A synchronous water exchange artificial intelligence control system for digestive endoscopy surgery and a use method thereof

Abstract

A kind of synchronous water exchange artificial intelligence control system of digestive endoscopic surgery, it is characterized in that, including visual field image real-time acquisition module, multi-dimensional quality quantitative evaluation module, adaptive parameter decision module, emergency mode switching module, man-machine interaction module, data analysis module, the multi-dimensional visual field quality evaluation system including field of vision definition, bleeding degree, bubble proportion, mucosa visibility is constructed by computer vision algorithm, the characteristics of endoscopic picture are extracted and quantitatively analyzed in real time;According to the definition threshold of pre-set, automatically trigger water injection and suction operation, realize the synchronous dynamic exchange of clean water and turbid water;While built-in operation stage recognition logic can be matched with water injection flow and suction negative pressure parameters for different operation stages such as mucosa marker, submucosal injection, lesion incision, lesion stripping, wound hemostasis, and set the emergency water exchange mode triggered by bleeding event, when detecting intraoperative bleeding, automatically increase water exchange rate to quickly flush blood, the surgeon automatically switches AI automatic control mode and manual mode by stepping on foot switch, independently decide water injection opportunity.

Description

Technical Field

[0001] This invention relates to the field of intelligent surgical assistance, and in particular to an artificial intelligence control system and method for synchronous water exchange during digestive endoscopy. Background Technology

[0002] With the continuous development and popularization of digestive endoscopy and minimally invasive surgery, endoscopic surgery, represented by endoscopic submucosal dissection and peroral endoscopic myotomy, has become the preferred treatment for early gastrointestinal tumors and submucosal lesions worldwide. The core operational logic of these surgeries involves a series of precise operations within the digestive tract lumen, using a high-frequency electrosurgical unit or other energy instruments to mark, inject, cut, dissect, and hemostatic the lesion. Compared to traditional air-medium surgery, liquid-medium surgery leverages the buoyancy and refractive properties of liquids to achieve clearer anatomical layer identification. Furthermore, the higher specific heat capacity of liquids effectively reduces the risk of thermal damage during electrosurgical resection. However, the liquid environment also presents a critical challenge throughout the entire procedure: maintaining a clear surgical field of view during uninterrupted surgical manipulation. The clarity of the surgical field directly affects the surgeon's judgment of the submucosal vessel course, tissue layer boundaries, and lesion infiltration depth. Any degree of obstruction or blurring of the field of view can lead to serious complications such as mis-cutting, perforation, or bleeding, increasing surgical risks and prolonging operation time.

[0003] In early endoscopic surgery, surgeons relied on their subjective judgment of the endoscopic image. When the field of vision became blurred, they would step on a foot pedal to trigger water injection to flush away turbid blood and fluid. Simultaneously, they would activate a suction device to remove the turbid fluid containing blood, tissue debris, and air bubbles, thus replacing the turbid fluid with clean water. This method of water exchange, entirely dependent on manual judgment and operation, has several inherent drawbacks. First, when performing delicate submucosal dissections, surgeons need to operate both the endoscope and electrosurgical instruments simultaneously. Frequent interruptions to step on the foot pedal to control water exchange not only distract the surgeon but also disrupt the surgical rhythm and prolong the operation time. Second, manually judging the clarity of the field of vision depends entirely on the surgeon's subjective experience, lacking objective and quantifiable standards for assessing the quality of the field of vision. This can lead to water injection being too early or too late. Injecting water too early can cause unnecessary fluid disturbances that interfere with delicate operations, while injecting water too late allows blood or air bubbles to form large areas of obstruction, requiring a large amount of water to restore the field of vision, resulting in ineffective waste of fluid and fluctuations in luminal pressure. Furthermore, in emergency situations such as sudden bleeding, the surgeon may be unable to respond promptly to the deterioration of the field of vision due to focusing on hemostasis, causing the blood to spread rapidly and cover the entire surgical area, forcing the surgery to be interrupted for extensive flushing and suction, which seriously affects the safety and efficiency of the surgery.

[0004] With the continuous improvement of computer vision technology and embedded image processing hardware capabilities, some automated water exchange control systems based on image analysis have emerged in recent years. These systems typically acquire endoscopic video images in real time, using methods such as image blur assessment, brightness analysis, or simple color threshold judgment to quantify the clarity of the current field of view. They automatically initiate water injection and suction operations when the field of view quality falls below a preset threshold, thus reducing the operator's workload to some extent. However, existing automated water exchange systems still have significant shortcomings in terms of the dimensions, accuracy, and reliability of field of view quality assessment. Most systems use only a single image clarity index, such as blur assessment based on grayscale gradients or judgment logic based on overall brightness. Such a single index cannot effectively distinguish the specific causes of decreased field of view quality. In underwater endoscopic surgery, field of view obstruction may originate from diffuse red blood due to active bleeding, suspended tissue debris generated during electrocautery, the accumulation of numerous microbubbles formed by water injection impact, or physical obstruction caused by mucosal flap folding. Different causes of obstruction correspond to different water exchange strategy requirements, and a single index confuses these factors, resulting in a lack of targeted water exchange actions.

