A laparoscopic surgery smoke adaptive control method and system

Abstract

The application discloses a kind of laparoscopic surgery smoke self-adaptive control method and system, the method includes: continuous real-time acquisition laparoscopic video stream, energy instrument operating parameter and smoke removal equipment operating state parameter;Video stream is input into space-time double-branch smoke state network, obtains smoke density value, field obscuration and smoke change rate;Based on the type of operation, operation step and smoke state, retrieve reference control group from historical record library;According to multidimensional rule base, match candidate control rule;By rule base intelligent agent fusion decision, generate control instruction including target negative pressure, valve opening, filter mode;By embedded controller, execute smoke removal operation, and write this control data back to historical record library.The method and system of the application construct the closed-loop control link of "recognition-decision-execution-feedback", can be adapted to different operation scene, improve smoke removal response speed and accuracy, guarantee the safety and equipment stability of operation.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for medical devices, and in particular to an adaptive control method and system for smoke during laparoscopic surgery. Background Technology

[0002] Laparoscopic surgery, due to its advantages of minimal trauma, rapid recovery, and clear surgical field, has been widely used in minimally invasive treatments in gynecology, urology, and general surgery. During laparoscopic surgery, the use of energy instruments such as electrocautery and ultrasonic scalpels to cut and coagulate tissue generates a large amount of surgical smoke. Studies have shown that surgical smoke contains various toxic and harmful substances such as formaldehyde, benzene compounds, carbon monoxide, and viral particles. This not only poses a potential threat to the health of operating room medical staff but also forms a floating smoke layer and localized hot fog in front of the laparoscopic lens, causing reduced brightness under the microscope, blurred tissue boundaries, and distorted depth judgment, seriously affecting the accuracy and safety of the surgical procedure.

[0003] To eliminate the interference of surgical smoke on the surgical field, existing laparoscopic systems are typically equipped with negative pressure suction smoke removal devices. However, current smoke removal devices mostly use manual control, meaning the surgeon or assistant manually adjusts the start / stop or level of negative pressure suction based on the smoke level in the surgical field. This control method has a significant response lag problem. When a large amount of smoke is generated, the operator often cannot adjust in time, leading to a decrease in surgical field clarity; conversely, if the negative pressure is not turned off or reduced in time after the smoke dissipates, unnecessary pressure fluctuations and gas loss within the cavity will occur. Furthermore, manual control increases the workload of the surgical team, distracts the surgeon, and is not conducive to focused surgical execution.

[0004] In recent years, some studies have attempted to introduce sensor technology into smoke removal control, using optical or aerosol sensors to detect smoke concentration and automatically trigger the start and stop of the smoke removal device based on a preset threshold. This type of automatic control scheme based on single threshold triggering is an improvement over purely manual control, but it still has the following shortcomings: First, the sensor detection point and the smoke source often have spatial differences, making it difficult for the detection results to accurately reflect the degree of smoke obstruction within the lens's field of view; second, fixed threshold control cannot adapt to the dynamic and rapid changes in surgical smoke, easily leading to frequent start-stops or untimely responses in scenarios with frequent smoke concentration fluctuations; third, the characteristics of smoke generated by different surgical types and stages vary significantly. For example, smoke from cholecystectomy is mostly intermittent and burst-like, while smoke from myomectomy is continuously generated, making a single sensor triggering strategy unable to provide differentiated adaptation for different surgical scenarios.

[0005] To address the aforementioned issues, some scholars have proposed using deep learning models to identify smoke in laparoscopic videos and controlling smoke removal equipment based on the identification results. While this type of vision-driven control scheme has made progress in smoke recognition accuracy, key technical shortcomings remain. Specifically, existing solutions mainly suffer from the following three problems:

[0006] First, there is a lack of structured historical data support. Existing systems only process real-time data from the current surgery, failing to systematically store and reuse smoke control experience accumulated from past surgeries. For similar surgical scenarios, parameter adjustments and adaptations must be repeated for each surgery, resulting in low efficiency and difficulty in forming standardized, optimized control schemes. More importantly, due to the lack of a historical data feedback mechanism, the system cannot achieve data-driven continuous learning and performance optimization, making it difficult for control effectiveness to gradually improve with the accumulation of surgical cases.

[0007] Second, the rule base design is rigid and simplistic. Existing solutions mostly use pre-set static logic for control rules, such as triggering different levels of negative pressure suction based on smoke concentration thresholds. This rule base fails to adequately consider multi-dimensional characteristics such as surgical type (e.g., the smoke generation patterns differ between cholecystectomy and prostate surgery), smoke composition characteristics, the operating status of energy devices (e.g., electrosurgical power and continuous operating time), and the status of the smoke removal equipment itself (e.g., remaining filter life and current negative pressure value). When the smoke generation pattern changes during surgery, the static rule base cannot dynamically adjust the control strategy, easily leading to a mismatch between the control strategy and the actual situation, resulting in poor smoke removal or wasted resources.

[0008] Third, the intelligent control chain is disconnected. The existing intelligent smoke removal system suffers from significant module fragmentation in its "perception-decision-execution" chain. The visual recognition module (such as a convolutional neural network model) and the control execution module (such as a microcontroller) are often developed and run independently, lacking a unified intelligent agent for collaborative decision-making and precise scheduling. Visual recognition results are typically only used to trigger switching actions, failing to transmit fine-grained quantitative information such as smoke density, obstruction rate, and changing trends to the control execution end. This results in key control parameters such as negative pressure parameters, valve opening, and filtration mode being unable to be finely adjusted based on the smoke state. This open-loop or semi-open-loop control architecture makes it difficult to achieve closed-loop collaborative control of smoke recognition, dynamic adjustment of negative pressure parameters, filtration mode switching, and system safety protection, thus limiting overall smoke removal efficiency and operational stability.

[0009] In summary, existing laparoscopic surgery smoke control technologies have significant shortcomings in areas such as historical experience reuse, multi-dimensional rule adaptation, and integrated perception, decision-making, and execution. Therefore, developing an adaptive smoke control method for laparoscopic surgery that integrates historical surgical data, a multi-dimensional dynamic rule base, and agent-based collaborative decision-making is crucial for improving surgical safety and operational efficiency, and is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] To address the technical problems existing in the background art, this invention proposes an adaptive smoke control method and system for laparoscopic surgery.

[0011] The present invention proposes an adaptive smoke control method for laparoscopic surgery, comprising the following steps:

[0012] S1. Acquire real-time video stream of laparoscopic surgery, and simultaneously acquire the operating parameters of the energy device and the operating status parameters of the smoke removal device. The operating parameters of the energy device include at least the electrosurgical power and the continuous working time of the electrosurgical unit; the operating status parameters of the smoke removal device include at least the current negative pressure value, the current flow rate value, and the remaining lifespan of the filter element.

[0013] S2. Input the real-time video stream into a pre-trained spatiotemporal dual-branch smoke state network for processing to obtain the smoke state quantization data at the current moment. The smoke state quantization data includes at least the smoke density value, field of view occlusion rate, and smoke change rate.

[0014] S3. Based on the current surgical type, current surgical step, and smoke state quantification data, retrieve the K historical control records with the highest similarity to the current scene from the historical record database to form a reference control group;

[0015] S4. Based on the current surgical type, current surgical step, smoke state quantification data, operating parameters of energy devices, and operating status parameters of smoke removal equipment, at least one candidate control rule is matched from the multi-dimensional rule base.

