Surgical robot OCT (optical coherence tomography) guided anterior retinal membrane self-adaptive stripping system

Abstract

The invention relates to the technical field of medical robots and intelligent surgical equipment, and particularly discloses a surgical robot OCT (optical coherence tomography) guided anterior retinal membrane self-adaptive stripping system. A tail end stripping instrument is driven by a surgical robot, and a biomechanical original vibration signal is generated by a micro-pulse exciter; further, a prospective ductile impedance fingerprint is generated to predict the tissue adhesion strength; meanwhile, a high-risk adhesion area is positioned according to the fingerprint, and the image is verified by obtaining a local microstructure; and finally, in combination with mechanical and morphological information, generating a comprehensive evaluation report of the peeling risk, and generating a composite self-adaptive regulation and control instruction according to the comprehensive evaluation report so as to realize intelligent adjustment of surgical parameters. According to the invention, the problem of iatrogenic retina injury caused by large adhesion strength difference in the anterior retina dissection operation is solved, the operation is converted from a reaction mode to a feedforward intelligent control mode, and the safety and success rate of the operation are improved.

Description

Technical Field

[0001] This invention relates to the field of medical robots and intelligent surgical equipment technology, and more specifically, to a surgical robot OCT-guided adaptive epiretinal membrane dissection system. Background Technology

[0002] Epiretinal membrane detachment is a crucial surgical procedure for treating various blinding fundus diseases. Its core challenge lies in the significant variations in the adhesion strength between the epiretinal membrane and the underlying fragile retinal tissue in different areas, which are difficult to predict accurately preoperatively. During the procedure, surgeons primarily rely on visual feedback under a microscope and extremely weak and distorted tactile feedback transmitted through surgical instruments to determine the detachment force and direction. This highly experience-dependent and subjective approach makes it extremely easy to cause iatrogenic retinal tears or damage when encountering areas of strong adhesion, leading to permanent visual impairment. This constitutes the core risk and technical bottleneck of the current procedure.

[0003] To address this challenge, the industry commonly employs methods such as enhancing surgeons' skills and utilizing intraoperative imaging equipment like optical coherence tomography (OCT) for auxiliary observation. During the dissection process, surgeons can intermittently observe the cross-sectional structure of the membrane and retina using OCT images to assess the level of dissection and tissue morphology. However, this method primarily provides static or quasi-static morphological information and cannot reflect the biomechanical characteristics of the tissue in front of the instrument, i.e., the toughness of adhesions, in real time and proactively. Surgeons can only react after observing excessive traction and deformation of the tissue, rather than anticipating the risks.

[0004] The drawbacks of traditional methods are obvious. Surgical safety and success rates are heavily dependent on the surgeon's personal experience and on-site condition, lacking objective and quantifiable risk assessment methods, resulting in poor consistency and repeatability of surgical outcomes. Furthermore, existing auxiliary imaging technologies merely provide passive observation tools; the core closed loop of surgical decision-making and execution still relies on humans, failing to achieve automated and intelligent responses from risk perception to avoidance actions. Therefore, the entire surgical process is essentially a reactive rather than predictive risk management model, unable to fundamentally and proactively avoid surgical complications caused by unknown areas of strong adhesion. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background art, a surgical robot OCT-guided adaptive epiretinal membrane dissection system is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: This invention provides a surgical robot OCT-guided adaptive epiretinal membrane dissection system, comprising: a raw vibration signal generation module, which generates a biomechanical raw vibration signal, wherein the biomechanical raw vibration signal characterizes the physical properties of the tissue in the area to be dissected.

[0007] The impedance eigenvector generation module generates a forward-looking toughness impedance eigenvector based on the generated original biomechanical vibration signal.

[0008] The adhesion early warning coordinate point positioning module locates high-risk adhesion early warning coordinate points based on the generated prospective toughness impedance feature vector. The high-risk adhesion early warning coordinate points indicate the location of the high-risk adhesion area in the three-dimensional space of the surgery.

[0009] The microstructure verification image acquisition module acquires local microstructure verification images based on the high-risk adhesion warning coordinate points. The local microstructure verification images show the thickness, layered structure, and fusion status of the anterior retinal membrane with the underlying retinal tissue at the high-risk adhesion warning coordinate points.

[0010] The comprehensive assessment report generation module generates a comprehensive assessment report on peeling risk based on local microstructure verification images and prospective toughness impedance feature vectors. The comprehensive assessment report on peeling risk includes a quantitative risk index.

[0011] The adaptive control instruction generation module generates composite adaptive control instructions based on the comprehensive risk assessment report of the stripping process.

[0012] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention, by introducing forward-looking biomechanical detection and multimodal information fusion, transforms surgical operation from a reactive mode that relies on the doctor's subjective experience to a feedforward intelligent control mode based on objective data prediction. The system can sense the adhesion strength of the tissue in front before the dissection instrument reaches the risk area, thereby providing early risk warning and actively adjusting the surgical strategy. This working method of changing from passive response to active avoidance can avoid tissue damage caused by untimely response when encountering unknown strong adhesion points, thus improving the safety of the operation.

