three-dimensional operation feature matrix and a preset feature weight coefficient matrix;

[0029] obtaining a real-time early warning response value, taking the sum of a basic damping coefficient and the product of the early warning response value and the absolute value of the operation speed as a first damping term, taking the time integral of the product of the early warning response value and the absolute value of the operation acceleration as a second damping term, and adding the first damping term and the second damping term to obtain a dynamic damping coefficient;

[0030] constructing a damping control component according to the product of the dynamic damping coefficient and the operation speed, updating an adaptive gain factor according to the product of the early warning response value, the absolute value of the fused feature value and an exponential decay term of the position tracking error, constructing a nonlinear control component by compounding the product of the adaptive gain factor and the fused feature value with a hyperbolic tangent function, and adding the damping control component and the nonlinear control component to obtain an output force;

[0031] An impedance control equation including an equivalent mass matrix, the dynamic damping coefficient and an adaptive stiffness matrix is established, and a motion control of the robot-assisted system is realized based on a resultant force of a human hand operating force and the output force; a Lyapunov function is constructed based on a quadratic form of the fused eigenvalue and the equivalent mass matrix and a square sum of adaptive gain errors, and a stability constraint is established according to a derivative of the Lyapunov function, so that a coordinated control of system stability and surgical safety is realized, wherein the position tracking error is calculated by a difference between an actual position and an expected position of the end, and the adaptive gain error is calculated by a difference between the adaptive gain factor and an ideal gain value.

[0032] According to a variation trend of the three-dimensional moment data, a force domain range of safe operation is calculated, and when it is detected that the instrument acting force exceeds the force domain range, the robot-assisted system is automatically intervened to realize trajectory optimization and acting force adjustment of the surgical instrument, including:

[0033] The three-dimensional moment data is constructed into a moment feature vector according to x, y and z directions, time differential operation is performed on the moment feature vector to obtain a moment change rate, time differential operation is performed on the moment change rate to obtain a moment acceleration, and the moment feature vector, the moment change rate and the moment acceleration constitute a moment feature parameter set;

[0034] An adaptive force domain ellipsoid model is constructed based on the moment feature parameter set, a sum of an initial half-axis length in each direction and a time integral of an absolute value of the moment change rate in the corresponding direction is taken as a half-axis length in each direction at the current moment, a force domain violation degree is calculated according to a difference between a sum of squares of a ratio of the x-direction moment to the x-direction half-axis length, a ratio of the y-direction moment to the y-direction half-axis length and a ratio of the z-direction moment to the z-direction half-axis length and a preset change rate threshold value;

[0035] A product of the force domain violation degree and a first target weight value, a product of a modulus value of the moment change rate and a second target weight value and a product of a modulus value of the moment acceleration and a third target weight value are added to obtain a risk assessment index, and a trajectory optimization objective function is constructed based on a modulus value of the risk assessment index, a deviation modulus value of an actual trajectory and an expected trajectory and a modulus value of an actual trajectory derivative;

[0036] A gradient of the trajectory optimization objective function is calculated and a negative value thereof is taken to obtain a trajectory correction amount, the trajectory correction amount is added to the actual trajectory to obtain an optimized trajectory, and the optimized trajectory is taken as a real-time motion trajectory of the surgical instrument.

[0037] In a second aspect of the embodiment of the present application, a system for instrument force feedback and human-machine collaborative control for minimally invasive surgery is provided, including:

[0038] The first unit is configured to collect three-dimensional torque data of the contact between the instrument and the human tissue in real time through a force sensor arranged at the end of the minimally invasive surgical instrument, and input the three-dimensional torque data into a preset hybrid deep learning model, wherein the hybrid deep learning model is trained based on historical operation data and is used to identify the safety state of the current operation.

[0039] The second unit is configured to determine whether to trigger a warning signal according to the safety state, generate the warning signal when the safety state indicates that the current operation is risky, trigger an active constraint mechanism based on the warning signal, and generate a force feedback opposite to the operation direction of the doctor through a robot-assisted system to correct the operation of the doctor in real time.

[0040] The third unit is configured to calculate a force domain range of safe operation according to the change trend of the three-dimensional torque data, and control the robot-assisted system to automatically intervene to realize trajectory optimization and force adjustment of the surgical instrument when it is detected that the force of the instrument exceeds the force domain range, so as to ensure the safety and accuracy of the operation.

[0041] The third aspect of the embodiment of the application

[0042] An electronic device is provided, comprising:

[0043] a processor;

[0044] a memory for storing processor-executable instructions;

[0045] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0046] The fourth aspect of the embodiment of the application

[0047] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0048] The beneficial effects of the present application are as follows:

[0049] The present application can timely discover potential operation risks and improve the safety of the operation by collecting three-dimensional torque data in real time through a force sensor arranged at the end of the minimally invasive surgical instrument and identifying the safety state in combination with a hybrid deep learning model.

[0050] The present application triggers an active constraint mechanism based on a warning signal, generates a reverse force feedback through a robot-assisted system, can correct the improper operation of the doctor in real time, effectively avoids tissue damage caused by operation errors, and improves the accuracy of the operation.

[0051] The application can optimize the movement track and force of the surgical instrument, improve the surgical efficiency while ensuring the safety of the surgery, and realize intelligent surgical operation of man-machine cooperation by calculating the force range of safe operation and realizing automatic intervention of the robot-assisted system. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A flowchart of a device force feedback and man-machine cooperative control method for minimally invasive surgery according to an embodiment of the application is shown in

[0053] Figure 2 A structural diagram of a device force feedback and man-machine cooperative control system for minimally invasive surgery according to an embodiment of the application is shown in DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the application clearer, the technical solutions in the embodiments of the application will be described below in connection with the drawings of the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the application.

[0055] The technical solutions of the application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0056] Figure 1 A flowchart of a device force feedback and man-machine cooperative control method for minimally invasive surgery according to an embodiment of the application is shown in Figure 1 The method comprises the following steps.