[0016] S5. The rule base agent makes decisions based on the reference control group and candidate control rules, and generates control instructions that include the target negative pressure value, valve opening value, filter mode identifier and operating mode identifier.

[0017] S6. Send the control command to the embedded controller to drive the negative pressure pump, proportional valve and filter components to perform smoke removal operation, and write back the scene characteristics, control command and execution effect of this control as a new structured history record to the history record library.

[0018] Preferably, the spatiotemporal dual-branch smoke state network includes a spatial branch, a temporal branch, a gating fusion unit, and a control output unit arranged in parallel;

[0019] The input of the spatial branch is used to receive a single frame image of the laparoscopic video at the current moment, and outputs the spatial morphological feature vector of the smoke in the current frame.

[0020] The input of the time branch is used to receive a laparoscopic video sequence containing the current frame and several consecutive previous frames, and outputs a feature vector of dynamic changes in smoke.

[0021] The first input of the gated fusion unit is connected to the output of the spatial branch, and the second input is connected to the output of the temporal branch. It is used to adaptively weight and fuse the spatial morphology feature vector of smoke and the dynamic change feature vector of smoke to output a spatiotemporal fusion feature vector.

[0022] The input of the control output unit is connected to the output of the gated fusion unit, and is used to map the spatiotemporal fusion feature vector into smoke state quantization data.

[0023] Preferably, step S3 specifically includes:

[0024] The current surgical type, current surgical step, smoke density value, field of view occlusion rate, and smoke change rate in the smoke state quantification data are combined to form the current scene feature vector;

[0025] Calculate the weighted similarity distance between the current scene feature vector and the historical scene feature vector stored in each structured historical record in the history database; wherein the calculation of the weighted similarity distance takes into account at least the differences in smoke density values, differences in field of view occlusion rates, differences in smoke change rates, surgical type matching degree, and surgical step matching degree.

[0026] Based on the weighted similarity distance in ascending order, the first K structured historical records are selected from the historical record database;

[0027] From the K selected structured historical records, extract the K sets of historical control parameter vectors associated with them. The historical control parameter vectors include at least historical negative pressure values, historical valve opening values, and historical filtering mode identifiers. The K sets of historical control parameter vectors are then output as a reference control group.

[0028] Preferably, the weighted similarity distance Calculated using the following formula:

[0029] ;

[0030] in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; The rate of change of smoke in the current scene; The historical smoke density value corresponding to the i-th structured historical record in the historical record database; The historical field of view occlusion rate corresponding to the i-th structured historical record in the historical record library; Let be the rate of change of the historical smoke screen corresponding to the i-th structured historical record in the historical record database; Identify the current surgical type; Identify historical surgical types; This indicates the current surgical step. This serves as an identifier for historical surgical steps; the function M(a,b) returns 0 if a and b are the same, and 1 if they are different. The preset weighting coefficients, and .

[0031] Preferably, step S4 specifically includes:

[0032] S41. Based on the current surgical type and the current surgical procedure, all rule entries that match the applicable surgical type and applicable surgical procedure are initially selected from the multi-dimensional rule base to form a primary candidate rule set.

[0033] S42. For each rule entry in the primary candidate rule set, determine in sequence whether the current smoke density value falls within its specified smoke density value range, whether the current field of view occlusion rate falls within its specified field of view occlusion rate range, whether the current smoke change rate falls within its specified smoke change rate range, and whether the current operating parameters of the energy device meet its specified device power conditions.

[0034] S43. Include the rule entries that satisfy all the judgment conditions in step S42 into the secondary candidate rule set;

[0035] S44. Calculate the matching score between each rule entry in the secondary candidate rule set and the current scene; the matching score is calculated based on at least the degree of closeness between the current smoke density value and the center value of the rule's smoke density value interval, the degree of closeness between the current field of view occlusion rate and the center value of the rule's field of view occlusion rate interval, and the degree of conformity between the current smoke removal equipment's operating status parameters and the safety boundary value defined by the rule.

[0036] S45. Based on the matching scores from high to low, select at least one rule entry from the secondary candidate rule set as the candidate control rule output.

[0037] Preferably, the matching score Calculated using the following formula:

[0038] ;

[0039] in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; Let be the center value of the smoke density value interval for the j-th rule entry; Let be the center value of the field of view occlusion rate interval for the j-th rule entry; This is the preset normalized range for smoke density values; The preset normalized range for field of view occlusion rate; This is the current negative pressure value; This represents the current traffic volume. This represents the remaining lifespan of the filter element. The upper limit of negative pressure safety is defined for the j-th rule entry; The lower limit value for traffic security defined for the j-th rule entry; The filter cartridge life safety lower limit is defined for the j-th rule entry; the function S(·) is used to evaluate the degree of compliance between the current equipment state and the safety boundary. Its output value range is [0, 1]. It returns 1 when the equipment state fully meets the safety boundary requirements and returns 0 when the equipment state deviates significantly from the safety boundary. The higher the degree of compliance, the larger the return value. The preset weighting coefficients, .

[0040] Preferably, step S5 specifically includes:

[0041] S51. Obtain at least one safety boundary constraint defined in the candidate control rules, wherein the safety boundary constraint includes at least a negative pressure safety upper limit value, a flow rate safety lower limit value, and a filter cartridge life safety lower limit value;

[0042] S52. Compare the current negative pressure value, current flow rate value, and remaining filter life value in the current operating status parameters of the smoke removal equipment with the upper limit of negative pressure safety, the lower limit of flow rate safety, and the lower limit of filter life safety, respectively. If any current value violates its corresponding safety boundary constraint, generate a control command containing an alarm flag and maintaining the current operating parameters, and end this step; otherwise, execute step S53.

[0043] S53. From the K sets of historical control parameter vectors contained in the reference control group, extract the historical negative pressure value with the highest frequency as the mainstream historical negative pressure recommendation value, and extract the historical valve opening value with the highest frequency as the mainstream historical valve opening recommendation value.

[0044] S54. Compare the mainstream historical negative pressure recommended value with the target negative pressure value defined in the candidate control rules, and compare the mainstream historical valve opening recommended value with the target valve opening value defined in the candidate control rules.

[0045] S55. If the absolute value of the difference between the mainstream historical negative pressure recommendation value and the target negative pressure value is less than the first preset threshold, and the absolute value of the difference between the mainstream historical valve opening recommendation value and the target valve opening value is less than the second preset threshold, then it is determined that the experience and rules are consistent, and a control command is generated based on the mainstream historical negative pressure recommendation value and the mainstream historical valve opening recommendation value.

[0046] S56. If the consistency condition in step S55 is not met, it is determined that there is a discrepancy between experience and rules. The safety boundary constraint is used as a hard constraint. A weighted calculation is performed between the mainstream historical negative pressure recommendation value and the target negative pressure value, and a weighted calculation is performed between the mainstream historical valve opening recommendation value and the target valve opening value. The calculation results are used as the final target negative pressure value and target valve opening value, respectively, and control commands are generated based on these.