[0013] (2) This invention achieves a high degree of automation and precision in the surgical process. By using a robotic system to replace human hands in performing delicate dissection operations, it reduces the uncertainty caused by human factors such as physiological tremors and fatigue. More importantly, the system automatically generates and executes optimal control instructions based on a quantified risk assessment report, and collaboratively optimizes multiple key parameters such as dissection speed, angle, and posture. This closed-loop adaptive control ensures that each dissection action is tailored to the current local tissue characteristics, thereby making the entire surgical process more precise, stable, and standardized, reducing reliance on the surgeon's individual skills, and improving the overall quality and success rate of the surgery.

[0014] (3) This invention creatively integrates two different information dimensions: biomechanical perception and microstructural imaging, forming a dual verification mechanism that complements mechanical prediction and morphological verification. Single mechanical signals or morphological images have limitations, but this invention combines the two. Mechanical signals quickly and proactively identify high-risk suspected areas, and high-precision optical imaging then accurately confirms the structure of these points, constructing a more comprehensive and reliable risk assessment model. This deep fusion of multimodal information makes the system's understanding and judgment of complex pathological conditions superior to that of a single information source, thus enabling more intelligent and accurate surgical decisions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the system module structure connection of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This invention provides a surgical robot OCT-guided adaptive epiretinal membrane dissection system, comprising: a raw vibration signal generation module, an impedance feature vector generation module, an adhesion early warning coordinate point positioning module, a microstructure verification image acquisition module, a comprehensive evaluation report generation module, and an adaptive control command generation module.

[0019] The original vibration signal generation module is connected to the impedance feature vector generation module, the impedance feature vector generation module is connected to the adhesion early warning coordinate point positioning module, the adhesion early warning coordinate point positioning module is connected to the microstructure verification image acquisition module, both the impedance feature vector generation module and the microstructure verification image acquisition module are connected to the comprehensive evaluation report generation module, and the comprehensive evaluation report generation module is connected to the adaptive control command generation module.

[0020] The original vibration signal generation module generates a biomechanical original vibration signal, which characterizes the physical properties of the tissue in the area to be stripped.

[0021] In a specific embodiment of the present invention, the specific steps for generating the original biomechanical vibration signal include: controlling the surgical robot to drive the end-effector, driving the micropulse exciter integrated on the effector, and applying a preset waveform of mechanical micropulse to the anterior retinal membrane and retinal tissue in front of the effector.

[0022] In a specific embodiment of the present invention, the control of the surgical robot to drive the end-effector dissection instrument and drive the micropulse exciter integrated on the dissection instrument to apply a preset waveform of mechanical micropulse to the preretinal membrane and retinal tissue in front of the dissection instrument includes: controlling the surgical robot to drive the end-effector dissection instrument, and while moving along a predetermined dissection path, driving the micropulse exciter integrated on the dissection instrument to apply a preset waveform of mechanical micropulse to the preretinal membrane and retinal tissue in front of the dissection instrument.

[0023] By utilizing a micro-vibration sensor array mounted on the dissection instrument, transient echoes generated by the transmission, reflection, and attenuation of mechanical micropulses in different tissue media are captured synchronously.

[0024] Integrate transient echoes to generate original biomechanical vibration signals.

[0025] It should be noted that the process begins with controlling a surgical robot. This robot, through its robotic arm, moves an end-effector instrument within the eye along a pre-planned dissection path. This path is a three-dimensional spatial curve planned before surgery based on the patient's fundus images, defining the starting point, direction of travel, and ending point of the dissection operation. During movement, the system simultaneously drives a micropulse exciter integrated into the tip of the end-effector instrument. This micropulse exciter is a piezoelectric ceramic element capable of generating minute mechanical vibrations. It applies a pre-defined waveform of mechanical micropulse to the epiretinal membrane and retinal tissue in the direction of instrument movement. The waveform, frequency, and energy of this pulse are pre-set to detect its physical properties without damaging the tissue. Next, a micro-vibration sensor array, also mounted on the end-effector instrument, composed of multiple miniature accelerometers, sensitively captures the vibration signals. When the mechanical micropulse passes through tissues with different physical properties, such as the relatively tough epiretinal membrane and the relatively soft retinal tissue, transmission, reflection, and energy attenuation occur, generating a brief vibration echo, or transient echo. A micro-vibration sensor array synchronously captures these echo signals carrying tissue structure information. Finally, the transient echo signals collected by each sensor in the array are arranged and combined according to time sequence and spatial location to form a single, continuous time-series data stream. This data stream completely records the interaction process between the mechanical pulse and the tissue at the detection point, ultimately generating a biomechanical raw vibration signal characterizing the physical properties of the tissue in the area to be dissected, which serves as input for the next step of analysis.