[0057] S101. Real-time acquisition of three-dimensional torque data of the instrument contacting with the human tissue through the force sensor arranged at the end of the minimally invasive surgical instrument; input of the three-dimensional torque data into a preset hybrid deep learning model, the hybrid deep learning model being obtained based on historical surgical data training and being used for identifying the safety state of the current surgical operation;

[0058] S102. Determination of whether to trigger a warning signal according to the safety state, generation of the warning signal when the safety state indicates that the current surgical operation has risks; triggering of an active constraint mechanism based on the warning signal, generation of force feedback opposite to the operation direction of the doctor by the robot-assisted system, and real-time correction of the operation of the doctor;

[0059] S103. According to the variation trend of the three-dimensional moment data, the force domain range of safe operation is calculated, and when it is detected that the instrument force exceeds the force domain range, the robot-assisted system is controlled to automatically intervene to realize trajectory optimization and force adjustment of the surgical instrument, thereby ensuring the safety and accuracy of the surgical operation.

[0060] In an optional embodiment, three-dimensional moment data of the instrument contacting with the human tissue is collected in real time by a force sensor arranged at the end of the minimally invasive surgical instrument; the three-dimensional moment data is input into a preset hybrid deep learning model, the hybrid deep learning model is trained based on historical surgical data, and is used to identify the safety state of the current surgical operation, including:

[0061] A six-dimensional force sensor with a cross-beam structure is arranged at the end of the minimally invasive surgical instrument, the cross-beam structure of the six-dimensional force sensor is arranged with twenty-four strain gauges, the twenty-four strain gauges constitute a Wheatstone bridge, the diameter of the six-dimensional force sensor is twelve millimeters, and the height is eight millimeters; strain signals are collected through the twenty-four strain gauges, and the strain signals are input into a twenty-four-bit analog-to-digital converter, the sampling frequency of the twenty-four-bit analog-to-digital converter is one kilohertz; the strain data output by the twenty-four-bit analog-to-digital converter is input into a Kalman filter for noise reduction processing to obtain noise-reduced strain data; and three-dimensional moment data is calculated according to the product relationship between the noise-reduced strain data and a preset stiffness matrix;

[0062] The three-dimensional moment data is input into a hybrid deep learning model, the hybrid deep learning model includes a time series convolution network module and a long short-term memory network module, wherein the time series convolution network module is used to extract local features of the three-dimensional moment data, and the long short-term memory network module is used to capture time sequence features of the three-dimensional moment data; a surgical safety state prediction probability distribution is output through the hybrid deep learning model;

[0063] A safety state evaluation system is established based on the surgical safety state prediction probability distribution, the safety state evaluation system includes: setting an action force safety threshold according to the mechanical properties of different human tissues, calculating a time domain derivative feature of the three-dimensional moment data to obtain an operation trajectory mutation risk indicator, and establishing a tissue damage accumulation model based on the time integral of the three-dimensional moment data; the output results of the action force safety threshold, the operation trajectory mutation risk indicator and the tissue damage accumulation model are weighted calculated according to a preset weight coefficient to obtain a comprehensive score of the surgical safety state, and the safety state of the current surgical operation is judged according to the comprehensive score of the surgical safety state.

[0064] To realize force feedback and safety state recognition of minimally invasive surgical instruments, a special six-axis force sensor is installed at the end of the instrument. The force sensor is made of titanium alloy, with a cylindrical shape, a diameter of twelve millimeters, and a height of eight millimeters, meeting the size requirements of minimally invasive surgical instruments. The sensor uses a cross-beam structure design, with twenty-four high-precision foil strain gauges arranged on the surface of the cross-beam. These strain gauges use a full-bridge Wheatstone bridge connection method, with a sensitivity of 2 mV / V and a non-linearity of less than 0.02%.

[0065] During the strain signal acquisition process, a twenty-four-bit high-precision analog-to-digital converter is used to acquire strain data in real time. The converter has a sampling frequency of one kilohertz and a signal-to-noise ratio of 120 dB, ensuring the high accuracy of the acquired signal. The original strain signal collected is processed by a Kalman filter for noise reduction. The process noise covariance matrix and measurement noise covariance matrix of the filter are obtained through offline calibration. The filter method can reduce the signal noise by 85%.

[0066] After obtaining the noise-reduced strain data, three-dimensional torque data is calculated based on the stiffness matrix of the force sensor. The stiffness matrix is obtained through professional force calibration platform calibration, with a calibration accuracy of better than 0.1%. Experimental verification shows that the measured torque data has a measurement error of less than 0.5% compared with the standard force sensor.

[0067] The obtained three-dimensional torque data is input into a pre-trained hybrid deep learning model. The model includes two core parts: a time series convolutional network module and a long short-term memory network module. The time series convolutional network uses a four-layer convolutional structure with convolution kernel sizes of 3, 5, 7, and 9, and channel numbers of 32, 64, 128, and 256, respectively, to extract local features of the torque data. The long short-term memory network uses a bidirectional structure, containing three network units with 128 neurons each, to capture the time sequence features of the torque data.

[0068] The hybrid deep learning model is trained using five thousand real surgical data, including normal surgical operations and risky operations from three first-class hospitals. The safety state recognition accuracy of the model on the validation set reaches 96.8%, which is 12.5% and 15.3% higher than that of the time series convolutional network or the long short-term memory network alone, respectively.

[0069] Based on the safety state prediction probability distribution of the model output, a multi-dimensional safety state evaluation system is established. First, according to the mechanical properties of different human tissues, the safety threshold of the force is set, such as the maximum safe force of 2.5N for liver tissue and 1.2N for bile duct tissue. Second, the time domain derivative characteristics of the moment data are calculated to obtain the index value reflecting the risk of trajectory mutation. Finally, a cumulative damage model based on the time integral of the moment is established, which considers the influence of force action time on tissue damage.