[0047] Preferably, the construction and process of the historical record library specifically includes:

[0048] Complete data from multiple typical laparoscopic surgeries were collected. The complete data included surgical videos, energy device parameters, smoke removal equipment control parameters and status parameters. Experts scored and labeled the smoke removal effect of each key control based on the postoperative video playback.

[0049] Each control operation, along with its corresponding scenario data, control parameters, and performance score, is stored as an initial structured historical record in the historical record library.

[0050] After each execution of step S6, the newly generated structured historical records are stored in the historical record library; the historical records under the same surgical type and surgical step label are periodically statistically analyzed, the average effect score of each combination of control parameters is calculated, and the similarity calculation weight of the historical records corresponding to the parameter combination with the highest average score is increased.

[0051] Preferably, the construction process of the multi-dimensional rule base specifically includes:

[0052] Data mining is performed on the structured historical records accumulated in the historical record database, and the records are grouped based on surgical type and surgical procedure;

[0053] For each group, a clustering algorithm is used to analyze the smoke state quantification data to form several typical smoke state intervals;

[0054] Each smoke state interval is associated with the historical control parameter with the highest effect score in that interval within the same group, and the statistical safety boundary of the equipment operating state parameters in that group is added to form an initial multi-dimensional rule entry.

[0055] Set a rule confidence index. When a rule is matched and executed, if the effect score of the new historical records it generates is consistently higher or lower than the historical average effect score of the rule, then fine-tuning of the target control parameters or applicable range of the rule will be triggered. When a new surgical type or procedure appears and sufficient data is accumulated, a new rule generation process will be automatically triggered.

[0056] This invention proposes an adaptive smoke control system for laparoscopic surgery, applied to the adaptive smoke control method for laparoscopic surgery as described in any of the above claims, comprising:

[0057] The video and status acquisition unit is used to continuously and in real-time acquire laparoscopic video streams, energy device operating parameters, and smoke removal equipment operating status parameters.

[0058] The spatiotemporal dual-branch smoke recognition module is used to process laparoscopic video streams and output smoke state quantification data;

[0059] The historical record library module is used to store structured historical records and retrieve and output reference control groups based on the current surgical type, current surgical step, and smoke status.

[0060] The multi-dimensional rule base module is used to store multi-dimensional rule entries and match and output candidate control rules based on the current surgical type, current surgical step, smoke state quantification data, energy device operating parameters and smoke removal equipment operating status parameters.

[0061] The rule-based intelligent agent decision-making module is used to generate standardized control instructions based on the reference control group and candidate control rules;

[0062] The execution control module is used to receive control commands and drive the negative pressure pump, proportional valve and filter components to perform smoke removal operations, and feed the execution results back to the historical record library module.

[0063] The proposed laparoscopic surgery smoke adaptive control method and system continuously and in real-time collects laparoscopic video streams and equipment status data, drives a spatiotemporal dual-branch smoke state network to accurately quantify smoke, and combines the experience reuse of historical records and dynamic matching of multi-dimensional rule bases. The rule base agent realizes closed-loop collaborative control of perception, decision-making and execution. Compared with the prior art, it improves the timeliness of smoke recognition and the fineness of control commands. It can adaptively output the optimal smoke removal strategy under different surgical types and stages. At the same time, it has built-in multi-level safety boundary constraints and a continuous optimization mechanism based on historical effect feedback, which takes into account smoke removal efficiency, equipment protection and surgical safety. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the adaptive smoke control method for laparoscopic surgery proposed in this invention.

[0065] Figure 2 This is a schematic diagram of the system architecture of an adaptive control system for laparoscopic surgery smoke proposed in this invention;

[0066] Figure 3 This is a schematic diagram of the workflow of Embodiment 1 of the laparoscopic surgery smoke adaptive control system proposed in this invention. Detailed Implementation

[0067] Reference Figure 1 The present invention proposes an adaptive smoke control method for laparoscopic surgery, comprising the following steps:

[0068] S1. Acquire the real-time video stream of the laparoscopic surgery and simultaneously acquire the operating parameters of the energy device and the operating status parameters of the smoke removal device. The operating parameters of the energy device include at least the electrosurgical power and the continuous working time of the electrosurgical unit; the operating status parameters of the smoke removal device include at least the current negative pressure value, the current flow rate value, and the remaining lifespan of the filter element.

[0069] It should be noted that the system consists of a high-definition laparoscopic camera system, a data interface for an energy platform (such as an electrosurgical unit), and sensors integrated into the smoke removal device (such as pressure sensors and flow sensors). It is responsible for collecting real-time video streams of laparoscopic surgery, electrosurgical power / continuous working time, current negative pressure value, current flow rate value, and remaining lifespan of the filter element.

[0070] S2. Input the real-time video stream into the pre-trained spatiotemporal dual-branch smoke state network for processing to obtain the smoke state quantization data at the current moment. The smoke state quantization data includes at least the smoke density value, field of view occlusion rate, and smoke change rate.

[0071] In this embodiment, the spatiotemporal dual-branch smoke state network includes a spatial branch, a temporal branch, a gating fusion unit, and a control output unit arranged in parallel;

[0072] The input of the spatial branch is used to receive a single frame image of the laparoscopic video at the current moment, and outputs the spatial morphological feature vector of the smoke in the current frame.

[0073] The input of the time branch is used to receive a laparoscopic video sequence containing the current frame and several consecutive previous frames, and outputs a feature vector of dynamic changes in smoke.

[0074] The first input of the gated fusion unit is connected to the output of the spatial branch, and the second input is connected to the output of the temporal branch. It is used to adaptively weight and fuse the spatial morphology feature vector of smoke and the dynamic change feature vector of smoke to output a spatiotemporal fusion feature vector.

[0075] The input of the control output unit is connected to the output of the gated fusion unit, and is used to map the spatiotemporal fusion feature vector into smoke state quantization data.

[0076] Specifically, the continuous real-time video stream during the operation is input into the spatiotemporal dual-branch smoke state network at 25 frames per second. For each frame, scaling, denoising, white balance correction, and normalization are performed to obtain the current frame image. Simultaneously extract the most recent 5 frame sequences. This serves as the input for the time branch. To balance real-time performance and stability, in one implementation, the recognition results are refreshed every 0.2 seconds.

[0077] The spatial branch (based on the ResNet-18 backbone network) is responsible for extracting the smoke morphology, edges, and occlusion regions in the current frame; the temporal branch is responsible for extracting the changing trends of smoke diffusion, aggregation, and dissipation in the last 5 frames; the gated fusion unit is responsible for weighted fusion of the static information of the current frame with the dynamic information of the continuous frames; and the control output unit is responsible for generating pixel-level smoke probability maps, density values, occlusion rates, and change rates, among other control variables.

[0078] In the temporal branch (based on the ConvGRU structure), the five most recent frames are processed by a shared-weight convolutional encoder to obtain five temporal features. and The preferred size of each time-series feature is... The five time-series features mentioned above are then input into the ConvGRU unit to obtain the time-state features. The time-state characteristics are used to characterize the diffusion direction, dissipation rate, and continuity of smoke changes within a short time window.

[0079] To achieve the fusion of spatial and temporal information, the spatial branches Mapped to temporal state features via 1×1 convolution With the same channel dimension, we get Then and After concatenation, the data is input into the gating fusion unit to obtain the gating weights. Thus, the fusion characteristics are obtained. Features after gating fusion Then input the coordinate attention module to obtain the enhanced features. It is used to retain information about the distribution of smoke in the horizontal and vertical directions.