[0005] In conclusion, there is an urgent clinical need for a new generation of intelligent control systems that can comprehensively, in real-time, and interpretably quantify and assess the multidimensional quality of the surgical field, dynamically and adaptively adjust water exchange parameters according to the surgical stage, and simultaneously combine the convenience of AI automatic control with the flexibility of manual operation, so as to fundamentally solve the aforementioned technical problems that have long plagued clinical practice. Summary of the Invention

[0006] The purpose of this invention is to provide an artificial intelligence control system and method for synchronous water exchange during digestive endoscopy, aiming to solve the problems mentioned in the background above.

[0007] To achieve the above objectives, this invention proposes an artificial intelligence control system for synchronous water exchange during digestive endoscopic surgery. The system comprises a real-time visual field image acquisition module, a multi-dimensional quality quantification assessment module, an adaptive parameter decision module, an emergency mode switching module, a human-computer interaction module, and a data analysis module. The real-time visual field image acquisition module acquires continuous video frames of the surgical area via endoscopy. The multi-dimensional quality quantification assessment module proposes a multi-dimensional quality assessment algorithm to quantify the clarity of the current surgical field. The adaptive parameter decision module proposes a parameter adaptive synchronous control algorithm to adaptively generate reference values ​​for water injection flow rate and suction negative pressure. The emergency mode switching module automatically switches to an emergency water exchange mode based on the degree of bleeding and a preset bleeding threshold, rapidly increasing the water exchange rate to flush out accumulated blood. The human-computer interaction module switches between manual and intelligent control modes via a foot switch. The data analysis module stores data from the entire surgical process.

[0008] Furthermore, the real-time visual field image acquisition module continuously acquires color video frames of the surgical area through the endoscopic camera and transmits each frame to the computing device in real time. The color video frame contains pixel value information of the red, green, and blue color channels.

[0009] Furthermore, the multidimensional quality assessment algorithm quantifies the clarity of the surgical field into local clarity values, bleeding severity scores, and the severity of air bubble obstruction. It analyzes and judges the clarity of the surgical field and automatically triggers water injection and suction operations based on the surgical field clarity index and preset clarity threshold, realizing the synchronous dynamic exchange of clean water and turbid water.

[0010] Further, the quantification of surgical field clarity includes the following steps: S1. Within the region of interest, extract the horizontal and vertical grayscale change values ​​of each pixel. Based on the standard deviation of the pixel values ​​of the red channel within the region of interest, obtain the local sharpness value by summing the gradients. S2. Calculate the pixel mean of each pixel channel, the area ratio of red pixels exceeding the standard, and the variance of red saturation to obtain the bleeding degree score. S3. Statistically calculate the proportion of specular reflection bright pixels within the field of view, and simultaneously calculate the continuous length of the connected edges of the bubbles. Add the proportion of bright pixels to the proportion of edge continuity to obtain the bubble occlusion severity index. The more bubbles there are and the more severe the connected areas are, the greater the bubble occlusion severity index will be. S4. Local visual clarity index, bleeding degree score, and bubble occlusion severity are used as three visibility influencing factors to calculate the comprehensive visibility index, which reflects the actual operable visual field quality. The deterioration of a single factor will lead to a decrease in the comprehensive visibility index. First, for all pixels within the area of ​​focus, the local field-of-view sharpness index is calculated based on the grayscale differences between adjacent pixels in the horizontal and vertical directions, and then using the standard deviation of pixel values ​​in the red channel. , ,in, This represents the pixel value standard deviation in the red channel. Indicates the area of ​​focus in the field of vision. This represents the horizontal grayscale gradient of a pixel. This represents the vertical grayscale gradient of a pixel. The standard deviation of the red channel is introduced into the denominator. When blood diffusion causes increased fluctuations in the red component, the sharpness index is automatically reduced. The mean values ​​of pixels within the area of ​​focus are calculated for the red, green, and blue channels. Then, the area of ​​pixels whose red component percentage exceeds a preset red threshold is used to calculate the degree of bleeding. , ,in, This indicates the average value of green channels within the area of ​​focus. This indicates the average value of the blue channel within the area of ​​visual focus. This indicates the average value of the red channel within the area of ​​visual focus. , These represent empirical weights, This indicates the proportion of pixel area where the red component exceeds a preset red threshold to the total area. The variance of red saturation is represented. Next, regarding the impact of bubbles within the region of interest, the proportion of specularly reflected bright pixels within this region is statistically analyzed. Then, by comparing the continuous length of the bubble's connected edges with the maximum perimeter of the region, the proportion of bright pixel ratios is directly added to the edge continuity ratio to calculate the bubble occlusion severity. , ,in, This indicates the number of pixels whose brightness exceeds the specular reflection threshold. Indicates the total number of pixels. This represents the total length of the continuous pixel segments at the edge of the bubble. Representing the maximum perimeter of the region, finally, based on the local field-of-view sharpness index... Bleeding severity score Severity of bubble occlusion Define the visibility comprehensive index , By comparing the comprehensive visibility index with a preset clarity threshold, the system automatically triggers water injection and suction operations to achieve synchronous dynamic exchange of clean and turbid water.