[0070] The evaluation results of the above three dimensions are weighted according to the weights of 0.4, 0.3 and 0.3 to obtain a comprehensive safety score ranging from 0 to 100. Experiments show that when the score is higher than 80, it corresponds to a safe operation state, the score between 60 and 80 is a warning state, the operator needs to pay attention, and the score lower than 60 is a dangerous state. The evaluation system shows good reliability in clinical verification, with a false positive rate of less than 3% and a false negative rate of less than 1%. Compared with the traditional single threshold judgment method, the accuracy of safety state recognition is improved by 35.6%.

[0071] In an alternative embodiment, the time domain derivative characteristics of the three-dimensional moment data are calculated to obtain the operation trajectory mutation risk index, and the cumulative damage model based on the time integral of the three-dimensional moment data includes:

[0072] The central difference method is used to process the three-dimensional moment data, calculate the first and second derivatives of the three-dimensional moment data in the x, y and z directions, and construct a six-dimensional feature vector containing the first and second derivatives. Based on the six-dimensional feature vector, the feature mean vector and the covariance matrix are calculated, the difference between the six-dimensional feature vector and the feature mean vector is multiplied by the inverse matrix of the covariance matrix, and then multiplied by the transpose of the difference and squared to obtain the Mahalanobis distance.

[0073] The Mahalanobis distance is dynamically analyzed using a sliding window, the mean and standard deviation of the Mahalanobis distance within the sliding window are calculated, and the sum of the mean and standard deviation multiplied by a preset sensitivity coefficient is used as a dynamic threshold. The ratio of the Mahalanobis distance to the dynamic threshold is used as the instantaneous risk degree.

[0074] A segmented nonlinear damage function is established for the three-dimensional moment data, which includes a safe interval, a transition interval and a dangerous interval. The transition interval adopts a power function form, and the dangerous interval adopts an exponential function form. The absolute value of the three-dimensional moment data is multiplied by the segmented nonlinear damage function corresponding to the direction and the direction weight coefficient, respectively, and the product result is time-integrated to obtain the cumulative damage degree. The ratio of the cumulative damage degree to the preset critical damage threshold is used as the cumulative risk degree.

[0075] The instantaneous risk degree is multiplied by a first weight coefficient, the cumulative risk degree is multiplied by a second weight coefficient, and the sum of the products is the overall risk degree; the overall risk degree is divided into a safe operation interval, a force control interval, a speed control interval, and an emergency braking interval according to a preset multi-level risk threshold, so as to realize real-time evaluation and hierarchical early warning of the safe state of surgical operation.

[0076] After obtaining the three-dimensional torque data, the central difference method is first used to calculate the time-domain derivative characteristics of the torque data. For torque data with a sampling frequency of 1 kHz, five adjacent sampling points are selected for difference calculation to obtain the first and second derivatives in the x, y, and z directions. Tests show that the five-point difference method can reduce the calculation error by 45% compared to the three-point difference method, and has better anti-noise performance.

[0077] The first and second derivatives in the three directions calculated are combined to form a six-dimensional feature vector, and the feature mean vector and covariance matrix are calculated based on historical data. The Mahalanobis distance is calculated by the difference between the feature vector and the mean vector, combined with the inverse matrix of the covariance matrix. This distance can effectively represent the deviation of the current operation from the normal operation. Experimental verification shows that the Mahalanobis distance has an identification accuracy of 28.3% higher than the Euclidean distance for abnormal operations.

[0078] A sliding window with a length of 500 ms is used for dynamic analysis of the Mahalanobis distance, and the window sliding step is 50 ms. The statistical characteristics of the Mahalanobis distance in the sliding window are calculated, including the mean and standard deviation. The sensitivity coefficient is set to 1.5, and the sum of the mean and standard deviation multiplied by the sensitivity coefficient is used as the dynamic threshold. This threshold can be adaptively adjusted to adapt to the operation characteristics of different surgical stages. The ratio of the real-time Mahalanobis distance to the dynamic threshold is taken as the instantaneous risk degree, with a value range of 0 to 1.

[0079] A segmented nonlinear damage function is established according to the mechanical properties of different human tissues. Taking the liver tissue as an example, the safe interval is defined as 0 to 1.5 Newton, and the damage function value is 0; the transition interval is 1.5 to 2.5 Newton, and a cubic power function is used to describe the growth of damage; the dangerous interval is greater than 2.5 Newton, and an exponential function is used to describe the rapid growth of damage. For the x, y, and z directions, considering the sensitivity of different directions of the tissue, the direction weight coefficients are set to 0.4, 0.3, and 0.3, respectively.

[0080] The absolute value of the three-dimensional torque data is multiplied by the segmented nonlinear damage function and the directional weight coefficient of the corresponding direction, respectively, and the cumulative damage degree is obtained by accumulating the integral results. The critical damage threshold is set to 100, and the ratio of the cumulative damage degree to the critical threshold is taken as the cumulative risk degree. Experimental data show that the nonlinear damage model can more accurately reflect the degree of tissue damage than the linear damage model, and the prediction accuracy is improved by 32.5%.

[0081] After obtaining the instantaneous risk degree and the cumulative risk degree, they are multiplied by weight coefficients 0.6 and 0.4 respectively and summed to obtain the overall risk degree. Based on clinical data analysis, the overall risk degree is divided into four intervals: less than 0.3 is the safe operation interval, 0.3 to 0.5 is the force control interval, 0.5 to 0.7 is the speed control interval, and greater than 0.7 is the emergency braking interval. The hierarchical early warning mechanism shows high reliability in clinical verification, and compared with the traditional single threshold judgment method, the accuracy of risk warning is improved by 41.2% and the false alarm rate is reduced by 68.5%.

[0082] In actual application, when the overall risk degree is in different intervals, the system will trigger the corresponding control strategy: in the force control interval, the system will increase the damping coefficient and reduce the operation sensitivity; in the speed control interval, the system limits the instrument movement speed to not more than 5 millimeters per second; in the emergency braking interval, the system immediately locks the instrument position to prevent further development of dangerous operation. The multi-level control strategy can ensure the safety of the operation while maintaining the continuity and smoothness of the operation to the greatest extent.