[0080] The control output unit includes a segmentation head, a density regression head, and a rate of change regression head. The segmentation head will... High-level features after upsampling Perform feature pyramid fusion to output pixel-level smoke probability maps. Density regression head pair After global average pooling, the output density prior is obtained through two fully connected layers. ; Rate of change regression head for time state characteristics Perform global pooling and linear mapping, and prioritize output changes. With confidence level .

[0081] The quantization of smoke screen parameters is performed according to the following rules: The pixel-level smoke screen probability map is satisfied with... The number of pixels is denoted as Field of view occlusion rate according to Calculate; denote the average probability of pixels satisfying the threshold as Then the smoke density value The result, obtained by combining the segmentation result and the density prior, can be expressed as: Its value ranges from 0 to 1, where, The activation function has an output range of (0, 1); the rate of change of the smoke screen. It is obtained by combining the change in occlusion rate with the prior a priori change, and can be expressed as: ;in, The activation function has an output range of (-1, 1). Using this method, the model output simultaneously considers both the current frame state and short-term trends.

[0082] To ensure that the model output can be directly fed into the control link, the system encapsulates the recognition result of each frame into {frame number, timestamp, surgery type, step label, pixel-level smoke probability map}. Confidence The structured message is sent to the rule-based agent. Instead of recalculating visual features, the rule-based agent directly uses the structured results, case parameter set, and candidate rule set to make a joint decision.

[0083] In this embodiment, the training samples for the spatiotemporal dual-branch smoke state network are derived from manually reviewed and annotated laparoscopic smoke video clips. The annotations include at least smoke region masks, concentration level labels, and continuous frame trend labels. The training set, validation set, and test set are preferably divided in a 7:2:1 ratio. Before training, random brightness perturbation, local reflection enhancement, random cropping, and mirror enhancement are applied to the images to improve the model's robustness to different lighting conditions, instrument reflections, and weak smoke scenes.

[0084] In one implementation, the optimizer uses AdamW, with a batch size of 12, an initial learning rate of 1e-4, and 90 training epochs; the loss function is... It can be represented as:

[0085] ;

[0086] in, For smoke segmentation loss, For concentration regression loss, For the regression loss of the diffusion trend, This is used to constrain the consistency between the current frame output and short-term sequence changes. Through the above training process, the model hierarchy, connection relationships, training steps, and key parameters can be clearly defined.

[0087] Compared with smoke screen observation models based solely on single-frame convolutional backbone networks, the spatiotemporal dual-branch smoke state network in this application emphasizes the joint modeling of "current frame spatial features + continuous frame dynamic features", which makes the output results naturally carry temporal information and is more suitable for directly serving rule matching and negative pressure control. Therefore, its model approach is significantly different from schemes that are only used for observation prompts or displays.

[0088] S3. Based on the current surgical type, current surgical step, and smoke state quantification data, retrieve the K historical control records with the highest similarity to the current scene from the historical record database to form a reference control group.

[0089] In this embodiment, step S3 specifically includes:

[0090] The current surgical type, current surgical step, smoke density value, field of view occlusion rate, and smoke change rate in the smoke state quantification data are combined to form the current scene feature vector;

[0091] Calculate the weighted similarity distance between the current scene feature vector and the historical scene feature vector stored in each structured historical record in the history database; wherein, the calculation of the weighted similarity distance takes into account at least the differences in smoke density values, differences in field of view occlusion rates, differences in smoke change rates, surgical type matching degree, and surgical step matching degree.

[0092] Based on the weighted similarity distance in ascending order, the first K structured historical records are selected from the historical record database;

[0093] From the selected K structured historical records, extract the K sets of historical control parameter vectors associated with them. The historical control parameter vectors include at least the historical negative pressure value, the historical valve opening value, and the historical filter mode identifier. Then, use these K sets of historical control parameter vectors as the output of the reference control group.

[0094] Specifically, weighted similarity distance Calculated using the following formula:

[0095] ;

[0096] in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; The rate of change of smoke in the current scene; The historical smoke density value corresponding to the i-th structured historical record in the historical record database; The historical field of view occlusion rate corresponding to the i-th structured historical record in the historical record library; Let be the rate of change of the historical smoke screen corresponding to the i-th structured historical record in the historical record database; Identify the current surgical type; Identify historical surgical types; This indicates the current surgical step. This serves as an identifier for historical surgical steps; the function M(a,b) returns 0 if a and b are the same, and 1 if they are different. The preset weighting coefficients, and .

[0097] In this embodiment, the construction and process of the historical record library specifically includes:

[0098] Complete data from multiple typical laparoscopic surgeries were collected. The complete data included surgical videos, energy device parameters, smoke removal equipment control parameters and status parameters. Experts scored and labeled the smoke removal effect of each key control based on the postoperative video playback.

[0099] Each control operation, along with its corresponding scenario data, control parameters, and performance score, is stored as an initial structured historical record in the historical record library.

[0100] After each execution of step S6, the newly generated structured historical records are stored in the historical record library; the historical records under the same surgical type and surgical step label are periodically statistically analyzed, the average effect score of each combination of control parameters is calculated, and the similarity calculation weight of the historical records corresponding to the parameter combination with the highest average score is increased.

[0101] S4. Based on the current surgical type, current surgical procedure, smoke state quantification data, operating parameters of the energy device, and operating status parameters of the smoke removal equipment, at least one candidate control rule is matched from the multi-dimensional rule base.

[0102] In this embodiment, step S4 specifically includes:

[0103] S41. Based on the current surgical type and the current surgical procedure, all rule entries that match the applicable surgical type and applicable surgical procedure are initially selected from the multi-dimensional rule base to form a primary candidate rule set.

[0104] S42. For each rule entry in the primary candidate rule set, determine in sequence whether the current smoke density value falls within its specified smoke density value range, whether the current field of view occlusion rate falls within its specified field of view occlusion rate range, whether the current smoke change rate falls within its specified smoke change rate range, and whether the current operating parameters of the energy device meet its specified device power conditions.

[0105] S43. Include the rule entries that satisfy all the judgment conditions in step S42 into the secondary candidate rule set;

[0106] S44. Calculate the matching score between each rule entry in the secondary candidate rule set and the current scene; the matching score is calculated based on at least the degree of closeness between the current smoke density value and the center value of the rule's smoke density value interval, the degree of closeness between the current field of view occlusion rate and the center value of the rule's field of view occlusion rate interval, and the degree of conformity between the current smoke removal equipment's operating status parameters and the safety boundary value defined by the rule.

[0107] S45. Based on the matching scores from high to low, select at least one rule entry from the secondary candidate rule set as the candidate control rule output.

[0108] Specifically, matching scores Calculated using the following formula:

[0109] ;

[0110] in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; Let be the center value of the smoke density value interval for the j-th rule entry; Let be the center value of the field of view occlusion rate interval for the j-th rule entry; This is the preset normalized range for smoke density values; This is the preset normalized range for field of view occlusion rate; This is the current negative pressure value; This represents the current traffic volume. This represents the remaining lifespan of the filter element. The upper limit value for negative pressure safety is defined for the j-th rule entry; The lower limit value for traffic security defined for the j-th rule entry; The filter cartridge life safety lower limit is defined for the j-th rule entry; the function S(·) is used to evaluate the degree of compliance between the current equipment state and the safety boundary. Its output value range is [0, 1]. It returns 1 when the equipment state fully meets the safety boundary requirements and returns 0 when the equipment state deviates significantly from the safety boundary. The higher the degree of compliance, the larger the return value. The preset weighting coefficients, .