[0026] The surgical robot is a high-precision, multi-degree-of-freedom robotic arm system used to assist surgeons in performing delicate operations. Its function is to achieve sub-millimeter-level stable positioning and motion control. The end-effector is a specialized tool, such as a micro-forceps or curette, installed at the end of the surgical robot's robotic arm. It integrates signal excitation and acquisition elements. The predetermined dissection path is a three-dimensional coordinate point sequence with a data structure of {(x1,y1,z1),(x2,y2,z2),…,(xn,yn,zn)}. It defines the spatial trajectory that the dissection operation should follow, and is based on preoperative optical coherence tomography (OCT) images. The micropulse exciter is a miniature piezoelectric transducer that converts electrical signals into precise mechanical vibrations. The preset waveform mechanical micropulse is a vibration signal of a specific form. Its characteristic attributes are defined by waveform type, center frequency, pulse width, and amplitude. It is set based on the analysis of experimental data from over 500 isolated pig eye models, selecting the parameter combination with the best energy penetration and no tissue damage. The epiretinal membrane is a pathologically proliferating fibrous membrane within the eye. The retina is the layer of photoreceptor nerve tissue at the back of the eyeball. A microvibration sensor array is an array of multiple microelectromechanical system (MEMS) accelerometers, its function being to synchronously acquire weak vibration signals at multiple points. Transient echoes are rapidly decaying reflected vibration signals generated after the interaction of mechanical micropulses with tissue; the data structure is a time-series data set containing amplitude and phase information.

[0027] For example, a surgical robot is controlled to move a 25G microforceps integrated with sensors as an end-effector. This forceps moves along a predetermined dissection path from the optic disc edge coordinates (1.5, 2.3, 0.1) to the vicinity of the fovea coordinates (4.8, 2.5, 0.1). When it reaches the detection point (2.0, 2.4, 0.1) on the path, a micropulse exciter at the tip of the forceps is activated to apply a mechanical micropulse with a center frequency of 1000 Hz, an amplitude of 1 μm, and a duration of 5 ms to the anterior retinal membrane and retinal tissue directly in front of it. A 2x2 micro-vibration sensor array consisting of 4 sensors mounted on the tip of the forceps synchronously captures 4 returning transient echoes. These four transient echo data streams, each with a duration of 20 milliseconds and a sampling frequency of 44.1 kHz, were sequentially spliced ​​together in the sensor spatial order (top left, top right, bottom left, bottom right) to form a single data stream with a total duration of 80 milliseconds. This data stream was then saved as a file as a biomechanical original vibration signal.

[0028] The impedance eigenvector generation module generates a forward-looking toughness impedance eigenvector based on the generated original biomechanical vibration signal.

[0029] In a specific embodiment of the present invention, the specific steps of generating a prospective toughness impedance feature vector based on the generated original biomechanical vibration signal include: performing joint time-domain and frequency-domain analysis on the original biomechanical vibration signal to extract the energy attenuation rate, dominant frequency offset, and harmonic component complexity from the original biomechanical vibration signal.

[0030] In a specific embodiment of the present invention, the step of performing joint time-domain and frequency-domain analysis on the original biomechanical vibration signal to extract the energy decay rate, dominant frequency offset, and harmonic component complexity in the original biomechanical vibration signal includes: performing time-domain analysis on the original biomechanical vibration signal to extract the energy decay rate.

[0031] Frequency domain analysis was performed on the original biomechanical vibration signal to extract the dominant frequency offset and harmonic component complexity.

[0032] It should be noted that this step receives the original biomechanical vibration signal generated in the previous step as input and analyzes it to generate a prospective toughness impedance eigenvector. First, a joint time-domain and frequency-domain analysis is performed on the input biomechanical vibration signal, a time-series data. In the time-domain analysis, the descent rate of the signal amplitude envelope is calculated to obtain the energy decay rate. In the frequency-domain analysis, a Fast Fourier Transform (FFT) is used to transform the signal from the time domain to the frequency domain, obtaining its spectrum. On the spectrum, the frequency point with the highest energy is located, and its difference from the original center frequency emitted by the micropulse exciter is calculated to obtain the dominant frequency offset. Simultaneously, the number and relative amplitude of each harmonic peak outside the dominant frequency are counted, and a weighted sum is used to obtain a quantized value as the harmonic component complexity. Through the above analysis, multiple characteristic parameters, such as the energy decay rate, dominant frequency offset, and harmonic component complexity, are extracted from the original signal.

[0033] It should also be noted that the energy decay rate is calculated as follows: First, the system automatically identifies and locates the peak point of the signal amplitude in the received biomechanical raw vibration signal, and records the amplitude and time of occurrence of the peak. Next, the system searches backward along the time axis to locate the point where the signal amplitude decays to a preset proportion (e.g., 10% of the peak amplitude), and records the time of that point. Finally, the system divides the difference between the peak amplitude and the decayed amplitude by the time difference between these two points to calculate an average decay rate. This rate is quantified as the energy decay rate. The harmonic component complexity is calculated as follows: First, after completing the Fast Fourier Transform and obtaining the spectrum, the system identifies a series of harmonic peaks other than the dominant frequency, i.e., those secondary energy peaks whose frequencies are integer multiples of the dominant frequency. Then, for each identified harmonic peak, the ratio of its amplitude to the amplitude of the dominant frequency peak is calculated to obtain a relative amplitude value. Finally, the relative amplitude values ​​of all harmonics are summed according to their harmonic orders to obtain a single quantized value, which is the harmonic component complexity. The formula can be described as follows: the harmonic component complexity equals the sum of the products of the orders of all harmonics and their relative amplitudes.