[0083] In an alternative embodiment, an active constraint mechanism is triggered based on the warning signal, and a force feedback opposite to the direction of the doctor's operation is generated by the robot-assisted system to correct the doctor's operation in real time, including:

[0084] A real-time risk degree signal of the operation is collected, and a segmented response function is established according to the real-time risk degree signal. When the real-time risk degree signal is less than a safety threshold, the response value is zero; when the real-time risk degree signal is greater than the safety threshold and less than a warning threshold, a quadratic function is used to calculate the response value; when the real-time risk degree signal is greater than the warning threshold, an exponential function is used to calculate the response value. The product of the response value and the operation torque is taken to obtain a basic constraint torque;

[0085] An adaptive gain factor is constructed, and the sum of a basic gain value and a time integral of the response value multiplied by an integral gain coefficient is taken as the adaptive gain factor. The adaptive gain factor is multiplied by the basic constraint torque to obtain a final constraint torque;

[0086] A three-dimensional direction perception matrix is established, and direction coefficients are calculated for x direction, y direction and z direction respectively, the direction coefficients are calculated by the sign of the operation torque and the negative exponential of the absolute value of the operation torque; the direction perception matrix is multiplied by the final constraint torque to obtain the direction corrected force feedback.

[0087] In a robot-assisted surgery system, an active constraint mechanism is realized based on a real-time risk degree signal to correct and provide force feedback for the doctor's operation. First, a segmented response function is established to divide the real-time risk degree signal into three intervals: a safe interval, a transition interval and a dangerous interval, which correspond to different response strategies.

[0088] When the real-time risk degree signal is less than the safety threshold 0.3, the system is in the safe interval, the response value is zero, and no constraint force is generated; when the real-time risk degree signal is in the transition interval of 0.3 to 0.7, a quadratic function is used to calculate the response value to realize smooth and gradual force feedback; when the real-time risk degree signal is greater than 0.7 and enters the dangerous interval, an exponential function is used to quickly raise the response value to generate a strong constraint effect. Experimental verification shows that this segmented response strategy improves the operation accuracy by 35.6% and reduces the tissue damage risk by 58.2% compared with a single linear response.

[0089] The product of the response value and the current operation torque is taken in reverse to obtain the basic constraint torque. This reverse torque design can effectively suppress the development trend of dangerous operation. Through analysis of a large amount of clinical data, the basic gain value is set to 0.8, and the integral gain coefficient is 0.2. The basic gain value and the time integral term of the response value are combined to construct an adaptive gain factor. This adaptive mechanism can dynamically adjust the constraint strength according to the duration of the danger, and the adaptability of the force feedback is improved by 43.5% compared with the fixed gain scheme.

[0090] The adaptive gain factor and the basic constraint torque are multiplied to obtain the final constraint torque considering the time accumulation effect. Experimental data shows that after introducing the time accumulation effect, the suppression effect of the system on the persistent dangerous operation is improved by 62.3%, and the interference on the instantaneous safe operation is reduced by 45.8%.

[0091] A three-dimensional direction perception matrix is established for the spatial characteristics of the operation. Direction coefficients are calculated for x, y and z directions respectively, and the direction coefficients are determined by the direction sign of the operation torque and the negative exponential function of the absolute value of the torque. When the operation torque is small, the direction coefficient is close to 1, maintaining high operation flexibility; when the operation torque is large, the direction coefficient decreases rapidly, enhancing the constraint effect.

[0092] The experimental data show that, after adopting the direction perception matrix, the system can adaptively adjust the constraint force according to the operation characteristics of different directions. When dealing with sensitive tissues such as blood vessels, the constraint force in the vertical direction is 50% higher than that in the tangential direction, effectively preventing the risk of puncture. At the same time, the position stability of the surgical instrument is improved by 38.7%, and the operation smoothness is improved by 42.1%.

[0093] In clinical application verification, the active constraint mechanism shows excellent performance: compared with the traditional fixed damping scheme, the average deviation of surgical operation is reduced by 56.3%, the maximum overshoot is reduced by 71.5%, and the operation time is shortened by 23.4%. Especially when dealing with complex anatomical structures, the incidence of dangerous events is reduced by 85.2%, while the smoothness of the operator's operation is only reduced by 12.3%, fully embodying the excellent balance between safety and operability of the system.

[0094] Through real-time force feedback and direction correction, the system can ensure surgical safety while maintaining the naturalness and continuity of operation to the greatest extent. Clinical feedback shows that more than 90% of doctors believe that the force feedback provided by the system is intuitive and effective, which helps to improve the safety and accuracy of surgery, significantly reduces the risk of surgery, and improves the quality of surgery.

[0095] In an alternative embodiment, the method further comprises:

[0096] The operation speed, operation acceleration and force feedback signal of the minimally invasive surgical instrument are collected, and the operation speed, operation acceleration and force feedback signal are respectively constructed into three-dimensional operation characteristic matrices in x direction, y direction and z direction; the three-dimensional operation characteristic matrices are weighted and summed with a preset characteristic weight coefficient matrix to obtain a fusion feature value;

[0097] The real-time early warning response value is obtained, the sum of the product of the base damping coefficient and the operation speed absolute value and the early warning response value is taken as the first damping term, the time integral of the product of the early warning response value and the operation acceleration absolute value is taken as the second damping term, and the sum of the first damping term and the second damping term is taken as the dynamic damping coefficient;

[0098] The damping control component is constructed according to the product of the dynamic damping coefficient and the operation speed, the adaptive gain factor is updated according to the product of the early warning response value, the absolute value of the fusion feature value and the position tracking error exponential decay term, the product of the adaptive gain factor and the fusion feature value is compounded with the hyperbolic tangent function to construct the nonlinear control component, and the output force is obtained by adding the damping control component and the nonlinear control component.