[0011] Furthermore, the parameter adaptive synchronous control algorithm matches feature scores based on experience using feature weights for each surgical stage, determines the current surgical stage based on the stage probability at consecutive moments, and obtains a reference value for suction negative pressure by using a linear gain term composed of the current bleeding degree score, lesion dissection stage probability, and visibility comprehensive index, and by using the gain of bubble occlusion severity, thus achieving parameter adaptive adjustment.

[0012] Furthermore, the parameter adaptive adjustment includes the following steps: S1. Extract five features from the image within the current field of view: presence or absence of instruments, degree of mucosal protrusion, rate of decay of red turbidity after water injection, current bleeding score, and texture complexity. Multiply these features by the fixed weights preset for each surgical stage to obtain the real-time matching score for each stage. S2. Define a smoothing factor, use the stage probability of the previous time step, add the real-time matching score of the current time step, and normalize it to obtain the smoothed stage probability distribution of the current time step. S3. Based on the baseline injection flow rate, a real-time injection flow rate reference value is generated by using the current bleeding severity score, the probability of lesion dissection stage, and the comprehensive visibility index as linear gain terms. The higher the probability of lesion dissection stage, the more severe the bleeding, and the worse the field of vision, the greater the flow rate. S4. Finally, based on the basic suction negative pressure, add the product of the negative pressure adjustment dynamic range and the normalized value of the injection flow rate, and then obtain the suction negative pressure reference value through the bubble blockage severity gain, so as to realize the synchronization of suction negative pressure and injection flow rate, and further enhance suction when the bubbles are severe to prevent bubble accumulation. First, based on experience, feature score matching was performed on the feature sets of five features: presence or absence of extraction instruments, degree of mucosal elevation, rate of decay of red turbidity after injection, current bleeding score, and texture complexity. Based on the feature matching scores under various conditions, the real-time matching score during the surgical phase was calculated. , ,in, Indicates the first The first surgical stage for the first The weight coefficients of each feature Indicates the first Features The feature matching score at time step 1 is used. Then, using the stage probability from the previous time step, the current stage probability is calculated using the matching score at the current time step and the smoothing factor. , ,in, Indicates the first Each surgical stage Probability at any given time stage Represents the smoothing factor. Indicates the first Each surgical stage Probability at any given time stage Indicates the first Each surgical stage Real-time matching and scoring. This represents the total number of stages. Based on the stage probability at the current moment, the current surgical stage is determined. Then, using the current bleeding severity score, the probability of lesion removal stage, and the comprehensive visibility index, a water injection flow rate gain term is constructed. When the patient is currently in the lesion removal stage, severe bleeding leads to poor visibility, which is addressed through the water injection flow rate gain term. With basic water injection flow rate Calculate the reference value of actual water injection flow rate , ,in, express Bleeding severity score at any given time express Probability of lesion removal at any given time. express Comprehensive visibility index at any time Indicates the gain coefficient for the degree of bleeding. This represents the gain coefficient during the surgical phase. This represents the visibility comprehensive index gain coefficient. Finally, based on the effect of bubble obstruction, a bubble obstruction severity gain is defined. The injection flow rate is then synchronously adjusted by regulating the suction negative pressure through the negative pressure adjustment dynamic range and the injection flow rate. ,in, express Constantly attract negative pressure reference values. This indicates that the base attracts negative pressure. This indicates the dynamic range of the suction negative pressure dynamic adjustment. Indicates the maximum water injection flow rate. express The severity of bubble occlusion at any given moment. This indicates the bubble blocking gain coefficient, which ensures that the suction negative pressure and the water injection flow rate increase synchronously, maintaining an inflow-outflow balance.

[0013] Furthermore, the emergency mode switching module has preset clarity thresholds and time thresholds. When the overall visibility index is lower than the clarity threshold and exceeds the preset time, it automatically switches to emergency water exchange mode, increasing the water flow reference value and suction negative pressure reference value to twice their respective preset maximum safety values ​​to quickly flush out accumulated blood. When the overall visibility index recovers to the preset clarity threshold, the emergency water exchange mode stops. Furthermore, the human-computer interaction module allows the surgeon to switch between manual mode and AI automatic control mode by triggering a foot switch. With a clear field of vision, the surgeon can autonomously decide when to inject water to provide underwater impact force, assisting in unfolding the partially cut lesion and completing a more precise dissection operation.