[0111] In this embodiment, the construction process of the multi-dimensional rule base specifically includes:

[0112] Data mining is performed on the structured historical records accumulated in the historical record database, and the records are grouped based on surgical type and surgical procedure;

[0113] For each group, a clustering algorithm is used to analyze the smoke state quantification data to form several typical smoke state intervals;

[0114] Each smoke state interval is associated with the historical control parameter with the highest effect score in that interval within the same group, and the statistical safety boundary of the equipment operating state parameters in that group is added to form an initial multi-dimensional rule entry.

[0115] Set a rule confidence index. When a rule is matched and executed, if the effect score of the new historical records it generates is consistently higher or lower than the historical average effect score of the rule, then fine-tuning of the target control parameters or applicable range of the rule will be triggered. When a new surgical type or procedure appears and sufficient data is accumulated, a new rule generation process will be automatically triggered.

[0116] Specifically, each historical record includes at least {surgery type, step label, smoke feature vector, control parameter vector, execution result vector, and safety label}. The smoke feature vector includes... The location of the smoke screen's center of gravity and the location of the energy device's heat source; the control parameter vector includes the target negative pressure. Valve opening Filtering mode Operating mode and execution time The execution result vector includes the sharpness recovery time. The decrease in occlusion rate Doctor adoption mark and abnormal alarm flags .

[0117] In one embodiment, the smoke density value The normalized concentration (0 to 1) obtained from model regression represents the field-of-view occlusion rate. according to Calculation, where Represents the number of pixels identified as smoke at time i; smoke change rate. according to The calculation describes the occlusion trend over the past 5 frames. Based on this unified definition, subsequent historical retrieval, rule matching, and control execution all revolve around the same set of control variables.

[0118] After standardizing the units in the fields, structured records are created. Pressure is standardized to mmHg, flow rate to L / min, valve opening to 0% to 100%, and image clarity recovery time to seconds. Before archiving, records are supplemented with surgical procedure tags, surgical step tags, instrument category tags, and markers indicating whether the doctor adopted the procedure, to improve the accuracy of similar scenario searches.

[0119] An example of a historical record can be represented as: {"Surgery Type":"Laparoscopic Myomectomy","Step Tags":"Membrane Removal and Hemostasis Stage","Smoke Screen Characteristics":{" ":0.49," ":32," ":0.024},"Control Parameters":{"Target Negative Pressure":118,"Valve Opening":57,"Filtering Mode":"Enhanced","Operating Mode":"Pulse"},"Execution Result":{"Clarity Recovery Time":1.4,"Obstruction Rate Reduction":18,"Doctor's Adoption":1},"Safety Tag":"Normal"}.

[0120] A combined index is built based on surgical type, procedure label, density range, and filtering mode; similarity distance is calculated for candidate records.

[0121] ;

[0122] Among them, the differences of each continuous feature are normalized before being included in the calculation; when the field of view occlusion rate , When expressed as a percentage, The item can be equivalently written as . This represents a function for matching surgical types. This represents a function that matches surgical steps; it returns 0 if they match and 1 if they don't. The preset weighting coefficients are used. In a preferred embodiment, the weights are set to 0.32, 0.28, 0.18, 0.12, and 0.10, respectively; the K records with the smallest distance are selected as the source of the reference control group.

[0123] The reference control set output by the historical data library includes at least the target negative pressure range, recommended valve opening, recommended filtration mode, recommended operating mode, corresponding clarity recovery time, and applicable constraints. For example, in the scenario of "tumor removal and hemostasis stage + moderate smoke screen obstruction", the target negative pressure range of [112mmHg, 120mmHg], the valve opening range of 52% to 60%, and the recommended combination of "enhanced filtration + pulse operation" can be directly output.

[0124] Records adopted by doctors and whose clarity recovery time is less than a preset threshold are given higher weights; records that trigger overpressure, insufficient flow, or are actively withdrawn by doctors are given lower weights. Preferably, every 50 similar surgeries are accumulated, the Top-K search center and empirical parameter weights are recalculated using "doctor adoption rate, coverage decline rate, recovery time, and safety alarm rate" as indicators, thereby achieving self-optimization of the historical record database.

[0125] The historical record database in this application differs from existing methods that only store logs or video clips in that it directly addresses control decisions by constructing field structures, retrieval logic, and weight update mechanisms. It can answer both "what does the empirical parameter include" and "how can the historical record database be continuously optimized".

[0126] S5. The rule-based intelligent agent makes decisions based on the reference control group and candidate control rules, and generates control instructions that include the target negative pressure value, valve opening value, filter mode identifier and operating mode identifier.

[0127] In this embodiment, step S5 specifically includes:

[0128] S51. Obtain at least one safety boundary constraint defined in the candidate control rules. The safety boundary constraint shall include at least the upper limit of negative pressure safety, the lower limit of flow rate safety, and the lower limit of filter life safety.

[0129] S52. Compare the current negative pressure value, current flow rate value, and remaining filter life value in the current operating status parameters of the smoke removal equipment with the upper limit of negative pressure safety, the lower limit of flow rate safety, and the lower limit of filter life safety, respectively. If any current value violates its corresponding safety boundary constraint, generate a control command containing an alarm flag and maintaining the current operating parameters, and end this step; otherwise, execute step S53.

[0130] S53. From the K sets of historical control parameter vectors contained in the reference control group, extract the historical negative pressure value with the highest frequency as the mainstream historical negative pressure recommendation value, and extract the historical valve opening value with the highest frequency as the mainstream historical valve opening recommendation value.

[0131] S54. Compare the mainstream historical negative pressure recommended value with the target negative pressure value defined in the candidate control rules, and compare the mainstream historical valve opening recommended value with the target valve opening value defined in the candidate control rules.

[0132] S55. If the absolute value of the difference between the mainstream historical negative pressure recommendation value and the target negative pressure value is less than the first preset threshold, and the absolute value of the difference between the mainstream historical valve opening recommendation value and the target valve opening value is less than the second preset threshold, then it is determined that the experience and rules are consistent, and a control command is generated based on the mainstream historical negative pressure recommendation value and the mainstream historical valve opening recommendation value.

[0133] S56. If the consistency condition in step S55 is not met, it is determined that there is a discrepancy between experience and rules. The safety boundary constraint is used as a hard constraint. A weighted calculation is performed between the mainstream historical negative pressure recommendation value and the target negative pressure value, and a weighted calculation is performed between the mainstream historical valve opening recommendation value and the target valve opening value. The calculation results are used as the final target negative pressure value and target valve opening value, respectively, and control commands are generated based on these.

