AI energy platform closed-loop control system for urinary intracavity operation

Abstract

The invention relates to the technical field of medical apparatus and instruments, and particularly discloses an AI energy platform closed-loop control system for urinary intracavity surgery, which comprises the following steps: by collecting visual and biological impedance data of a surgical area in real time, constructing a surgical area topological relation graph taking tissue partitions as nodes and taking a spatial adjacency and function communication relation as edges; according to the method, a key node group representing key blood vessels is identified, and by analyzing all associated paths from non-key nodes to the key node group, the propagation risk of energy operation is calculated, quantitative risk assessment parameters are generated, and risk assessment parameters of target nodes, tissue type probability and real-time biological impedance change rate are fused. By generating a matched energy control instruction, a laser or plasma generator is driven to execute precise energy output; the tissue state change after action is captured again and fed back, and closed-loop control is formed; according to the invention, active quantitative protection of a key anatomical structure and self-adaptive precise regulation and control of energy are realized, and the safety and operation efficiency of an operation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical devices, in particular to an AI energy platform closed-loop control system for urinary endoluminal surgery. BACKGROUND

[0002] Currently, in the endoluminal surgery of urology, the use of energy platforms (such as laser, plasma) has become very common. In order to improve the precision and safety of the operation, the existing technology begins to introduce artificial intelligence assistance, which identifies the tissue type by analyzing the endoscope visual image or single modal sensing data (such as spectrum), and attempts to automatically adjust the energy output power to a certain extent according to the identification result. However, such systems usually simply weight and fuse different sensing data (such as vision, spectrum, bioimpedance) or analyze them independently, and their decision logic focuses on the classification of the tissue properties of the "current contact point", but lacks overall modeling and understanding of the spatial topological relationship and functional association between different tissue segments in the surgical area.

[0003] The existing technology has the following shortcomings: In the endoluminal surgery of urology, the existing AI energy control system faces the dilemma of multi-modal sensing data conflict and clinical safety priority decision. Specifically, when the system judges that the current area is tumor tissue that needs to be removed based on visual and spectral data, it may also detect abnormal impedance signals due to the proximity of large blood vessels. The existing technology lacks a mechanism that can deeply integrate the global safety constraint of "protecting key blood vessel structures" beyond the local resection goal into the real-time control logic, resulting in the system making dangerous energy decisions that conform to the local identification logic but violate the overall surgical safety principle when data conflicts occur, and failing to autonomously avoid the potential damage risk to large blood vessels within the control cycle. SUMMARY

[0004] The purpose of the present application is to provide an AI energy platform closed-loop control system for urinary endoluminal surgery to solve the problems in the above background.

[0005] The purpose of the present application can be achieved by the following technical solutions: An AI energy platform closed-loop control system for urinary endoluminal surgery, comprising: a topological relationship graph construction module for constructing a topological relationship graph of the surgical area based on multi-modal sensing data containing vision and bioimpedance obtained from the surgical area in real time; a key structure identification module for identifying a key node group corresponding to a preset key anatomical structure from all nodes based on the feature vectors of the nodes in the topological relationship graph; The risk propagation analysis module is configured to receive a topological relationship graph containing a critical node group identifier, and based on a topological structure of the corresponding topological relationship graph, calculate a potential influence of an energy operation applied to any non-critical node on the critical node group by analyzing an associated path connected by edges between the non-critical node and the critical node group, and generate a risk assessment parameter corresponding to each non-critical node; The safety control instruction generation module is configured to fuse the feature vector of the target node and the risk assessment parameter in real time for a target node on which the current surgical instrument acts, and generate an energy control instruction matched with the target node; The closed-loop control execution module is configured to adjust laser or plasma parameters output in real time according to the energy control instruction, and act on a surgical region; a state change of the surgical region after the action is captured by multi-modal sensing data and fed back to the topological relationship graph construction module to form a closed-loop control.

[0006] As a further scheme of the present application, the surgical region topological relationship graph specifically comprises: The surgical topological relationship graph is composed of nodes and edges connecting the nodes, the nodes correspond to physical partitions of the surgical region, each node has a feature vector fused based on multi-modal sensing data of the corresponding partition, and the edges are used to represent spatial adjacency association or functional connectivity association between the physical partitions corresponding to the nodes.