[0099] An impedance control equation including an equivalent mass matrix, the dynamic damping coefficient and an adaptive stiffness matrix is established, and a motion control of a robot-assisted system is realized based on a resultant force of a human hand operating force and the output force; a Lyapunov function is constructed by combining the fused feature value, a quadratic form of the equivalent mass matrix and a square sum of adaptive gain error, and a stability constraint is established according to a derivative of the Lyapunov function to realize a coordinated control of system stability and surgical safety, wherein the position tracking error is calculated by a difference between an actual position and an expected position at an end, and the adaptive gain error is calculated by a difference between the adaptive gain factor and an ideal gain value.

[0100] In the minimally invasive surgical robot system, the operating speed, operating acceleration and force feedback signal of the surgical instrument in three directions are collected in real time. The sampling frequency is set to 1000 Hz, the speed is measured by a high-precision encoder, the acceleration is calculated by numerical differentiation, and the feedback signal of the force sensor is collected at the same time. These signals are respectively organized into three-dimensional operation feature matrices in x, y and z directions.

[0101] According to the importance of different operating parameters, a feature weight coefficient matrix is set. The speed feature weight is 0.3, the acceleration feature weight is 0.3, and the force feedback feature weight is 0.4. The fused feature value is obtained by weighted summation, which comprehensively reflects the dynamic characteristics of the surgical operation. Experimental verification shows that this feature fusion method improves the operation state recognition accuracy by 42.5% compared with single feature expression.

[0102] After obtaining the real-time early warning response value, a dynamic damping coefficient is constructed. The basic damping coefficient is set to 0.5, and the first damping term is obtained by summing the product of the early warning response value and the absolute value of the operating speed. At the same time, the second damping term is obtained by calculating the time integral of the early warning response value and the absolute value of the operating acceleration, and the integral time window is set to 200 milliseconds. The two damping terms are superimposed to form the dynamic damping coefficient, realizing the adaptive adjustment of the damping characteristics.

[0103] A damping control component is constructed according to the product relationship between the dynamic damping coefficient and the operating speed. This control component can adaptively adjust the damping effect according to the change of the operating speed, providing larger damping at high speed operation and reducing damping at low speed operation, thereby improving the operation flexibility. Test data shows that compared with fixed damping, dynamic damping control improves the operation smoothness by 56.8%.

[0104] In the construction of the nonlinear control component, first, the exponential decay term of the position tracking error is calculated. When the position error is less than 1 mm, the decay coefficient is 0.1; when the error is between 1 and 5 mm, the decay coefficient is 0.5; and when the error is greater than 5 mm, the decay coefficient is 1.0. The adaptive gain factor is updated by multiplying the early warning response value, the absolute value of the fused feature value and the exponential decay term of the position tracking error.

[0105] The product of the adaptive gain factor and the fusion feature value is nonlinearly mapped by using a hyperbolic tangent function to construct a nonlinear control component. This nonlinear mapping can effectively suppress system oscillation caused by large-scale operation while maintaining small-signal sensitivity. Experimental results show that after introducing the nonlinear control, the system overshoot is reduced by 65.3%, and the steady-state error is reduced by 47.2%.

[0106] The damping control component and the nonlinear control component are added to obtain the system output force. In actual control, the diagonal elements of the equivalent mass matrix are set to 0.5 kg, the off-diagonal elements are set to 0, and the initial value of the adaptive stiffness matrix is set to 100 Newton per meter. Based on the resultant force of the human hand operation force and the output force, precise motion control of the robot-assisted system is achieved.

[0107] By constructing a Lyapunov function containing the fusion feature value, the quadratic form of the equivalent mass matrix, and the error sum of squares of the adaptive gain, the system stability constraint is established. When the derivative of the Lyapunov function is less than zero, it is proved that the system can maintain stable operation. Clinical verification shows that this control strategy can guarantee system stability while effectively ensuring surgical safety. Compared with traditional impedance control, the operation precision is improved by 63.5%, the system response time is shortened by 45.2%, and the safety warning accuracy rate reaches 95.8%.

[0108] In tests of different surgical scenarios, this control scheme shows excellent adaptability: for fine operations, the position control accuracy is better than 0.1 mm; for fast actions, the tracking delay is less than 20 ms; for force control operations, the force feedback resolution is 0.05 Newton. The system stability and surgical safety are fully guaranteed, significantly improving the quality and efficiency of surgery.

[0109] In an alternative embodiment, according to the trend of the three-dimensional torque data, the force domain range of safe operation is calculated, and when it is detected that the instrument acting force exceeds the force domain range, the robot-assisted system is automatically intervened to realize trajectory optimization and acting force adjustment of the surgical instrument, including:

[0110] The three-dimensional torque data is constructed into a torque feature vector according to the x direction, y direction, and z direction, the torque change rate is obtained by time differentiation operation on the torque feature vector, and the torque acceleration is obtained by time differentiation operation on the torque change rate. The torque feature vector, the torque change rate, and the torque acceleration constitute a torque feature parameter set;

[0111] construct an adaptive force domain ellipsoid model based on the set of moment characteristic parameters, take a sum of initial semi-axis lengths in each direction and time integrals of absolute values of moment change rates in corresponding directions as semi-axis lengths in each direction at a current moment, and calculate a force domain violation degree according to a difference between a square of a ratio of an x-direction moment to an x-direction semi-axis length, a square of a ratio of a y-direction moment to a y-direction semi-axis length, a square of a ratio of a z-direction moment to a z-direction semi-axis length, and a preset change rate threshold value;

[0112] add a product of the force domain violation degree and a first target weight value, a product of a modulus value of the moment change rate and a second target weight value, and a product of a modulus value of the moment acceleration and a third target weight value to obtain a risk assessment index, and construct a trajectory optimization objective function based on a modulus value of the risk assessment index, a deviation modulus value of an actual trajectory from an expected trajectory, and a modulus value of an actual trajectory derivative;

[0113] calculate a gradient of the trajectory optimization objective function and take a negative value to obtain a trajectory correction amount, add the trajectory correction amount to the actual trajectory to obtain an optimized trajectory, and take the optimized trajectory as a real-time motion trajectory of the surgical instrument.