[0034] The extracted energy decay rate, main frequency offset, and harmonic component complexity are compared with the preset benchmark parameter model to calculate the difference.

[0035] The calculated difference is quantized into a multi-dimensional feature vector, which is the forward toughness impedance feature vector.

[0036] It should be noted that the extracted feature parameters are then compared one by one with a pre-defined baseline parameter model. This model stores a set of ideal parameter values, representing the standard vibrational feedback of the epiretinal membrane and retinal tissue in a healthy, non-adhesive, and easily separable state. The differences between each parameter are calculated by subtracting the baseline values ​​from the measured feature parameter values. Finally, these differences, such as differences in energy decay rate, dominant frequency offset, and harmonic component complexity, are combined into an ordered list of values, i.e., a multi-dimensional feature vector. This feature vector quantitatively reflects the degree of deviation of the tissue in the current detection area from the ideal state, thus predicting the adhesion strength and dissection difficulty of the tissue ahead. This feature vector is the prospective toughness impedance feature vector output by this step.

[0037] It should also be noted that the formula for calculating the difference is: The formula for constructing the eigenvector of forward toughness impedance is: , where the symbol Representing the A measured value of a characteristic parameter, such as the measured value of the energy decay rate. Symbol The corresponding first parameter in the baseline parameter model A reference parameter value. Symbol It is the calculated number of The difference between the feature parameters. (Symbol) This represents the final generated forward toughness impedance eigenvector, which is a vector composed of... Individual differences The eigenvectors formed.

[0038] The prospective toughness impedance eigenvector is a multi-dimensional eigenvector with a data structure of a one-dimensional floating-point array. Its function is to quantitatively describe the mechanical properties and potential dissection risks of the tissue to be dissected. The energy decay rate is a scalar value used to measure the rate at which mechanical pulse energy is lost during propagation in the tissue. It is based on over 300 sets of preclinical experimental data and is positively correlated with the degree of tissue fibrosis. The dominant frequency offset is a scalar value representing the change in the dominant frequency of the echo signal relative to the center frequency of the excitation pulse; it reflects changes in tissue stiffness. The harmonic component complexity is a dimensionless scalar value used to characterize the abundance of higher harmonic components in the echo signal. The characteristic parameters are a set of the energy decay rate, dominant frequency offset, and harmonic component complexity. The baseline parameter model is a reference dataset pre-stored in the system. Its data structure is a vector containing standard characteristic parameter values. It is based on the analysis of the original biomechanical vibration signals of normally separated areas from 50 successful dissection surgeries, extracting the average value of the characteristic parameters.

[0039] The adhesion early warning coordinate point positioning module locates high-risk adhesion early warning coordinate points based on the generated prospective toughness impedance feature vector. The high-risk adhesion early warning coordinate points indicate the location of the high-risk adhesion area in the three-dimensional space of the surgery.

[0040] In a specific embodiment of the present invention, the specific steps for locating the high-risk adhesion warning coordinate point based on the generated forward toughness impedance feature vector include: real-time monitoring of the feature values ​​of each dimension in the forward toughness impedance feature vector and comparing them with a preset toughness safety threshold.

[0041] When at least one eigenvalue in the prospective toughness impedance eigenvector exceeds its corresponding toughness safety threshold, the current detection area is determined to be a high-risk adhesion area.

[0042] Record the spatial position of the current dissection instrument, and combine the detection direction and depth information of the micropulse to calculate and output the high-risk adhesion warning coordinates of the high-risk adhesion area in the three-dimensional space of the surgery.

[0043] It should be noted that this step receives and processes each prospective toughness impedance feature vector generated in the previous step in real time to identify and locate potential surgical risk points. First, the system continuously monitors each input prospective toughness impedance feature vector, a multi-dimensional feature vector containing values ​​such as energy decay rate difference and main frequency offset difference. The system compares each feature value in this vector with its corresponding preset toughness safety threshold. This set of toughness safety thresholds is derived in advance based on statistical analysis of a large amount of clinical data, setting a critical value for each feature dimension, indicating that the degree of tissue adhesion has reached a level that may cause risks such as retinal tears. Next, the system makes a judgment based on the comparison results. If at least one feature value in the prospective toughness impedance feature vector exceeds its corresponding toughness safety threshold, for example, if the energy decay rate difference is greater than the preset toughness safety threshold, the system immediately identifies the tissue area currently being probed by the mechanical micropulse as a high-risk adhesion area. Once identified as a high-risk adhesion area, the system immediately performs a positioning operation. It reads and records the precise position of the tip of the distal dissection instrument in the three-dimensional surgical space from the surgical robot's control system, i.e., the spatial position of the dissection instrument. Simultaneously, combining the known detection direction of the mechanical micropulse emitted by the micropulse exciter (usually a directional vector aligned with the instrument tip) and the effective detection depth information of the pulse in the tissue, a vector operation is used to calculate the precise location of the risk point in front of the dissection instrument. Finally, the system outputs this calculated three-dimensional coordinate, the high-risk adhesion warning coordinate point, to guide subsequent refined exploration.