[0007] As a further scheme of the present application, the critical node group corresponding to the preset critical anatomical structure is identified from all nodes, specifically comprising: Based on a preset critical anatomical structure spectral feature threshold, the feature vectors of all nodes in the topological relationship graph are initially screened, and the nodes whose spectral feature vectors satisfy the threshold are marked as candidate critical nodes; A biological impedance dynamic change curve corresponding to the candidate critical node is obtained, a periodic pulsation amplitude of the biological impedance dynamic change curve in a unit time is calculated, the pulsation amplitude is compared with a preset blood vessel feature amplitude range, and nodes meeting the requirements are screened to form a set of to-be-verified critical nodes; For each node in the set of to-be-verified critical nodes, the feature vectors of all first-order adjacent nodes of the node in the topological relationship graph are extracted, a cooperative consistency parameter of the corresponding node and its adjacent nodes in the spectral and impedance features is calculated, and if the cooperative consistency parameter is higher than a preset cooperation threshold, the corresponding node is determined as a final critical node, and the corresponding node is added to the critical node group.

[0008] As a further scheme of the present application, the generation process of the risk assessment parameter specifically comprises: For each non-critical node in the topological relationship graph, all associated paths of the non-critical node to all critical nodes in the critical node group are enumerated, and the associated path is composed of a series of edges connected at the head and tail. For each enumerated associated path, according to the preset weight of each edge on the associated path and the total number of edges contained by the path, a path attenuation factor representing the degree of attenuation of the energy influence on the corresponding associated path is calculated; The path attenuation factors of different associated paths from the same target non-critical node to all critical nodes are superimposed to obtain a comprehensive attenuation value, and a risk assessment parameter is generated according to the size of the comprehensive attenuation value.

[0009] As a further scheme of the present application, the calculation process of the path attenuation factor is: According to the type of each edge constituting the associated path, an initial weight is set for each edge; wherein the edge connecting the spatial adjacent physical partitions is assigned a first weight value, and the edge connecting the physical partitions with functional connectivity is assigned a second weight value greater than the first weight value; According to the type of each edge constituting the associated path, an initial weight is set for each edge; wherein the edge connecting the spatial adjacent physical partitions is assigned a first weight value, and the edge connecting the physical partitions with functional connectivity is assigned a second weight value greater than the first weight value; The equivalent weights of all edges on the associated path are multiplied, and the product is divided by the total number of edges contained by the corresponding associated path, and the quotient value is the path attenuation factor.

[0010] As a further scheme of the present application, the generation of the energy control instruction matched with the target node specifically includes: The received risk assessment parameter of the target node is compared with a plurality of preset continuous safety interval thresholds to determine the safety level to which the risk assessment parameter belongs; From the feature vector of the target node, the probability distribution representing the type of organization and the real-time biological impedance change rate are parsed out; The safety level, the probability distribution of the type of organization and the biological impedance change rate are combined to form a multi-dimensional query index; Based on the multi-dimensional query index, a pre-defined energy parameter mapping table is searched and matched to directly map an energy control instruction composed of a specific power value, an action time and an energy waveform mode.

[0011] As a further scheme of the present application, the parsing process of the biological impedance change rate is: The plurality of numerical components in the feature vector representing the spectral response of different tissues are compared with the spectral discrimination thresholds corresponding to the plurality of preset tissue types, respectively; According to the comparison result, an initial probability value is calculated for each preset tissue type; The dispersion degree between the initial probability values corresponding to all tissue types is calculated to obtain a total confidence index; When the overall confidence index is lower than the preset confidence threshold, normalization smoothing processing is performed on all initial probability values based on the corresponding confidence index to obtain a final tissue type probability distribution; the bioimpedance values at the current and previous time points are extracted from the feature vector, and the change amount of the bioimpedance values in a unit time is calculated to obtain a bioimpedance change rate.

[0012] As a further scheme of the present application: the laser or plasma parameters are adjusted and an effect is generated on the surgical area, and specifically, the method comprises: The energy control instruction is received, and the energy control instruction contains the final energy form identifier, power value, energy action time and specific waveform mode code determined after the safety decision; If the energy form identifier is laser, a set of wavelength tuning signals and pulse frequency modulation signals for controlling the laser generator are generated according to the power value and the waveform mode code to adjust the laser parameters; If the energy form identifier is plasma, a set of radio frequency power driving signals and working gas flow adjustment signals for controlling the plasma generator are generated according to the power value and the waveform mode code to adjust the plasma parameters.