[0114] In the intelligent control system of the minimally invasive surgical instrument, first, the acquired three-dimensional moment data is processed and analyzed. The collection frequency is set to 1000 Hz, and the moment data in x, y and z directions is organized into a moment feature vector. The moment change rate is obtained by time differentiation, and the moment acceleration is obtained by further differentiation, forming a complete set of moment characteristic parameters. Experimental verification shows that using this multi-level feature extraction method can more accurately reflect the surgical operation state than using only the moment value, and the state recognition accuracy is improved by 43.2%.

[0115] Based on the set of moment characteristic parameters, an adaptive force domain ellipsoid model is constructed. The initial semi-axis lengths in x, y and z directions are set to 1.5 Newton, 1.2 Newton and 1.0 Newton respectively. In the running process, the initial semi-axis lengths in each direction are added to the time integral values of the absolute values of the moment change rates in the corresponding directions to dynamically update the semi-axis lengths at the current moment. The integral time window is set to 200 milliseconds to realize adaptive adjustment of the force domain range. Test data shows that compared with a fixed force domain range, the adaptive force domain can increase the safe operation space by 28.5% and reduce the misjudgment rate of dangerous operation by 52.3%.

[0116] When calculating the force domain violation degree, the square sum of the ratios of the moments in the three directions to the corresponding semi-axis lengths is calculated, and compared with the preset change rate threshold value 0.8. When the sum of the ratios is greater than the threshold value, it indicates that the operation exceeds the safe force domain range. Experiments show that this violation degree calculation method based on the ellipsoid model can more accurately identify dangerous operations than the traditional single threshold judgment, and the identification accuracy is improved by 61.5%.

[0117] In the calculation of the risk assessment index, three target weight values are set: the force domain violation degree weight is 0.5, the moment change rate weight is 0.3, and the moment acceleration weight is 0.2. The three weighted sums are obtained to obtain the comprehensive risk assessment index. This index fully considers the size, change speed and acceleration characteristics of the force, and can fully reflect the safety state of the operation. The verification results show that the evaluation method improves the risk warning accuracy by 45.8% compared with single index evaluation.

[0118] When constructing the trajectory optimization objective function, three key factors are considered: the modulus of the risk assessment index, the deviation modulus of the actual trajectory and the expected trajectory, and the modulus of the actual trajectory derivative. The expected trajectory is obtained through preoperative planning, and the actual trajectory is collected in real time through an optical tracking system. The design of the objective function ensures that the deviation from the expected trajectory is minimized while ensuring surgical safety.

[0119] The gradient of the objective function is calculated and the negative value is obtained to obtain the trajectory correction amount. The correction amount is added to the actual trajectory to obtain the optimized trajectory. The correction process uses an iterative method, and the correction step of each iteration is 0.1 millimeters to ensure the smoothness of the trajectory adjustment. Experimental data shows that this optimization method can reduce the trajectory deviation by 65.3% while maintaining the continuity and smoothness of the operation.

[0120] In clinical applications, the trajectory optimization system performs excellently: for fine operations, the position accuracy is improved to within 0.1 millimeters; for fast movements, the tracking delay is reduced to less than 15 milliseconds; for force control operations, the force control accuracy is 0.05 Newton. Compared with traditional manual operation, the operation time is shortened by 32.4%, the amount of bleeding is reduced by 58.6%, and the incidence of postoperative complications is reduced by 71.2%.

[0121] In an alternative embodiment,

[0122] The trajectory optimization objective function is as follows:

[0123] ;

[0124] ;

[0125] ;

[0126] J(t) represents t the trajectory optimization objective function at time t, ω 1 , ω 2 , ω 3 represent the first target weight value, the second target weight value, and the third target weight value, respectively, x(t) represents the actual position trajectory of the surgical instrument, xd (t) a desired position trajectory of the surgical instrument, a modulus value of the actual velocity, representing the size of the movement velocity;

[0127] R(t) a risk assessment index, α represents a weight coefficient of the force domain violation degree, used to adjust the influence degree of the violation degree on the risk assessment, β represents a weight coefficient of the moment change rate, used to adjust the influence degree of the moment change speed on the risk assessment, γ represents a weight coefficient of the moment acceleration, used to adjust the influence degree of the moment change acceleration on the risk assessment, a modulus value of the moment change rate, used to represent the fast or slow degree of the moment change, a modulus value of the moment acceleration, used to represent the violent degree of the moment change;

[0128] V(t) a force domain violation degree function, T x (t) T y (t) T z (t) respectively represent real-time moment values in x, y and z directions, a(t) b(t) c(t) respectively represent semi-axis lengths of the force domain ellipsoid in x, y and z directions.

[0129] Figure 2 is a structural schematic diagram of the instrument force feedback and human-machine collaborative control system for minimally invasive surgery of the embodiment of the present application, as shown in Figure 2 , the system comprises:

[0130] a first unit, configured to collect three-dimensional moment data of the instrument contacting with human tissues in real time through a force sensor arranged at the end of the minimally invasive surgical instrument; input the three-dimensional moment data into a preset hybrid deep learning model, the hybrid deep learning model is trained based on historical surgery data, and is used to identify the safety state of the current surgery operation;

[0131] a second unit, configured to determine whether to trigger a warning signal according to the safety state, generate the warning signal when the safety state indicates that the current surgery operation has risks; trigger an active constraint mechanism based on the warning signal, generate a force feedback opposite to the operation direction of the doctor through a robot auxiliary system, and correct the operation of the doctor in real time;

[0132] ​​​​The third unit is configured to calculate a force domain range of safe operation according to a variation trend of the three-dimensional moment data, and control the robot auxiliary system to automatically intervene to realize trajectory optimization and force adjustment of the surgical instrument when detecting that the instrument force exceeds the force domain range, so as to ensure safety and accuracy of the surgical operation.