[0134] In this embodiment, the rule-based intelligent agent serves as the core decision-making hub between "perception" and "execution." It receives reference control groups from the historical data repository, candidate control rules from the multi-dimensional rule base, and real-time operating status parameters of the current smoke removal equipment. First, it performs a safety boundary check, comparing the current negative pressure, flow rate, and remaining filter lifespan with the upper limit of negative pressure, lower limit of flow rate, and lower limit of filter lifespan defined in the candidate control rules. If any parameter exceeds the safety boundary, a control command to maintain the status quo with an alarm indicator is directly generated. If the safety check passes, it further compares the consistency of mainstream historical control suggestions in the reference control group with the target actions in the candidate control rules. When the deviation is less than a preset threshold, historical experience parameters are prioritized to generate the control command. When there is a discrepancy, the safety boundary is used as a hard constraint to weightedly fuse the two, thereby outputting a standardized control command containing fields such as target negative pressure, valve opening, filtration mode, and operating mode. The decision basis and execution results are then written back to the historical data repository for subsequent optimization. Through this safety-first, experience-and-rule-coordinated decision-making mechanism, the rule-based intelligent agent achieves a complete mapping from multi-source inputs to reliable control commands.

[0135] It should be noted that the input to the rule-based intelligent agent includes at least {surgery type, step label, ...} , , Electrosurgical power Continuous working time of electrosurgical unit Current negative pressure value Current traffic value Current remaining lifespan of the filter element The set of case parameters and the set of candidate rules are defined as follows: the "case parameter set" is the Top-K reference control group output by the historical record database, and the "candidate rule set" is the set of candidate rules output by the rule database.

[0136] The rule-based intelligent agent only outputs control results in JSON format. The output fields must include at least {"Target Negative Pressure", "Valve Opening", "Filtering Mode", "Operating Mode", "Pulse Interval", "Re-identification Cycle", "Warning Level", "Warning Reason"}. The target negative pressure is in mmHg, the valve opening is in % (percentage), and the pulse interval and re-identification cycle are in seconds. For example, in a medium-to-high concentration smoke scenario, the output could be {"Target Negative Pressure":130,"Valve Opening":62,"Filtering Mode":"High Efficiency","Operating Mode":"Continuous","Pulse Interval":0,"Re-identification Cycle":0.5,"Warning Level":"None","Warning Reason":""}.

[0137] The preferred decision-making order for the rule-based agent is as follows: first, check the security boundary, then compare whether the reference control group given by the historical record database is consistent with the main rule item; if consistent, select the set of parameters with the shortest recovery time and lowest security risk within the interval; if inconsistent, use the security rule as the superior constraint to truncate and correct the reference control group, and generate the final control command.

[0138] Specifically, each rule must include at least {procedure label, step label, density range, occlusion rate range, change rate range, instrument power range, equipment status constraint, target action, priority, and safety boundary}. Among them, the target action must include at least the target negative pressure, valve opening, filtration mode, operating mode, and re-identification cycle; the safety boundary must include at least the upper limit of negative pressure, the lower limit of flow rate, and the lower limit of filter life.

[0139] In one embodiment, a rule entry can be represented as: ={Surgical procedure tag="Laparoscopic myomectomy", Procedure tag="Molecule removal and hemostasis stage", Target negative pressure = 118 mmHg, valve opening = 57%, filtration mode = "enhanced", operating mode = "pulse", priority = 2, maximum negative pressure value Minimum flow rate Minimum remaining filter life }

[0140] First, a preliminary screening is conducted based on the surgical type and procedure labels, eliminating rules irrelevant to the current surgical procedure; second, the current smoke density value is considered. Field of view occlusion rate Smoke screen variation rate Instrument power The reference control group output from the equipment status and historical records database is used for fine screening; finally, the matching score is calculated for the remaining candidate rules and the rule with the highest score is selected as the current rule main item.

[0141] The preferred method for rule-based scoring is:

[0142] ;

[0143] in This indicates the degree of matching after the current value falls within the rule range. Indicates the power matching degree of the equipment. Indicates the degree to which the equipment status meets requirements. This indicates the degree of consistency between the rule and the historical record database output reference control group, with each value ranging from 0 to 1.

[0144] When multiple rule scores are close, conflict resolution follows the order of "safety rules take precedence over control efficiency rules, equipment protection rules take precedence over conventional optimization rules, and technique-specific rules take precedence over general rules." For example, when the current negative pressure is close to the upper limit or the remaining filter life is below the lower limit, even if the smoke concentration continues to rise, degraded control or alarm rules will be implemented first, rather than continuing to increase the negative pressure.

[0145] The rule base matching results should at least output the target negative pressure, valve opening degree, filtering mode, operating mode, pulse interval, re-identification cycle, and safety warning level, and send these fields to the rule base agent in the form of a structured message. This clearly answers the questions of "what are the decision rules" and "what control commands should be generated."

[0146] Statistical analysis is performed on historical records accumulated under the same surgical procedures and steps. If the average clarity recovery time corresponding to a certain rule in the last 50 surgeries is consistently greater than 1.5 seconds and the doctor adoption rate is higher than 85%, the target negative pressure is increased by 5 mmHg or the valve opening is increased by 3%. If the proportion of the same rule triggering overpressure or low flow alarms exceeds 3%, the corresponding upper limit is lowered and the priority of the rule is reduced.

[0147] S6. Send the control command to the embedded controller to drive the negative pressure pump, proportional valve and filter components to perform smoke removal operation, and write back the scene characteristics, control command and execution effect of this control as a new structured history record to the history record library.

[0148] Reference Figure 2 The present invention proposes an adaptive control system for laparoscopic surgery smoke, applied to the adaptive control method for laparoscopic surgery smoke as described in any of the above claims, comprising:

[0149] The video and status acquisition unit is used to continuously and in real-time acquire laparoscopic video streams, energy device operating parameters, and smoke removal equipment operating status parameters.

[0150] The spatiotemporal dual-branch smoke recognition module is used to process laparoscopic video streams and output smoke state quantification data;

[0151] The historical record library module is used to store structured historical records and retrieve and output reference control groups based on the current surgical type, current surgical step, and smoke status.

[0152] The multi-dimensional rule base module is used to store multi-dimensional rule entries and match and output candidate control rules based on the current surgical type, current surgical step, smoke state quantification data, energy device operating parameters and smoke removal equipment operating status parameters.

[0153] The rule-based intelligent agent decision-making module is used to generate standardized control instructions based on the reference control group and candidate control rules;

[0154] The execution control module is used to receive control commands and drive the negative pressure pump, proportional valve and filter components to perform smoke removal operations, and feed the execution results back to the historical record library module.

[0155] Example 1:

[0156] like Figure 3 As shown, this embodiment takes the "tumor dissection and hemostasis stage" in laparoscopic myomectomy as an example to illustrate in detail the specific implementation process of the laparoscopic surgery smoke adaptive control method and system of the present invention.

[0157] Before the surgery begins, the operator selects the "Laparoscopic Myomectomy" control template in the system interface and loads the rule group corresponding to the "tumor removal and hemostasis stage". The system automatically connects and self-tests the laparoscopic video link, the energy device communication link, and the smoke removal equipment sensor link.

[0158] Once the surgery begins, the video and status acquisition unit starts working, continuously acquiring the following data in real time:

[0159] Video stream data: Real-time surgical video frames are continuously acquired through the laparoscopic main video stream interface at a sampling frequency of 25 frames / second, with each frame image resolution normalized to 512×512 pixels.