[0044] It should also be noted that the formula for calculating the coordinates of high-risk adhesion warning points is as follows: , where the symbol The recorded spatial location of the dissection instrument is represented by a three-dimensional coordinate vector (in millimeters). Symbol The detection depth information representing the micropulse is a scalar (unit: millimeters). Symbol The direction of detection for the micropulse is represented by a three-dimensional unit vector, which has no dimension. (Symbol) This is the final calculated and output high-risk adhesion warning coordinate point, which is also a three-dimensional coordinate vector (unit: millimeters). Detection depth information. The setting is based on the effective range determined after calibration experiments on the attenuation characteristics of mechanical micropulses of different energies in intraocular tissues.

[0045] The toughness safety threshold is a pre-defined set of reference values. Its data structure is the same as the prospective toughness impedance eigenvector, which is a one-dimensional floating-point array. Its function is to serve as a standard for judging whether the risk of the dissection operation has increased. The setting is based on the analysis of surgical videos of more than 100 cases of intraoperative complications, and the deduction of the critical range of the characteristic values ​​of each dimension of the prospective toughness impedance eigenvector before tissue tearing. A typical toughness safety threshold can be set to [1.5, 40, 0.7]. In this setting, the first value "1.5" corresponds to the threshold of the energy decay rate difference, in decibels per millisecond, which means that when the measured decay rate is more than 1.5 decibels per millisecond faster than the reference, it is abnormal. The second value "40" corresponds to the threshold of the main frequency offset difference, in Hertz, which means that when the echo main frequency deviates from the original frequency by more than 40 Hertz, the tissue hardness changes significantly. The third value "0.7" corresponds to the threshold of harmonic component complexity, which is a dimensionless value. When the complexity exceeds 0.7, it indicates that the signal distortion is severe and the tissue structure is complex. The system triggers a high-risk warning when at least one dimension of the real-time monitored prospective toughness impedance characteristic vector exceeds this set of typical thresholds. The high-risk adhesion area is an internal logical state identifier used to mark areas of tissue with strong adhesions in front of the probe. The spatial position of the dissection instrument is provided in real-time by the encoder built into the surgical robot, providing precise three-dimensional coordinate data indicating the position of the instrument tip in the ocular coordinate system. The detection direction and depth information of the micropulse are two combined parameters: direction is a unit vector representing the direction of instrument movement, and depth is a scalar representing the distance at which the mechanical wave can effectively transmit information. The high-risk adhesion warning coordinate point is the final output of this step, a precise three-dimensional coordinate indicating the specific spatial location of risk points requiring further investigation.

[0046] The microstructure verification image acquisition module acquires local microstructure verification images based on the high-risk adhesion warning coordinate points. The local microstructure verification images show the thickness, layered structure, and fusion status of the anterior retinal membrane with the underlying retinal tissue at the high-risk adhesion warning coordinate points.

[0047] In a specific embodiment of the present invention, the specific steps of obtaining a local microstructure verification image based on the high-risk adhesion warning coordinate points include: transmitting the high-risk adhesion warning coordinate points to the scanning controller of the optical coherence tomography system.

[0048] The optical coherence tomography (OCT) system immediately interrupts its conventional scanning mode and instead precisely aligns its scanning center with the high-risk adhesion warning coordinates, performing a high-density, high-resolution local fine scan to acquire and generate local microstructure verification images.

[0049] It should be noted that this step receives the high-risk adhesion warning coordinates from the previous step and initiates a precise focusing scan based on these coordinates to obtain image evidence. First, the system transmits the input high-risk adhesion warning coordinates, this three-dimensional coordinate data, to the scan controller of an optical coherence tomography (OCT) system integrated into the surgical system. This scan controller is a dedicated module for handling scan commands and controlling the scan hardware. Upon receiving the coordinate data, the scan controller immediately generates and sends an interrupt command to the OCT system, stopping the currently executing conventional scan mode used for global observation. Next, the scan controller issues a new positioning command, driving the scanning galvanometer within the system to precisely move and align the scan center of the OCT system—its beam focus—to the newly received high-risk adhesion warning coordinates. Once positioning is complete, the system performs a pre-set high-density, high-resolution local fine scan within a tiny area centered on these coordinates. Compared to a conventional scan, this scan acquires more scan lines within a smaller area and improves axial resolution to obtain richer microstructural details. Finally, the optical coherence tomography system processes the optical interference signals acquired in this refined scan. After data reconstruction, a high-resolution cross-sectional image is generated. This image is a local microstructure verification image that can clearly show the thickness, layered structure, and fusion state of the epiretinal membrane with the underlying retinal tissue at the warning coordinate point, and serves as the final output of this step.