[0013] The present application has the following beneficial effects: (1) Traditional surgery relies on the experience of doctors to make risk judgments, while the present application constructs a topological relationship diagram of the surgical area to digitally model the spatial and functional correlation between tissues. The system not only accurately identifies key node groups representing blood vessels according to spectral and bioimpedance characteristics, but also introduces risk propagation analysis. This analysis enumerates all possible conduction paths of energy from the operation point to the key nodes and calculates the comprehensive attenuation value to quantify the potential risk (risk assessment parameter). This makes the clinical principle of "protecting blood vessels" change from an abstract concept relying on subjective experience to an objective parameter that can be quantified in real time based on topological network calculation. When generating energy instructions, this risk assessment parameter serves as the primary constraint condition, which can directly force the system to suppress high-energy cutting mode when the risk is too high and preferentially adopt protective strategies, thereby actively avoiding the risk of misinjury to large blood vessels at the decision-making source, and moving safety control from "after-the-fact remedy" to "pre-emptive prevention".

[0014] (2) The present application can dynamically match the microstate of the organization by fusing multi-modal perception data and real-time risk analysis to enable energy control instructions. The system analyzes the probability distribution of the organization type and the real-time bioimpedance change rate from the feature vector of the target node, which reflects the water content state and the degree of thermal effect of the organization. Combined with the safety level, the optimal power, action time and waveform mode combination are directly mapped through multi-dimensional query index. This closed-loop control based on real-time physiological feedback enables the energy output to be adaptive to different types of tissue (such as tumors, normal mucosa) and their instantaneous state changes (such as drying and carbonization). Reducing the absolute dependence on the experience of the operator, reducing the problem of excessive tissue damage or low cutting efficiency caused by improper manual adjustment, through automatic optimization, maintaining the best energy effect, thereby improving the overall operation cutting efficiency, coagulation effect and operation consistency under the premise of ensuring the safety boundary. BRIEF DESCRIPTION OF DRAWINGS

[0015] The present application will be further described below in conjunction with the accompanying drawings.

[0016] Figure 1 is a system block diagram of the present application. DETAILED DESCRIPTION

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

[0018] Please refer to Figure 1 The present application is an AI energy platform closed-loop control system for urinary endoluminal surgery, which comprises: Topology relationship graph construction module: for constructing a topology relationship graph of the surgical area based on multi-modal sensing data containing vision and bioimpedance acquired in real time from the surgical area; Key structure identification module: based on the feature vector of the nodes in the topology relationship graph, the key node group corresponding to the preset key anatomical structure is identified from all nodes; Risk propagation analysis module: for receiving the topology relationship graph containing the identification of the key node group, and based on the topology structure of the corresponding topology relationship graph, the potential influence of the energy operation applied to any non-key node on the key node group is calculated by analyzing the associated path connected by the edges between the non-key nodes and the key node group, and the risk assessment parameters corresponding to each non-key node are generated; The safety control instruction generation module is configured to generate energy control instructions matched with the target node by fusing the feature vector of the target node and the risk assessment parameters in real time for the target node acted on by the current surgical instrument. The closed-loop control execution module is configured to adjust the output laser or plasma parameters in real time according to the energy control instructions, and to act on the surgical region; the state change of the surgical region after the action is captured by the multi-modal sensing data and fed back to the topological relationship graph construction module to form a closed-loop control.

[0019] In the topological relationship graph construction module, the construction of the topological relationship graph of the surgical region starts from the real-time acquisition of the multi-modal sensing data. The visual data is acquired by a miniature camera integrated in the front end of the endoscope. The bioimpedance data is acquired by a pair of microelectrodes installed on the working end of the surgical instrument. When the instrument contacts the tissue, the electrode pair applies a set of safe alternating current with an amplitude of 1 volt and a frequency range of 5 kHz to 500 kHz to the local tissue at a frequency of 1000 times per second, and synchronously measures the voltage response to obtain the complex impedance value at the contact point. The modulus of the complex impedance value is taken as the real-time bioimpedance data.

[0020] Based on the synchronized multi-modal data, the physical partitioning and node generation of the surgical region are performed. For each frame of real-time image, a simple linear iterative clustering algorithm is used for superpixel segmentation to divide the image into about 500 regions with similar size, continuous color and texture, and each region is defined as a physical partition. Each physical partition corresponds to a node in the topological relationship graph. A feature vector is generated for each node, and the dimension of the vector is 8. The construction process is as follows: from the image superpixel region corresponding to the node, the average values of the red channel, green channel and blue channel are calculated as the first three components; from the spectral sensor data corresponding to the same spatial coordinates, the relative reflection intensities near the three characteristic wavelengths of 540 nm, 577 nm and 600 nm are extracted as the fourth to sixth components; the bioimpedance modulus of all measurement points in the physical partition is averaged to obtain the seventh component; the standard deviation of the bioimpedance in the last 100 ms time window is calculated as the eighth component.