[0160] Operating parameters of the energy device: The electrosurgical power and continuous working time of the electrosurgical unit are synchronously collected at a sampling frequency of 10Hz through the communication interface of the high-frequency electrosurgical unit. At a certain control time t in this embodiment, the collected data are: electrosurgical power is 46W and continuous working time is 5.8s.

[0161] Operating parameters of the smoke removal equipment: The current negative pressure value, current flow rate value, and remaining filter life value are synchronously collected by the sensor integrated in the negative pressure suction system at a sampling frequency of 10Hz. At the same control time t in this embodiment, the collected data are: current negative pressure value is 86mmHg, current flow rate value is 29L / min, and remaining filter life value is 53%.

[0162] All the above data is aligned with a uniform timestamp and encapsulated into a data packet with a sliding step of 0.2 seconds for use by subsequent processing modules.

[0163] The acquired video frame at the current moment and the preceding consecutive frames are input into a pre-trained spatiotemporal bi-branch smoke state network. At a certain control time t, the spatiotemporal bi-branch smoke state network outputs a smoke density value of 0.52, a field-of-view occlusion rate of 34%, and a smoke change rate of 0.027.

[0164] The history database module receives the current surgery type (laparoscopic myomectomy), surgical steps (tumor removal and hemostasis stage), and output smoke status quantification data.

[0165] In this embodiment, a large number of structured historical records are already stored in the historical record database. The historical record database was searched according to the criteria of "laparoscopic myomectomy + tumor removal and hemostasis stage + moderate smoke screen obscuring the view," resulting in three sets of similar scenario reference control groups: the first group was 112 mmHg, 52% valve opening, medium filtration, with a clarity recovery time of 1.8 seconds; the second group was 118 mmHg, 57% valve opening, enhanced filtration, with a clarity recovery time of 1.4 seconds; and the third group was 126 mmHg, 61% valve opening, high-efficiency filtration, with a clarity recovery time of 1.3 seconds but with a low-flow edge risk.

[0166] The rule base matches "0.40≤" based on the current data. <0.58, 18%≤ <38%, 0.01≤ <0.04, 38W≤ The specific rules for the procedure “≤52W” yield the following main rules: target negative pressure 118mmHg, valve opening 57%, filtration mode enhanced, operating mode pulse, and safety boundary [missing information]. =150mmHg =26L / min =8%.

[0167] After comparing historical search results with the main rule item, the rule-based agent found that the second reference control group was consistent with the main rule item, and therefore generated a JSON control instruction: {"Target negative pressure":118,"Valve opening":57,"Filtering mode":"Enhanced","Operating mode":"Pulse","Pulse interval":0.8,"Re-identification cycle":0.4,"Warning level":"None","Warning reason":""}.

[0168] After receiving the above control commands, the STM32H743VIT6 controller increases the duty cycle of the negative pressure pump through PWM1, adjusts the opening of the proportional valve to 57% through PWM2, and switches the filter component to enhanced mode. The control cycle is 50ms, and the execution quantity is continuously corrected based on the feedback from the pressure sensor and flow sensor to stabilize the actual negative pressure between 116mmHg and 120mmHg.

[0169] The system re-identifies the control command 1.4 seconds after it is issued. =0.24、 =14% =0.018, surgical field clarity recovered to 97%, no overpressure or low flow alarms were triggered, indicating that the control action was effective. If it is still detected at this time... Higher than 32% or If the pressure continues to rise, the rule-based agent will continue to increase the negative pressure or shorten the pulse interval.

[0170] After the surgery, the record is written back to the history database in a structured format, saving at least the surgical procedure label, step label, identification result, control instructions, execution result, and whether the doctor adopted the instructions. If, after accumulating 50 similar cases, the 118mmHg setting consistently has the shortest average recovery time and the lowest safety alarm rate, it is upgraded to the priority reference control group for that scenario, and the priority of the corresponding main rule item is increased accordingly.

[0171] As can be seen from the above embodiments, the method of this application can directly obtain control commands from real-time video and device data, and realize an interpretable, executable, and traceable adaptive control process for laparoscopic smoke through a historical record library, a rule-based intelligent agent, and an STM32 controller.

[0172] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for adaptive smoke control in laparoscopic surgery, characterized in that, Includes the following steps: S1. Acquire real-time video stream of laparoscopic surgery, and simultaneously acquire the operating parameters of the energy device and the operating status parameters of the smoke removal device. The operating parameters of the energy device include at least the electrosurgical power and the continuous working time of the electrosurgical unit; the operating status parameters of the smoke removal device include at least the current negative pressure value, the current flow rate value, and the remaining lifespan of the filter element. S2. Input the real-time video stream into a pre-trained spatiotemporal dual-branch smoke state network for processing to obtain the smoke state quantization data at the current moment. The smoke state quantization data includes at least the smoke density value, field of view occlusion rate, and smoke change rate. S3. Based on the current surgical type, current surgical step, and smoke state quantification data, retrieve the K historical control records with the highest similarity to the current scene from the historical record database to form a reference control group; S4. Based on the current surgical type, current surgical step, smoke state quantification data, operating parameters of energy devices, and operating status parameters of smoke removal equipment, at least one candidate control rule is matched from the multi-dimensional rule base. S5. The rule base agent makes decisions based on the reference control group and candidate control rules, and generates control instructions that include the target negative pressure value, valve opening value, filter mode identifier and operating mode identifier. S6. Send the control command to the embedded controller to drive the negative pressure pump, proportional valve and filter components to perform smoke removal operation, and write back the scene characteristics, control command and execution effect of this control as a new structured history record to the history record library.

2. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, The spatiotemporal dual-branch smoke state network includes parallel spatial branches, temporal branches, gating fusion units, and control output units. The input of the spatial branch is used to receive a single frame image of the laparoscopic video at the current moment, and outputs the spatial morphological feature vector of the smoke in the current frame. The input of the time branch is used to receive a laparoscopic video sequence containing the current frame and several consecutive previous frames, and outputs a feature vector of dynamic changes in smoke. The first input of the gated fusion unit is connected to the output of the spatial branch, and the second input is connected to the output of the temporal branch. It is used to adaptively weight and fuse the spatial morphology feature vector of smoke and the dynamic change feature vector of smoke to output a spatiotemporal fusion feature vector. The input of the control output unit is connected to the output of the gated fusion unit, and is used to map the spatiotemporal fusion feature vector into smoke state quantization data.

3. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, Step S3 specifically includes: The current surgical type, current surgical step, smoke density value, field of view occlusion rate, and smoke change rate in the smoke state quantification data are combined to form the current scene feature vector; Calculate the weighted similarity distance between the current scene feature vector and the historical scene feature vector stored in each structured historical record in the history database; wherein the calculation of the weighted similarity distance takes into account at least the differences in smoke density values, differences in field of view occlusion rates, differences in smoke change rates, surgical type matching degree, and surgical step matching degree. Based on the weighted similarity distance in ascending order, the first K structured historical records are selected from the historical record database; From the K selected structured historical records, extract the K sets of historical control parameter vectors associated with them. The historical control parameter vectors include at least historical negative pressure values, historical valve opening values, and historical filtering mode identifiers. The K sets of historical control parameter vectors are then output as a reference control group.