[0014] Furthermore, the data analysis module stores structured data streams, recording frame-by-frame the local visual field clarity, bleeding severity score, air bubble obstruction severity, visibility index, stage, measured water injection flow rate, and measured suction negative pressure for each frame of the image. It also automatically marks the start and end times and duration of manual control periods and emergency bleeding response events. Postoperatively, based on the structured data stream, it generates an operation quality report that includes a visual field quality change curve, surgical stage time distribution, frequency of manual intervention, and statistics on bleeding events, enabling quantitative analysis of surgical operation efficiency and surgeon skills training.

[0015] Furthermore, a method for using an AI-powered control system for synchronous water exchange during digestive endoscopic surgery includes three core stages: preoperative parameter calibration, intraoperative real-time closed-loop control, and postoperative operational data review. Preoperatively, the system adapts and calibrates the visual field assessment weights and surgical stage identification parameters based on individual patient differences. During the surgery, the system automatically performs multi-dimensional quantitative assessment of the visual field and inference of the surgical stage, generating and dynamically adjusting the injection flow rate and suction negative pressure in real time to achieve synchronous water exchange. Simultaneously, it monitors the foot pedal signal throughout the procedure, immediately switching to manual control mode when the pedal is pressed and automatically reverting to AI automatic control mode after release. Furthermore, when intraoperative bleeding is detected, the system forcibly increases the water exchange rate to quickly clear accumulated blood. Postoperatively, the system exports the entire operational data to generate a quality report, supporting surgical optimization and skills training. Beneficial effects

[0016] 1. The multidimensional quality assessment algorithm proposed in this invention calculates the horizontal and vertical grayscale gradients of pixels within the region of interest in the field of view. Based on the standard deviation of pixel values ​​in the red channel within the region of interest, the algorithm sums these gradients to obtain a local sharpness value. It then calculates the pixel mean of each pixel channel, the area proportion of red-excess pixels, and the variance of red saturation to obtain a bleed score. The algorithm also statistically analyzes the proportion of specularly reflected bright pixels within the region of interest and calculates the continuous length of the bubble's connected edges. The proportion of bright pixel highlights is directly added to the edge continuity proportion to obtain a bubble occlusion severity index; the more bubbles and the more severe the connected areas, the higher the bubble occlusion severity index. The local sharpness index, bleed score, and bubble occlusion severity are used as three... This invention proposes a multidimensional quality assessment algorithm that uses several visibility influencing factors to calculate a comprehensive visibility index, reflecting the actual operational visual field quality. Deterioration of any single factor will lead to a decrease in the comprehensive visibility index, overcoming the technical limitation of traditional single indicators that cannot distinguish the causes of visual field deterioration. The algorithm utilizes the standard deviation of the red channel as a sharpness penalty to suppress blood artifacts, constructs an adaptive basis for the bleeding score using the sum of the green and blue channel means as the denominator, achieving nonlinear sensitivity, and explicitly incorporates the spatial connectivity of bubbles into the occlusion assessment through edge continuity length, combining it with the bright spot ratio to form a unified index. Finally, it couples three independent physical dimensions in a product form to form a highly realistic and fully interpretable visual field quality criterion, significantly improving the accuracy and reliability of visual field quality assessment.

[0017] 2. The parameter adaptive synchronous control algorithm proposed in this invention extracts five features from the image within the current field of view: presence or absence of instruments, degree of mucosal bulging, rate of red turbidity decay after injection, current bleeding score, and texture complexity. These features are multiplied by preset fixed weights for each surgical stage to obtain the real-time matching score for each stage. A smoothing factor is defined, and the stage probability from the previous moment is added to the current real-time matching score and normalized to obtain the smoothed stage probability distribution for the current moment. Based on the baseline injection flow rate, a linear gain term is generated using the current bleeding score, lesion dissection stage probability, and visibility comprehensive index to generate a real-time injection flow rate reference value. The higher the lesion dissection stage probability, the more severe the bleeding, and the worse the field of view, the greater the flow rate. Finally, based on the baseline suction negative pressure, the dynamic range of negative pressure adjustment and the injection flow rate are added. The product of the normalized flow rate values, combined with the gain from the severity of bubble blockage, yields the reference value for suction negative pressure. This synchronizes the suction negative pressure with the injection flow rate, and provides additional suction when bubbles are severe to prevent bubble accumulation. The parameter adaptive synchronous control algorithm proposed in this invention uses linear weighted scores to distinguish between the bleeding stage and the static stage, constructing a normalized fractional form with memory smoothing to achieve time-stability stage probabilities. For the first time, the continuous surgical stage probabilities are directly used as continuous gain factors for flow rate regulation, eliminating hard switching jitter. Furthermore, the bubble blockage severity index is directly coupled to the suction negative pressure control formula, simultaneously solving the balance between inflow and outflow and bubble removal. Water exchange smoothly transitions with the surgical progress, allowing for advance adjustment before pressure fluctuations due to changes in the surgical field, ensuring the clarity of the surgical field and avoiding the water flow impact and tissue disturbance caused by traditional hard switching methods.