[0050] Optical coherence tomography (OCT) is a non-invasive optical imaging device that functions similarly to ultrasound in optical systems. It generates cross-sectional images of tissue with micron-level resolution by emitting near-infrared light and analyzing its reflection signals at different depths within the tissue. The scan controller is the internal control unit of the OCT system, receiving external commands and controlling the position, scanning mode, and parameters of the scanning beam. The standard scan mode is the default operating state of the OCT system, characterized by a large scanning range and high speed, but relatively low resolution, used for quickly providing an overview of the tissue. The scan center refers to the focal point of the OCT system's scanning beam on the target tissue. High-density, high-resolution local fine-tuning scanning is a special scanning mode characterized by dense, high-precision scanning within a small area to obtain microstructural details at specific locations. The local microstructure verification image is the final data generated in this step, providing morphological evidence for the previous risk assessment based on mechanical signals. The image content must clearly show information such as the thickness, layering structure, and fusion status of the anterior retinal membrane. The thickness, layered structure, and fusion status of the epiretinal membrane with the underlying retinal tissue are key imaging features for assessing the severity of the pathology.

[0051] The comprehensive assessment report generation module generates a comprehensive assessment report on peeling risk based on local microstructure verification images and prospective toughness impedance feature vectors. The comprehensive assessment report on peeling risk includes a quantitative risk index.

[0052] In a specific embodiment of the present invention, the specific steps of generating a comprehensive assessment report on peeling risk based on local microstructure verification images and prospective toughness impedance feature vectors include: extracting morphological parameters from the local microstructure verification images, wherein the morphological parameters include the membrane-retina fusion area, changes in the curvature of the anterior retinal surface, and the presence or absence of fibrosis.

[0053] It should be noted that this step receives the local microstructure verification image generated in the previous step and integrates it with the previous mechanical prediction information to complete a comprehensive assessment of the peeling risk. First, the system performs image analysis on the input high-resolution local microstructure verification image to quantify and extract the key morphological parameters contained in the image. This includes automatically measuring the boundary of the region where the epiretinal membrane fuses with the underlying retinal tissue and calculating the fusion area. The calculation can be simply described as follows: an image recognition algorithm finds the boundary of the fusion region, then calculates the sum of all pixels within this boundary to obtain the area value; by identifying and fitting the contour line of the anterior retinal surface, the image recognition algorithm analyzes the curvature of this contour line, obtaining the curvature value, then calculates the curvature value of the adjacent region's contour line, and finally subtracts these two curvature values ​​to obtain a difference, which is the change in curvature of the anterior retinal surface; and using grayscale analysis technology, the local microstructure verification image is automatically converted into a grayscale image. The grayscale values ​​of each grayscale region are compared with the normal grayscale range. If the grayscale value of a certain grayscale region is not within the normal grayscale range, it is determined that the grayscale region has signs of fibrosis and hyperplasia; conversely, if the grayscale value of a certain grayscale region is within the normal grayscale range, it is determined that the grayscale region does not have signs of fibrosis and hyperplasia.

[0054] By correlating morphological parameters with the prospective toughness impedance eigenvector that triggered this focused scan, a dual information pair containing both mechanical prediction and morphological evidence is constructed.

[0055] It should be noted that, next, the system correlates these extracted morphological parameters with the prospective toughness impedance feature vector that triggered this focused scan. Specifically, a new data structure is created that combines the prospective toughness impedance feature vector representing mechanical properties with multiple morphological parameter values ​​representing morphological features, thereby constructing a dual information pair containing both "mechanical prediction" and "morphological evidence".

[0056] Based on the dual information pairs, a quantitative final risk index is generated.

[0057] Based on the generated final risk index, the final stripping risk level, type, and probability of retinal damage in high-risk areas are comprehensively assessed, thereby generating a comprehensive stripping risk assessment report.

[0058] It should be noted that after the system completes the prospective toughness impedance feature vector analysis based on biomechanical signals and the prospective structural feature analysis based on medical images, it will initiate the final comprehensive assessment process, which is explained as follows: The system first fuses and weights these two sets of analysis results—the "dual information pairs"—to generate a quantified final risk index, and based on this, classifies the risk level, determines the risk type, and predicts the probability of injury. The calculation of the final risk index can be understood as a comprehensive scoring process. The system assigns different weights to the toughness impedance feature vector from biomechanical signals ("mechanical risk score") and the structural features from medical images ("morphological risk score"). For example, the weight of the mechanical risk score might be set to 0.6, and the weight of the morphological risk score to 0.4, because mechanical signals better reflect the current physical state in real time. The calculation formula can be simply described as: the final risk index equals the mechanical risk score multiplied by its weight, plus the morphological risk score multiplied by its weight. This index is a value between 0 and 100; the higher the score, the greater the overall risk. The final risk level classification is as follows: a final risk index below 40 points is rated as "low risk"; 40 to 70 points is rated as "moderate risk"; and above 70 points is rated as "high risk." This level will be displayed on the screen in a prominent color (such as green, yellow, or red) to provide doctors with the most intuitive risk warning. Risk type determination: The system will further analyze which specific indicator(s) contribute most to the high risk, thereby determining the root cause of the risk. For example, if the risk mainly stems from a rapid energy decay rate and a large image fusion area, the system will classify the risk type as "adhesion risk," indicating that the main problem lies in the tight adhesion between the membrane and the tissue. If the risk mainly comes from a large dominant frequency offset and fibrosis in the image, it will be classified as "sclerosis risk," indicating that the tissue itself has become hardened and brittle. If the complexity of harmonic components and drastic curvature changes are the main causes, it may be classified as "traction risk." Calculation of retinal damage probability: This is a prediction based on historical data and statistical models. The system internally stores a large amount of data on past surgical cases, including the correlation between various risk indices and final surgical outcomes (such as whether a retinal tear occurred). After calculating the current final risk index, the system queries this database to find cases with similar risk indices and calculates the proportion of retinal damage among these cases. The calculation method can be described as follows: a similarity interval is defined around the final risk index of the new case; then, the number of historical cases with retinal damage within this interval is divided by the total number of all historical cases within that interval to obtain a percentage; this percentage is the predicted probability of damage.