4. The adaptive smoke control method for laparoscopic surgery according to claim 3, characterized in that, The weighted similarity distance Calculated using the following formula: ； in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; The rate of change of smoke in the current scene; The historical smoke density value corresponding to the i-th structured historical record in the historical record database; The historical field of view occlusion rate corresponding to the i-th structured historical record in the historical record library; Let be the rate of change of the historical smoke screen corresponding to the i-th structured historical record in the historical record database; Identify the current surgical type; Identify historical surgical types; This indicates the current surgical step. This serves as an identifier for historical surgical steps; the function M(a,b) returns 0 if a and b are the same, and 1 if they are different. The preset weighting coefficients, and .

5. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, Step S4 specifically includes: S41. Based on the current surgical type and the current surgical procedure, all rule entries that match the applicable surgical type and applicable surgical procedure are initially selected from the multi-dimensional rule base to form a primary candidate rule set. S42. For each rule entry in the primary candidate rule set, determine in sequence whether the current smoke density value falls within its specified smoke density value range, whether the current field of view occlusion rate falls within its specified field of view occlusion rate range, whether the current smoke change rate falls within its specified smoke change rate range, and whether the current operating parameters of the energy device meet its specified device power conditions. S43. Include the rule entries that satisfy all the judgment conditions in step S42 into the secondary candidate rule set; S44. Calculate the matching score between each rule entry in the secondary candidate rule set and the current scene; the matching score is calculated based on at least the degree of closeness between the current smoke density value and the center value of the rule's smoke density value interval, the degree of closeness between the current field of view occlusion rate and the center value of the rule's field of view occlusion rate interval, and the degree of conformity between the current smoke removal equipment's operating status parameters and the safety boundary value defined by the rule. S45. Based on the matching scores from high to low, select at least one rule entry from the secondary candidate rule set as the candidate control rule output.

6. The adaptive smoke control method for laparoscopic surgery according to claim 5, characterized in that, The matching score Calculated using the following formula: ； in, This represents the smoke density value for the current scene. The current scene's field of view occlusion rate; Let be the center value of the smoke density value interval for the j-th rule entry; Let be the center value of the field of view occlusion rate interval for the j-th rule entry; This is the preset normalized range for smoke density values; The preset normalized range for field of view occlusion rate; This is the current negative pressure value; This represents the current traffic volume. This represents the remaining lifespan of the filter element. The upper limit of negative pressure safety is defined for the j-th rule entry; The lower limit value for traffic security defined for the j-th rule entry; The filter cartridge life safety lower limit is defined for the j-th rule entry; the function S(·) is used to evaluate the degree of compliance between the current equipment state and the safety boundary. Its output value range is [0, 1]. It returns 1 when the equipment state fully meets the safety boundary requirements and returns 0 when the equipment state deviates significantly from the safety boundary. The higher the degree of compliance, the larger the return value. The preset weighting coefficients, .

7. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, Step S5 specifically includes: S51. Obtain at least one safety boundary constraint defined in the candidate control rules, wherein the safety boundary constraint includes at least a negative pressure safety upper limit value, a flow rate safety lower limit value, and a filter cartridge life safety lower limit value; S52. Compare the current negative pressure value, current flow rate value, and remaining filter life value in the current operating status parameters of the smoke removal equipment with the upper limit of negative pressure safety, the lower limit of flow rate safety, and the lower limit of filter life safety, respectively. If any current value violates its corresponding safety boundary constraint, generate a control command containing an alarm flag and maintaining the current operating parameters, and end this step; otherwise, execute step S53. S53. From the K sets of historical control parameter vectors contained in the reference control group, extract the historical negative pressure value with the highest frequency as the mainstream historical negative pressure recommendation value, and extract the historical valve opening value with the highest frequency as the mainstream historical valve opening recommendation value. S54. Compare the mainstream historical negative pressure recommended value with the target negative pressure value defined in the candidate control rules, and compare the mainstream historical valve opening recommended value with the target valve opening value defined in the candidate control rules. S55. If the absolute value of the difference between the mainstream historical negative pressure recommendation value and the target negative pressure value is less than the first preset threshold, and the absolute value of the difference between the mainstream historical valve opening recommendation value and the target valve opening value is less than the second preset threshold, then it is determined that the experience and rules are consistent, and a control command is generated based on the mainstream historical negative pressure recommendation value and the mainstream historical valve opening recommendation value. S56. If the consistency condition in step S55 is not met, it is determined that there is a discrepancy between experience and rules. The safety boundary constraint is used as a hard constraint. A weighted calculation is performed between the mainstream historical negative pressure recommendation value and the target negative pressure value, and a weighted calculation is performed between the mainstream historical valve opening recommendation value and the target valve opening value. The calculation results are used as the final target negative pressure value and target valve opening value, respectively, and control commands are generated based on these.

8. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, The construction and process of the historical record database specifically includes: Complete data from multiple typical laparoscopic surgeries were collected. The complete data included surgical videos, energy device parameters, smoke removal equipment control parameters and status parameters. Experts scored and labeled the smoke removal effect of each key control based on the postoperative video playback. Each control operation, along with its corresponding scenario data, control parameters, and performance score, is stored as an initial structured historical record in the historical record library. After each execution of step S6, the newly generated structured historical records are stored in the historical record library; the historical records under the same surgical type and surgical step label are periodically statistically analyzed, the average effect score of each combination of control parameters is calculated, and the similarity calculation weight of the historical records corresponding to the parameter combination with the highest average score is increased.

9. The adaptive smoke control method for laparoscopic surgery according to claim 1, characterized in that, The construction process of the multi-dimensional rule base specifically includes: Data mining is performed on the structured historical records accumulated in the historical record database, and the records are grouped based on surgical type and surgical procedure; For each group, a clustering algorithm is used to analyze the smoke state quantification data to form several typical smoke state intervals; Each smoke state interval is associated with the historical control parameter with the highest effect score in that interval within the same group, and the statistical safety boundary of the equipment operating state parameters in that group is added to form an initial multi-dimensional rule entry. Set a rule confidence index. When a rule is matched and executed, if the effect score of the new historical records it generates is consistently higher or lower than the historical average effect score of the rule, then fine-tuning of the target control parameters or applicable range of the rule will be triggered. When a new surgical type or procedure appears and sufficient data is accumulated, a new rule generation process will be automatically triggered.

10. A laparoscopic surgery smoke adaptive control system, applied to the laparoscopic surgery smoke adaptive control method as described in any one of claims 1-9, characterized in that, include: The video and status acquisition unit is used to continuously and in real-time acquire laparoscopic video streams, energy device operating parameters, and smoke removal equipment operating status parameters. The spatiotemporal dual-branch smoke recognition module is used to process laparoscopic video streams and output smoke state quantification data; The historical record library module is used to store structured historical records and retrieve and output reference control groups based on the current surgical type, current surgical step, and smoke status. The multi-dimensional rule base module is used to store multi-dimensional rule entries and match and output candidate control rules based on the current surgical type, current surgical step, smoke state quantification data, energy device operating parameters and smoke removal equipment operating status parameters. The rule-based intelligent agent decision-making module is used to generate standardized control instructions based on the reference control group and candidate control rules; The execution control module is used to receive control commands and drive the negative pressure pump, proportional valve and filter components to perform smoke removal operations, and feed the execution results back to the historical record library module.

Images