[0018] 3. By setting clarity and time thresholds, when active bleeding is detected and causes severe obstruction of the field of vision, the injection flow rate and suction negative pressure can be automatically increased to several times the maximum safe value to achieve rapid removal of blood accumulation. After the emergency mode is deactivated, the normal control parameters are restored by a smooth decay method, avoiding the risk of lumen pressure oscillation and mucosal adhesion caused by instantaneous parameter jumps, thus balancing rapid response and safe stability. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0020] Figure 1 This is a system schematic diagram of the present invention; Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] To achieve the above objectives, this invention proposes an artificial intelligence control system for synchronous water exchange during digestive endoscopic surgery. The system comprises a real-time visual field image acquisition module, a multi-dimensional quality quantification assessment module, an adaptive parameter decision module, an emergency mode switching module, a human-computer interaction module, and a data analysis module. The real-time visual field image acquisition module acquires continuous video frames of the surgical area via endoscopy. The multi-dimensional quality quantification assessment module proposes a multi-dimensional quality assessment algorithm to quantify the clarity of the current surgical field. The adaptive parameter decision module proposes a parameter adaptive synchronous control algorithm to adaptively generate reference values ​​for water injection flow rate and suction negative pressure. The emergency mode switching module automatically switches to an emergency water exchange mode based on the degree of bleeding and a preset bleeding threshold, rapidly increasing the water exchange rate to flush out accumulated blood. The human-computer interaction module switches between manual and intelligent control modes via a foot switch. The data analysis module stores data from the entire surgical process.

[0023] Specifically, the real-time visual field image acquisition module continuously acquires color video frames of the surgical area through an endoscope camera and transmits each frame to a computing device in real time. The color video frame contains pixel value information of the red, green, and blue color channels.

[0024] Specifically, the multidimensional quality assessment algorithm quantifies surgical field clarity into local sharpness values, bleeding severity scores, and the degree of air bubble obstruction. It analyzes and judges the clarity of the surgical field, and based on the surgical field clarity index and a preset sharpness threshold, automatically triggers water injection and suction operations to achieve synchronous dynamic exchange of clean and turbid water. First, for all pixels within the area of ​​focus, the algorithm calculates the local visual field clarity index based on the grayscale differences between adjacent pixels in the horizontal and vertical directions, and then uses the standard deviation of pixel values ​​in the red channel. , ,in, This represents the pixel value standard deviation in the red channel. Indicates the area of ​​focus in the field of vision. This represents the horizontal grayscale gradient of a pixel. This represents the vertical grayscale gradient of a pixel. The standard deviation of the red channel is introduced into the denominator. When blood diffusion causes increased fluctuations in the red component, the sharpness index is automatically reduced. The mean values ​​of pixels within the area of ​​focus are calculated for the red, green, and blue channels. Then, the area of ​​pixels whose red component percentage exceeds a preset red threshold is used to calculate the degree of bleeding. , ,in, This indicates the average value of green channels within the area of ​​focus. This indicates the average value of the blue channel within the area of ​​visual focus. This indicates the average value of the red channel within the area of ​​visual focus. , These represent empirical weights, This indicates the proportion of pixel area where the red component exceeds a preset red threshold to the total area. The variance of red saturation is represented. Next, regarding the impact of bubbles within the region of interest, the proportion of specularly reflected bright pixels within this region is statistically analyzed. Then, by comparing the continuous length of the bubble's connected edges with the maximum perimeter of the region, the proportion of bright pixel ratios is directly added to the edge continuity ratio to calculate the bubble occlusion severity. , ,in, This indicates the number of pixels whose brightness exceeds the specular reflection threshold. Indicates the total number of pixels. This represents the total length of the continuous pixel segments at the edge of the bubble. Representing the maximum perimeter of the region, finally, based on the local field-of-view sharpness index... Bleeding severity score Severity of bubble occlusion Define the visibility comprehensive index , By comparing the comprehensive visibility index with a preset clarity threshold, the system automatically triggers water injection and suction operations to achieve synchronous dynamic exchange of clean and turbid water.