[0021] The edges in the topological graph are established based on two kinds of associations. The first one is spatial adjacency association: if two nodes represent physical partitions that share a boundary in the image plane, an edge is established between them, which is labeled as "spatial adjacency edge". The second one is functional connectivity association: the Euclidean distance between the feature vectors of any two nodes is calculated, especially the dynamic trend of their bioimpedance components (7th and 8th components). If the bioimpedance values of two nodes show highly synchronized rising or falling trend (the correlation coefficient of their change direction is greater than 0.8) in the last 5 consecutive sampling periods, it is considered that there may be blood flow or electrical physiological connectivity between the two partitions, and an edge is established between the corresponding two nodes, which is labeled as "functional connectivity edge".

[0022] During the surgery, the above topological graph is dynamically updated. Each time a new video frame is captured and processed, superpixel segmentation is re-executed, and the spatial position and visual feature vector of the node are updated. Bioimpedance data is refreshed at a higher frequency, and the bioimpedance-related feature components (7th and 8th components) of the node are updated in real time. The set of edges is also adjusted: the spatial adjacency edge is reconstructed according to the new superpixel adjacency relationship; the functional connectivity edge is updated or removed according to the correlation coefficient recalculated according to the latest continuous bioimpedance data.

[0023] In the key structure identification module, the key node group is identified from all nodes in the topological graph, and a preliminary screening based on spectral features is first performed. The preset key anatomical structure is blood vessels, and the spectral feature threshold is set according to the absorption characteristics of oxyhemoglobin at a specific wavelength. Specifically, for the feature vector of a node, the 4th to 6th components are extracted, i.e. the relative reflection intensity near the wavelengths of 540 nm, 577 nm and 600 nm. The preset threshold condition is that the 5th component (577 nm) value needs to be lower than the 4th component (540 nm) and the 6th component (600 nm) value at the same time, and the difference between the three needs to be greater than 0.15. Traverse each node in the topological graph, compare the corresponding components of its feature vector with the threshold condition. All nodes that meet the above conditions are marked as candidate key nodes and enter the next step of verification.

[0024] Next, secondary screening is performed based on the pulsatile characteristics of bioimpedance. The modulus sequence of bioimpedance of each candidate key node within the last 1000 ms time window is obtained, which constitutes the bioimpedance dynamic curve. The periodic pulsatile amplitude of the curve is calculated as follows: first, all local maximum and minimum values within the time window are found, then the absolute value of the difference between each adjacent local maximum and minimum value is calculated, and finally the average of these absolute values is calculated, which is the periodic pulsatile amplitude. The preset blood vessel feature amplitude range is 0.5 ohm to 3.0 ohm. The calculated pulsatile amplitude of each candidate key node is compared with the range, and only the nodes with amplitudes within the range are retained. These nodes through comparison together constitute a set of key nodes to be verified.

[0025] Finally, the final determination is made by analyzing the consistency of the node and its neighborhood. For each node in the set of key nodes to be verified, all first-order adjacent nodes directly connected to it through an edge in the topological relationship graph are found. The feature vector of the node and each adjacent node is extracted. The collaborative consistency parameter is calculated as follows: first, the Pearson correlation coefficient of the node and each adjacent node on the spectral feature component (4th to 6th component) is calculated, obtaining a set of spectral correlation coefficients; at the same time, the Pearson correlation coefficient of the node and each adjacent node on the bioimpedance component (7th component) is calculated, obtaining a set of impedance correlation coefficients. Then, the average of all spectral correlation coefficients and the average of all impedance correlation coefficients are added, and the sum is the collaborative consistency parameter of the node. The preset collaborative threshold is 0.7. If the collaborative consistency parameter of a node is higher than 0.7, the node is determined as the final key node and is added to the key node group. This step ensures that the identified key nodes not only meet the blood vessel feature in their own characteristics, but also have coherent tissue characteristics in the local area where they are located, improving the accuracy of identification.

[0026] In the risk propagation analysis module, the generation of risk assessment parameters begins with the analysis of the topological relationship graph containing the identification of the key node group. First, for each non-key node in the topological relationship graph, enumerate all associated paths to each key node in the key node group. The enumeration process uses a depth-first search strategy, starting from the current non-key node, traversing along the connected nodes, and searching all paths that can reach the target key node. In this process, the maximum depth of path search is set to 5, i.e. the total number of edges contained in an associated path does not exceed 5, to avoid infinite growth of computational complexity, and based on the physical law of energy influence decay with distance, to ignore long propagation paths. All searched connected sequences composed of a series of connected edges are recorded as the associated path of the non-key node to a specific key node.