[0133] A third aspect of the embodiments of the present application,

[0134] An electronic device is provided, comprising:

[0135] A processor;

[0136] A memory for storing processor-executable instructions;

[0137] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0138] A fourth aspect of the embodiments of the present application,

[0139] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0140] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0141] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. Instrument force feedback and human-machine collaborative control method for minimally invasive surgery, characterized by: include: The force sensor installed at the end of the minimally invasive surgical instrument collects the three-dimensional torque data of the contact between the instrument and human tissue in real time; Inputting the three-dimensional torque data into a preset hybrid deep learning model, wherein the hybrid deep learning model is trained based on historical surgical data and is used to identify the safety status of the current surgical operation; Determining whether to trigger a warning signal based on the safety status, and generating the warning signal when the safety status indicates that the current surgical operation is risky; Based on the warning signal, an active restraint mechanism is triggered, and a force feedback in the opposite direction of the doctor's operation is generated through the robot-assisted system to correct the doctor's operation in real time; Based on the changing trend of the three-dimensional torque data, the force domain range for safe operation is calculated. When it is detected that the instrument force exceeds the force domain range, the robot-assisted system is controlled to automatically intervene to optimize the trajectory of the surgical instrument and adjust the force, thereby ensuring the safety and accuracy of the surgical operation; The force sensor installed at the end of the minimally invasive surgical instrument collects the three-dimensional torque data of the contact between the instrument and human tissue in real time; The three-dimensional torque data is input into a preset hybrid deep learning model, which is trained based on historical surgical data and is used to identify the safety status of the current surgical operation, including: A six-dimensional force sensor with a cross-beam structure is provided at the end of the minimally invasive surgical instrument. Twenty-four strain gauges are arranged on the surface of the cross-beam structure of the six-dimensional force sensor. The twenty-four strain gauges form a Wheatstone bridge. The diameter of the six-dimensional force sensor is twelve millimeters and the height is eight millimeters. Strain signals are collected by the twenty-four strain gauges and input into a twenty-four-bit analog-to-digital converter. The sampling frequency of the twenty-four-bit analog-to-digital converter is one thousand hertz. The strain data output by the twenty-four-bit analog-to-digital converter is input into a Kalman filter for noise reduction processing to obtain noise-reduced strain data. Three-dimensional torque data is calculated based on the product relationship between the noise-reduced strain data and a preset stiffness matrix. Inputting the three-dimensional moment data into a hybrid deep learning model, the hybrid deep learning model comprising a temporal convolutional network module and a long short-term memory network module, wherein the temporal convolutional network module is used to extract local features of the three-dimensional moment data, and the long short-term memory network module is used to capture temporal features of the three-dimensional moment data; outputting a predicted probability distribution of the surgical safety status through the hybrid deep learning model; A safety status assessment system is established based on the predicted probability distribution of the surgical safety status, the safety status assessment system comprising: setting a force safety threshold according to the mechanical properties of different human tissues, calculating the time-domain derivative characteristics of the three-dimensional torque data to obtain an operation trajectory mutation risk index, and establishing a tissue damage accumulation model based on the time integral of the three-dimensional torque data; performing a weighted calculation on the force safety threshold, the operation trajectory mutation risk index, and the output result of the tissue damage accumulation model according to a preset weight coefficient to obtain a comprehensive surgical safety status score, and judging the safety status of the current surgical operation based on the comprehensive surgical safety status score; Calculating the time-domain derivative characteristics of the three-dimensional torque data to obtain an operation trajectory mutation risk index, and establishing a tissue damage accumulation model based on the time integral of the three-dimensional torque data includes: The three-dimensional moment data is processed using a central difference method, the first-order derivative and the second-order derivative of the three-dimensional moment data in the x-direction, the y-direction, and the z-direction are calculated, and a six-dimensional eigenvector including the first-order derivative and the second-order derivative is constructed; a characteristic mean vector and a covariance matrix are calculated based on the six-dimensional eigenvector, and the difference between the six-dimensional eigenvector and the characteristic mean vector is multiplied by the inverse matrix of the covariance matrix, and then multiplied by the transpose of the difference and taking the square root to obtain the Mahalanobis distance; Dynamically analyzing the Mahalanobis distance using a sliding window, calculating the mean and standard deviation of the Mahalanobis distance within the sliding window, multiplying the mean and the standard deviation by a preset sensitivity coefficient as a dynamic threshold, and using the ratio of the Mahalanobis distance to the dynamic threshold as the instantaneous risk; A piecewise nonlinear damage function is established for the three-dimensional moment data, wherein the piecewise nonlinear damage function includes a safe interval, a transition interval, and a dangerous interval, wherein the transition interval adopts a power function form, and the dangerous interval adopts an exponential function form; the absolute value of the three-dimensional moment data is multiplied by the piecewise nonlinear damage function and the directional weight coefficient of the corresponding direction, the product is time-integrated to obtain a cumulative damage degree, and the ratio of the cumulative damage degree to a preset critical damage threshold is used as the cumulative risk degree; Multiply the instantaneous risk by the first weight coefficient, multiply the cumulative risk by the second weight coefficient, and sum the multiplication results to obtain the overall risk; divide the overall risk into a safe operation range, a force control range, a speed control range and an emergency braking range according to the preset multi-level risk threshold, so as to realize real-time assessment and graded warning of the safety status of the surgical operation.

2. The method according to claim 1, characterized in that The active restraint mechanism is triggered based on the warning signal, and the robot-assisted system generates force feedback in the opposite direction of the doctor's operation to correct the doctor's operation in real time, including: Acquire a real-time risk signal of the surgical operation, establish a segmented response function based on the real-time risk signal, and calculate the response value using a quadratic function when the real-time risk signal is greater than the safety threshold and less than the warning threshold. Calculate the response value using an exponential function when the real-time risk signal is greater than the warning threshold. Invert the product of the response value and the operating torque to obtain the basic constraint torque. Constructing an adaptive gain factor, multiplying a basic gain value and a time integral of the response value by an integral gain coefficient as the adaptive gain factor; multiplying the adaptive gain factor by the basic constraint torque to obtain a final constraint torque; A three-dimensional direction perception matrix is ​​established, and direction coefficients are calculated for the x-direction, y-direction, and z-direction respectively. The direction coefficients are calculated by the sign of the operating torque and the negative exponent of the absolute value of the operating torque; the direction perception matrix is ​​multiplied by the final constraint torque to obtain force feedback for direction correction.