[0025] Specifically, the parameter adaptive synchronous control algorithm performs feature score matching based on experience using feature weights for each surgical stage. Based on the stage probability at consecutive time points, it determines the current surgical stage. A linear gain term is constructed using the current bleeding severity score, the probability of lesion removal stage, and the visibility index. A reference value for suction negative pressure is obtained through a gain based on the severity of bubble occlusion, thus achieving adaptive parameter adjustment. First, based on experience, feature score matching is performed on the feature set of five features: presence or absence of extraction instruments, degree of mucosal bulging, rate of red turbidity decay after water injection, current bleeding score, and texture complexity. Based on the feature matching scores under various conditions, the real-time matching score for the surgical stage is calculated. , ,in, Indicates the first The first surgical stage for the first The weight coefficients of each feature Indicates the first Features The feature matching score at time step 1 is used. Then, using the stage probability from the previous time step, the current stage probability is calculated using the matching score at the current time step and the smoothing factor. , ,in, Indicates the first Each surgical stage Probability at any given time stage Represents the smoothing factor. Indicates the first Each surgical stage Probability at any given time stage Indicates the first Each surgical stage Real-time matching and scoring. This represents the total number of stages. Based on the stage probability at the current moment, the current surgical stage is determined. Then, using the current bleeding severity score, the probability of lesion removal stage, and the comprehensive visibility index, a water injection flow rate gain term is constructed. When the patient is currently in the lesion removal stage, severe bleeding leads to poor visibility, which is addressed through the water injection flow rate gain term. With basic water injection flow rate Calculate the reference value of actual water injection flow rate , ,in, express Bleeding severity score at any given time express Probability of lesion removal at any given time. express Comprehensive visibility index at any time Indicates the gain coefficient for the degree of bleeding. This represents the gain coefficient during the surgical phase. This represents the visibility comprehensive index gain coefficient. Finally, based on the effect of bubble obstruction, a bubble obstruction severity gain is defined. The injection flow rate is then synchronously adjusted by regulating the suction negative pressure through the negative pressure adjustment dynamic range and the injection flow rate. ,in, express Constantly attract negative pressure reference values. This indicates that the base attracts negative pressure. This indicates the dynamic range of the suction negative pressure dynamic adjustment. Indicates the maximum water injection flow rate. express The severity of bubble occlusion at any given moment. This indicates the bubble blocking gain coefficient, which ensures that the suction negative pressure and the water injection flow rate increase synchronously, maintaining an inflow-outflow balance.

[0026] Specifically, the emergency mode switching module has preset clarity thresholds and time thresholds. When the overall visibility index is lower than the clarity threshold and exceeds the preset time, it automatically switches to emergency water exchange mode, increasing the water injection flow reference value and the suction negative pressure reference value to twice their respective preset maximum safety values ​​to quickly flush out accumulated blood. When the overall visibility index recovers to the preset clarity threshold, the emergency water exchange mode stops. Specifically, the human-computer interaction module allows the surgeon to switch between manual mode and AI automatic control mode by triggering a foot switch. With a clear field of vision, the surgeon can autonomously decide when to inject water to provide underwater impact force, assisting in unfolding the partially cut lesion and completing a more precise dissection operation.

[0027] Specifically, the data analysis module stores a structured data stream, recording frame-by-frame the local visual clarity, bleeding severity score, air bubble obstruction severity, visibility index, stage, measured water injection flow rate, and measured suction negative pressure for each frame of the image. It also automatically marks the start and end times and duration of manual control periods and emergency bleeding response events. Postoperatively, based on the structured data stream, it generates an operation quality report that includes a visual field quality change curve, surgical stage time distribution, frequency of manual intervention, and statistics on bleeding events, thereby achieving quantitative analysis of surgical operation efficiency and surgeon skills training.

[0028] Specifically, a method for using an AI-powered control system for synchronized water exchange in digestive endoscopic surgery includes three core stages: preoperative parameter calibration, intraoperative real-time closed-loop control, and postoperative operational data review. Preoperatively, the system adapts and calibrates the visual field assessment weights and surgical stage identification parameters based on individual patient differences. During the procedure, the system automatically performs multi-dimensional quantitative assessment of the visual field and inference of the surgical stage, generating and dynamically adjusting the injection flow rate and suction negative pressure in real time to achieve synchronized water exchange. Simultaneously, it monitors the foot pedal signal throughout the procedure, immediately switching to manual control mode when the pedal is pressed and automatically reverting to AI automatic control mode upon release. Furthermore, it forcibly increases the water exchange rate to quickly clear accumulated blood when intraoperative bleeding is detected. Postoperatively, the system exports the entire operational data to generate a quality report, supporting surgical optimization and skills training.

[0029] This embodiment describes the complete procedure of an underwater endoscopic submucosal dissection. During the preoperative parameter calibration phase, system parameters were adapted on an external simulation platform based on the patient's gastrointestinal volume, lesion size, and location. A quantitative amount of anticoagulated blood and microbubbles were injected into a fluid-filled simulated cavity to recreate the typical field of view during underwater surgery. In this environment, the region of interest for the multi-dimensional quality quantification assessment module was calibrated as a circular area in the center of the endoscopic field of view. Weights were adjusted to maintain a low bleeding score for minor oozing and a rapid increase for active bleeding. For the five surgical stages—mucosal marking, submucosal injection, lesion incision, lesion dissection, and wound hemostasis—feature sequences were extracted to determine the optimal feature weight coefficients for each stage. Simultaneously, based on individual patient differences, the baseline injection flow rate, maximum safe flow rate, baseline suction negative pressure, and upper limit of safe negative pressure were set. Furthermore, the visibility threshold for triggering water exchange, emergency response criteria, and time constants for parameter smoothing transitions were determined.