[0027] For each enumerated associated path, a path attenuation factor is calculated to quantify the degree of attenuation of energy influence as it conducts along the path. The calculation process first assigns an initial weight to each edge on the path. The edges are distinguished by their types: if an edge is marked as a "spatial adjacency edge", it is assigned a first weight value, which is preset as 0.7; if an edge is marked as a "functional connectivity edge", it is assigned a second weight value, which is preset as 0.9. The second weight value is greater than the first weight value, reflecting that the energy conduction efficiency through functional connectivity is higher than that of pure spatial adjacency.

[0028] Subsequently, an energy attenuation correction coefficient is introduced to the above initial weight according to the current planning energy type, to obtain the equivalent weight of each edge under the action of a specific energy. The current energy type is divided into two categories: laser and plasma. If laser energy is used, the correction coefficient is set as: for spatial adjacency edges, the coefficient is 1.0; for functional connectivity edges, the coefficient is 1.1. If plasma energy is used, the correction coefficient is set as: for spatial adjacency edges, the coefficient is 0.9; for functional connectivity edges, the coefficient is 1.2. The calculation method of the equivalent weight is: multiply the initial weight value of a certain edge by the correction coefficient corresponding to the type of the edge and the current energy type, and the product obtained is the equivalent weight of the edge.

[0029] After obtaining the equivalent weights of all edges on an associated path, the path attenuation factor of the path is calculated. The specific calculation process is: multiply the equivalent weights of all edges on this path from the starting point to the end point in order, to obtain a continuous product. Then, divide this continuous product by the total number of edges contained in the associated path. The final quotient obtained is defined as the path attenuation factor of the associated path. The physical meaning of this calculation process is that the continuous multiplication simulates the cumulative effect of the attenuation of energy as it conducts along the path through each edge, and dividing by the total number of edges is a normalization process for paths of different lengths, so that the attenuation factors of long and short paths are comparable.

[0030] After completing the calculation of all path attenuation factors, for the same target non-critical node, the influence of different paths to all key nodes needs to be integrated. The specific method is: sum all the path attenuation factors of the associated paths from the target non-critical node to each key node in the key node group, to obtain a total sum, which is called the integrated attenuation value. This value represents the total potential strength of the influence of the energy operation from the non-critical node to the entire key node group through all possible paths.

[0031] Finally, according to the size of the calculated comprehensive attenuation value, the final risk assessment parameter of the target non-critical node is generated through a preset linear mapping relationship. The mapping relationship is defined as follows: the minimum possible value of the comprehensive attenuation value is set to 0, and the maximum possible experience value is set to 10. When the comprehensive attenuation value is less than or equal to 2, the risk assessment parameter is mapped to 0, representing negligible risk; when the comprehensive attenuation value is greater than 2 and less than or equal to 5, the risk assessment parameter is mapped to the difference between the comprehensive attenuation value and 2; when the comprehensive attenuation value is greater than 5, the risk assessment parameter is uniformly mapped to 3, representing the highest level of risk. Through the above steps, a quantitative risk assessment parameter is generated for each non-critical node, which is directly used for subsequent energy control decision-making.

[0032] In the safety control instruction generation module, the generation of the energy control instruction starts with the fusion processing of the risk assessment parameter and the feature vector of the target node. First, the received risk assessment parameter of the target node (the value is derived from the aforementioned risk propagation analysis) is compared with the preset three consecutive safety interval thresholds. The three thresholds are set as: , , ; thereby defining four safety levels . The decision rule is: if , (safe); if , (low risk); if , (medium risk); if , (high risk). This safety level constitutes the primary constraint condition for subsequent instruction generation.

[0033] Subsequently, the probability distribution of the organization type is analyzed from the feature vector of the target node. The feature vector contains multiple components, of which the 4th to 6th components , , correspond to the spectral response values at wavelengths of 540 nm, 577 nm and 600 nm. The preset organization types include normal mucosa , tumor tissue and bleeding area . Each type of tissue corresponds to a set of preset spectral discrimination thresholds, including the lower limit and the upper limit . An initial matching score is calculated for each type of tissue, and the calculation expression of the initial matching score is: ; wherein, represents the initial matching score, is 1 when the condition in the bracket is true, otherwise 0, represents the lower bound of the th tissue type, represents the upper bound of the th tissue type, th component of the feature vector. This score represents the number of spectral components of the feature vector that fall within the preset threshold range of the tissue type. The initial probability value is calculated by: ; where, is the matching score of the th tissue type; This calculation converts the matching scores into a probability distribution using the Softmax function, such that the sum of the initial probabilities of all tissue types is 1.