[0059] The adaptive control instruction generation module generates composite adaptive control instructions based on the comprehensive risk assessment report of the stripping process.

[0060] In a specific embodiment of the present invention, the specific steps of generating a composite adaptive control instruction based on the comprehensive risk assessment report of retinal stripping include: reading the final risk index, risk type, and probability of retinal damage from the comprehensive risk assessment report of retinal stripping.

[0061] Based on the risk type, a core risk avoidance plan is initially matched from a pre-set control strategy library, and then the execution intensity of the core risk avoidance plan is finely adjusted based on the final risk index.

[0062] The probability of retinal damage is used as the final decision-making basis for triggering emergency intervention, thereby generating a composite adaptive control command and sending it to the surgical robot.

[0063] It's important to note that after the system generates a comprehensive risk assessment report, it enters the intelligent control instruction generation stage. First, the system reads three core data points from the report: risk type, final risk index, and probability of retinal damage. The system doesn't just look at a single number; it performs a three-step intelligent decision-making process: Step 1: Selecting a core strategy based on the "risk type," addressing the "what to do" question. The system will retrieve targeted basic solutions from its strategy library based on the identified "adhesion risk," "hardening risk," or "traction risk." For example, for "adhesion risk," the core strategy is to adjust the dissection angle to find the optimal separation surface and activate micropulse excitation to gently "loosen" the adhesion points; while for "hardening risk," the core strategy is to adjust the instrument posture, using the instrument's side for gentle tangential dissection to avoid "hard contact" that could cause tissue fragmentation. Step 2: Fine-tuning the strategy intensity based on the final risk index, addressing the "how much to do" question. This adjustment can be simply described as multiplying the final risk index as a percentage by a preset maximum adjustment amount. For example, for "adhesion-type risks," the preset maximum reduction speed is 80%. If the current risk index is 50, the robot's actual dissection speed will be reduced by 50% multiplied by 80%, or 40%. Similarly, the higher the risk index, the wider the adjustment range of the dissection angle, and the higher the energy and frequency of the micropulses. The third step: Making a final decision and safety lock based on the probability of retinal damage, addressing the "can it be done?" question. The system sets a critical damage probability, such as 85%. If the probability of retinal damage is below this value, the robot executes the combined instructions generated in the first two steps. However, if the probability of retinal damage exceeds 85%, the system will determine that continuing the original plan is highly likely to cause a retinal tear. At this point, it will trigger an emergency intervention mechanism: immediately pausing the robot's dissection action, slightly retracting the instrument tip to a safe position, and issuing the highest-level alarm to the surgeon, requiring manual confirmation or modification of the surgical plan, thus adding a final layer of protection for surgical safety.

[0064] It should also be noted that the maximum adjustment values ​​preset in the system's control strategy library for different risk avoidance schemes are empirical safety values, typically set as follows: For peeling speed, the maximum adjustment value is usually set to reduce by 80%, meaning that under the highest risk, the robot's peeling action will go from an already extremely slow speed to almost static, thereby minimizing the dynamic pulling force on the tissue; For peeling angle, the typical maximum adjustment value is a fluctuation range of ±15 degrees based on the initial entry angle. This range allows the robot to make fine exploratory swings within a small sector to find the tissue layer that is easiest to separate. To avoid harsh vertical lifting, the maximum adjustment for the instrument posture can be from a posture perpendicular to the retinal surface (90 degrees) to a tangential posture with a maximum tilt of 45 degrees. The purpose of this adjustment is to transform the direct "lifting" force into a gentler lateral "scraping" force to deal with hardened tissue. Finally, the maximum adjustment for the micropulse excitation parameters is reflected in two aspects: the pulse frequency can be excited up to 200 Hz, and the pulse energy can be increased from a base value of zero or very low to the maximum set energy within the device's safety threshold. The aim is to loosen tight adhesions through high-frequency micro-mechanical vibrations, rather than relying on a single large force.

[0065] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A surgical robot OCT-guided adaptive epiretinal membrane dissection system, characterized in that, include: The original vibration signal generation module generates a biomechanical original vibration signal, which characterizes the physical properties of the tissue in the area to be stripped. The impedance eigenvector generation module generates a forward-looking toughness impedance eigenvector based on the generated original biomechanical vibration signal. The adhesion early warning coordinate point positioning module locates high-risk adhesion early warning coordinate points based on the generated prospective toughness impedance feature vector. The high-risk adhesion early warning coordinate points indicate the position of the high-risk adhesion area in the three-dimensional space of the surgery. The microstructure verification image acquisition module acquires local microstructure verification images based on the high-risk adhesion warning coordinate points. The local microstructure verification images show the thickness, layered structure, and fusion status of the epiretinal membrane with the underlying retinal tissue at the high-risk adhesion warning coordinate points. The comprehensive assessment report generation module generates a comprehensive assessment report on peeling risk based on local microstructure verification images and prospective toughness impedance feature vectors. The comprehensive assessment report on peeling risk includes a quantitative risk index. The adaptive control instruction generation module generates composite adaptive control instructions based on the comprehensive risk assessment report of the stripping process.

2. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 1, characterized in that, The specific steps for generating the original biomechanical vibration signal include: The surgical robot is controlled to drive the end-effector, which in turn drives the micropulse exciter integrated on the effector to apply mechanical micropulses of a preset waveform to the epiretinal membrane and retinal tissue in front of the effector. By using a micro-vibration sensor array mounted on the dissection instrument, transient echoes generated by the transmission, reflection, and attenuation of mechanical micropulses in different tissue media are captured synchronously. Integrate transient echoes to generate original biomechanical vibration signals.

3. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 2, characterized in that, The control of the surgical robot to drive the end-effector dissection instrument and drive the micropulse exciter integrated on the dissection instrument to apply mechanical micropulses of a preset waveform to the preretinal membrane and retinal tissue in front of the dissection instrument includes: controlling the surgical robot to drive the end-effector dissection instrument, and while moving along a predetermined dissection path, driving the micropulse exciter integrated on the dissection instrument to apply mechanical micropulses of a preset waveform to the preretinal membrane and retinal tissue in front of the dissection instrument.

4. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 1, characterized in that, The specific steps for generating a prospective ductile impedance feature vector based on the generated original biomechanical vibration signal include: We perform joint time-domain and frequency-domain analysis on the original biomechanical vibration signal to extract the energy decay rate, dominant frequency offset, and harmonic component complexity from the original biomechanical vibration signal. The extracted energy decay rate, main frequency offset, and harmonic component complexity are compared with the preset benchmark parameter model, and the difference is calculated. The calculated difference is quantized into a multi-dimensional feature vector, which is the forward toughness impedance feature vector.

5. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 4, characterized in that, The steps of performing joint time-domain and frequency-domain analysis on the original biomechanical vibration signal to extract the energy attenuation rate, dominant frequency offset, and harmonic component complexity from the original biomechanical vibration signal include: Time-domain analysis was performed on the original biomechanical vibration signal to extract the energy decay rate; Frequency domain analysis was performed on the original biomechanical vibration signal to extract the dominant frequency offset and harmonic component complexity.

6. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 1, characterized in that, The specific steps for locating high-risk adhesion warning coordinate points based on the generated forward-looking toughness impedance feature vector include: Real-time monitoring of the eigenvalues ​​of each dimension in the prospective toughness impedance eigenvector, and comparison with the preset toughness safety threshold; When at least one eigenvalue in the prospective toughness impedance eigenvector exceeds its corresponding toughness safety threshold, the current detection area is determined to be a high-risk adhesion area. Record the spatial position of the current dissection instrument, and combine the detection direction and depth information of the micropulse to calculate and output the high-risk adhesion warning coordinates of the high-risk adhesion area in the three-dimensional space of the surgery.

7. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 1, characterized in that, The specific steps for obtaining local microstructure verification images based on high-risk adhesion warning coordinate points include: The high-risk adhesion warning coordinates are transmitted to the scanning controller of the optical coherence tomography system. The optical coherence tomography (OCT) system immediately interrupts its conventional scanning mode and instead precisely aligns its scanning center with the high-risk adhesion warning coordinates, performing a high-density, high-resolution local fine scan to acquire and generate local microstructure verification images.

8. The surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 1, characterized in that, The specific steps for generating a comprehensive assessment report on peeling risk based on local microstructure verification images and prospective toughness impedance feature vectors include: Morphological parameters were extracted from the local microstructure verification images, including the membrane-retina fusion area, changes in the curvature of the anterior retinal surface, and the presence or absence of fibrosis. By correlating morphological parameters with the prospective toughness impedance eigenvector that triggered this focused scan, a dual information pair containing both mechanical prediction and morphological evidence is constructed. Based on the dual information pairs, a quantitative final risk index is generated; Based on the generated final risk index, the final stripping risk level, type, and probability of retinal damage in high-risk areas are comprehensively assessed, thereby generating a comprehensive stripping risk assessment report.

9. A surgical robot OCT-guided adaptive epiretinal membrane dissection system according to claim 8, characterized in that, The specific steps for generating composite adaptive control instructions based on the comprehensive risk assessment report for stripping include: Read the final risk index, risk type, and probability of retinal damage from the comprehensive risk assessment report for retinal stripping. Based on the risk type, a core risk avoidance plan is initially matched from a pre-set control strategy library, and then the execution intensity of the core risk avoidance plan is finely adjusted based on the final risk index. The probability of retinal damage is used as the final decision-making basis for triggering emergency intervention, thereby generating a composite adaptive control command and sending it to the surgical robot.

Images