[0030] During the intraoperative real-time closed-loop control phase, after the surgery begins, the system enters AI automatic control mode, and the real-time image acquisition module continuously acquires tens of frames of color video stream per second. In the mucosal marking phase, the field of view is clean, the system calculates a high value for local visual clarity, the bleeding severity score is close to zero, air bubble obstruction is slight, and the overall mucosal visibility index is far above the preset threshold; the system does not trigger water exchange. The phase identification module, based on characteristics such as no instruments, low elevation, and low texture, determines the phase with high probability, generates lower baseline irrigation flow rate and suction negative pressure reference values, and maintains weak circulation to keep the optical window clear.

[0031] After entering the lesion dissection stage, the electrocautery knife is inserted into the submucosal layer, and bleeding and electrocautery bubbles appear in the field of view. At this time, the standard deviation of the red channel increases, and the clarity index decreases compared to before; the bleeding score increases due to the increase in the red channel mean and the decrease in the green and blue channel mean; the severity of bubble obstruction also increases due to the increase in specular reflection points and the increase in the continuous length of the edge. When the overall visibility index falls below the trigger threshold, the system automatically starts synchronous water exchange. The stage identification module identifies the dissection stage with high probability based on characteristics such as the presence of instruments, a negative turbidity attenuation rate, and increased texture complexity. The reference value of water injection flow rate is automatically increased accordingly, and the suction negative pressure is simultaneously enhanced, which not only compensates for the blurred field of view caused by bleeding, but also provides appropriate water flow to assist in the separation of the mucosal flap.

[0032] During surgery, a sudden small artery hemorrhage occurred, instantly filling the screen with blood. At this moment, the standard deviation of the red channel increased sharply, causing a sharp drop in clarity. The bleeding score exceeded the emergency threshold due to red mean saturation, extremely low green and blue components, and increased red saturation variance, resulting in a very low visibility index. If this state persisted beyond the preset time, the bleeding emergency response module forcibly increased the injection flow rate and suction negative pressure to twice the maximum safe value, entering an emergency water exchange mode to quickly flush away the accumulated blood. Once the bleeding score returned to a safe range and visibility improved, the control parameters gradually returned to their automatically calculated values ​​with a smooth decay.

[0033] During the procedure, the surgeon needs to use pulsed water flow to unfold the mucosal flap. Therefore, the surgeon presses the foot switch to switch to manual control mode. When the foot switch is pressed, water is continuously injected, with a maximum continuous injection time of 20 seconds. The water injection is stopped immediately after the surgeon releases the foot. The water pump and suction device are directly linearly controlled by the foot depth signal. After the water injection stops and the surgeon releases the foot, the system automatically switches back to AI automatic control without delay. The water injection and suction parameters are immediately restored to the automatically calculated values ​​corresponding to the current stage. The water flow is continuous and stable throughout the entire process.

[0034] During the postoperative data review phase, after the surgery, the system stores structured data with timestamps throughout the entire process, including the clarity of each frame, bleeding score, degree of air bubble obstruction, visibility index, probability of each stage, measured fluid injection flow rate and suction negative pressure, as well as markers for human intervention and emergency events. By exporting and analyzing the data, a surgical quality report is generated. The visual field visibility curve clearly shows the time of visual field deterioration and recovery, the stage probability curve corresponds to the surgeon's operational steps, and the periods of human intervention and emergency events are automatically statistically analyzed.

[0035] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0036] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An artificial intelligence control system for synchronized water exchange during digestive endoscopy, characterized in that, It includes a real-time visual field image acquisition module, a multi-dimensional quality quantification assessment module, an adaptive parameter decision module, an emergency mode switching module, a human-computer interaction module, and a data analysis module; the real-time visual field image acquisition module acquires continuous video frames of the surgical area through the endoscope; the multi-dimensional quality quantification assessment module proposes a multi-dimensional quality assessment algorithm to quantify the clarity of the current surgical field; The adaptive parameter decision module proposes an adaptive synchronous control algorithm for parameters, which adaptively generates reference values ​​for water injection flow rate and suction negative pressure. The emergency mode switching module automatically switches to emergency water exchange mode based on the degree of bleeding and a preset bleeding threshold, quickly increasing the water exchange rate to flush out accumulated blood; the human-computer interaction module switches between manual mode and intelligent control mode via a foot switch; and the data analysis module stores data from the entire surgical process.

2. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 1, characterized in that, The real-time visual field image acquisition module continuously acquires color video frames of the surgical area through an endoscope camera and transmits each frame to a computing device in real time. The color video frame contains pixel value information of the red, green, and blue color channels.

3. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 1, characterized in that, The multidimensional quality assessment algorithm quantifies the clarity of the surgical field into local clarity values, bleeding severity scores, and the severity of air bubble obstruction. It analyzes and judges the clarity of the surgical field and automatically triggers water injection and suction operations based on the surgical field clarity index and preset clarity threshold, realizing the synchronous dynamic exchange of clean water and turbid water.

4. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 3, characterized in that, The quantitative assessment of surgical field clarity includes the following steps: S1. Within the region of interest, extract the horizontal and vertical grayscale change values ​​of each pixel. Based on the standard deviation of the pixel values ​​of the red channel within the region of interest, obtain the local sharpness value by summing the gradients. S2. Calculate the pixel mean of each pixel channel, the area ratio of red pixels exceeding the standard, and the variance of red saturation to obtain the bleeding degree score. S3. Statistically calculate the proportion of specular reflection bright pixels within the field of view, and simultaneously calculate the continuous length of the connected edges of the bubbles. Add the proportion of bright pixels to the proportion of edge continuity to obtain the bubble occlusion severity index. The more bubbles there are and the more severe the connected areas are, the greater the bubble occlusion severity index will be. S4. Using local visual clarity index, bleeding degree score, and bubble occlusion severity as three visibility influencing factors, calculate the comprehensive visibility index to reflect the actual operable visual field quality. Deterioration of a single factor will lead to a decrease in the comprehensive visibility index.

5. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 1, characterized in that, The parameter adaptive synchronous control algorithm matches feature scores based on experience using feature weights for each surgical stage, determines the current surgical stage based on the stage probability at consecutive moments, and obtains a reference value for suction negative pressure by forming a linear gain term using the current bleeding degree score, lesion dissection stage probability, and visibility comprehensive index, and by using the gain of bubble occlusion severity.

6. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 5, characterized in that, The adaptive adjustment of the parameters includes the following steps: S1. Extract five features from the image within the current field of view: presence or absence of instruments, degree of mucosal protrusion, rate of decay of red turbidity after water injection, current bleeding score, and texture complexity. Multiply these features by the fixed weights preset for each surgical stage to obtain the real-time matching score for each stage. S2. Define a smoothing factor, use the stage probability of the previous time step, add the real-time matching score of the current time step, and normalize it to obtain the smoothed stage probability distribution of the current time step. S3. Based on the baseline injection flow rate, a real-time injection flow rate reference value is generated by using the current bleeding severity score, the probability of lesion dissection stage, and the comprehensive visibility index as linear gain terms. The higher the probability of lesion dissection stage, the more severe the bleeding, and the worse the field of vision, the greater the flow rate. S4. Finally, based on the basic suction negative pressure, the product of the negative pressure adjustment dynamic range and the normalized value of the injection flow rate is added, and then the suction negative pressure reference value is obtained through the gain of the severity of bubble blockage. This achieves synchronization between suction negative pressure and injection flow rate, and additional suction is strengthened when the bubbles are severe to prevent bubble accumulation.

7. The artificial intelligence control system for synchronized water exchange during digestive endoscopy as described in claim 1, characterized in that, The emergency mode switching module has preset clarity thresholds and time thresholds. When the overall visibility index is lower than the clarity threshold and exceeds the preset time, it automatically switches to emergency water exchange mode, increasing the water flow reference value and suction negative pressure reference value to twice their respective preset maximum safety values ​​to quickly flush out accumulated blood. When the overall visibility index returns to the preset clarity threshold, the emergency water exchange mode stops.

8. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 1, characterized in that, The human-computer interaction module allows the surgeon to switch between manual mode and AI automatic control mode by triggering a foot switch. With a clear field of vision, the surgeon can autonomously decide when to inject water to provide underwater impact force, assisting in unfolding the partially cut lesion and completing a more precise dissection operation.

9. The artificial intelligence control system for synchronous water exchange during digestive endoscopy as described in claim 1, characterized in that, The data analysis module stores a structured data stream, recording frame-by-frame the local visual clarity, bleeding severity score, air bubble obstruction severity, visibility index, stage, measured water injection flow rate, and measured suction negative pressure for each frame of the image. It also automatically marks the start and end times and duration of manual control periods and emergency bleeding response events. Postoperatively, based on the structured data stream, it generates an operation quality report that includes a visual field quality change curve, surgical stage time distribution, frequency of manual intervention, and statistics on bleeding events, enabling quantitative analysis of surgical operation efficiency and surgeon skills training.

10. A method for using an artificial intelligence control system for synchronous water exchange in digestive endoscopy, comprising three core stages: preoperative parameter calibration, intraoperative real-time closed-loop control, and postoperative operational data review; preoperatively, the system adapts and calibrates the visual field assessment weights and surgical stage identification parameters according to individual patient differences; intraoperatively, the system automatically performs multidimensional quantitative assessment of the visual field and inference of the surgical stage, generates and dynamically adjusts the injection flow rate and suction negative pressure in real time to achieve synchronous water exchange, and simultaneously monitors the foot pedal signal throughout the process, immediately switching to manual control mode when the foot pedal is pressed and automatically reverting to AI automatic control mode after release, and forcibly increasing the water exchange rate to quickly clear accumulated blood when an intraoperative bleeding event is detected; postoperatively, the system exports the entire process operation data to generate a quality report, supporting surgical optimization and skills training.

Images