[0034] Next, the reliability of this initial probability distribution is evaluated. The degree of dispersion between all tissue type corresponding initial probability values is calculated, resulting in an overall confidence index . Here, the entropy value is used to measure the degree of dispersion (i.e. uncertainty): ; The overall confidence index is defined as: ; where, is the maximum entropy value of the three-class distribution. The value of ranges between 0 and 1, with a larger value indicating a higher degree of certainty of the probability distribution. The preset confidence threshold is set to 0.6. When , it is considered that the initial classification confidence is insufficient, and the normalization smoothing processing based on the confidence index needs to be performed on all initial probability values to obtain the final tissue type probability distribution. The calculation expression of the tissue type probability distribution is: ; where, represents the final tissue type probability distribution, is the total number of tissue types, is the confidence weight used to smooth the initial probability distribution to a uniform distribution when the initial classification confidence is insufficient, and its value is equal to the overall confidence index . This processing performs linear interpolation between the initial probability and the uniform distribution. The lower the confidence, the more the result tends to be a uniform distribution, embodying the conservative principle in high uncertainty. If , then

[0035] ​Meanwhile, the rate of bioimpedance change is extracted from the feature vector. The method is as follows: from the 7th component of the feature vector and the stored history values of the previous 4 time points, a sequence of bioimpedance values of the last 5 time points is obtained . The rate of bioimpedance change is obtained by calculating the linear regression slope of the sequence in unit time (500 ms). That is, a linear regression is performed with the time point as the independent variable and the impedance value as the dependent variable, and the obtained slope is , in ohm / s.

[0036] Then, the above-obtained safety level , the final tissue type probability distribution (typically, the tissue type with the highest probability is taken as the representative) and the rate of bioimpedance change are combined to form a multi-dimensional query index: . Among them, is quantified into three categories: (if ), (if ), (if ).

[0037] Finally, based on the multi-dimensional query index, a lookup and matching are performed in a pre-defined energy parameter mapping table. The mapping table takes the three-tuple of as the key, and its corresponding value is a specific energy control instruction three-tuple (power, action time, energy waveform mode). For example, the index may be mapped to the instruction , indicating a power of 80 watts, an action time of 100 ms, and a pulse waveform mode. Through direct table mapping, the energy control instruction that strictly matches the real-time state of the target node can be obtained, thereby realizing the comprehensive decision of safety constraints, tissue characteristics and dynamic physiological responses.

[0038] In the closed-loop control execution module, the execution of closed-loop control begins with the reception and analysis of the energy control instruction. The instruction is received in the form of a data packet, which explicitly contains four key parameters: the final energy form identifier (represented by integer 1 for laser and integer 2 for plasma), the power value (in watts), the energy action time (in milliseconds), and a specific waveform mode code (represented by an integer, for example, 1 for continuous wave and 2 for pulse wave). The instruction is the direct output result of the previous safety decision process.

[0039] If the energy form is identified as laser (i.e., value 1), the specific laser control signals are generated according to the power value and the waveform mode code in the instruction. First, the power value is converted to the required injection current value for driving the laser diode, according to a pre-defined power-current lookup table. For example, for a thulium laser generator with a wavelength of 1940 nm, 80 W of power can correspond to 2.5 A of driving current. An analog voltage signal proportional to the current value is generated as the wavelength tuning signal, which is fed into a proportional-integral controller to maintain the wavelength stability of the laser output. Meanwhile, a pulse frequency modulation signal is generated according to the waveform mode code: if the code is 2 (pulse wave), a square wave digital signal with the corresponding frequency and duty cycle (usually set to 50%) is generated according to a pre-defined optimized frequency matching the target tissue type (e.g., 30-50 Hz for prostate tissue), which is used to modulate the continuous laser output into pulse mode. These two sets of signals are sent synchronously to the hardware control interface of the laser generator to precisely adjust the laser parameters of its output.