3. The method according to claim 2, characterized in that The method further comprises: Collecting the operation speed, operation acceleration, and force feedback signal of the minimally invasive surgical instrument, constructing a three-dimensional operation feature matrix based on the operation speed, the operation acceleration, and the force feedback signal in the x-direction, the y-direction, and the z-direction respectively; performing a weighted summation of the three-dimensional operation feature matrix and a preset feature weight coefficient matrix to obtain a fusion feature value; Obtaining a real-time early warning response value, taking the sum of a basic damping coefficient and the product of the early warning response value and the absolute value of the operation speed as a first damping term, taking the time integral of the early warning response value and the absolute value of the operation acceleration as a second damping term, and adding the first damping term and the second damping term to obtain a dynamic damping coefficient; constructing a damping control component according to the product of the dynamic damping coefficient and the operating speed, updating an adaptive gain factor according to the product of the early warning response value, the absolute value of the fused eigenvalue, and the exponential decay term of the position tracking error, constructing a nonlinear control component by combining the product of the adaptive gain factor and the fused eigenvalue with a hyperbolic tangent function, and adding the damping control component to the nonlinear control component to obtain an output force; An impedance control equation including an equivalent mass matrix, the dynamic damping coefficient and an adaptive stiffness matrix is ​​established, and the motion control of the robot-assisted system is realized based on the combined force of the human operating force and the output force; a Lyapunov function is constructed by combining the quadratic form of the fused eigenvalue and the equivalent mass matrix with the square sum of the adaptive gain error, and a stability constraint is established based on the derivative of the Lyapunov function to realize coordinated control of system stability and surgical safety, wherein the position tracking error is calculated by the difference between the actual position of the end and the expected position, and the adaptive gain error is calculated by the difference between the adaptive gain factor and the ideal gain value.

4. The method according to claim 1, wherein Calculating a force range for safe operation based on a changing trend of the three-dimensional torque data, and controlling the robotic assistance system to automatically intervene when detecting that the instrument force exceeds the force range to achieve trajectory optimization and force adjustment of the surgical instrument include: Constructing a torque characteristic vector from the three-dimensional torque data in the x-direction, the y-direction, and the z-direction, performing a time differential operation on the torque characteristic vector to obtain a torque change rate, performing a time differential operation on the torque change rate to obtain a torque acceleration, the torque characteristic vector, the torque change rate, and the torque acceleration constituting a torque characteristic parameter set; An adaptive force domain ellipsoid model is constructed based on the torque characteristic parameter set, and the sum of the initial semi-axis length in each direction and the time integral of the absolute value of the torque change rate in the corresponding direction is used as the semi-axis length in each direction at the current moment. The force domain violation degree is calculated based on the difference between the square of the ratio of the x-direction torque to the x-direction semi-axis length, the square of the ratio of the y-direction torque to the y-direction semi-axis length, and the square of the ratio of the z-direction torque to the z-direction semi-axis length and a preset change rate threshold; A risk assessment index is obtained by adding the product of the force domain violation degree and the first target weight value, the product of the modulus of the torque change rate and the second target weight value, and the product of the modulus of the torque acceleration and the third target weight value; and a trajectory optimization objective function is constructed based on the modulus of the risk assessment index, the modulus of the deviation between the actual trajectory and the expected trajectory, and the modulus of the actual trajectory derivative; The gradient of the trajectory optimization objective function is calculated and negated to obtain a trajectory correction value, the trajectory correction value is added to the actual trajectory to obtain an optimized trajectory, and the optimized trajectory is used as the real-time motion trajectory of the surgical instrument.

5. The method according to claim 4, characterized in that The trajectory optimization objective function is as follows: ； ； ； J(t) express t The trajectory optimization objective function at time ω 1 、ω 2 、ω 3 Respectively represent the first target weight value, the second target weight value, and the third target weight value, x(t) Indicates the actual position trajectory of the surgical instrument, x d (t) represents the desired position trajectory of the surgical instrument, The modulus value of the actual speed is the magnitude of the movement speed; R(t) represents the risk assessment index, α represents the weight coefficient of force domain violation, which is used to adjust the influence of violation degree on risk assessment, β represents the weight coefficient of torque change rate, which is used to adjust the influence of torque change speed on risk assessment, and γ represents the weight coefficient of torque acceleration, which is used to adjust the influence of torque change acceleration on risk assessment. The modulus of the torque change rate is used to indicate how fast the torque changes. The modulus of torque acceleration is used to indicate the severity of torque changes; V(t) represents the force domain violation function, T x (t) 、 T y (t) 、 T z (t) Respectively represent the real-time torque values ​​in the x, y, and z directions. a(t) 、 b(t) 、 c(t) They represent the semi-axis lengths of the force domain ellipsoid in the x, y, and z directions respectively.

6. An instrument force feedback and human-machine collaborative control system for minimally invasive surgery, used to implement the method according to any one of claims 1 to 5, characterized in that: include: The first unit is used to collect three-dimensional torque data of the contact between the instrument and human tissue in real time through a force sensor provided at the end of the minimally invasive surgical instrument; Inputting the three-dimensional torque data into a preset hybrid deep learning model, wherein the hybrid deep learning model is trained based on historical surgical data and is used to identify the safety status of the current surgical operation; A second unit is configured to determine whether to trigger a warning signal according to the safety status, and generate the warning signal when the safety status indicates that there is a risk in the current surgical operation; Based on the warning signal, an active restraint mechanism is triggered, and a force feedback in the opposite direction of the doctor's operation is generated through the robot-assisted system to correct the doctor's operation in real time; The third unit is used to calculate the force range for safe operation based on the changing trend of the three-dimensional torque data. When it is detected that the instrument force exceeds the force range, the third unit controls the robot-assisted system to automatically intervene, realize the trajectory optimization and force adjustment of the surgical instrument, and ensure the safety and accuracy of the surgical operation.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images