[0040] If the energy form is identified as plasma (i.e., value 2), the control signals for the plasma generator are generated according to the power value and the waveform mode code in the instruction. First, the power value is converted to a radio frequency power driving signal. The process is as follows: the power value is divided by a conversion coefficient (e.g., 10, representing 10 W per volt), to obtain a base voltage value. Then the base value is modulated according to the waveform mode code: if the mode is continuous wave, the constant voltage is output; if the mode is pulse wave, a low-frequency pulse signal (e.g., 10 Hz) is used to switch the voltage. The modulated voltage signal is used to control the output amplitude of the radio frequency power supply. Meanwhile, a working gas flow rate adjustment signal is generated. The gas flow rate (in liters per minute) is linearly determined according to the power value, with the calculation formula: the base flow rate equals the power value multiplied by a gas flow rate coefficient (e.g., 0.05 L / min / W), and is ensured to be within a safe range (e.g., 1-5 L / min). The signal controls the opening degree of a proportional electromagnetic valve to precisely adjust the flow rate of argon or mixed gas delivered to the tip of the surgical instrument. These two sets of signals work together to adjust the intensity, stability, and cutting-coagulation characteristics of the plasma plume.

[0041] The generated control signals drive the energy generator, so that the laser or plasma energy can be applied to the physical sub-zone corresponding to the current target node in the surgical area through the tip of the surgical instrument. The application of energy causes the local tissue to undergo the expected changes, such as cutting, vaporization, or coagulation.

[0042] Following the application of energy, the new state of the surgical area is captured in real time by multimodal sensors. An endoscopic camera captures visual data of changes in color and morphology within the treated area; contact electrodes measure new bioimpedance values ​​caused by tissue dehydration, carbonization, or coagulation. This newly acquired sensor data is immediately transmitted to the initial data processing step of this method.

[0043] The newly acquired sensor data was used to update the initially constructed surgical area topology map. Specifically, the visual and bioimpedance components in the feature vectors of the target node and its surrounding affected nodes were recalculated and replaced based on the new data. Simultaneously, based on the new spatial adjacency relationships and bioimpedance synchronicity after tissue changes, the "edges" between nodes may also be added, deleted, or relabeled. This updated topology map, containing the latest state of the tissue after energy intervention, will serve as a new input for the next control cycle, thus forming a complete, real-time closed-loop control system from "perception-decision-execution" to "re-perception."

[0044] The working principle of this invention is as follows: First, visual and bioimpedance data of the surgical area are acquired in real time using an endoscopic camera and microelectrode pairs on surgical instruments, and a dynamically updated topological graph of the surgical area is constructed based on this data. This graph uses superpixel partitions as nodes, integrates color, spectral, and bioimpedance features, and establishes edges between nodes through spatial adjacency and functional connectivity. Next, based on the unique spectral absorption characteristics and bioimpedance pulsation patterns of blood vessels, the system identifies key node groups representing critical blood vessels from all nodes. Then, by analyzing all associated paths from any non-critical node to the key node group to be operated on, and combining the type and weight of the edges on the paths with the currently used energy type, the system calculates the attenuation degree of energy influence propagating along these paths, and comprehensively generates risk assessment parameters for the operation location. Subsequently, the system integrates the risk assessment parameters of the target node, the probability distribution of tissue type, and the real-time bioimpedance change rate, and generates specific energy control instructions containing power, action time, and waveform mode by querying a preset energy parameter mapping table. Finally, the instruction is converted into a control signal to drive the laser generator or plasma generator, applying precise energy to the surgical area. The changes in the tissue state after the action are captured by the sensor again and used to update the topology map in real time, thereby starting the next control cycle and forming a closed loop of continuous perception, analysis, decision-making and execution.

[0045] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A closed-loop control system for an AI energy platform used in urological surgery, characterized in that, include: Topology graph construction module: used to construct a topology graph of the surgical area based on multimodal sensor data including visual and bioimpedance data acquired in real time from the surgical area; Key structure identification module: Based on the feature vectors of nodes in the topology graph, identify key node groups corresponding to preset key anatomical structures from all nodes; Risk propagation analysis module: It is used to receive the topology graph containing the key node group identifier, and based on the topology of the corresponding topology graph, it analyzes the associated paths between non-key nodes and key node groups through edge connections, calculates the potential impact of energy operations applied to any non-key node on the key node group, and generates risk assessment parameters corresponding to each non-key node. Safety control command generation module: used to generate energy control commands that match the target node by real-time fusion of the target node's own feature vector and risk assessment parameters for the target node being acted upon by the current surgical instrument; Closed-loop control execution module: Based on energy control commands, it adjusts the output laser or plasma parameters in real time and acts on the surgical area; the changes in the state of the surgical area after the action are captured by multimodal sensor data and fed back to the topology graph construction module to form closed-loop control.

2. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 1, characterized in that, The surgical area topology diagram specifically includes: The surgical topology graph consists of nodes and edges connecting the nodes. Nodes correspond to physical partitions of the surgical area. Each node has a feature vector formed by fusing multimodal sensor data of the corresponding partition. Edges are used to represent the spatial adjacency or functional connectivity between the physical partitions corresponding to the nodes.

3. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 1, characterized in that, The identification of key node groups corresponding to preset key anatomical structures from all nodes specifically includes: Based on the preset key anatomical structure spectral feature threshold, the feature vectors of all nodes in the topological relationship graph are initially screened, and nodes whose spectral feature vectors meet the threshold are marked as candidate key nodes. Obtain the bioimpedance dynamic change curves corresponding to candidate key nodes, calculate the periodic pulsation amplitude of the bioimpedance dynamic change curves per unit time; compare the pulsation amplitude with the preset vascular characteristic amplitude range, screen out nodes that meet the requirements, and form a set of key nodes to be verified. For each node in the set of key nodes to be verified, extract the feature vectors of all first-order adjacent nodes in the topology graph, and calculate the coordination consistency parameter of the corresponding node and its adjacent nodes in terms of spectral and impedance characteristics. If the coordination consistency parameter is higher than the preset coordination threshold, the corresponding node is determined to be the final key node and is added to the key node group.

4. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 1, characterized in that, The process of generating the risk assessment parameters specifically includes: For each non-critical node in the topological graph, enumerate all associated paths from the non-critical node to all critical nodes in the critical node group. Each associated path consists of a series of edges connected end to end. For each enumerated associated path, a path attenuation factor is calculated based on the preset weights of each edge on the associated path and the total number of edges contained in the path, which characterizes the degree of attenuation of energy influence on the corresponding associated path. The path attenuation factors of different associated paths from the same non-critical node to all critical nodes are superimposed to obtain a comprehensive attenuation value. Risk assessment parameters are generated based on the magnitude of the comprehensive attenuation value.

5. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 4, characterized in that, The calculation process for the path attenuation factor is as follows: Based on the type of each edge that constitutes the associated path, an initial weight is assigned to each edge; wherein, the edge connecting adjacent physical partitions in space is assigned a first weight value, and the edge connecting physical partitions with functional connectivity is assigned a second weight value greater than the first weight value. Based on the energy type used in the current plan, energy attenuation correction coefficients are introduced for the first and second weight values ​​respectively to obtain the equivalent weight of each edge under the action of a specific energy. The path decay factor is the quotient obtained by multiplying the equivalent weights of all edges on the associated path and then dividing the result by the total number of edges in the corresponding associated path.

6. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 1, characterized in that, The generation of energy control commands that match the target node specifically includes: The received risk assessment parameters of the target node are compared with multiple preset consecutive security interval thresholds to determine the security level to which the risk assessment parameters belong. From the feature vector of the target node, the probability distribution representing the tissue type and the real-time rate of change of bioimpedance are extracted; The probability distribution of safety level, tissue type, and rate of change of bioimpedance are combined to form a multidimensional query index. Based on a multidimensional query index, a search and match is performed in a predefined energy parameter mapping table to directly map an energy control command composed of a specific power value, application time, and energy waveform pattern.

7. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 6, characterized in that, The analytical process for the rate of change of bioimpedance is as follows: The numerical components in the feature vector that represent the spectral response of different tissues are compared with the preset spectral discrimination thresholds corresponding to various tissue types. Based on the comparison results, an initial probability value is calculated for each preset tissue type; Calculate the degree of dispersion among the initial probability values ​​corresponding to all organization types to obtain the overall confidence index; When the overall confidence index is lower than the preset confidence threshold, all initial probability values ​​are normalized and smoothed based on the corresponding confidence index to obtain the final organization type probability distribution. The bioimpedance values ​​at the current and previous times are extracted from the feature vector, and the change in bioimpedance values ​​per unit time is calculated to obtain the rate of change of bioimpedance.

8. The closed-loop control system for an AI energy platform used in urinary tract surgery according to claim 1, characterized in that, The adjustment of the output laser or plasma parameters, and its effect on the surgical area, specifically includes: Receive energy control commands, which include the final energy form identifier, power value, energy application time, and specific waveform pattern code determined after safety decisions. If the energy form is identified as laser, a set of wavelength tuning signals and pulse frequency modulation signals are generated based on the power value and waveform pattern encoding to control the laser generator and adjust the laser parameters. If the energy form is identified as plasma, a set of radio frequency power drive signals and working gas flow regulation signals are generated based on the power value and waveform pattern encoding to control the plasma generator and adjust the plasma parameters.

Images